4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/string.h>
9 #include <linux/file.h>
10 #include <linux/fdtable.h>
11 #include <linux/fsnotify.h>
12 #include <linux/module.h>
13 #include <linux/tty.h>
14 #include <linux/namei.h>
15 #include <linux/backing-dev.h>
16 #include <linux/capability.h>
17 #include <linux/securebits.h>
18 #include <linux/security.h>
19 #include <linux/mount.h>
20 #include <linux/fcntl.h>
21 #include <linux/slab.h>
22 #include <asm/uaccess.h>
24 #include <linux/personality.h>
25 #include <linux/pagemap.h>
26 #include <linux/syscalls.h>
27 #include <linux/rcupdate.h>
28 #include <linux/audit.h>
29 #include <linux/falloc.h>
30 #include <linux/fs_struct.h>
31 #include <linux/ima.h>
32 #include <linux/dnotify.h>
36 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
40 struct iattr newattrs;
42 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46 newattrs.ia_size = length;
47 newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 newattrs.ia_file = filp;
50 newattrs.ia_valid |= ATTR_FILE;
53 /* Remove suid/sgid on truncate too */
54 ret = should_remove_suid(dentry);
56 newattrs.ia_valid |= ret | ATTR_FORCE;
58 mutex_lock(&dentry->d_inode->i_mutex);
59 ret = notify_change(dentry, &newattrs);
60 mutex_unlock(&dentry->d_inode->i_mutex);
64 static long do_sys_truncate(const char __user *pathname, loff_t length)
71 if (length < 0) /* sorry, but loff_t says... */
74 error = user_path(pathname, &path);
77 inode = path.dentry->d_inode;
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
81 if (S_ISDIR(inode->i_mode))
85 if (!S_ISREG(inode->i_mode))
88 error = mnt_want_write(path.mnt);
92 error = inode_permission(inode, MAY_WRITE);
94 goto mnt_drop_write_and_out;
98 goto mnt_drop_write_and_out;
100 error = get_write_access(inode);
102 goto mnt_drop_write_and_out;
105 * Make sure that there are no leases. get_write_access() protects
106 * against the truncate racing with a lease-granting setlease().
108 error = break_lease(inode, O_WRONLY);
110 goto put_write_and_out;
112 error = locks_verify_truncate(inode, NULL, length);
114 error = security_path_truncate(&path);
116 error = do_truncate(path.dentry, length, 0, NULL);
119 put_write_access(inode);
120 mnt_drop_write_and_out:
121 mnt_drop_write(path.mnt);
128 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
130 return do_sys_truncate(path, length);
133 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
136 struct dentry *dentry;
148 /* explicitly opened as large or we are on 64-bit box */
149 if (f.file->f_flags & O_LARGEFILE)
152 dentry = f.file->f_path.dentry;
153 inode = dentry->d_inode;
155 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
159 /* Cannot ftruncate over 2^31 bytes without large file support */
160 if (small && length > MAX_NON_LFS)
164 if (IS_APPEND(inode))
167 sb_start_write(inode->i_sb);
168 error = locks_verify_truncate(inode, f.file, length);
170 error = security_path_truncate(&f.file->f_path);
172 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
173 sb_end_write(inode->i_sb);
180 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
182 long ret = do_sys_ftruncate(fd, length, 1);
183 /* avoid REGPARM breakage on x86: */
184 asmlinkage_protect(2, ret, fd, length);
188 /* LFS versions of truncate are only needed on 32 bit machines */
189 #if BITS_PER_LONG == 32
190 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
192 return do_sys_truncate(path, length);
194 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
195 asmlinkage long SyS_truncate64(long path, loff_t length)
197 return SYSC_truncate64((const char __user *) path, length);
199 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
202 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
204 long ret = do_sys_ftruncate(fd, length, 0);
205 /* avoid REGPARM breakage on x86: */
206 asmlinkage_protect(2, ret, fd, length);
209 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
210 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
212 return SYSC_ftruncate64((unsigned int) fd, length);
214 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
216 #endif /* BITS_PER_LONG == 32 */
219 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
221 struct inode *inode = file->f_path.dentry->d_inode;
224 if (offset < 0 || len <= 0)
227 /* Return error if mode is not supported */
228 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
231 /* Punch hole must have keep size set */
232 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
233 !(mode & FALLOC_FL_KEEP_SIZE))
236 if (!(file->f_mode & FMODE_WRITE))
239 /* It's not possible punch hole on append only file */
240 if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
243 if (IS_IMMUTABLE(inode))
247 * Revalidate the write permissions, in case security policy has
248 * changed since the files were opened.
