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>
39 int do_truncate(struct user_namespace
*mnt_userns
, 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(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(mnt_userns
, dentry
, &newattrs
, NULL
);
66 inode_unlock(dentry
->d_inode
);
70 long vfs_truncate(const struct path
*path
, loff_t length
)
72 struct user_namespace
*mnt_userns
;
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 mnt_userns
= mnt_user_ns(path
->mnt
);
89 error
= inode_permission(mnt_userns
, 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(mnt_userns
, 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_sys_ftruncate(unsigned int fd
, loff_t length
, int small
)
159 struct dentry
*dentry
;
171 /* explicitly opened as large or we are on 64-bit box */
172 if (f
.file
->f_flags
& O_LARGEFILE
)
175 dentry
= f
.file
->f_path
.dentry
;
176 inode
= dentry
->d_inode
;
178 if (!S_ISREG(inode
->i_mode
) || !(f
.file
->f_mode
& FMODE_WRITE
))
182 /* Cannot ftruncate over 2^31 bytes without large file support */
183 if (small
&& length
> MAX_NON_LFS
)
187 /* Check IS_APPEND on real upper inode */
188 if (IS_APPEND(file_inode(f
.file
)))
190 sb_start_write(inode
->i_sb
);
191 error
= security_path_truncate(&f
.file
->f_path
);
193 error
= do_truncate(file_mnt_user_ns(f
.file
), dentry
, length
,
194 ATTR_MTIME
| ATTR_CTIME
, f
.file
);
195 sb_end_write(inode
->i_sb
);
202 SYSCALL_DEFINE2(ftruncate
, unsigned int, fd
, unsigned long, length
)
204 return do_sys_ftruncate(fd
, length
, 1);
208 COMPAT_SYSCALL_DEFINE2(ftruncate
, unsigned int, fd
, compat_ulong_t
, length
)
210 return do_sys_ftruncate(fd
, length
, 1);
214 /* LFS versions of truncate are only needed on 32 bit machines */
215 #if BITS_PER_LONG == 32
216 SYSCALL_DEFINE2(truncate64
, const char __user
*, path
, loff_t
, length
)
218 return do_sys_truncate(path
, length
);
221 SYSCALL_DEFINE2(ftruncate64
, unsigned int, fd
, loff_t
, length
)
223 return do_sys_ftruncate(fd
, length
, 0);
225 #endif /* BITS_PER_LONG == 32 */
228 int vfs_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
230 struct inode
*inode
= file_inode(file
);
233 if (offset
< 0 || len
<= 0)
236 /* Return error if mode is not supported */
237 if (mode
& ~FALLOC_FL_SUPPORTED_MASK
)
240 /* Punch hole and zero range are mutually exclusive */
241 if ((mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
)) ==
242 (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
))
245 /* Punch hole must have keep size set */
246 if ((mode
& FALLOC_FL_PUNCH_HOLE
) &&
247 !(mode
& FALLOC_FL_KEEP_SIZE
))
250 /* Collapse range should only be used exclusively. */
251 if ((mode
& FALLOC_FL_COLLAPSE_RANGE
) &&
252 (mode
& ~FALLOC_FL_COLLAPSE_RANGE
))
255 /* Insert range should only be used exclusively. */
256 if ((mode
& FALLOC_FL_INSERT_RANGE
) &&
257 (mode
& ~FALLOC_FL_INSERT_RANGE
))
260 /* Unshare range should only be used with allocate mode. */
261 if ((mode
& FALLOC_FL_UNSHARE_RANGE
) &&
262 (mode
& ~(FALLOC_FL_UNSHARE_RANGE
| FALLOC_FL_KEEP_SIZE
)))
265 if (!(file
->f_mode
& FMODE_WRITE
))
269 * We can only allow pure fallocate on append only files
271 if ((mode
& ~FALLOC_FL_KEEP_SIZE
) && IS_APPEND(inode
))
274 if (IS_IMMUTABLE(inode
))
278 * We cannot allow any fallocate operation on an active swapfile
280 if (IS_SWAPFILE(inode
))
284 * Revalidate the write permissions, in case security policy has
285 * changed since the files were opened.
