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 */
227 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
228 COMPAT_SYSCALL_DEFINE3(truncate64
, const char __user
*, pathname
,
229 compat_arg_u64_dual(length
))
231 return ksys_truncate(pathname
, compat_arg_u64_glue(length
));
235 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
236 COMPAT_SYSCALL_DEFINE3(ftruncate64
, unsigned int, fd
,
237 compat_arg_u64_dual(length
))
239 return ksys_ftruncate(fd
, compat_arg_u64_glue(length
));
243 int vfs_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
245 struct inode
*inode
= file_inode(file
);
248 if (offset
< 0 || len
<= 0)
251 /* Return error if mode is not supported */
252 if (mode
& ~FALLOC_FL_SUPPORTED_MASK
)
255 /* Punch hole and zero range are mutually exclusive */
256 if ((mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
)) ==
257 (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
))
260 /* Punch hole must have keep size set */
261 if ((mode
& FALLOC_FL_PUNCH_HOLE
) &&
262 !(mode
& FALLOC_FL_KEEP_SIZE
))
265 /* Collapse range should only be used exclusively. */
266 if ((mode
& FALLOC_FL_COLLAPSE_RANGE
) &&
267 (mode
& ~FALLOC_FL_COLLAPSE_RANGE
))
270 /* Insert range should only be used exclusively. */
271 if ((mode
& FALLOC_FL_INSERT_RANGE
) &&
272 (mode
& ~FALLOC_FL_INSERT_RANGE
))
275 /* Unshare range should only be used with allocate mode. */
276 if ((mode
& FALLOC_FL_UNSHARE_RANGE
) &&
277 (mode
& ~(FALLOC_FL_UNSHARE_RANGE
| FALLOC_FL_KEEP_SIZE
)))
280 if (!(file
->f_mode
& FMODE_WRITE
))
284 * We can only allow pure fallocate on append only files
286 if ((mode
& ~FALLOC_FL_KEEP_SIZE
) && IS_APPEND(inode
))
289 if (IS_IMMUTABLE(inode
))
293 * We cannot allow any fallocate operation on an active swapfile
295 if (IS_SWAPFILE(inode
))
299 * Revalidate the write permissions, in case security policy has
300 * changed since the files were opened.
302 ret
= security_file_permission(file
, MAY_WRITE
);
306 if (S_ISFIFO(inode
->i_mode
))
309 if (S_ISDIR(inode
->i_mode
))
312 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
315 /* Check for wrap through zero too */
316 if (((offset
+ len
) > inode
->i_sb
->s_maxbytes
) || ((offset
+ len
) < 0))
319 if (!file
->f_op
->fallocate
)
322 file_start_write(file
);
323 ret
= file
->f_op
->fallocate(file
, mode
, offset
, len
);
326 * Create inotify and fanotify events.
328 * To keep the logic simple always create events if fallocate succeeds.
329 * This implies that events are even created if the file size remains
330 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
333 fsnotify_modify(file
);
335 file_end_write(file
);
338 EXPORT_SYMBOL_GPL(vfs_fallocate
);
340 int ksys_fallocate(int fd
, int mode
, loff_t offset
, loff_t len
)
342 struct fd f
= fdget(fd
);
346 error
= vfs_fallocate(f
.file
, mode
, offset
, len
);
352 SYSCALL_DEFINE4(fallocate
, int, fd
, int, mode
, loff_t
, offset
, loff_t
, len
)
354 return ksys_fallocate(fd
, mode
, offset
, len
);
357 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
358 COMPAT_SYSCALL_DEFINE6(fallocate
, int, fd
, int, mode
, compat_arg_u64_dual(offset
),
359 compat_arg_u64_dual(len
))
361 return ksys_fallocate(fd
, mode
, compat_arg_u64_glue(offset
),
362 compat_arg_u64_glue(len
));
367 * access() needs to use the real uid/gid, not the effective uid/gid.
368 * We do this by temporarily clearing all FS-related capabilities and
369 * switching the fsuid/fsgid around to the real ones.
