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>
38 int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
39 loff_t length, unsigned int time_attrs, struct file *filp)
42 struct iattr newattrs;
44 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
48 newattrs.ia_size = length;
49 newattrs.ia_valid = ATTR_SIZE | time_attrs;
51 newattrs.ia_file = filp;
52 newattrs.ia_valid |= ATTR_FILE;
55 /* Remove suid, sgid, and file capabilities on truncate too */
56 ret = dentry_needs_remove_privs(dentry);
60 newattrs.ia_valid |= ret | ATTR_FORCE;
62 inode_lock(dentry->d_inode);
63 /* Note any delegations or leases have already been broken: */
64 ret = notify_change(mnt_userns, dentry, &newattrs, NULL);
65 inode_unlock(dentry->d_inode);
69 long vfs_truncate(const struct path *path, loff_t length)
71 struct user_namespace *mnt_userns;
75 inode = path->dentry->d_inode;
77 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
78 if (S_ISDIR(inode->i_mode))
80 if (!S_ISREG(inode->i_mode))
83 error = mnt_want_write(path->mnt);
87 mnt_userns = mnt_user_ns(path->mnt);
88 error = inode_permission(mnt_userns, inode, MAY_WRITE);
90 goto mnt_drop_write_and_out;
94 goto mnt_drop_write_and_out;
96 error = get_write_access(inode);
98 goto mnt_drop_write_and_out;
101 * Make sure that there are no leases. get_write_access() protects
102 * against the truncate racing with a lease-granting setlease().
104 error = break_lease(inode, O_WRONLY);
106 goto put_write_and_out;
108 error = locks_verify_truncate(inode, NULL, length);
110 error = security_path_truncate(path);
112 error = do_truncate(mnt_userns, 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 = locks_verify_truncate(inode, f.file, length);
194 error = security_path_truncate(&f.file->f_path);
196 error = do_truncate(file_mnt_user_ns(f.file), dentry, length,
197 ATTR_MTIME | ATTR_CTIME, f.file);
198 sb_end_write(inode->i_sb);
205 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
207 return do_sys_ftruncate(fd, length, 1);
211 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_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 */
231 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
233 struct inode *inode = file_inode(file);
236 if (offset < 0 || len <= 0)
239 /* Return error if mode is not supported */
240 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
243 /* Punch hole and zero range are mutually exclusive */
244 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
245 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
248 /* Punch hole must have keep size set */
249 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
250 !(mode & FALLOC_FL_KEEP_SIZE))
253 /* Collapse range should only be used exclusively. */
254 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
255 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
258 /* Insert range should only be used exclusively. */
259 if ((mode & FALLOC_FL_INSERT_RANGE) &&
260 (mode & ~FALLOC_FL_INSERT_RANGE))
263 /* Unshare range should only be used with allocate mode. */
264 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
265 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
268 if (!(file->f_mode & FMODE_WRITE))
272 * We can only allow pure fallocate on append only files
274 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
277 if (IS_IMMUTABLE(inode))
281 * We cannot allow any fallocate operation on an active swapfile
283 if (IS_SWAPFILE(inode))
287 * Revalidate the write permissions, in case security policy has
288 * changed since the files were opened.
290 ret = security_file_permission(file, MAY_WRITE);
294 if (S_ISFIFO(inode->i_mode))
297 if (S_ISDIR(inode->i_mode))
300 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
303 /* Check for wrap through zero too */
304 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
307 if (!file->f_op->fallocate)
310 file_start_write(file);
311 ret = file->f_op->fallocate(file, mode, offset, len);
314 * Create inotify and fanotify events.
316 * To keep the logic simple always create events if fallocate succeeds.
317 * This implies that events are even created if the file size remains
318 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
321 fsnotify_modify(file);
323 file_end_write(file);
326 EXPORT_SYMBOL_GPL(vfs_fallocate);
328 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
330 struct fd f = fdget(fd);
334 error = vfs_fallocate(f.file, mode, offset, len);
340 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
342 return ksys_fallocate(fd, mode, offset, len);
346 * access() needs to use the real uid/gid, not the effective uid/gid.
347 * We do this by temporarily clearing all FS-related capabilities and
348 * switching the fsuid/fsgid around to the real ones.