287 ret
= security_file_permission(file
, MAY_WRITE
);
291 if (S_ISFIFO(inode
->i_mode
))
294 if (S_ISDIR(inode
->i_mode
))
297 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
300 /* Check for wrap through zero too */
301 if (((offset
+ len
) > inode
->i_sb
->s_maxbytes
) || ((offset
+ len
) < 0))
304 if (!file
->f_op
->fallocate
)
307 file_start_write(file
);
308 ret
= file
->f_op
->fallocate(file
, mode
, offset
, len
);
311 * Create inotify and fanotify events.
313 * To keep the logic simple always create events if fallocate succeeds.
314 * This implies that events are even created if the file size remains
315 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
318 fsnotify_modify(file
);
320 file_end_write(file
);
323 EXPORT_SYMBOL_GPL(vfs_fallocate
);
325 int ksys_fallocate(int fd
, int mode
, loff_t offset
, loff_t len
)
327 struct fd f
= fdget(fd
);
331 error
= vfs_fallocate(f
.file
, mode
, offset
, len
);
337 SYSCALL_DEFINE4(fallocate
, int, fd
, int, mode
, loff_t
, offset
, loff_t
, len
)
339 return ksys_fallocate(fd
, mode
, offset
, len
);
343 * access() needs to use the real uid/gid, not the effective uid/gid.
344 * We do this by temporarily clearing all FS-related capabilities and
345 * switching the fsuid/fsgid around to the real ones.
347 static const struct cred
*access_override_creds(void)
349 const struct cred
*old_cred
;
350 struct cred
*override_cred
;
352 override_cred
= prepare_creds();
356 override_cred
->fsuid
= override_cred
->uid
;
357 override_cred
->fsgid
= override_cred
->gid
;
359 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
360 /* Clear the capabilities if we switch to a non-root user */
361 kuid_t root_uid
= make_kuid(override_cred
->user_ns
, 0);
362 if (!uid_eq(override_cred
->uid
, root_uid
))
363 cap_clear(override_cred
->cap_effective
);
365 override_cred
->cap_effective
=
366 override_cred
->cap_permitted
;
370 * The new set of credentials can *only* be used in
371 * task-synchronous circumstances, and does not need
372 * RCU freeing, unless somebody then takes a separate
375 * NOTE! This is _only_ true because this credential
376 * is used purely for override_creds() that installs
377 * it as the subjective cred. Other threads will be
378 * accessing ->real_cred, not the subjective cred.
380 * If somebody _does_ make a copy of this (using the
381 * 'get_current_cred()' function), that will clear the
382 * non_rcu field, because now that other user may be
383 * expecting RCU freeing. But normal thread-synchronous
384 * cred accesses will keep things non-RCY.
386 override_cred
->non_rcu
= 1;
388 old_cred
= override_creds(override_cred
);
390 /* override_cred() gets its own ref */
391 put_cred(override_cred
);
396 static long do_faccessat(int dfd
, const char __user
*filename
, int mode
, int flags
)
401 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
402 const struct cred
*old_cred
= NULL
;
404 if (mode
& ~S_IRWXO
) /* where's F_OK, X_OK, W_OK, R_OK? */
407 if (flags
& ~(AT_EACCESS
| AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
))
410 if (flags
& AT_SYMLINK_NOFOLLOW
)
411 lookup_flags
&= ~LOOKUP_FOLLOW
;
412 if (flags
& AT_EMPTY_PATH
)
413 lookup_flags
|= LOOKUP_EMPTY
;
415 if (!(flags
& AT_EACCESS
)) {
416 old_cred
= access_override_creds();
422 res
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
426 inode
= d_backing_inode(path
.dentry
);
428 if ((mode
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
430 * MAY_EXEC on regular files is denied if the fs is mounted
431 * with the "noexec" flag.
434 if (path_noexec(&path
))
435 goto out_path_release
;
438 res
= inode_permission(mnt_user_ns(path
.mnt
), inode
, mode
| MAY_ACCESS
);
439 /* SuS v2 requires we report a read only fs too */
440 if (res
|| !(mode
& S_IWOTH
) || special_file(inode
->i_mode
))
441 goto out_path_release
;
443 * This is a rare case where using __mnt_is_readonly()
444 * is OK without a mnt_want/drop_write() pair. Since
445 * no actual write to the fs is performed here, we do
446 * not need to telegraph to that to anyone.
448 * By doing this, we accept that this access is
449 * inherently racy and know that the fs may change
450 * state before we even see this result.
