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/dnotify.h>
33 #include <linux/compat.h>
34 #include <linux/mnt_idmapping.h>
35 #include <linux/filelock.h>
39 int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
40 loff_t length, unsigned int time_attrs, struct file *filp)
43 struct iattr newattrs;
45 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
49 newattrs.ia_size = length;
50 newattrs.ia_valid = ATTR_SIZE | time_attrs;
52 newattrs.ia_file = filp;
53 newattrs.ia_valid |= ATTR_FILE;
56 /* Remove suid, sgid, and file capabilities on truncate too */
57 ret = dentry_needs_remove_privs(idmap, dentry);
61 newattrs.ia_valid |= ret | ATTR_FORCE;
63 inode_lock(dentry->d_inode);
64 /* Note any delegations or leases have already been broken: */
65 ret = notify_change(idmap, dentry, &newattrs, NULL);
66 inode_unlock(dentry->d_inode);
70 long vfs_truncate(const struct path *path, loff_t length)
72 struct mnt_idmap *idmap;
76 inode = path->dentry->d_inode;
78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 if (S_ISDIR(inode->i_mode))
81 if (!S_ISREG(inode->i_mode))
84 error = mnt_want_write(path->mnt);
88 idmap = mnt_idmap(path->mnt);
89 error = inode_permission(idmap, inode, MAY_WRITE);
91 goto mnt_drop_write_and_out;
95 goto mnt_drop_write_and_out;
97 error = get_write_access(inode);
99 goto mnt_drop_write_and_out;
102 * Make sure that there are no leases. get_write_access() protects
103 * against the truncate racing with a lease-granting setlease().
105 error = break_lease(inode, O_WRONLY);
107 goto put_write_and_out;
109 error = security_path_truncate(path);
111 error = do_truncate(idmap, path->dentry, length, 0, NULL);
114 put_write_access(inode);
115 mnt_drop_write_and_out:
116 mnt_drop_write(path->mnt);
120 EXPORT_SYMBOL_GPL(vfs_truncate);
122 long do_sys_truncate(const char __user *pathname, loff_t length)
124 unsigned int lookup_flags = LOOKUP_FOLLOW;
128 if (length < 0) /* sorry, but loff_t says... */
132 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
134 error = vfs_truncate(&path, length);
137 if (retry_estale(error, lookup_flags)) {
138 lookup_flags |= LOOKUP_REVAL;
144 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
146 return do_sys_truncate(path, length);
150 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
152 return do_sys_truncate(path, length);
156 long do_ftruncate(struct file *file, loff_t length, int small)
159 struct dentry *dentry;
162 /* explicitly opened as large or we are on 64-bit box */
163 if (file->f_flags & O_LARGEFILE)
166 dentry = file->f_path.dentry;
167 inode = dentry->d_inode;
168 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
171 /* Cannot ftruncate over 2^31 bytes without large file support */
172 if (small && length > MAX_NON_LFS)
175 /* Check IS_APPEND on real upper inode */
176 if (IS_APPEND(file_inode(file)))
178 sb_start_write(inode->i_sb);
179 error = security_file_truncate(file);
181 error = do_truncate(file_mnt_idmap(file), dentry, length,
182 ATTR_MTIME | ATTR_CTIME, file);
183 sb_end_write(inode->i_sb);
188 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
199 error = do_ftruncate(fd_file(f), length, small);
205 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
207 return do_sys_ftruncate(fd, length, 1);
211 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
213 return do_sys_ftruncate(fd, length, 1);
217 /* LFS versions of truncate are only needed on 32 bit machines */
218 #if BITS_PER_LONG == 32
219 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
221 return do_sys_truncate(path, length);
224 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
226 return do_sys_ftruncate(fd, length, 0);
228 #endif /* BITS_PER_LONG == 32 */
230 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
231 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
232 compat_arg_u64_dual(length))
234 return ksys_truncate(pathname, compat_arg_u64_glue(length));
238 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
239 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
240 compat_arg_u64_dual(length))
242 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
246 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
248 struct inode *inode = file_inode(file);
252 if (offset < 0 || len <= 0)
255 if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
259 * Modes are exclusive, even if that is not obvious from the encoding
260 * as bit masks and the mix with the flag in the same namespace.