250 ret = security_file_permission(file, MAY_WRITE);
254 if (S_ISFIFO(inode->i_mode))
258 * Let individual file system decide if it supports preallocation
259 * for directories or not.
261 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
264 /* Check for wrap through zero too */
265 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
268 if (!file->f_op->fallocate)
271 sb_start_write(inode->i_sb);
272 ret = file->f_op->fallocate(file, mode, offset, len);
273 sb_end_write(inode->i_sb);
277 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
279 struct fd f = fdget(fd);
283 error = do_fallocate(f.file, mode, offset, len);
289 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
290 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
292 return SYSC_fallocate((int)fd, (int)mode, offset, len);
294 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
298 * access() needs to use the real uid/gid, not the effective uid/gid.
299 * We do this by temporarily clearing all FS-related capabilities and
300 * switching the fsuid/fsgid around to the real ones.
302 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
304 const struct cred *old_cred;
305 struct cred *override_cred;
310 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
313 override_cred = prepare_creds();
317 override_cred->fsuid = override_cred->uid;
318 override_cred->fsgid = override_cred->gid;
320 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
321 /* Clear the capabilities if we switch to a non-root user */
322 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
323 if (!uid_eq(override_cred->uid, root_uid))
324 cap_clear(override_cred->cap_effective);
326 override_cred->cap_effective =
327 override_cred->cap_permitted;
330 old_cred = override_creds(override_cred);
332 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
336 inode = path.dentry->d_inode;
338 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
340 * MAY_EXEC on regular files is denied if the fs is mounted
341 * with the "noexec" flag.
344 if (path.mnt->mnt_flags & MNT_NOEXEC)
345 goto out_path_release;
348 res = inode_permission(inode, mode | MAY_ACCESS);
349 /* SuS v2 requires we report a read only fs too */
350 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
351 goto out_path_release;
353 * This is a rare case where using __mnt_is_readonly()
354 * is OK without a mnt_want/drop_write() pair. Since
355 * no actual write to the fs is performed here, we do
356 * not need to telegraph to that to anyone.
358 * By doing this, we accept that this access is
359 * inherently racy and know that the fs may change
360 * state before we even see this result.