452 if (__mnt_is_readonly(path
.mnt
))
457 if (retry_estale(res
, lookup_flags
)) {
458 lookup_flags
|= LOOKUP_REVAL
;
463 revert_creds(old_cred
);
468 SYSCALL_DEFINE3(faccessat
, int, dfd
, const char __user
*, filename
, int, mode
)
470 return do_faccessat(dfd
, filename
, mode
, 0);
473 SYSCALL_DEFINE4(faccessat2
, int, dfd
, const char __user
*, filename
, int, mode
,
476 return do_faccessat(dfd
, filename
, mode
, flags
);
479 SYSCALL_DEFINE2(access
, const char __user
*, filename
, int, mode
)
481 return do_faccessat(AT_FDCWD
, filename
, mode
, 0);
484 SYSCALL_DEFINE1(chdir
, const char __user
*, filename
)
488 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
490 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
494 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
498 set_fs_pwd(current
->fs
, &path
);
502 if (retry_estale(error
, lookup_flags
)) {
503 lookup_flags
|= LOOKUP_REVAL
;
510 SYSCALL_DEFINE1(fchdir
, unsigned int, fd
)
512 struct fd f
= fdget_raw(fd
);
520 if (!d_can_lookup(f
.file
->f_path
.dentry
))
523 error
= file_permission(f
.file
, MAY_EXEC
| MAY_CHDIR
);
525 set_fs_pwd(current
->fs
, &f
.file
->f_path
);
532 SYSCALL_DEFINE1(chroot
, const char __user
*, filename
)
536 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
538 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
542 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
547 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT
))
549 error
= security_path_chroot(&path
);
553 set_fs_root(current
->fs
, &path
);
557 if (retry_estale(error
, lookup_flags
)) {
558 lookup_flags
|= LOOKUP_REVAL
;
565 int chmod_common(const struct path
*path
, umode_t mode
)
567 struct inode
*inode
= path
->dentry
->d_inode
;
568 struct inode
*delegated_inode
= NULL
;
569 struct iattr newattrs
;
572 error
= mnt_want_write(path
->mnt
);
577 error
= security_path_chmod(path
, mode
);
580 newattrs
.ia_mode
= (mode
& S_IALLUGO
) | (inode
->i_mode
& ~S_IALLUGO
);
581 newattrs
.ia_valid
= ATTR_MODE
| ATTR_CTIME
;
582 error
= notify_change(mnt_user_ns(path
->mnt
), path
->dentry
,
583 &newattrs
, &delegated_inode
);
586 if (delegated_inode
) {
587 error
= break_deleg_wait(&delegated_inode
);
591 mnt_drop_write(path
->mnt
);
595 int vfs_fchmod(struct file
*file
, umode_t mode
)
598 return chmod_common(&file
->f_path
, mode
);
601 SYSCALL_DEFINE2(fchmod
, unsigned int, fd
, umode_t
, mode
)
603 struct fd f
= fdget(fd
);
607 err
= vfs_fchmod(f
.file
, mode
);
613 static int do_fchmodat(int dfd
, const char __user
*filename
, umode_t mode
)
617 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
619 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
621 error
= chmod_common(&path
, mode
);
623 if (retry_estale(error
, lookup_flags
)) {
624 lookup_flags
|= LOOKUP_REVAL
;
631 SYSCALL_DEFINE3(fchmodat
, int, dfd
, const char __user
*, filename
,
634 return do_fchmodat(dfd
, filename
, mode
);
637 SYSCALL_DEFINE2(chmod
, const char __user
*, filename
, umode_t
, mode
)
639 return do_fchmodat(AT_FDCWD
, filename
, mode
);
642 int chown_common(const struct path
*path
, uid_t user
, gid_t group
)
644 struct user_namespace
*mnt_userns
, *fs_userns
;
645 struct inode
*inode
= path
->dentry
->d_inode
;
646 struct inode
*delegated_inode
= NULL
;
648 struct iattr newattrs
;
652 uid
= make_kuid(current_user_ns(), user
);
653 gid
= make_kgid(current_user_ns(), group
);
655 mnt_userns
= mnt_user_ns(path
->mnt
);
656 fs_userns
= i_user_ns(inode
);
657 uid
= mapped_kuid_user(mnt_userns
, fs_userns
, uid
);
658 gid
= mapped_kgid_user(mnt_userns
, fs_userns
, gid
);
661 newattrs
.ia_valid
= ATTR_CTIME
;
662 if (user
!= (uid_t
) -1) {
665 newattrs
.ia_valid
|= ATTR_UID
;
666 newattrs
.ia_uid
= uid
;
668 if (group
!= (gid_t
) -1) {
671 newattrs
.ia_valid
|= ATTR_GID
;
672 newattrs
.ia_gid
= gid
;