262 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
263 * encoded as no bit set.
265 switch (mode & FALLOC_FL_MODE_MASK) {
266 case FALLOC_FL_ALLOCATE_RANGE:
267 case FALLOC_FL_UNSHARE_RANGE:
268 case FALLOC_FL_ZERO_RANGE:
270 case FALLOC_FL_PUNCH_HOLE:
271 if (!(mode & FALLOC_FL_KEEP_SIZE))
274 case FALLOC_FL_COLLAPSE_RANGE:
275 case FALLOC_FL_INSERT_RANGE:
276 if (mode & FALLOC_FL_KEEP_SIZE)
283 if (!(file->f_mode & FMODE_WRITE))
287 * On append-only files only space preallocation is supported.
289 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
292 if (IS_IMMUTABLE(inode))
296 * We cannot allow any fallocate operation on an active swapfile
298 if (IS_SWAPFILE(inode))
302 * Revalidate the write permissions, in case security policy has
303 * changed since the files were opened.
305 ret = security_file_permission(file, MAY_WRITE);
309 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
313 if (S_ISFIFO(inode->i_mode))
316 if (S_ISDIR(inode->i_mode))
319 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
322 /* Check for wraparound */
323 if (check_add_overflow(offset, len, &sum))
326 if (sum > inode->i_sb->s_maxbytes)
329 if (!file->f_op->fallocate)
332 file_start_write(file);
333 ret = file->f_op->fallocate(file, mode, offset, len);
336 * Create inotify and fanotify events.
338 * To keep the logic simple always create events if fallocate succeeds.
339 * This implies that events are even created if the file size remains
340 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
343 fsnotify_modify(file);
345 file_end_write(file);
348 EXPORT_SYMBOL_GPL(vfs_fallocate);
350 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
352 struct fd f = fdget(fd);
356 error = vfs_fallocate(fd_file(f), mode, offset, len);
362 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
364 return ksys_fallocate(fd, mode, offset, len);
367 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
368 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
369 compat_arg_u64_dual(len))
371 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
372 compat_arg_u64_glue(len));
377 * access() needs to use the real uid/gid, not the effective uid/gid.
378 * We do this by temporarily clearing all FS-related capabilities and
379 * switching the fsuid/fsgid around to the real ones.
381 * Creating new credentials is expensive, so we try to skip doing it,
382 * which we can if the result would match what we already got.
384 static bool access_need_override_creds(int flags)
386 const struct cred *cred;
388 if (flags & AT_EACCESS)
391 cred = current_cred();
392 if (!uid_eq(cred->fsuid, cred->uid) ||
393 !gid_eq(cred->fsgid, cred->gid))
396 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
397 kuid_t root_uid = make_kuid(cred->user_ns, 0);
398 if (!uid_eq(cred->uid, root_uid)) {
399 if (!cap_isclear(cred->cap_effective))
402 if (!cap_isidentical(cred->cap_effective,
403 cred->cap_permitted))
411 static const struct cred *access_override_creds(void)
413 const struct cred *old_cred;
414 struct cred *override_cred;
416 override_cred = prepare_creds();
421 * XXX access_need_override_creds performs checks in hopes of skipping
422 * this work. Make sure it stays in sync if making any changes in this
426 override_cred->fsuid = override_cred->uid;
427 override_cred->fsgid = override_cred->gid;
429 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
430 /* Clear the capabilities if we switch to a non-root user */
431 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
432 if (!uid_eq(override_cred->uid, root_uid))
433 cap_clear(override_cred->cap_effective);
435 override_cred->cap_effective =
436 override_cred->cap_permitted;
440 * The new set of credentials can *only* be used in
441 * task-synchronous circumstances, and does not need
442 * RCU freeing, unless somebody then takes a separate
445 * NOTE! This is _only_ true because this credential
446 * is used purely for override_creds() that installs
447 * it as the subjective cred. Other threads will be
448 * accessing ->real_cred, not the subjective cred.