371 static const struct cred
*access_override_creds(void)
373 const struct cred
*old_cred
;
374 struct cred
*override_cred
;
376 override_cred
= prepare_creds();
380 override_cred
->fsuid
= override_cred
->uid
;
381 override_cred
->fsgid
= override_cred
->gid
;
383 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
384 /* Clear the capabilities if we switch to a non-root user */
385 kuid_t root_uid
= make_kuid(override_cred
->user_ns
, 0);
386 if (!uid_eq(override_cred
->uid
, root_uid
))
387 cap_clear(override_cred
->cap_effective
);
389 override_cred
->cap_effective
=
390 override_cred
->cap_permitted
;
394 * The new set of credentials can *only* be used in
395 * task-synchronous circumstances, and does not need
396 * RCU freeing, unless somebody then takes a separate
399 * NOTE! This is _only_ true because this credential
400 * is used purely for override_creds() that installs
401 * it as the subjective cred. Other threads will be
402 * accessing ->real_cred, not the subjective cred.
404 * If somebody _does_ make a copy of this (using the
405 * 'get_current_cred()' function), that will clear the
406 * non_rcu field, because now that other user may be
407 * expecting RCU freeing. But normal thread-synchronous
408 * cred accesses will keep things non-RCY.
410 override_cred
->non_rcu
= 1;
412 old_cred
= override_creds(override_cred
);
414 /* override_cred() gets its own ref */
415 put_cred(override_cred
);
420 static long do_faccessat(int dfd
, const char __user
*filename
, int mode
, int flags
)
425 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
426 const struct cred
*old_cred
= NULL
;
428 if (mode
& ~S_IRWXO
) /* where's F_OK, X_OK, W_OK, R_OK? */
431 if (flags
& ~(AT_EACCESS
| AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
))
434 if (flags
& AT_SYMLINK_NOFOLLOW
)
435 lookup_flags
&= ~LOOKUP_FOLLOW
;
436 if (flags
& AT_EMPTY_PATH
)
437 lookup_flags
|= LOOKUP_EMPTY
;
439 if (!(flags
& AT_EACCESS
)) {
440 old_cred
= access_override_creds();
446 res
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
450 inode
= d_backing_inode(path
.dentry
);
452 if ((mode
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
454 * MAY_EXEC on regular files is denied if the fs is mounted
455 * with the "noexec" flag.
458 if (path_noexec(&path
))
459 goto out_path_release
;
462 res
= inode_permission(mnt_user_ns(path
.mnt
), inode
, mode
| MAY_ACCESS
);
463 /* SuS v2 requires we report a read only fs too */
464 if (res
|| !(mode
& S_IWOTH
) || special_file(inode
->i_mode
))
465 goto out_path_release
;
467 * This is a rare case where using __mnt_is_readonly()
468 * is OK without a mnt_want/drop_write() pair. Since
469 * no actual write to the fs is performed here, we do
470 * not need to telegraph to that to anyone.
472 * By doing this, we accept that this access is
473 * inherently racy and know that the fs may change
474 * state before we even see this result.
476 if (__mnt_is_readonly(path
.mnt
))
481 if (retry_estale(res
, lookup_flags
)) {
482 lookup_flags
|= LOOKUP_REVAL
;
487 revert_creds(old_cred
);
492 SYSCALL_DEFINE3(faccessat
, int, dfd
, const char __user
*, filename
, int, mode
)
494 return do_faccessat(dfd
, filename
, mode
, 0);
497 SYSCALL_DEFINE4(faccessat2
, int, dfd
, const char __user
*, filename
, int, mode
,
500 return do_faccessat(dfd
, filename
, mode
, flags
);
503 SYSCALL_DEFINE2(access
, const char __user
*, filename
, int, mode
)
505 return do_faccessat(AT_FDCWD
, filename
, mode
, 0);
508 SYSCALL_DEFINE1(chdir
, const char __user
*, filename
)
512 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
514 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
518 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
522 set_fs_pwd(current