674 if (!S_ISDIR(inode
->i_mode
))
676 ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
678 error
= security_path_chown(path
, uid
, gid
);
680 error
= notify_change(mnt_userns
, path
->dentry
, &newattrs
,
683 if (delegated_inode
) {
684 error
= break_deleg_wait(&delegated_inode
);
691 int do_fchownat(int dfd
, const char __user
*filename
, uid_t user
, gid_t group
,
698 if ((flag
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
701 lookup_flags
= (flag
& AT_SYMLINK_NOFOLLOW
) ? 0 : LOOKUP_FOLLOW
;
702 if (flag
& AT_EMPTY_PATH
)
703 lookup_flags
|= LOOKUP_EMPTY
;
705 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
708 error
= mnt_want_write(path
.mnt
);
711 error
= chown_common(&path
, user
, group
);
712 mnt_drop_write(path
.mnt
);
715 if (retry_estale(error
, lookup_flags
)) {
716 lookup_flags
|= LOOKUP_REVAL
;
723 SYSCALL_DEFINE5(fchownat
, int, dfd
, const char __user
*, filename
, uid_t
, user
,
724 gid_t
, group
, int, flag
)
726 return do_fchownat(dfd
, filename
, user
, group
, flag
);
729 SYSCALL_DEFINE3(chown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
731 return do_fchownat(AT_FDCWD
, filename
, user
, group
, 0);
734 SYSCALL_DEFINE3(lchown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
736 return do_fchownat(AT_FDCWD
, filename
, user
, group
,
737 AT_SYMLINK_NOFOLLOW
);
740 int vfs_fchown(struct file
*file
, uid_t user
, gid_t group
)
744 error
= mnt_want_write_file(file
);
748 error
= chown_common(&file
->f_path
, user
, group
);
749 mnt_drop_write_file(file
);
753 int ksys_fchown(unsigned int fd
, uid_t user
, gid_t group
)
755 struct fd f
= fdget(fd
);
759 error
= vfs_fchown(f
.file
, user
, group
);
765 SYSCALL_DEFINE3(fchown
, unsigned int, fd
, uid_t
, user
, gid_t
, group
)
767 return ksys_fchown(fd
, user
, group
);
770 static int do_dentry_open(struct file
*f
,
772 int (*open
)(struct inode
*, struct file
*))
774 static const struct file_operations empty_fops
= {};
777 path_get(&f
->f_path
);
779 f
->f_mapping
= inode
->i_mapping
;
780 f
->f_wb_err
= filemap_sample_wb_err(f
->f_mapping
);
781 f
->f_sb_err
= file_sample_sb_err(f
);
783 if (unlikely(f
->f_flags
& O_PATH
)) {
784 f
->f_mode
= FMODE_PATH
| FMODE_OPENED
;
785 f
->f_op
= &empty_fops
;
789 if (f
->f_mode
& FMODE_WRITE
&& !special_file(inode
->i_mode
)) {
790 error
= get_write_access(inode
);
793 error
= __mnt_want_write(f
->f_path
.mnt
);
794 if (unlikely(error
)) {
795 put_write_access(inode
);
798 f
->f_mode
|= FMODE_WRITER
;
801 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
802 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
))
803 f
->f_mode
|= FMODE_ATOMIC_POS
;
805 f
->f_op
= fops_get(inode
->i_fop
);
806 if (WARN_ON(!f
->f_op
)) {
811 error
= security_file_open(f
);
815 error
= break_lease(locks_inode(f
), f
->f_flags
);
819 /* normally all 3 are set; ->open() can clear them if needed */
820 f
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
822 open
= f
->f_op
->open
;
824 error
= open(inode
, f
);
828 f
->f_mode
|= FMODE_OPENED
;
829 if ((f
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) == FMODE_READ
)
830 i_readcount_inc(inode
);
831 if ((f
->f_mode
& FMODE_READ
) &&
832 likely(f
->f_op
->read
|| f
->f_op
->read_iter
))
833 f
->f_mode
|= FMODE_CAN_READ
;
834 if ((f
->f_mode
& FMODE_WRITE
) &&
835 likely(f
->f_op
->write
|| f
->f_op
->write_iter
))
836 f
->f_mode
|= FMODE_CAN_WRITE
;
838 f
->f_flags
&= ~(O_CREAT
| O_EXCL
| O_NOCTTY
| O_TRUNC
);
840 file_ra_state_init(&f
->f_ra
, f
->f_mapping
->host
->i_mapping
);
842 /* NB: we're sure to have correct a_ops only after f_op->open */
843 if (f
->f_flags
& O_DIRECT
) {
844 if (!f
->f_mapping
->a_ops
|| !f
->f_mapping
->a_ops
->direct_IO
)
849 * XXX: Huge page cache doesn't support writing yet. Drop all page
850 * cache for this file before processing writes.