450 * If somebody _does_ make a copy of this (using the
451 * 'get_current_cred()' function), that will clear the
452 * non_rcu field, because now that other user may be
453 * expecting RCU freeing. But normal thread-synchronous
454 * cred accesses will keep things non-racy to avoid RCU
457 override_cred->non_rcu = 1;
459 old_cred = override_creds(override_cred);
461 /* override_cred() gets its own ref */
462 put_cred(override_cred);
467 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
472 unsigned int lookup_flags = LOOKUP_FOLLOW;
473 const struct cred *old_cred = NULL;
475 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
478 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
481 if (flags & AT_SYMLINK_NOFOLLOW)
482 lookup_flags &= ~LOOKUP_FOLLOW;
483 if (flags & AT_EMPTY_PATH)
484 lookup_flags |= LOOKUP_EMPTY;
486 if (access_need_override_creds(flags)) {
487 old_cred = access_override_creds();
493 res = user_path_at(dfd, filename, lookup_flags, &path);
497 inode = d_backing_inode(path.dentry);
499 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
501 * MAY_EXEC on regular files is denied if the fs is mounted
502 * with the "noexec" flag.
505 if (path_noexec(&path))
506 goto out_path_release;
509 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
510 /* SuS v2 requires we report a read only fs too */
511 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
512 goto out_path_release;
514 * This is a rare case where using __mnt_is_readonly()
515 * is OK without a mnt_want/drop_write() pair. Since
516 * no actual write to the fs is performed here, we do
517 * not need to telegraph to that to anyone.
519 * By doing this, we accept that this access is
520 * inherently racy and know that the fs may change
521 * state before we even see this result.
523 if (__mnt_is_readonly(path.mnt))
528 if (retry_estale(res, lookup_flags)) {
529 lookup_flags |= LOOKUP_REVAL;
534 revert_creds(old_cred);
539 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
541 return do_faccessat(dfd, filename, mode, 0);
544 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
547 return do_faccessat(dfd, filename, mode, flags);
550 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
552 return do_faccessat(AT_FDCWD, filename, mode, 0);
555 SYSCALL_DEFINE1(chdir, const char __user *, filename)
559 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
561 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
565 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
569 set_fs_pwd(current->fs, &path);
573 if (retry_estale(error, lookup_flags)) {
574 lookup_flags |= LOOKUP_REVAL;
581 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
583 struct fd f = fdget_raw(fd);
591 if (!d_can_lookup(fd_file(f)->f_path.dentry))
594 error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
596 set_fs_pwd(current->fs, &fd_file(f)->f_path);
603 SYSCALL_DEFINE1(chroot, const char __user *, filename)
607 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
609 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
613 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
618 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
620 error = security_path_chroot(&path);
624 set_fs_root(current->fs, &path);
628 if (retry_estale(error, lookup_flags)) {
629 lookup_flags |= LOOKUP_REVAL;
636 int chmod_common(const struct path *path, umode_t mode)
638 struct inode *inode = path->dentry->d_inode;
639 struct inode *delegated_inode = NULL;
640 struct iattr newattrs;
643 error = mnt_want_write(path->mnt);
648 error = security_path_chmod(path, mode);
651 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
652 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
653 error = notify_change(mnt_idmap(path->mnt), path->dentry,
654 &newattrs, &delegated_inode);
657 if (delegated_inode) {
658 error = break_deleg_wait(&delegated_inode);
662 mnt_drop_write(path->mnt);
666 int vfs_fchmod(struct file *file, umode_t mode)
669 return chmod_common(&file->f_path, mode);
672 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
674 struct fd f = fdget(fd);
678 err = vfs_fchmod(fd_file(f), mode);
684 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
689 unsigned int lookup_flags;
691 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
694 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
695 if (flags & AT_EMPTY_PATH)
696 lookup_flags |= LOOKUP_EMPTY;
699 error = user_path_at(dfd, filename, lookup_flags, &path);
701 error = chmod_common(&path, mode);
703 if (retry_estale(error, lookup_flags)) {
704 lookup_flags |= LOOKUP_REVAL;
711 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
712 umode_t, mode, unsigned int, flags)
714 return do_fchmodat(dfd, filename, mode, flags);
717 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
720 return do_fchmodat(dfd, filename, mode, 0);
723 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
725 return do_fchmodat(AT_FDCWD, filename, mode, 0);
729 * Check whether @kuid is valid and if so generate and set vfsuid_t in
732 * Return: true if @kuid is valid, false if not.