350 static const struct cred *access_override_creds(void)
352 const struct cred *old_cred;
353 struct cred *override_cred;
355 override_cred = prepare_creds();
359 override_cred->fsuid = override_cred->uid;
360 override_cred->fsgid = override_cred->gid;
362 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
363 /* Clear the capabilities if we switch to a non-root user */
364 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
365 if (!uid_eq(override_cred->uid, root_uid))
366 cap_clear(override_cred->cap_effective);
368 override_cred->cap_effective =
369 override_cred->cap_permitted;
373 * The new set of credentials can *only* be used in
374 * task-synchronous circumstances, and does not need
375 * RCU freeing, unless somebody then takes a separate
378 * NOTE! This is _only_ true because this credential
379 * is used purely for override_creds() that installs
380 * it as the subjective cred. Other threads will be
381 * accessing ->real_cred, not the subjective cred.
383 * If somebody _does_ make a copy of this (using the
384 * 'get_current_cred()' function), that will clear the
385 * non_rcu field, because now that other user may be
386 * expecting RCU freeing. But normal thread-synchronous
387 * cred accesses will keep things non-RCY.
389 override_cred->non_rcu = 1;
391 old_cred = override_creds(override_cred);
393 /* override_cred() gets its own ref */
394 put_cred(override_cred);
399 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
404 unsigned int lookup_flags = LOOKUP_FOLLOW;
405 const struct cred *old_cred = NULL;
407 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
410 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
413 if (flags & AT_SYMLINK_NOFOLLOW)
414 lookup_flags &= ~LOOKUP_FOLLOW;
415 if (flags & AT_EMPTY_PATH)
416 lookup_flags |= LOOKUP_EMPTY;
418 if (!(flags & AT_EACCESS)) {
419 old_cred = access_override_creds();
425 res = user_path_at(dfd, filename, lookup_flags, &path);
429 inode = d_backing_inode(path.dentry);
431 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
433 * MAY_EXEC on regular files is denied if the fs is mounted
434 * with the "noexec" flag.
437 if (path_noexec(&path))
438 goto out_path_release;
441 res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
442 /* SuS v2 requires we report a read only fs too */
443 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
444 goto out_path_release;
446 * This is a rare case where using __mnt_is_readonly()
447 * is OK without a mnt_want/drop_write() pair. Since
448 * no actual write to the fs is performed here, we do
449 * not need to telegraph to that to anyone.
451 * By doing this, we accept that this access is
452 * inherently racy and know that the fs may change
453 * state before we even see this result.
455 if (__mnt_is_readonly(path.mnt))
460 if (retry_estale(res, lookup_flags)) {
461 lookup_flags |= LOOKUP_REVAL;
466 revert_creds(old_cred);
471 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
473 return do_faccessat(dfd, filename, mode, 0);
476 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
479 return do_faccessat(dfd, filename, mode, flags);
482 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
484 return do_faccessat(AT_FDCWD, filename, mode, 0);
487 SYSCALL_DEFINE1(chdir, const char __user *, filename)
491 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
493 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
497 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
501 set_fs_pwd(current->fs, &path);
505 if (retry_estale(error, lookup_flags)) {
506 lookup_flags |= LOOKUP_REVAL;
513 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
515 struct fd f = fdget_raw(fd);
523 if (!d_can_lookup(f.file->f_path.dentry))
526 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
528 set_fs_pwd(current->fs, &f.file->f_path);
535 SYSCALL_DEFINE1(chroot, const char __user *, filename)
539 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
541 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
545 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
550 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
552 error = security_path_chroot(&path);
556 set_fs_root(current->fs, &path);
560 if (retry_estale(error, lookup_flags)) {
561 lookup_flags |= LOOKUP_REVAL;
568 int chmod_common(const struct path *path, umode_t mode)
570 struct inode *inode = path->dentry->d_inode;
571 struct inode *delegated_inode = NULL;
572 struct iattr newattrs;
575 error = mnt_want_write(path->mnt);
580 error = security_path_chmod(path, mode);
583 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
584 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
585 error = notify_change(mnt_user_ns(path->mnt), path->dentry,
586 &newattrs, &delegated_inode);
589 if (delegated_inode) {
590 error = break_deleg_wait(&delegated_inode);
594 mnt_drop_write(path->mnt);
598 int vfs_fchmod(struct file *file, umode_t mode)