852 if (f
->f_mode
& FMODE_WRITE
) {
854 * Paired with smp_mb() in collapse_file() to ensure nr_thps
855 * is up to date and the update to i_writecount by
856 * get_write_access() is visible. Ensures subsequent insertion
857 * of THPs into the page cache will fail.
860 if (filemap_nr_thps(inode
->i_mapping
)) {
861 struct address_space
*mapping
= inode
->i_mapping
;
863 filemap_invalidate_lock(inode
->i_mapping
);
865 * unmap_mapping_range just need to be called once
866 * here, because the private pages is not need to be
867 * unmapped mapping (e.g. data segment of dynamic
868 * shared libraries here).
870 unmap_mapping_range(mapping
, 0, 0, 0);
871 truncate_inode_pages(mapping
, 0);
872 filemap_invalidate_unlock(inode
->i_mapping
);
879 if (WARN_ON_ONCE(error
> 0))
882 if (f
->f_mode
& FMODE_WRITER
) {
883 put_write_access(inode
);
884 __mnt_drop_write(f
->f_path
.mnt
);
887 path_put(&f
->f_path
);
888 f
->f_path
.mnt
= NULL
;
889 f
->f_path
.dentry
= NULL
;
895 * finish_open - finish opening a file
896 * @file: file pointer
897 * @dentry: pointer to dentry
898 * @open: open callback
899 * @opened: state of open
901 * This can be used to finish opening a file passed to i_op->atomic_open().
903 * If the open callback is set to NULL, then the standard f_op->open()
904 * filesystem callback is substituted.
906 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
907 * the return value of d_splice_alias(), then the caller needs to perform dput()
908 * on it after finish_open().
910 * Returns zero on success or -errno if the open failed.
912 int finish_open(struct file
*file
, struct dentry
*dentry
,
913 int (*open
)(struct inode
*, struct file
*))
915 BUG_ON(file
->f_mode
& FMODE_OPENED
); /* once it's opened, it's opened */
917 file
->f_path
.dentry
= dentry
;
918 return do_dentry_open(file
, d_backing_inode(dentry
), open
);
920 EXPORT_SYMBOL(finish_open
);
923 * finish_no_open - finish ->atomic_open() without opening the file
925 * @file: file pointer
926 * @dentry: dentry or NULL (as returned from ->lookup())
928 * This can be used to set the result of a successful lookup in ->atomic_open().
930 * NB: unlike finish_open() this function does consume the dentry reference and
931 * the caller need not dput() it.
933 * Returns "0" which must be the return value of ->atomic_open() after having
934 * called this function.