734 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
736 if (!uid_valid(kuid))
738 attr->ia_valid |= ATTR_UID;
739 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
744 * Check whether @kgid is valid and if so generate and set vfsgid_t in
747 * Return: true if @kgid is valid, false if not.
749 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
751 if (!gid_valid(kgid))
753 attr->ia_valid |= ATTR_GID;
754 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
758 int chown_common(const struct path *path, uid_t user, gid_t group)
760 struct mnt_idmap *idmap;
761 struct user_namespace *fs_userns;
762 struct inode *inode = path->dentry->d_inode;
763 struct inode *delegated_inode = NULL;
765 struct iattr newattrs;
769 uid = make_kuid(current_user_ns(), user);
770 gid = make_kgid(current_user_ns(), group);
772 idmap = mnt_idmap(path->mnt);
773 fs_userns = i_user_ns(inode);
776 newattrs.ia_vfsuid = INVALID_VFSUID;
777 newattrs.ia_vfsgid = INVALID_VFSGID;
778 newattrs.ia_valid = ATTR_CTIME;
779 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
781 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
784 if (!S_ISDIR(inode->i_mode))
785 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
786 setattr_should_drop_sgid(idmap, inode);
787 /* Continue to send actual fs values, not the mount values. */
788 error = security_path_chown(
790 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
791 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
793 error = notify_change(idmap, path->dentry, &newattrs,
796 if (delegated_inode) {
797 error = break_deleg_wait(&delegated_inode);
804 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
811 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
814 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
815 if (flag & AT_EMPTY_PATH)
816 lookup_flags |= LOOKUP_EMPTY;
818 error = user_path_at(dfd, filename, lookup_flags, &path);
821 error = mnt_want_write(path.mnt);
824 error = chown_common(&path, user, group);
825 mnt_drop_write(path.mnt);
828 if (retry_estale(error, lookup_flags)) {
829 lookup_flags |= LOOKUP_REVAL;
836 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
837 gid_t, group, int, flag)
839 return do_fchownat(dfd, filename, user, group, flag);
842 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
844 return do_fchownat(AT_FDCWD, filename, user, group, 0);
847 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
849 return do_fchownat(AT_FDCWD, filename, user, group,
850 AT_SYMLINK_NOFOLLOW);
853 int vfs_fchown(struct file *file, uid_t user, gid_t group)
857 error = mnt_want_write_file(file);
861 error = chown_common(&file->f_path, user, group);
862 mnt_drop_write_file(file);
866 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
868 struct fd f = fdget(fd);
872 error = vfs_fchown(fd_file(f), user, group);
878 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
880 return ksys_fchown(fd, user, group);
883 static inline int file_get_write_access(struct file *f)
887 error = get_write_access(f->f_inode);
890 error = mnt_get_write_access(f->f_path.mnt);
893 if (unlikely(f->f_mode & FMODE_BACKING)) {
894 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
901 mnt_put_write_access(f->f_path.mnt);
903 put_write_access(f->f_inode);
907 static int do_dentry_open(struct file *f,
908 int (*open)(struct inode *, struct file *))
910 static const struct file_operations empty_fops = {};
911 struct inode *inode = f->f_path.dentry->d_inode;
914 path_get(&f->f_path);
916 f->f_mapping = inode->i_mapping;
917 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
918 f->f_sb_err = file_sample_sb_err(f);
920 if (unlikely(f->f_flags & O_PATH)) {
921 f->f_mode = FMODE_PATH | FMODE_OPENED;
922 f->f_op = &empty_fops;
926 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
927 i_readcount_inc(inode);
928 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
929 error = file_get_write_access(f);
932 f->f_mode |= FMODE_WRITER;
935 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
936 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
937 f->f_mode |= FMODE_ATOMIC_POS;
939 f->f_op = fops_get(inode->i_fop);
940 if (WARN_ON(!f->f_op)) {
945 error = security_file_open(f);
949 error = break_lease(file_inode(f), f->f_flags);
953 /* normally all 3 are set; ->open() can clear them if needed */
954 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
956 open = f->f_op->open;
958 error = open(inode, f);
962 f->f_mode |= FMODE_OPENED;
963 if ((f->f_mode & FMODE_READ) &&
964 likely(f->f_op->read || f->f_op->read_iter))
965 f->f_mode |= FMODE_CAN_READ;
966 if ((f->f_mode & FMODE_WRITE) &&
967 likely(f->f_op->write || f->f_op->write_iter))
968 f->f_mode |= FMODE_CAN_WRITE;
969 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
970 f->f_mode &= ~FMODE_LSEEK;
971 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
972 f->f_mode |= FMODE_CAN_ODIRECT;
974 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
975 f->f_iocb_flags = iocb_flags(f);
977 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
979 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
983 * XXX: Huge page cache doesn't support writing yet. Drop all page
984 * cache for this file before processing writes.