601 return chmod_common(&file->f_path, mode);
604 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
606 struct fd f = fdget(fd);
610 err = vfs_fchmod(f.file, mode);
616 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
620 unsigned int lookup_flags = LOOKUP_FOLLOW;
622 error = user_path_at(dfd, filename, lookup_flags, &path);
624 error = chmod_common(&path, mode);
626 if (retry_estale(error, lookup_flags)) {
627 lookup_flags |= LOOKUP_REVAL;
634 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
637 return do_fchmodat(dfd, filename, mode);
640 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
642 return do_fchmodat(AT_FDCWD, filename, mode);
645 int chown_common(const struct path *path, uid_t user, gid_t group)
647 struct user_namespace *mnt_userns;
648 struct inode *inode = path->dentry->d_inode;
649 struct inode *delegated_inode = NULL;
651 struct iattr newattrs;
655 uid = make_kuid(current_user_ns(), user);
656 gid = make_kgid(current_user_ns(), group);
658 mnt_userns = mnt_user_ns(path->mnt);
659 uid = kuid_from_mnt(mnt_userns, uid);
660 gid = kgid_from_mnt(mnt_userns, gid);
663 newattrs.ia_valid = ATTR_CTIME;
664 if (user != (uid_t) -1) {
667 newattrs.ia_valid |= ATTR_UID;
668 newattrs.ia_uid = uid;
670 if (group != (gid_t) -1) {
673 newattrs.ia_valid |= ATTR_GID;
674 newattrs.ia_gid = gid;
676 if (!S_ISDIR(inode->i_mode))
678 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
680 error = security_path_chown(path, uid, gid);
682 error = notify_change(mnt_userns, path->dentry, &newattrs,
685 if (delegated_inode) {
686 error = break_deleg_wait(&delegated_inode);
693 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
700 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
703 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
704 if (flag & AT_EMPTY_PATH)
705 lookup_flags |= LOOKUP_EMPTY;
707 error = user_path_at(dfd, filename, lookup_flags, &path);
710 error = mnt_want_write(path.mnt);
713 error = chown_common(&path, user, group);
714 mnt_drop_write(path.mnt);
717 if (retry_estale(error, lookup_flags)) {
718 lookup_flags |= LOOKUP_REVAL;
725 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
726 gid_t, group, int, flag)
728 return do_fchownat(dfd, filename, user, group, flag);
731 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
733 return do_fchownat(AT_FDCWD, filename, user, group, 0);
736 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
738 return do_fchownat(AT_FDCWD, filename, user, group,
739 AT_SYMLINK_NOFOLLOW);
742 int vfs_fchown(struct file *file, uid_t user, gid_t group)
746 error = mnt_want_write_file(file);
750 error = chown_common(&file->f_path, user, group);
751 mnt_drop_write_file(file);
755 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
757 struct fd f = fdget(fd);
761 error = vfs_fchown(f.file, user, group);
767 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
769 return ksys_fchown(fd, user, group);
772 static int do_dentry_open(struct file *f,
774 int (*open)(struct inode *, struct file *))
776 static const struct file_operations empty_fops = {};
779 path_get(&f->f_path);
781 f->f_mapping = inode->i_mapping;
782 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
783 f->f_sb_err = file_sample_sb_err(f);
785 if (unlikely(f->f_flags & O_PATH)) {
786 f->f_mode = FMODE_PATH | FMODE_OPENED;
787 f->f_op = &empty_fops;
791 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
792 error = get_write_access(inode);
795 error = __mnt_want_write(f->f_path.mnt);
796 if (unlikely(error)) {
797 put_write_access(inode);
800 f->f_mode |= FMODE_WRITER;
803 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
804 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
805 f->f_mode |= FMODE_ATOMIC_POS;
807 f->f_op = fops_get(inode->i_fop);
808 if (WARN_ON(!f->f_op)) {
813 error = security_file_open(f);
817 error = break_lease(locks_inode(f), f->f_flags);
821 /* normally all 3 are set; ->open() can clear them if needed */
822 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
824 open = f->f_op->open;
826 error = open(inode, f);
830 f->f_mode |= FMODE_OPENED;
831 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
832 i_readcount_inc(inode);
833 if ((f->f_mode & FMODE_READ) &&
834 likely(f->f_op->read || f->f_op->read_iter))
835 f->f_mode |= FMODE_CAN_READ;
836 if ((f->f_mode & FMODE_WRITE) &&
837 likely(f->f_op->write || f->f_op->write_iter))
838 f->f_mode |= FMODE_CAN_WRITE;
840 f->f_write_hint = WRITE_LIFE_NOT_SET;
841 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
843 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
845 /* NB: we're sure to have correct a_ops only after f_op->open */
846 if (f->f_flags & O_DIRECT) {
847 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
852 * XXX: Huge page cache doesn't support writing yet. Drop all page
853 * cache for this file before processing writes.