936 int finish_no_open(struct file
*file
, struct dentry
*dentry
)
938 file
->f_path
.dentry
= dentry
;
941 EXPORT_SYMBOL(finish_no_open
);
943 char *file_path(struct file
*filp
, char *buf
, int buflen
)
945 return d_path(&filp
->f_path
, buf
, buflen
);
947 EXPORT_SYMBOL(file_path
);
950 * vfs_open - open the file at the given path
951 * @path: path to open
952 * @file: newly allocated file with f_flag initialized
953 * @cred: credentials to use
955 int vfs_open(const struct path
*path
, struct file
*file
)
957 file
->f_path
= *path
;
958 return do_dentry_open(file
, d_backing_inode(path
->dentry
), NULL
);
961 struct file
*dentry_open(const struct path
*path
, int flags
,
962 const struct cred
*cred
)
967 validate_creds(cred
);
969 /* We must always pass in a valid mount pointer. */
972 f
= alloc_empty_file(flags
, cred
);
974 error
= vfs_open(path
, f
);
982 EXPORT_SYMBOL(dentry_open
);
984 struct file
*open_with_fake_path(const struct path
*path
, int flags
,
985 struct inode
*inode
, const struct cred
*cred
)
987 struct file
*f
= alloc_empty_file_noaccount(flags
, cred
);
992 error
= do_dentry_open(f
, inode
, NULL
);
1000 EXPORT_SYMBOL(open_with_fake_path
);
1002 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1003 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1005 inline struct open_how
build_open_how(int flags
, umode_t mode
)
1007 struct open_how how
= {
1008 .flags
= flags
& VALID_OPEN_FLAGS
,
1009 .mode
= mode
& S_IALLUGO
,
1012 /* O_PATH beats everything else. */
1013 if (how
.flags
& O_PATH
)
1014 how
.flags
&= O_PATH_FLAGS
;
1015 /* Modes should only be set for create-like flags. */
1016 if (!WILL_CREATE(how
.flags
))
1021 inline int build_open_flags(const struct open_how
*how
, struct open_flags
*op
)
1023 u64 flags
= how
->flags
;
1024 u64 strip
= FMODE_NONOTIFY
| O_CLOEXEC
;
1025 int lookup_flags
= 0;
1026 int acc_mode
= ACC_MODE(flags
);
1028 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS
),
1029 "struct open_flags doesn't yet handle flags > 32 bits");
1032 * Strip flags that either shouldn't be set by userspace like
1033 * FMODE_NONOTIFY or that aren't relevant in determining struct
1034 * open_flags like O_CLOEXEC.
1039 * Older syscalls implicitly clear all of the invalid flags or argument
1040 * values before calling build_open_flags(), but openat2(2) checks all
1043 if (flags
& ~VALID_OPEN_FLAGS
)
1045 if (how
->resolve
& ~VALID_RESOLVE_FLAGS
)
1048 /* Scoping flags are mutually exclusive. */
1049 if ((how
->resolve
& RESOLVE_BENEATH
) && (how
->resolve
& RESOLVE_IN_ROOT
))
1052 /* Deal with the mode. */
1053 if (WILL_CREATE(flags
)) {
1054 if (how
->mode
& ~S_IALLUGO
)
1056 op
->mode
= how
->mode
| S_IFREG
;
1064 * In order to ensure programs get explicit errors when trying to use
1065 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1066 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1067 * have to require userspace to explicitly set it.
1069 if (flags
& __O_TMPFILE
) {
1070 if ((flags
& O_TMPFILE_MASK
) != O_TMPFILE
)
1072 if (!(acc_mode
& MAY_WRITE
))
1075 if (flags
& O_PATH
) {
1076 /* O_PATH only permits certain other flags to be set. */
1077 if (flags
& ~O_PATH_FLAGS
)
1083 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1084 * check for O_DSYNC if the need any syncing at all we enforce it's
1085 * always set instead of having to deal with possibly weird behaviour
1086 * for malicious applications setting only __O_SYNC.
1088 if (flags
& __O_SYNC
)
1091 op
->open_flag
= flags
;
1093 /* O_TRUNC implies we need access checks for write permissions */
1094 if (flags
& O_TRUNC
)
1095 acc_mode
|= MAY_WRITE
;
1097 /* Allow the LSM permission hook to distinguish append
1098 access from general write access. */
1099 if (flags
& O_APPEND
)
1100 acc_mode
|= MAY_APPEND
;
1102 op
->acc_mode
= acc_mode
;
1104 op
->intent
= flags
& O_PATH
? 0 : LOOKUP_OPEN
;
1106 if (flags
& O_CREAT
) {
1107 op
->intent
|= LOOKUP_CREATE
;
1108 if (flags
& O_EXCL
) {
1109 op
->intent
|= LOOKUP_EXCL
;
1110 flags
|= O_NOFOLLOW
;
1114 if (flags
& O_DIRECTORY
)
1115 lookup_flags
|= LOOKUP_DIRECTORY
;
1116 if (!(flags
& O_NOFOLLOW
))
1117 lookup_flags
|= LOOKUP_FOLLOW
;
1119 if (how
->resolve
& RESOLVE_NO_XDEV
)
1120 lookup_flags
|= LOOKUP_NO_XDEV
;
1121 if (how
->resolve
& RESOLVE_NO_MAGICLINKS
)
1122 lookup_flags
|= LOOKUP_NO_MAGICLINKS
;
1123 if (how
->resolve
& RESOLVE_NO_SYMLINKS
)
1124 lookup_flags
|= LOOKUP_NO_SYMLINKS
;
1125 if (how
->resolve
& RESOLVE_BENEATH
)
1126 lookup_flags
|= LOOKUP_BENEATH
;
1127 if (how
->resolve
& RESOLVE_IN_ROOT
)
1128 lookup_flags
|= LOOKUP_IN_ROOT
;
1129 if (how
->resolve
& RESOLVE_CACHED
) {
1130 /* Don't bother even trying for create/truncate/tmpfile open */
1131 if (flags
& (O_TRUNC
| O_CREAT
| O_TMPFILE
))
1133 lookup_flags
|= LOOKUP_CACHED
;
1136 op
->lookup_flags
= lookup_flags
;
1141 * file_open_name - open file and return file pointer
1143 * @name: struct filename containing path to open
1144 * @flags: open flags as per the open(2) second argument
1145 * @mode: mode for the new file if O_CREAT is set, else ignored
1147 * This is the helper to open a file from kernelspace if you really
1148 * have to. But in generally you should not do this, so please move
1149 * along, nothing to see here..