->fs
, &path
);
526 if (retry_estale(error
, lookup_flags
)) {
527 lookup_flags
|= LOOKUP_REVAL
;
534 SYSCALL_DEFINE1(fchdir
, unsigned int, fd
)
536 struct fd f
= fdget_raw(fd
);
544 if (!d_can_lookup(f
.file
->f_path
.dentry
))
547 error
= file_permission(f
.file
, MAY_EXEC
| MAY_CHDIR
);
549 set_fs_pwd(current
->fs
, &f
.file
->f_path
);
556 SYSCALL_DEFINE1(chroot
, const char __user
*, filename
)
560 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
562 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
566 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
571 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT
))
573 error
= security_path_chroot(&path
);
577 set_fs_root(current
->fs
, &path
);
581 if (retry_estale(error
, lookup_flags
)) {
582 lookup_flags
|= LOOKUP_REVAL
;
589 int chmod_common(const struct path
*path
, umode_t mode
)
591 struct inode
*inode
= path
->dentry
->d_inode
;
592 struct inode
*delegated_inode
= NULL
;
593 struct iattr newattrs
;
596 error
= mnt_want_write(path
->mnt
);
601 error
= security_path_chmod(path
, mode
);
604 newattrs
.ia_mode
= (mode
& S_IALLUGO
) | (inode
->i_mode
& ~S_IALLUGO
);
605 newattrs
.ia_valid
= ATTR_MODE
| ATTR_CTIME
;
606 error
= notify_change(mnt_user_ns(path
->mnt
), path
->dentry
,
607 &newattrs
, &delegated_inode
);
610 if (delegated_inode
) {
611 error
= break_deleg_wait(&delegated_inode
);
615 mnt_drop_write(path
->mnt
);
619 int vfs_fchmod(struct file
*file
, umode_t mode
)
622 return chmod_common(&file
->f_path
, mode
);
625 SYSCALL_DEFINE2(fchmod
, unsigned int, fd
, umode_t
, mode
)
627 struct fd f
= fdget(fd
);
631 err
= vfs_fchmod(f
.file
, mode
);
637 static int do_fchmodat(int dfd
, const char __user
*filename
, umode_t mode
)
641 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
643 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
645 error
= chmod_common(&path
, mode
);
647 if (retry_estale(error
, lookup_flags
)) {
648 lookup_flags
|= LOOKUP_REVAL
;
655 SYSCALL_DEFINE3(fchmodat
, int, dfd
, const char __user
*, filename
,
658 return do_fchmodat(dfd
, filename
, mode
);
661 SYSCALL_DEFINE2(chmod
, const char __user
*, filename
, umode_t
, mode
)
663 return do_fchmodat(AT_FDCWD
, filename
, mode
);
666 int chown_common(const struct path
*path
, uid_t user
, gid_t group
)
668 struct user_namespace
*mnt_userns
, *fs_userns
;
669 struct inode
*inode
= path
->dentry
->d_inode
;
670 struct inode
*delegated_inode
= NULL
;
672 struct iattr newattrs
;
676 uid
= make_kuid(current_user_ns(), user
);
677 gid
= make_kgid(current_user_ns(), group
);
679 mnt_userns
= mnt_user_ns(path
->mnt
);
680 fs_userns
= i_user_ns(inode
);
681 uid
= mapped_kuid_user(mnt_userns
, fs_userns
, uid
);
682 gid
= mapped_kgid_user(mnt_userns
, fs_userns
, gid
);
685 newattrs
.ia_valid
= ATTR_CTIME
;
686 if (user
!= (uid_t
) -1) {
689 newattrs
.ia_valid
|= ATTR_UID
;
690 newattrs
.ia_uid
= uid
;
692 if (group
!= (gid_t
) -1) {
695 newattrs
.ia_valid
|= ATTR_GID
;
696 newattrs
.ia_gid
= gid
;
698 if (!S_ISDIR(inode
->i_mode
))
700 ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
702 error
= security_path_chown(path
, uid
, gid
);
704 error
= notify_change(mnt_userns
, path
->dentry
, &newattrs
,
707 if (delegated_inode
) {
708 error
= break_deleg_wait(&delegated_inode
);
715 int do_fchownat(int dfd
, const char __user
*filename
, uid_t user
, gid_t group
,
722 if ((flag
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
725 lookup_flags
= (flag
& AT_SYMLINK_NOFOLLOW
) ? 0 : LOOKUP_FOLLOW
;
726 if (flag
& AT_EMPTY_PATH
)
727 lookup_flags
|= LOOKUP_EMPTY
;
729 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
732 error
= mnt_want_write(path
.mnt
);
735 error
= chown_common(&path
, user