855 if (f->f_mode & FMODE_WRITE) {
857 * Paired with smp_mb() in collapse_file() to ensure nr_thps
858 * is up to date and the update to i_writecount by
859 * get_write_access() is visible. Ensures subsequent insertion
860 * of THPs into the page cache will fail.
863 if (filemap_nr_thps(inode->i_mapping))
864 truncate_pagecache(inode, 0);
870 if (WARN_ON_ONCE(error > 0))
873 if (f->f_mode & FMODE_WRITER) {
874 put_write_access(inode);
875 __mnt_drop_write(f->f_path.mnt);
878 path_put(&f->f_path);
879 f->f_path.mnt = NULL;
880 f->f_path.dentry = NULL;
886 * finish_open - finish opening a file
887 * @file: file pointer
888 * @dentry: pointer to dentry
889 * @open: open callback
890 * @opened: state of open
892 * This can be used to finish opening a file passed to i_op->atomic_open().
894 * If the open callback is set to NULL, then the standard f_op->open()
895 * filesystem callback is substituted.
897 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
898 * the return value of d_splice_alias(), then the caller needs to perform dput()
899 * on it after finish_open().
901 * Returns zero on success or -errno if the open failed.
903 int finish_open(struct file *file, struct dentry *dentry,
904 int (*open)(struct inode *, struct file *))
906 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
908 file->f_path.dentry = dentry;
909 return do_dentry_open(file, d_backing_inode(dentry), open);
911 EXPORT_SYMBOL(finish_open);
914 * finish_no_open - finish ->atomic_open() without opening the file
916 * @file: file pointer
917 * @dentry: dentry or NULL (as returned from ->lookup())
919 * This can be used to set the result of a successful lookup in ->atomic_open().
921 * NB: unlike finish_open() this function does consume the dentry reference and
922 * the caller need not dput() it.
924 * Returns "0" which must be the return value of ->atomic_open() after having
925 * called this function.
927 int finish_no_open(struct file *file, struct dentry *dentry)
929 file->f_path.dentry = dentry;
932 EXPORT_SYMBOL(finish_no_open);
934 char *file_path(struct file *filp, char *buf, int buflen)
936 return d_path(&filp->f_path, buf, buflen);
938 EXPORT_SYMBOL(file_path);
941 * vfs_open - open the file at the given path
942 * @path: path to open
943 * @file: newly allocated file with f_flag initialized
944 * @cred: credentials to use
946 int vfs_open(const struct path *path, struct file *file)
948 file->f_path = *path;
949 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
952 struct file *dentry_open(const struct path *path, int flags,
953 const struct cred *cred)
958 validate_creds(cred);
960 /* We must always pass in a valid mount pointer. */
963 f = alloc_empty_file(flags, cred);
965 error = vfs_open(path, f);
973 EXPORT_SYMBOL(dentry_open);
975 struct file *open_with_fake_path(const struct path *path, int flags,
976 struct inode *inode, const struct cred *cred)
978 struct file *f = alloc_empty_file_noaccount(flags, cred);
983 error = do_dentry_open(f, inode, NULL);
991 EXPORT_SYMBOL(open_with_fake_path);
993 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
994 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
996 inline struct open_how build_open_how(int flags, umode_t mode)
998 struct open_how how = {
999 .flags = flags & VALID_OPEN_FLAGS,
1000 .mode = mode & S_IALLUGO,
1003 /* O_PATH beats everything else. */
1004 if (how.flags & O_PATH)
1005 how.flags &= O_PATH_FLAGS;
1006 /* Modes should only be set for create-like flags. */
1007 if (!WILL_CREATE(how.flags))
1012 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1014 u64 flags = how->flags;
1015 u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1016 int lookup_flags = 0;
1017 int acc_mode = ACC_MODE(flags);
1019 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1020 "struct open_flags doesn't yet handle flags > 32 bits");
1023 * Strip flags that either shouldn't be set by userspace like
1024 * FMODE_NONOTIFY or that aren't relevant in determining struct
1025 * open_flags like O_CLOEXEC.
1030 * Older syscalls implicitly clear all of the invalid flags or argument
1031 * values before calling build_open_flags(), but openat2(2) checks all
1034 if (flags & ~VALID_OPEN_FLAGS)
1036 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1039 /* Scoping flags are mutually exclusive. */
1040 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1043 /* Deal with the mode. */
1044 if (WILL_CREATE(flags)) {
1045 if (how->mode & ~S_IALLUGO)
1047 op->mode = how->mode | S_IFREG;
1055 * In order to ensure programs get explicit errors when trying to use
1056 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1057 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1058 * have to require userspace to explicitly set it.