1151 struct file
*file_open_name(struct filename
*name
, int flags
, umode_t mode
)
1153 struct open_flags op
;
1154 struct open_how how
= build_open_how(flags
, mode
);
1155 int err
= build_open_flags(&how
, &op
);
1157 return ERR_PTR(err
);
1158 return do_filp_open(AT_FDCWD
, name
, &op
);
1162 * filp_open - open file and return file pointer
1164 * @filename: path to open
1165 * @flags: open flags as per the open(2) second argument
1166 * @mode: mode for the new file if O_CREAT is set, else ignored
1168 * This is the helper to open a file from kernelspace if you really
1169 * have to. But in generally you should not do this, so please move
1170 * along, nothing to see here..
1172 struct file
*filp_open(const char *filename
, int flags
, umode_t mode
)
1174 struct filename
*name
= getname_kernel(filename
);
1175 struct file
*file
= ERR_CAST(name
);
1177 if (!IS_ERR(name
)) {
1178 file
= file_open_name(name
, flags
, mode
);
1183 EXPORT_SYMBOL(filp_open
);
1185 struct file
*file_open_root(const struct path
*root
,
1186 const char *filename
, int flags
, umode_t mode
)
1188 struct open_flags op
;
1189 struct open_how how
= build_open_how(flags
, mode
);
1190 int err
= build_open_flags(&how
, &op
);
1192 return ERR_PTR(err
);
1193 return do_file_open_root(root
, filename
, &op
);
1195 EXPORT_SYMBOL(file_open_root
);
1197 static long do_sys_openat2(int dfd
, const char __user
*filename
,
1198 struct open_how
*how
)
1200 struct open_flags op
;
1201 int fd
= build_open_flags(how
, &op
);
1202 struct filename
*tmp
;
1207 tmp
= getname(filename
);
1209 return PTR_ERR(tmp
);
1211 fd
= get_unused_fd_flags(how
->flags
);
1213 struct file
*f
= do_filp_open(dfd
, tmp
, &op
);
1226 long do_sys_open(int dfd
, const char __user
*filename
, int flags
, umode_t mode
)
1228 struct open_how how
= build_open_how(flags
, mode
);
1229 return do_sys_openat2(dfd
, filename
, &how
);
1233 SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1235 if (force_o_largefile())
1236 flags
|= O_LARGEFILE
;
1237 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1240 SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
,
1243 if (force_o_largefile())
1244 flags
|= O_LARGEFILE
;
1245 return do_sys_open(dfd
, filename
, flags
, mode
);
1248 SYSCALL_DEFINE4(openat2
, int, dfd
, const char __user
*, filename
,
1249 struct open_how __user
*, how
, size_t, usize
)
1252 struct open_how tmp
;
1254 BUILD_BUG_ON(sizeof(struct open_how
) < OPEN_HOW_SIZE_VER0
);
1255 BUILD_BUG_ON(sizeof(struct open_how
) != OPEN_HOW_SIZE_LATEST
);
1257 if (unlikely(usize
< OPEN_HOW_SIZE_VER0
))
1260 err
= copy_struct_from_user(&tmp
, sizeof(tmp
), how
, usize
);
1264 audit_openat2_how(&tmp
);
1266 /* O_LARGEFILE is only allowed for non-O_PATH. */
1267 if (!(tmp
.flags
& O_PATH
) && force_o_largefile())
1268 tmp
.flags
|= O_LARGEFILE
;
1270 return do_sys_openat2(dfd
, filename
, &tmp
);
1273 #ifdef CONFIG_COMPAT
1275 * Exactly like sys_open(), except that it doesn't set the
1278 COMPAT_SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1280 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1284 * Exactly like sys_openat(), except that it doesn't set the
1287 COMPAT_SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1289 return do_sys_open(dfd
, filename
, flags
, mode
);
1296 * For backward compatibility? Maybe this should be moved
1297 * into arch/i386 instead?