986 if (f->f_mode & FMODE_WRITE) {
988 * Depends on full fence from get_write_access() to synchronize
989 * against collapse_file() regarding i_writecount and nr_thps
990 * updates. Ensures subsequent insertion of THPs into the page
993 if (filemap_nr_thps(inode->i_mapping)) {
994 struct address_space *mapping = inode->i_mapping;
996 filemap_invalidate_lock(inode->i_mapping);
998 * unmap_mapping_range just need to be called once
999 * here, because the private pages is not need to be
1000 * unmapped mapping (e.g. data segment of dynamic
1001 * shared libraries here).
1003 unmap_mapping_range(mapping, 0, 0, 0);
1004 truncate_inode_pages(mapping, 0);
1005 filemap_invalidate_unlock(inode->i_mapping);
1012 if (WARN_ON_ONCE(error > 0))
1017 path_put(&f->f_path);
1018 f->f_path.mnt = NULL;
1019 f->f_path.dentry = NULL;
1025 * finish_open - finish opening a file
1026 * @file: file pointer
1027 * @dentry: pointer to dentry
1028 * @open: open callback
1030 * This can be used to finish opening a file passed to i_op->atomic_open().
1032 * If the open callback is set to NULL, then the standard f_op->open()
1033 * filesystem callback is substituted.
1035 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1036 * the return value of d_splice_alias(), then the caller needs to perform dput()
1037 * on it after finish_open().
1039 * Returns zero on success or -errno if the open failed.
1041 int finish_open(struct file *file, struct dentry *dentry,
1042 int (*open)(struct inode *, struct file *))
1044 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1046 file->f_path.dentry = dentry;
1047 return do_dentry_open(file, open);
1049 EXPORT_SYMBOL(finish_open);
1052 * finish_no_open - finish ->atomic_open() without opening the file
1054 * @file: file pointer
1055 * @dentry: dentry or NULL (as returned from ->lookup())
1057 * This can be used to set the result of a successful lookup in ->atomic_open().
1059 * NB: unlike finish_open() this function does consume the dentry reference and
1060 * the caller need not dput() it.
1062 * Returns "0" which must be the return value of ->atomic_open() after having
1063 * called this function.
1065 int finish_no_open(struct file *file, struct dentry *dentry)
1067 file->f_path.dentry = dentry;
1070 EXPORT_SYMBOL(finish_no_open);
1072 char *file_path(struct file *filp, char *buf, int buflen)
1074 return d_path(&filp->f_path, buf, buflen);
1076 EXPORT_SYMBOL(file_path);
1079 * vfs_open - open the file at the given path
1080 * @path: path to open
1081 * @file: newly allocated file with f_flag initialized
1083 int vfs_open(const struct path *path, struct file *file)
1087 file->f_path = *path;
1088 ret = do_dentry_open(file, NULL);
1091 * Once we return a file with FMODE_OPENED, __fput() will call
1092 * fsnotify_close(), so we need fsnotify_open() here for
1095 fsnotify_open(file);
1100 struct file *dentry_open(const struct path *path, int flags,
1101 const struct cred *cred)
1106 /* We must always pass in a valid mount pointer. */
1109 f = alloc_empty_file(flags, cred);
1111 error = vfs_open(path, f);
1119 EXPORT_SYMBOL(dentry_open);
1122 * dentry_create - Create and open a file
1123 * @path: path to create
1125 * @mode: mode bits for new file
1126 * @cred: credentials to use
1128 * Caller must hold the parent directory's lock, and have prepared
1129 * a negative dentry, placed in @path->dentry, for the new file.