1060 if (flags & __O_TMPFILE) {
1061 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1063 if (!(acc_mode & MAY_WRITE))
1066 if (flags & O_PATH) {
1067 /* O_PATH only permits certain other flags to be set. */
1068 if (flags & ~O_PATH_FLAGS)
1074 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1075 * check for O_DSYNC if the need any syncing at all we enforce it's
1076 * always set instead of having to deal with possibly weird behaviour
1077 * for malicious applications setting only __O_SYNC.
1079 if (flags & __O_SYNC)
1082 op->open_flag = flags;
1084 /* O_TRUNC implies we need access checks for write permissions */
1085 if (flags & O_TRUNC)
1086 acc_mode |= MAY_WRITE;
1088 /* Allow the LSM permission hook to distinguish append
1089 access from general write access. */
1090 if (flags & O_APPEND)
1091 acc_mode |= MAY_APPEND;
1093 op->acc_mode = acc_mode;
1095 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1097 if (flags & O_CREAT) {
1098 op->intent |= LOOKUP_CREATE;
1099 if (flags & O_EXCL) {
1100 op->intent |= LOOKUP_EXCL;
1101 flags |= O_NOFOLLOW;
1105 if (flags & O_DIRECTORY)
1106 lookup_flags |= LOOKUP_DIRECTORY;
1107 if (!(flags & O_NOFOLLOW))
1108 lookup_flags |= LOOKUP_FOLLOW;
1110 if (how->resolve & RESOLVE_NO_XDEV)
1111 lookup_flags |= LOOKUP_NO_XDEV;
1112 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1113 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1114 if (how->resolve & RESOLVE_NO_SYMLINKS)
1115 lookup_flags |= LOOKUP_NO_SYMLINKS;
1116 if (how->resolve & RESOLVE_BENEATH)
1117 lookup_flags |= LOOKUP_BENEATH;
1118 if (how->resolve & RESOLVE_IN_ROOT)
1119 lookup_flags |= LOOKUP_IN_ROOT;
1120 if (how->resolve & RESOLVE_CACHED) {
1121 /* Don't bother even trying for create/truncate/tmpfile open */
1122 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1124 lookup_flags |= LOOKUP_CACHED;
1127 op->lookup_flags = lookup_flags;
1132 * file_open_name - open file and return file pointer
1134 * @name: struct filename containing path to open
1135 * @flags: open flags as per the open(2) second argument
1136 * @mode: mode for the new file if O_CREAT is set, else ignored
1138 * This is the helper to open a file from kernelspace if you really
1139 * have to. But in generally you should not do this, so please move
1140 * along, nothing to see here..
1142 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1144 struct open_flags op;
1145 struct open_how how = build_open_how(flags, mode);
1146 int err = build_open_flags(&how, &op);
1148 return ERR_PTR(err);
1149 return do_filp_open(AT_FDCWD, name, &op);
1153 * filp_open - open file and return file pointer
1155 * @filename: path to open
1156 * @flags: open flags as per the open(2) second argument
1157 * @mode: mode for the new file if O_CREAT is set, else ignored
1159 * This is the helper to open a file from kernelspace if you really
1160 * have to. But in generally you should not do this, so please move
1161 * along, nothing to see here..