, group
);
736 mnt_drop_write(path
.mnt
);
739 if (retry_estale(error
, lookup_flags
)) {
740 lookup_flags
|= LOOKUP_REVAL
;
747 SYSCALL_DEFINE5(fchownat
, int, dfd
, const char __user
*, filename
, uid_t
, user
,
748 gid_t
, group
, int, flag
)
750 return do_fchownat(dfd
, filename
, user
, group
, flag
);
753 SYSCALL_DEFINE3(chown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
755 return do_fchownat(AT_FDCWD
, filename
, user
, group
, 0);
758 SYSCALL_DEFINE3(lchown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
760 return do_fchownat(AT_FDCWD
, filename
, user
, group
,
761 AT_SYMLINK_NOFOLLOW
);
764 int vfs_fchown(struct file
*file
, uid_t user
, gid_t group
)
768 error
= mnt_want_write_file(file
);
772 error
= chown_common(&file
->f_path
, user
, group
);
773 mnt_drop_write_file(file
);
777 int ksys_fchown(unsigned int fd
, uid_t user
, gid_t group
)
779 struct fd f
= fdget(fd
);
783 error
= vfs_fchown(f
.file
, user
, group
);
789 SYSCALL_DEFINE3(fchown
, unsigned int, fd
, uid_t
, user
, gid_t
, group
)
791 return ksys_fchown(fd
, user
, group
);
794 static int do_dentry_open(struct file
*f
,
796 int (*open
)(struct inode
*, struct file
*))
798 static const struct file_operations empty_fops
= {};
801 path_get(&f
->f_path
);
803 f
->f_mapping
= inode
->i_mapping
;
804 f
->f_wb_err
= filemap_sample_wb_err(f
->f_mapping
);
805 f
->f_sb_err
= file_sample_sb_err(f
);
807 if (unlikely(f
->f_flags
& O_PATH
)) {
808 f
->f_mode
= FMODE_PATH
| FMODE_OPENED
;
809 f
->f_op
= &empty_fops
;
813 if (f
->f_mode
& FMODE_WRITE
&& !special_file(inode
->i_mode
)) {
814 error
= get_write_access(inode
);
817 error
= __mnt_want_write(f
->f_path
.mnt
);
818 if (unlikely(error
)) {
819 put_write_access(inode
);
822 f
->f_mode
|= FMODE_WRITER
;
825 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
826 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
))
827 f
->f_mode
|= FMODE_ATOMIC_POS
;
829 f
->f_op
= fops_get(inode
->i_fop
);
830 if (WARN_ON(!f
->f_op
)) {
835 error
= security_file_open(f
);
839 error
= break_lease(locks_inode(f
), f
->f_flags
);
843 /* normally all 3 are set; ->open() can clear them if needed */
844 f
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
846 open
= f
->f_op
->open
;
848 error
= open(inode
, f
);
852 f
->f_mode
|= FMODE_OPENED
;
853 if ((f
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) == FMODE_READ
)
854 i_readcount_inc(inode
);
855 if ((f
->f_mode
& FMODE_READ
) &&
856 likely(f
->f_op
->read
|| f
->f_op
->read_iter
))
857 f
->f_mode
|= FMODE_CAN_READ
;
858 if ((f
->f_mode
& FMODE_WRITE
) &&
859 likely(f
->f_op
->write
|| f
->f_op
->write_iter
))
860 f
->f_mode
|= FMODE_CAN_WRITE
;
861 if (f
->f_mapping
->a_ops
&& f
->f_mapping
->a_ops
->direct_IO
)
862 f
->f_mode
|= FMODE_CAN_ODIRECT
;
864 f
->f_flags
&= ~(O_CREAT
| O_EXCL
| O_NOCTTY
| O_TRUNC
);
866 file_ra_state_init(&f
->f_ra
, f
->f_mapping
->host
->i_mapping
);
868 if ((f
->f_flags
& O_DIRECT
) && !(f
->f_mode
& FMODE_CAN_ODIRECT
))
872 * XXX: Huge page cache doesn't support writing yet. Drop all page
873 * cache for this file before processing writes.
875 if (f
->f_mode
& FMODE_WRITE
) {
877 * Paired with smp_mb() in collapse_file() to ensure nr_thps
878 * is up to date and the update to i_writecount by
879 * get_write_access() is visible. Ensures subsequent insertion
880 * of THPs into the page cache will fail.
883 if (filemap_nr_thps(inode
->i_mapping
)) {
884 struct address_space
*mapping
= inode
->i_mapping
;
886 filemap_invalidate_lock(inode
->i_mapping
);
888 * unmap_mapping_range just need to be called once
889 * here, because the private pages is not need to be
890 * unmapped mapping (e.g. data segment of dynamic
891 * shared libraries here).