1131 * Caller sets @path->mnt to the vfsmount of the filesystem where
1132 * the new file is to be created. The parent directory and the
1133 * negative dentry must reside on the same filesystem instance.
1135 * On success, returns a "struct file *". Otherwise a ERR_PTR
1138 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1139 const struct cred *cred)
1144 f = alloc_empty_file(flags, cred);
1148 error = vfs_create(mnt_idmap(path->mnt),
1149 d_inode(path->dentry->d_parent),
1150 path->dentry, mode, true);
1152 error = vfs_open(path, f);
1154 if (unlikely(error)) {
1156 return ERR_PTR(error);
1160 EXPORT_SYMBOL(dentry_create);
1163 * kernel_file_open - open a file for kernel internal use
1164 * @path: path of the file to open
1165 * @flags: open flags
1166 * @cred: credentials for open
1168 * Open a file for use by in-kernel consumers. The file is not accounted
1169 * against nr_files and must not be installed into the file descriptor
1172 * Return: Opened file on success, an error pointer on failure.
1174 struct file *kernel_file_open(const struct path *path, int flags,
1175 const struct cred *cred)
1180 f = alloc_empty_file_noaccount(flags, cred);
1185 error = do_dentry_open(f, NULL);
1188 return ERR_PTR(error);
1194 EXPORT_SYMBOL_GPL(kernel_file_open);
1196 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1197 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1199 inline struct open_how build_open_how(int flags, umode_t mode)
1201 struct open_how how = {
1202 .flags = flags & VALID_OPEN_FLAGS,
1203 .mode = mode & S_IALLUGO,
1206 /* O_PATH beats everything else. */
1207 if (how.flags & O_PATH)
1208 how.flags &= O_PATH_FLAGS;
1209 /* Modes should only be set for create-like flags. */
1210 if (!WILL_CREATE(how.flags))
1215 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1217 u64 flags = how->flags;
1218 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1219 int lookup_flags = 0;
1220 int acc_mode = ACC_MODE(flags);
1222 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1223 "struct open_flags doesn't yet handle flags > 32 bits");
1226 * Strip flags that either shouldn't be set by userspace like
1227 * FMODE_NONOTIFY or that aren't relevant in determining struct
1228 * open_flags like O_CLOEXEC.
1233 * Older syscalls implicitly clear all of the invalid flags or argument
1234 * values before calling build_open_flags(), but openat2(2) checks all
1237 if (flags & ~VALID_OPEN_FLAGS)
1239 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1242 /* Scoping flags are mutually exclusive. */
1243 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1246 /* Deal with the mode. */
1247 if (WILL_CREATE(flags)) {
1248 if (how->mode & ~S_IALLUGO)
1250 op->mode = how->mode | S_IFREG;
1258 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1259 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1260 * O_TMPFILE below which requires O_DIRECTORY being raised.
1262 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1265 /* Now handle the creative implementation of O_TMPFILE. */
1266 if (flags & __O_TMPFILE) {
1268 * In order to ensure programs get explicit errors when trying
1269 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1270 * is raised alongside __O_TMPFILE.
1272 if (!(flags & O_DIRECTORY))
1274 if (!(acc_mode & MAY_WRITE))
1277 if (flags & O_PATH) {
1278 /* O_PATH only permits certain other flags to be set. */
1279 if (flags & ~O_PATH_FLAGS)
1285 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1286 * check for O_DSYNC if the need any syncing at all we enforce it's
1287 * always set instead of having to deal with possibly weird behaviour
1288 * for malicious applications setting only __O_SYNC.