1163 struct file *filp_open(const char *filename, int flags, umode_t mode)
1165 struct filename *name = getname_kernel(filename);
1166 struct file *file = ERR_CAST(name);
1168 if (!IS_ERR(name)) {
1169 file = file_open_name(name, flags, mode);
1174 EXPORT_SYMBOL(filp_open);
1176 struct file *file_open_root(const struct path *root,
1177 const char *filename, int flags, umode_t mode)
1179 struct open_flags op;
1180 struct open_how how = build_open_how(flags, mode);
1181 int err = build_open_flags(&how, &op);
1183 return ERR_PTR(err);
1184 return do_file_open_root(root, filename, &op);
1186 EXPORT_SYMBOL(file_open_root);
1188 static long do_sys_openat2(int dfd, const char __user *filename,
1189 struct open_how *how)
1191 struct open_flags op;
1192 int fd = build_open_flags(how, &op);
1193 struct filename *tmp;
1198 tmp = getname(filename);
1200 return PTR_ERR(tmp);
1202 fd = get_unused_fd_flags(how->flags);
1204 struct file *f = do_filp_open(dfd, tmp, &op);
1217 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1219 struct open_how how = build_open_how(flags, mode);
1220 return do_sys_openat2(dfd, filename, &how);
1224 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1226 if (force_o_largefile())
1227 flags |= O_LARGEFILE;
1228 return do_sys_open(AT_FDCWD, filename, flags, mode);
1231 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1234 if (force_o_largefile())
1235 flags |= O_LARGEFILE;
1236 return do_sys_open(dfd, filename, flags, mode);
1239 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1240 struct open_how __user *, how, size_t, usize)
1243 struct open_how tmp;
1245 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1246 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1248 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1251 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1255 /* O_LARGEFILE is only allowed for non-O_PATH. */
1256 if (!(tmp.flags & O_PATH) && force_o_largefile())
1257 tmp.flags |= O_LARGEFILE;
1259 return do_sys_openat2(dfd, filename, &tmp);
1262 #ifdef CONFIG_COMPAT
1264 * Exactly like sys_open(), except that it doesn't set the
1267 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1269 return do_sys_open(AT_FDCWD, filename, flags, mode);
1273 * Exactly like sys_openat(), except that it doesn't set the
1276 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1278 return do_sys_open(dfd, filename, flags, mode);
1285 * For backward compatibility? Maybe this should be moved
1286 * into arch/i386 instead?
1288 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1290 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1292 if (force_o_largefile())
1293 flags |= O_LARGEFILE;
1294 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1299 * "id" is the POSIX thread ID. We use the
1300 * files pointer for this..
1302 int filp_close(struct file *filp, fl_owner_t id)
1306 if (!file_count(filp)) {
1307 printk(KERN_ERR "VFS: Close: file count is 0\n");
1311 if (filp->f_op->flush)
1312 retval = filp->f_op->flush(filp, id);
1314 if (likely(!(filp->f_mode & FMODE_PATH))) {
1315 dnotify_flush(filp, id);
1316 locks_remove_posix(filp, id);
1322 EXPORT_SYMBOL(filp_close);
1325 * Careful here! We test whether the file pointer is NULL before
1326 * releasing the fd. This ensures that one clone task can't release
1327 * an fd while another clone is opening it.
1329 SYSCALL_DEFINE1(close, unsigned int, fd)
1331 int retval = close_fd(fd);
1333 /* can't restart close syscall because file table entry was cleared */
1334 if (unlikely(retval == -ERESTARTSYS ||
1335 retval == -ERESTARTNOINTR ||
1336 retval == -ERESTARTNOHAND ||
1337 retval == -ERESTART_RESTARTBLOCK))
1344 * close_range() - Close all file descriptors in a given range.
1346 * @fd: starting file descriptor to close
1347 * @max_fd: last file descriptor to close
1348 * @flags: reserved for future extensions
1350 * This closes a range of file descriptors. All file descriptors
1351 * from @fd up to and including @max_fd are closed.
1352 * Currently, errors to close a given file descriptor are ignored.
1354 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1355 unsigned int, flags)
1357 return __close_range(fd, max_fd, flags);
1361 * This routine simulates a hangup on the tty, to arrange that users
1362 * are given clean terminals at login time.
1364 SYSCALL_DEFINE0(vhangup)
1366 if (capable(CAP_SYS_TTY_CONFIG)) {
1374 * Called when an inode is about to be open.
1375 * We use this to disallow opening large files on 32bit systems if
1376 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1377 * on this flag in sys_open.
1379 int generic_file_open(struct inode * inode, struct file * filp)
1381 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1386 EXPORT_SYMBOL(generic_file_open);
1389 * This is used by subsystems that don't want seekable
1390 * file descriptors. The function is not supposed to ever fail, the only
1391 * reason it returns an 'int' and not 'void' is so that it can be plugged
1392 * directly into file_operations structure.
1394 int nonseekable_open(struct inode *inode, struct file *filp)
1396 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1400 EXPORT_SYMBOL(nonseekable_open);
1403 * stream_open is used by subsystems that want stream-like file descriptors.
1404 * Such file descriptors are not seekable and don't have notion of position
1405 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1406 * Contrary to file descriptors of other regular files, .read() and .write()
1407 * can run simultaneously.
1409 * stream_open never fails and is marked to return int so that it could be
1410 * directly used as file_operations.open .
1412 int stream_open(struct inode *inode, struct file *filp)
1414 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1415 filp->f_mode |= FMODE_STREAM;
1419 EXPORT_SYMBOL(stream_open);