893 unmap_mapping_range(mapping
, 0, 0, 0);
894 truncate_inode_pages(mapping
, 0);
895 filemap_invalidate_unlock(inode
->i_mapping
);
902 if (WARN_ON_ONCE(error
> 0))
905 if (f
->f_mode
& FMODE_WRITER
) {
906 put_write_access(inode
);
907 __mnt_drop_write(f
->f_path
.mnt
);
910 path_put(&f
->f_path
);
911 f
->f_path
.mnt
= NULL
;
912 f
->f_path
.dentry
= NULL
;
918 * finish_open - finish opening a file
919 * @file: file pointer
920 * @dentry: pointer to dentry
921 * @open: open callback
922 * @opened: state of open
924 * This can be used to finish opening a file passed to i_op->atomic_open().
926 * If the open callback is set to NULL, then the standard f_op->open()
927 * filesystem callback is substituted.
929 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
930 * the return value of d_splice_alias(), then the caller needs to perform dput()
931 * on it after finish_open().
933 * Returns zero on success or -errno if the open failed.
935 int finish_open(struct file
*file
, struct dentry
*dentry
,
936 int (*open
)(struct inode
*, struct file
*))
938 BUG_ON(file
->f_mode
& FMODE_OPENED
); /* once it's opened, it's opened */
940 file
->f_path
.dentry
= dentry
;
941 return do_dentry_open(file
, d_backing_inode(dentry
), open
);
943 EXPORT_SYMBOL(finish_open
);
946 * finish_no_open - finish ->atomic_open() without opening the file
948 * @file: file pointer
949 * @dentry: dentry or NULL (as returned from ->lookup())
951 * This can be used to set the result of a successful lookup in ->atomic_open().
953 * NB: unlike finish_open() this function does consume the dentry reference and
954 * the caller need not dput() it.
956 * Returns "0" which must be the return value of ->atomic_open() after having
957 * called this function.
959 int finish_no_open(struct file
*file
, struct dentry
*dentry
)
961 file
->f_path
.dentry
= dentry
;
964 EXPORT_SYMBOL(finish_no_open
);
966 char *file_path(struct file
*filp
, char *buf
, int buflen
)
968 return d_path(&filp
->f_path
, buf
, buflen
);
970 EXPORT_SYMBOL(file_path
);
973 * vfs_open - open the file at the given path
974 * @path: path to open
975 * @file: newly allocated file with f_flag initialized
976 * @cred: credentials to use
978 int vfs_open(const struct path
*path
, struct file
*file
)
980 file
->f_path
= *path
;
981 return do_dentry_open(file
, d_backing_inode(path
->dentry
), NULL
);
984 struct file
*dentry_open(const struct path
*path
, int flags
,
985 const struct cred
*cred
)
990 validate_creds(cred
);
992 /* We must always pass in a valid mount pointer. */
995 f
= alloc_empty_file(flags
, cred
);
997 error
= vfs_open(path
, f
);
1005 EXPORT_SYMBOL(dentry_open
);
1008 * dentry_create - Create and open a file
1009 * @path: path to create
1011 * @mode: mode bits for new file
1012 * @cred: credentials to use
1014 * Caller must hold the parent directory's lock, and have prepared
1015 * a negative dentry, placed in @path->dentry, for the new file.
1017 * Caller sets @path->mnt to the vfsmount of the filesystem where
1018 * the new file is to be created. The parent directory and the
1019 * negative dentry must reside on the same filesystem instance.