362 if (__mnt_is_readonly(path.mnt))
368 revert_creds(old_cred);
369 put_cred(override_cred);
373 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
375 return sys_faccessat(AT_FDCWD, filename, mode);
378 SYSCALL_DEFINE1(chdir, const char __user *, filename)
383 error = user_path_dir(filename, &path);
387 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
391 set_fs_pwd(current->fs, &path);
399 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
401 struct fd f = fdget_raw(fd);
409 inode = f.file->f_path.dentry->d_inode;
412 if (!S_ISDIR(inode->i_mode))
415 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
417 set_fs_pwd(current->fs, &f.file->f_path);
424 SYSCALL_DEFINE1(chroot, const char __user *, filename)
429 error = user_path_dir(filename, &path);
433 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
438 if (!capable(CAP_SYS_CHROOT))
440 error = security_path_chroot(&path);
444 set_fs_root(current->fs, &path);
452 static int chmod_common(struct path *path, umode_t mode)
454 struct inode *inode = path->dentry->d_inode;
455 struct iattr newattrs;
458 error = mnt_want_write(path->mnt);
461 mutex_lock(&inode->i_mutex);
462 error = security_path_chmod(path, mode);
465 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
466 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
467 error = notify_change(path->dentry, &newattrs);
469 mutex_unlock(&inode->i_mutex);
470 mnt_drop_write(path->mnt);
474 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
481 audit_inode(NULL, file->f_path.dentry);
482 err = chmod_common(&file->f_path, mode);
488 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
493 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
495 error = chmod_common(&path, mode);
501 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
503 return sys_fchmodat(AT_FDCWD, filename, mode);
506 static int chown_common(struct path *path, uid_t user, gid_t group)
508 struct inode *inode = path->dentry->d_inode;
510 struct iattr newattrs;
514 uid = make_kuid(current_user_ns(), user);
515 gid = make_kgid(current_user_ns(), group);
517 newattrs.ia_valid = ATTR_CTIME;
518 if (user != (uid_t) -1) {
521 newattrs.ia_valid |= ATTR_UID;
522 newattrs.ia_uid = uid;
524 if (group != (gid_t) -1) {
527 newattrs.ia_valid |= ATTR_GID;
528 newattrs.ia_gid = gid;
530 if (!S_ISDIR(inode->i_mode))
532 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
533 mutex_lock(&inode->i_mutex);
534 error = security_path_chown(path, uid, gid);
536 error = notify_change(path->dentry, &newattrs);
537 mutex_unlock(&inode->i_mutex);
542 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
543 gid_t, group, int, flag)
549 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
552 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
553 if (flag & AT_EMPTY_PATH)
554 lookup_flags |= LOOKUP_EMPTY;
555 error = user_path_at(dfd, filename, lookup_flags, &path);
558 error = mnt_want_write(path.mnt);
561 error = chown_common(&path, user, group);
562 mnt_drop_write(path.mnt);
569 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
571 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
574 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
576 return sys_fchownat(AT_FDCWD, filename, user, group,
577 AT_SYMLINK_NOFOLLOW);
580 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
582 struct fd f = fdget(fd);
588 error = mnt_want_write_file(f.file);
591 audit_inode(NULL, f.file->f_path.dentry);
592 error = chown_common(&f.file->f_path, user, group);
593 mnt_drop_write_file(f.file);
601 * You have to be very careful that these write
602 * counts get cleaned up in error cases and
603 * upon __fput(). This should probably never
604 * be called outside of __dentry_open().
606 static inline int __get_file_write_access(struct inode *inode,
607 struct vfsmount *mnt)
610 error = get_write_access(inode);
614 * Do not take mount writer counts on
615 * special files since no writes to
616 * the mount itself will occur.
618 if (!special_file(inode->i_mode)) {
620 * Balanced in __fput()
622 error = __mnt_want_write(mnt);
624 put_write_access(inode);
629 int open_check_o_direct(struct file *f)
631 /* NB: we're sure to have correct a_ops only after f_op->open */
632 if (f->f_flags & O_DIRECT) {
633 if (!f->f_mapping->a_ops ||
634 ((!f->f_mapping->a_ops->direct_IO) &&
635 (!f->f_mapping->a_ops->get_xip_mem))) {
642 static int do_dentry_open(struct file *f,
643 int (*open)(struct inode *, struct file *),
644 const struct cred *cred)
646 static const struct file_operations empty_fops = {};
650 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
651 FMODE_PREAD | FMODE_PWRITE;
653 if (unlikely(f->f_flags & O_PATH))
654 f->f_mode = FMODE_PATH;
656 path_get(&f->f_path);
657 inode = f->f_path.dentry->d_inode;
658 if (f->f_mode & FMODE_WRITE) {
659 error = __get_file_write_access(inode, f->f_path.mnt);
662 if (!special_file(inode->i_mode))
666 f->f_mapping = inode->i_mapping;
668 file_sb_list_add(f, inode->i_sb);
670 if (unlikely(f->f_mode & FMODE_PATH)) {
671 f->f_op = &empty_fops;
675 f->f_op = fops_get(inode->i_fop);
677 error = security_file_open(f, cred);
681 error = break_lease(inode, f->f_flags);
685 if (!open && f->f_op)
686 open = f->f_op->open;
688 error = open(inode, f);
692 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
693 i_readcount_inc(inode);
695 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
697 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
704 if (f->f_mode & FMODE_WRITE) {
705 put_write_access(inode);
706 if (!special_file(inode->i_mode)) {
708 * We don't consider this a real
709 * mnt_want/drop_write() pair
710 * because it all happenend right
711 * here, so just reset the state.