1290 if (flags & __O_SYNC)
1293 op->open_flag = flags;
1295 /* O_TRUNC implies we need access checks for write permissions */
1296 if (flags & O_TRUNC)
1297 acc_mode |= MAY_WRITE;
1299 /* Allow the LSM permission hook to distinguish append
1300 access from general write access. */
1301 if (flags & O_APPEND)
1302 acc_mode |= MAY_APPEND;
1304 op->acc_mode = acc_mode;
1306 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1308 if (flags & O_CREAT) {
1309 op->intent |= LOOKUP_CREATE;
1310 if (flags & O_EXCL) {
1311 op->intent |= LOOKUP_EXCL;
1312 flags |= O_NOFOLLOW;
1316 if (flags & O_DIRECTORY)
1317 lookup_flags |= LOOKUP_DIRECTORY;
1318 if (!(flags & O_NOFOLLOW))
1319 lookup_flags |= LOOKUP_FOLLOW;
1321 if (how->resolve & RESOLVE_NO_XDEV)
1322 lookup_flags |= LOOKUP_NO_XDEV;
1323 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1324 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1325 if (how->resolve & RESOLVE_NO_SYMLINKS)
1326 lookup_flags |= LOOKUP_NO_SYMLINKS;
1327 if (how->resolve & RESOLVE_BENEATH)
1328 lookup_flags |= LOOKUP_BENEATH;
1329 if (how->resolve & RESOLVE_IN_ROOT)
1330 lookup_flags |= LOOKUP_IN_ROOT;
1331 if (how->resolve & RESOLVE_CACHED) {
1332 /* Don't bother even trying for create/truncate/tmpfile open */
1333 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1335 lookup_flags |= LOOKUP_CACHED;
1338 op->lookup_flags = lookup_flags;
1343 * file_open_name - open file and return file pointer
1345 * @name: struct filename containing path to open
1346 * @flags: open flags as per the open(2) second argument
1347 * @mode: mode for the new file if O_CREAT is set, else ignored
1349 * This is the helper to open a file from kernelspace if you really
1350 * have to. But in generally you should not do this, so please move
1351 * along, nothing to see here..
1353 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1355 struct open_flags op;
1356 struct open_how how = build_open_how(flags, mode);
1357 int err = build_open_flags(&how, &op);
1359 return ERR_PTR(err);
1360 return do_filp_open(AT_FDCWD, name, &op);
1364 * filp_open - open file and return file pointer
1366 * @filename: path to open
1367 * @flags: open flags as per the open(2) second argument
1368 * @mode: mode for the new file if O_CREAT is set, else ignored
1370 * This is the helper to open a file from kernelspace if you really
1371 * have to. But in generally you should not do this, so please move
1372 * along, nothing to see here..
1374 struct file *filp_open(const char *filename, int flags, umode_t mode)
1376 struct filename *name = getname_kernel(filename);
1377 struct file *file = ERR_CAST(name);
1379 if (!IS_ERR(name)) {
1380 file = file_open_name(name, flags, mode);
1385 EXPORT_SYMBOL(filp_open);
1387 struct file *file_open_root(const struct path *root,
1388 const char *filename, int flags, umode_t mode)
1390 struct open_flags op;
1391 struct open_how how = build_open_how(flags, mode);
1392 int err = build_open_flags(&how, &op);
1394 return ERR_PTR(err);
1395 return do_file_open_root(root, filename, &op);
1397 EXPORT_SYMBOL(file_open_root);
1399 static long do_sys_openat2(int dfd, const char __user *filename,
1400 struct open_how *how)
1402 struct open_flags op;
1403 int fd = build_open_flags(how, &op);
1404 struct filename *tmp;
1409 tmp = getname(filename);
1411 return PTR_ERR(tmp);
1413 fd = get_unused_fd_flags(how->flags);
1415 struct file *f = do_filp_open(dfd, tmp, &op);
1427 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1429 struct open_how how = build_open_how(flags, mode);
1430 return do_sys_openat2(dfd, filename, &how);
1434 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1436 if (force_o_largefile())
1437 flags |= O_LARGEFILE;
1438 return do_sys_open(AT_FDCWD, filename, flags, mode);
1441 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1444 if (force_o_largefile())
1445 flags |= O_LARGEFILE;
1446 return do_sys_open(dfd, filename, flags, mode);
1449 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1450 struct open_how __user *, how, size_t, usize)
1453 struct open_how tmp;
1455 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1456 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1458 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1461 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1465 audit_openat2_how(&tmp);
1467 /* O_LARGEFILE is only allowed for non-O_PATH. */
1468 if (!(tmp.flags & O_PATH) && force_o_largefile())
1469 tmp.flags |= O_LARGEFILE;
1471 return do_sys_openat2(dfd, filename, &tmp);
1474 #ifdef CONFIG_COMPAT
1476 * Exactly like sys_open(), except that it doesn't set the
1479 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1481 return do_sys_open(AT_FDCWD, filename, flags, mode);
1485 * Exactly like sys_openat(), except that it doesn't set the
1488 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1490 return do_sys_open(dfd, filename, flags, mode);
1497 * For backward compatibility? Maybe this should be moved
1498 * into arch/i386 instead?