1021 * On success, returns a "struct file *". Otherwise a ERR_PTR
1024 struct file
*dentry_create(const struct path
*path
, int flags
, umode_t mode
,
1025 const struct cred
*cred
)
1030 validate_creds(cred
);
1031 f
= alloc_empty_file(flags
, cred
);
1035 error
= vfs_create(mnt_user_ns(path
->mnt
),
1036 d_inode(path
->dentry
->d_parent
),
1037 path
->dentry
, mode
, true);
1039 error
= vfs_open(path
, f
);
1041 if (unlikely(error
)) {
1043 return ERR_PTR(error
);
1047 EXPORT_SYMBOL(dentry_create
);
1049 struct file
*open_with_fake_path(const struct path
*path
, int flags
,
1050 struct inode
*inode
, const struct cred
*cred
)
1052 struct file
*f
= alloc_empty_file_noaccount(flags
, cred
);
1057 error
= do_dentry_open(f
, inode
, NULL
);
1065 EXPORT_SYMBOL(open_with_fake_path
);
1067 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1068 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1070 inline struct open_how
build_open_how(int flags
, umode_t mode
)
1072 struct open_how how
= {
1073 .flags
= flags
& VALID_OPEN_FLAGS
,
1074 .mode
= mode
& S_IALLUGO
,
1077 /* O_PATH beats everything else. */
1078 if (how
.flags
& O_PATH
)
1079 how
.flags
&= O_PATH_FLAGS
;
1080 /* Modes should only be set for create-like flags. */
1081 if (!WILL_CREATE(how
.flags
))
1086 inline int build_open_flags(const struct open_how
*how
, struct open_flags
*op
)
1088 u64 flags
= how
->flags
;
1089 u64 strip
= FMODE_NONOTIFY
| O_CLOEXEC
;
1090 int lookup_flags
= 0;
1091 int acc_mode
= ACC_MODE(flags
);
1093 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS
),
1094 "struct open_flags doesn't yet handle flags > 32 bits");
1097 * Strip flags that either shouldn't be set by userspace like
1098 * FMODE_NONOTIFY or that aren't relevant in determining struct
1099 * open_flags like O_CLOEXEC.
1104 * Older syscalls implicitly clear all of the invalid flags or argument
1105 * values before calling build_open_flags(), but openat2(2) checks all
1108 if (flags
& ~VALID_OPEN_FLAGS
)
1110 if (how
->resolve
& ~VALID_RESOLVE_FLAGS
)
1113 /* Scoping flags are mutually exclusive. */
1114 if ((how
->resolve
& RESOLVE_BENEATH
) && (how
->resolve
& RESOLVE_IN_ROOT
))
1117 /* Deal with the mode. */
1118 if (WILL_CREATE(flags
)) {
1119 if (how
->mode
& ~S_IALLUGO
)
1121 op
->mode
= how
->mode
| S_IFREG
;
1129 * In order to ensure programs get explicit errors when trying to use
1130 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1131 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1132 * have to require userspace to explicitly set it.
1134 if (flags
& __O_TMPFILE
) {
1135 if ((flags
& O_TMPFILE_MASK
) != O_TMPFILE
)
1137 if (!(acc_mode
& MAY_WRITE
))
1140 if (flags
& O_PATH
) {
1141 /* O_PATH only permits certain other flags to be set. */
1142 if (flags
& ~O_PATH_FLAGS
)
1148 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1149 * check for O_DSYNC if the need any syncing at all we enforce it's
1150 * always set instead of having to deal with possibly weird behaviour
1151 * for malicious applications setting only __O_SYNC.
1153 if (flags
& __O_SYNC
)
1156 op
->open_flag
= flags
;
1158 /* O_TRUNC implies we need access checks for write permissions */
1159 if (flags
& O_TRUNC
)
1160 acc_mode
|= MAY_WRITE
;
1162 /* Allow the LSM permission hook to distinguish append
1163 access from general write access. */
1164 if (flags
& O_APPEND
)
1165 acc_mode
|= MAY_APPEND
;
1167 op
->acc_mode
= acc_mode
;
1169 op
->intent
= flags
& O_PATH
? 0 : LOOKUP_OPEN
;
1171 if (flags
& O_CREAT
) {
1172 op
->intent
|= LOOKUP_CREATE
;
1173 if (flags
& O_EXCL
) {
1174 op
->intent
|= LOOKUP_EXCL
;
1175 flags
|= O_NOFOLLOW
;
1179 if (flags
& O_DIRECTORY
)
1180 lookup_flags
|= LOOKUP_DIRECTORY
;
1181 if (!(flags
& O_NOFOLLOW
))
1182 lookup_flags
|= LOOKUP_FOLLOW
;
1184 if (how
->resolve
& RESOLVE_NO_XDEV
)
1185 lookup_flags
|= LOOKUP_NO_XDEV
;
1186 if (how
->resolve
& RESOLVE_NO_MAGICLINKS
)
1187 lookup_flags
|= LOOKUP_NO_MAGICLINKS
;
1188 if (how
->resolve
& RESOLVE_NO_SYMLINKS
)
1189 lookup_flags
|= LOOKUP_NO_SYMLINKS
;
1190 if (how
->resolve
& RESOLVE_BENEATH
)
1191 lookup_flags
|= LOOKUP_BENEATH
;
1192 if (how
->resolve
& RESOLVE_IN_ROOT
)
1193 lookup_flags
|= LOOKUP_IN_ROOT
;
1194 if (how
->resolve
& RESOLVE_CACHED
) {
1195 /* Don't bother even trying for create/truncate/tmpfile open */
1196 if (flags
& (O_TRUNC
| O_CREAT
| O_TMPFILE
))
1198 lookup_flags
|= LOOKUP_CACHED
;
1201 op
->lookup_flags
= lookup_flags
;
1206 * file_open_name - open file and return file pointer
1208 * @name: struct filename containing path to open
1209 * @flags: open flags as per the open(2) second argument
1210 * @mode: mode for the new file if O_CREAT is set, else ignored
1212 * This is the helper to open a file from kernelspace if you really
1213 * have to. But in generally you should not do this, so please move
1214 * along, nothing to see here..