714 __mnt_drop_write(f->f_path.mnt);
718 path_put(&f->f_path);
719 f->f_path.mnt = NULL;
720 f->f_path.dentry = NULL;
725 * finish_open - finish opening a file
726 * @od: opaque open data
727 * @dentry: pointer to dentry
728 * @open: open callback
730 * This can be used to finish opening a file passed to i_op->atomic_open().
732 * If the open callback is set to NULL, then the standard f_op->open()
733 * filesystem callback is substituted.
735 int finish_open(struct file *file, struct dentry *dentry,
736 int (*open)(struct inode *, struct file *),
740 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
742 file->f_path.dentry = dentry;
743 error = do_dentry_open(file, open, current_cred());
745 *opened |= FILE_OPENED;
749 EXPORT_SYMBOL(finish_open);
752 * finish_no_open - finish ->atomic_open() without opening the file
754 * @od: opaque open data
755 * @dentry: dentry or NULL (as returned from ->lookup())
757 * This can be used to set the result of a successful lookup in ->atomic_open().
758 * The filesystem's atomic_open() method shall return NULL after calling this.
760 int finish_no_open(struct file *file, struct dentry *dentry)
762 file->f_path.dentry = dentry;
765 EXPORT_SYMBOL(finish_no_open);
767 struct file *dentry_open(const struct path *path, int flags,
768 const struct cred *cred)
773 validate_creds(cred);
775 /* We must always pass in a valid mount pointer. */
779 f = get_empty_filp();
781 return ERR_PTR(error);
785 error = do_dentry_open(f, NULL, cred);
787 error = open_check_o_direct(f);
798 EXPORT_SYMBOL(dentry_open);
800 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
802 int lookup_flags = 0;
806 op->mode = (mode & S_IALLUGO) | S_IFREG;
810 /* Must never be set by userspace */
811 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
814 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
815 * check for O_DSYNC if the need any syncing at all we enforce it's
816 * always set instead of having to deal with possibly weird behaviour
817 * for malicious applications setting only __O_SYNC.
819 if (flags & __O_SYNC)
823 * If we have O_PATH in the open flag. Then we
824 * cannot have anything other than the below set of flags
826 if (flags & O_PATH) {
827 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
830 acc_mode = MAY_OPEN | ACC_MODE(flags);
833 op->open_flag = flags;
835 /* O_TRUNC implies we need access checks for write permissions */
837 acc_mode |= MAY_WRITE;
839 /* Allow the LSM permission hook to distinguish append
840 access from general write access. */
841 if (flags & O_APPEND)
842 acc_mode |= MAY_APPEND;
844 op->acc_mode = acc_mode;
846 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
848 if (flags & O_CREAT) {
849 op->intent |= LOOKUP_CREATE;
851 op->intent |= LOOKUP_EXCL;
854 if (flags & O_DIRECTORY)
855 lookup_flags |= LOOKUP_DIRECTORY;
856 if (!(flags & O_NOFOLLOW))
857 lookup_flags |= LOOKUP_FOLLOW;
862 * filp_open - open file and return file pointer
864 * @filename: path to open
865 * @flags: open flags as per the open(2) second argument
866 * @mode: mode for the new file if O_CREAT is set, else ignored
868 * This is the helper to open a file from kernelspace if you really
869 * have to. But in generally you should not do this, so please move
870 * along, nothing to see here..