1500 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1502 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1504 if (force_o_largefile())
1505 flags |= O_LARGEFILE;
1506 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1511 * "id" is the POSIX thread ID. We use the
1512 * files pointer for this..
1514 static int filp_flush(struct file *filp, fl_owner_t id)
1518 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1519 "VFS: Close: file count is 0 (f_op=%ps)",
1524 if (filp->f_op->flush)
1525 retval = filp->f_op->flush(filp, id);
1527 if (likely(!(filp->f_mode & FMODE_PATH))) {
1528 dnotify_flush(filp, id);
1529 locks_remove_posix(filp, id);
1534 int filp_close(struct file *filp, fl_owner_t id)
1538 retval = filp_flush(filp, id);
1543 EXPORT_SYMBOL(filp_close);
1546 * Careful here! We test whether the file pointer is NULL before
1547 * releasing the fd. This ensures that one clone task can't release
1548 * an fd while another clone is opening it.
1550 SYSCALL_DEFINE1(close, unsigned int, fd)
1555 file = file_close_fd(fd);
1559 retval = filp_flush(file, current->files);
1562 * We're returning to user space. Don't bother
1563 * with any delayed fput() cases.
1567 /* can't restart close syscall because file table entry was cleared */
1568 if (unlikely(retval == -ERESTARTSYS ||
1569 retval == -ERESTARTNOINTR ||
1570 retval == -ERESTARTNOHAND ||
1571 retval == -ERESTART_RESTARTBLOCK))
1578 * sys_close_range() - Close all file descriptors in a given range.
1580 * @fd: starting file descriptor to close
1581 * @max_fd: last file descriptor to close
1582 * @flags: reserved for future extensions
1584 * This closes a range of file descriptors. All file descriptors
1585 * from @fd up to and including @max_fd are closed.
1586 * Currently, errors to close a given file descriptor are ignored.
1588 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1589 unsigned int, flags)
1591 return __close_range(fd, max_fd, flags);
1595 * This routine simulates a hangup on the tty, to arrange that users
1596 * are given clean terminals at login time.
1598 SYSCALL_DEFINE0(vhangup)
1600 if (capable(CAP_SYS_TTY_CONFIG)) {
1608 * Called when an inode is about to be open.
1609 * We use this to disallow opening large files on 32bit systems if
1610 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1611 * on this flag in sys_open.
1613 int generic_file_open(struct inode * inode, struct file * filp)
1615 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1620 EXPORT_SYMBOL(generic_file_open);
1623 * This is used by subsystems that don't want seekable
1624 * file descriptors. The function is not supposed to ever fail, the only
1625 * reason it returns an 'int' and not 'void' is so that it can be plugged
1626 * directly into file_operations structure.
1628 int nonseekable_open(struct inode *inode, struct file *filp)
1630 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1634 EXPORT_SYMBOL(nonseekable_open);
1637 * stream_open is used by subsystems that want stream-like file descriptors.
1638 * Such file descriptors are not seekable and don't have notion of position
1639 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1640 * Contrary to file descriptors of other regular files, .read() and .write()
1641 * can run simultaneously.
1643 * stream_open never fails and is marked to return int so that it could be
1644 * directly used as file_operations.open .
1646 int stream_open(struct inode *inode, struct file *filp)
1648 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1649 filp->f_mode |= FMODE_STREAM;
1653 EXPORT_SYMBOL(stream_open);