1216 struct file
*file_open_name(struct filename
*name
, int flags
, umode_t mode
)
1218 struct open_flags op
;
1219 struct open_how how
= build_open_how(flags
, mode
);
1220 int err
= build_open_flags(&how
, &op
);
1222 return ERR_PTR(err
);
1223 return do_filp_open(AT_FDCWD
, name
, &op
);
1227 * filp_open - open file and return file pointer
1229 * @filename: path to open
1230 * @flags: open flags as per the open(2) second argument
1231 * @mode: mode for the new file if O_CREAT is set, else ignored
1233 * This is the helper to open a file from kernelspace if you really
1234 * have to. But in generally you should not do this, so please move
1235 * along, nothing to see here..
1237 struct file
*filp_open(const char *filename
, int flags
, umode_t mode
)
1239 struct filename
*name
= getname_kernel(filename
);
1240 struct file
*file
= ERR_CAST(name
);
1242 if (!IS_ERR(name
)) {
1243 file
= file_open_name(name
, flags
, mode
);
1248 EXPORT_SYMBOL(filp_open
);
1250 struct file
*file_open_root(const struct path
*root
,
1251 const char *filename
, int flags
, umode_t mode
)
1253 struct open_flags op
;
1254 struct open_how how
= build_open_how(flags
, mode
);
1255 int err
= build_open_flags(&how
, &op
);
1257 return ERR_PTR(err
);
1258 return do_file_open_root(root
, filename
, &op
);
1260 EXPORT_SYMBOL(file_open_root
);
1262 static long do_sys_openat2(int dfd
, const char __user
*filename
,
1263 struct open_how
*how
)
1265 struct open_flags op
;
1266 int fd
= build_open_flags(how
, &op
);
1267 struct filename
*tmp
;
1272 tmp
= getname(filename
);
1274 return PTR_ERR(tmp
);
1276 fd
= get_unused_fd_flags(how
->flags
);
1278 struct file
*f
= do_filp_open(dfd
, tmp
, &op
);
1291 long do_sys_open(int dfd
, const char __user
*filename
, int flags
, umode_t mode
)
1293 struct open_how how
= build_open_how(flags
, mode
);
1294 return do_sys_openat2(dfd
, filename
, &how
);
1298 SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1300 if (force_o_largefile())
1301 flags
|= O_LARGEFILE
;
1302 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1305 SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
,
1308 if (force_o_largefile())
1309 flags
|= O_LARGEFILE
;
1310 return do_sys_open(dfd
, filename
, flags
, mode
);
1313 SYSCALL_DEFINE4(openat2
, int, dfd
, const char __user
*, filename
,
1314 struct open_how __user
*, how
, size_t, usize
)
1317 struct open_how tmp
;
1319 BUILD_BUG_ON(sizeof(struct open_how
) < OPEN_HOW_SIZE_VER0
);
1320 BUILD_BUG_ON(sizeof(struct open_how
) != OPEN_HOW_SIZE_LATEST
);
1322 if (unlikely(usize
< OPEN_HOW_SIZE_VER0
))
1325 err
= copy_struct_from_user(&tmp
, sizeof(tmp
), how
, usize
);
1329 audit_openat2_how(&tmp
);
1331 /* O_LARGEFILE is only allowed for non-O_PATH. */
1332 if (!(tmp
.flags
& O_PATH
) && force_o_largefile())
1333 tmp
.flags
|= O_LARGEFILE
;
1335 return do_sys_openat2(dfd
, filename
, &tmp
);
1338 #ifdef CONFIG_COMPAT
1340 * Exactly like sys_open(), except that it doesn't set the
1343 COMPAT_SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1345 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1349 * Exactly like sys_openat(), except that it doesn't set the
1352 COMPAT_SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1354 return do_sys_open(dfd
, filename
, flags
, mode
);
1361 * For backward compatibility? Maybe this should be moved
1362 * into arch/i386 instead?