872 struct file *filp_open(const char *filename, int flags, umode_t mode)
874 struct open_flags op;
875 int lookup = build_open_flags(flags, mode, &op);
876 return do_filp_open(AT_FDCWD, filename, &op, lookup);
878 EXPORT_SYMBOL(filp_open);
880 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
881 const char *filename, int flags)
883 struct open_flags op;
884 int lookup = build_open_flags(flags, 0, &op);
886 return ERR_PTR(-EINVAL);
887 if (!filename && (flags & O_DIRECTORY))
888 if (!dentry->d_inode->i_op->lookup)
889 return ERR_PTR(-ENOTDIR);
890 return do_file_open_root(dentry, mnt, filename, &op, lookup);
892 EXPORT_SYMBOL(file_open_root);
894 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
896 struct open_flags op;
897 int lookup = build_open_flags(flags, mode, &op);
898 char *tmp = getname(filename);
899 int fd = PTR_ERR(tmp);
902 fd = get_unused_fd_flags(flags);
904 struct file *f = do_filp_open(dfd, tmp, &op, lookup);
918 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
922 if (force_o_largefile())
923 flags |= O_LARGEFILE;
925 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
926 /* avoid REGPARM breakage on x86: */
927 asmlinkage_protect(3, ret, filename, flags, mode);
931 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
936 if (force_o_largefile())
937 flags |= O_LARGEFILE;
939 ret = do_sys_open(dfd, filename, flags, mode);
940 /* avoid REGPARM breakage on x86: */
941 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
948 * For backward compatibility? Maybe this should be moved
949 * into arch/i386 instead?
951 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
953 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
959 * "id" is the POSIX thread ID. We use the
960 * files pointer for this..
962 int filp_close(struct file *filp, fl_owner_t id)
966 if (!file_count(filp)) {
967 printk(KERN_ERR "VFS: Close: file count is 0\n");
971 if (filp->f_op && filp->f_op->flush)
972 retval = filp->f_op->flush(filp, id);
974 if (likely(!(filp->f_mode & FMODE_PATH))) {
975 dnotify_flush(filp, id);
976 locks_remove_posix(filp, id);
982 EXPORT_SYMBOL(filp_close);
985 * Careful here! We test whether the file pointer is NULL before
986 * releasing the fd. This ensures that one clone task can't release
987 * an fd while another clone is opening it.
989 SYSCALL_DEFINE1(close, unsigned int, fd)
991 int retval = __close_fd(current->files, fd);
993 /* can't restart close syscall because file table entry was cleared */
994 if (unlikely(retval == -ERESTARTSYS ||
995 retval == -ERESTARTNOINTR ||
996 retval == -ERESTARTNOHAND ||
997 retval == -ERESTART_RESTARTBLOCK))
1002 EXPORT_SYMBOL(sys_close);
1005 * This routine simulates a hangup on the tty, to arrange that users
1006 * are given clean terminals at login time.
1008 SYSCALL_DEFINE0(vhangup)
1010 if (capable(CAP_SYS_TTY_CONFIG)) {
1018 * Called when an inode is about to be open.
1019 * We use this to disallow opening large files on 32bit systems if
1020 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1021 * on this flag in sys_open.
1023 int generic_file_open(struct inode * inode, struct file * filp)
1025 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1030 EXPORT_SYMBOL(generic_file_open);
1033 * This is used by subsystems that don't want seekable
1034 * file descriptors. The function is not supposed to ever fail, the only
1035 * reason it returns an 'int' and not 'void' is so that it can be plugged
1036 * directly into file_operations structure.
1038 int nonseekable_open(struct inode *inode, struct file *filp)
1040 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1044 EXPORT_SYMBOL(nonseekable_open);