1299 SYSCALL_DEFINE2(creat
, const char __user
*, pathname
, umode_t
, mode
)
1301 int flags
= O_CREAT
| O_WRONLY
| O_TRUNC
;
1303 if (force_o_largefile())
1304 flags
|= O_LARGEFILE
;
1305 return do_sys_open(AT_FDCWD
, pathname
, flags
, mode
);
1310 * "id" is the POSIX thread ID. We use the
1311 * files pointer for this..
1313 int filp_close(struct file
*filp
, fl_owner_t id
)
1317 if (!file_count(filp
)) {
1318 printk(KERN_ERR
"VFS: Close: file count is 0\n");
1322 if (filp
->f_op
->flush
)
1323 retval
= filp
->f_op
->flush(filp
, id
);
1325 if (likely(!(filp
->f_mode
& FMODE_PATH
))) {
1326 dnotify_flush(filp
, id
);
1327 locks_remove_posix(filp
, id
);
1333 EXPORT_SYMBOL(filp_close
);
1336 * Careful here! We test whether the file pointer is NULL before
1337 * releasing the fd. This ensures that one clone task can't release
1338 * an fd while another clone is opening it.
1340 SYSCALL_DEFINE1(close
, unsigned int, fd
)
1342 int retval
= close_fd(fd
);
1344 /* can't restart close syscall because file table entry was cleared */
1345 if (unlikely(retval
== -ERESTARTSYS
||
1346 retval
== -ERESTARTNOINTR
||
1347 retval
== -ERESTARTNOHAND
||
1348 retval
== -ERESTART_RESTARTBLOCK
))
1355 * close_range() - Close all file descriptors in a given range.
1357 * @fd: starting file descriptor to close
1358 * @max_fd: last file descriptor to close
1359 * @flags: reserved for future extensions
1361 * This closes a range of file descriptors. All file descriptors
1362 * from @fd up to and including @max_fd are closed.
1363 * Currently, errors to close a given file descriptor are ignored.
1365 SYSCALL_DEFINE3(close_range
, unsigned int, fd
, unsigned int, max_fd
,
1366 unsigned int, flags
)
1368 return __close_range(fd
, max_fd
, flags
);
1372 * This routine simulates a hangup on the tty, to arrange that users
1373 * are given clean terminals at login time.
1375 SYSCALL_DEFINE0(vhangup
)
1377 if (capable(CAP_SYS_TTY_CONFIG
)) {
1385 * Called when an inode is about to be open.
1386 * We use this to disallow opening large files on 32bit systems if
1387 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1388 * on this flag in sys_open.
1390 int generic_file_open(struct inode
* inode
, struct file
* filp
)
1392 if (!(filp
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1397 EXPORT_SYMBOL(generic_file_open
);
1400 * This is used by subsystems that don't want seekable
1401 * file descriptors. The function is not supposed to ever fail, the only
1402 * reason it returns an 'int' and not 'void' is so that it can be plugged
1403 * directly into file_operations structure.
1405 int nonseekable_open(struct inode
*inode
, struct file
*filp
)
1407 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
);
1411 EXPORT_SYMBOL(nonseekable_open
);
1414 * stream_open is used by subsystems that want stream-like file descriptors.
1415 * Such file descriptors are not seekable and don't have notion of position
1416 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1417 * Contrary to file descriptors of other regular files, .read() and .write()
1418 * can run simultaneously.
1420 * stream_open never fails and is marked to return int so that it could be
1421 * directly used as file_operations.open .
1423 int stream_open(struct inode
*inode
, struct file
*filp
)
1425 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
| FMODE_ATOMIC_POS
);
1426 filp
->f_mode
|= FMODE_STREAM
;
1430 EXPORT_SYMBOL(stream_open
);