1364 SYSCALL_DEFINE2(creat
, const char __user
*, pathname
, umode_t
, mode
)
1366 int flags
= O_CREAT
| O_WRONLY
| O_TRUNC
;
1368 if (force_o_largefile())
1369 flags
|= O_LARGEFILE
;
1370 return do_sys_open(AT_FDCWD
, pathname
, flags
, mode
);
1375 * "id" is the POSIX thread ID. We use the
1376 * files pointer for this..
1378 int filp_close(struct file
*filp
, fl_owner_t id
)
1382 if (!file_count(filp
)) {
1383 printk(KERN_ERR
"VFS: Close: file count is 0\n");
1387 if (filp
->f_op
->flush
)
1388 retval
= filp
->f_op
->flush(filp
, id
);
1390 if (likely(!(filp
->f_mode
& FMODE_PATH
))) {
1391 dnotify_flush(filp
, id
);
1392 locks_remove_posix(filp
, id
);
1398 EXPORT_SYMBOL(filp_close
);
1401 * Careful here! We test whether the file pointer is NULL before
1402 * releasing the fd. This ensures that one clone task can't release
1403 * an fd while another clone is opening it.
1405 SYSCALL_DEFINE1(close
, unsigned int, fd
)
1407 int retval
= close_fd(fd
);
1409 /* can't restart close syscall because file table entry was cleared */
1410 if (unlikely(retval
== -ERESTARTSYS
||
1411 retval
== -ERESTARTNOINTR
||
1412 retval
== -ERESTARTNOHAND
||
1413 retval
== -ERESTART_RESTARTBLOCK
))
1420 * close_range() - Close all file descriptors in a given range.
1422 * @fd: starting file descriptor to close
1423 * @max_fd: last file descriptor to close
1424 * @flags: reserved for future extensions
1426 * This closes a range of file descriptors. All file descriptors
1427 * from @fd up to and including @max_fd are closed.
1428 * Currently, errors to close a given file descriptor are ignored.
1430 SYSCALL_DEFINE3(close_range
, unsigned int, fd
, unsigned int, max_fd
,
1431 unsigned int, flags
)
1433 return __close_range(fd
, max_fd
, flags
);
1437 * This routine simulates a hangup on the tty, to arrange that users
1438 * are given clean terminals at login time.
1440 SYSCALL_DEFINE0(vhangup
)
1442 if (capable(CAP_SYS_TTY_CONFIG
)) {
1450 * Called when an inode is about to be open.
1451 * We use this to disallow opening large files on 32bit systems if
1452 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1453 * on this flag in sys_open.
1455 int generic_file_open(struct inode
* inode
, struct file
* filp
)
1457 if (!(filp
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1462 EXPORT_SYMBOL(generic_file_open
);
1465 * This is used by subsystems that don't want seekable
1466 * file descriptors. The function is not supposed to ever fail, the only
1467 * reason it returns an 'int' and not 'void' is so that it can be plugged
1468 * directly into file_operations structure.
1470 int nonseekable_open(struct inode
*inode
, struct file
*filp
)
1472 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
);
1476 EXPORT_SYMBOL(nonseekable_open
);
1479 * stream_open is used by subsystems that want stream-like file descriptors.
1480 * Such file descriptors are not seekable and don't have notion of position
1481 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1482 * Contrary to file descriptors of other regular files, .read() and .write()
1483 * can run simultaneously.
1485 * stream_open never fails and is marked to return int so that it could be
1486 * directly used as file_operations.open .
1488 int stream_open(struct inode
*inode
, struct file
*filp
)
1490 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
| FMODE_ATOMIC_POS
);
1491 filp
->f_mode
|= FMODE_STREAM
;
1495 EXPORT_SYMBOL(stream_open
);