1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/string.h>
10 #include <linux/file.h>
11 #include <linux/fdtable.h>
12 #include <linux/fsnotify.h>
13 #include <linux/module.h>
14 #include <linux/tty.h>
15 #include <linux/namei.h>
16 #include <linux/backing-dev.h>
17 #include <linux/capability.h>
18 #include <linux/securebits.h>
19 #include <linux/security.h>
20 #include <linux/mount.h>
21 #include <linux/fcntl.h>
22 #include <linux/slab.h>
23 #include <linux/uaccess.h>
25 #include <linux/personality.h>
26 #include <linux/pagemap.h>
27 #include <linux/syscalls.h>
28 #include <linux/rcupdate.h>
29 #include <linux/audit.h>
30 #include <linux/falloc.h>
31 #include <linux/fs_struct.h>
32 #include <linux/dnotify.h>
33 #include <linux/compat.h>
34 #include <linux/mnt_idmapping.h>
35 #include <linux/filelock.h>
39 int do_truncate(struct mnt_idmap
*idmap
, struct dentry
*dentry
,
40 loff_t length
, unsigned int time_attrs
, struct file
*filp
)
43 struct iattr newattrs
;
45 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
49 newattrs
.ia_size
= length
;
50 newattrs
.ia_valid
= ATTR_SIZE
| time_attrs
;
52 newattrs
.ia_file
= filp
;
53 newattrs
.ia_valid
|= ATTR_FILE
;
56 /* Remove suid, sgid, and file capabilities on truncate too */
57 ret
= dentry_needs_remove_privs(idmap
, dentry
);
61 newattrs
.ia_valid
|= ret
| ATTR_FORCE
;
63 inode_lock(dentry
->d_inode
);
64 /* Note any delegations or leases have already been broken: */
65 ret
= notify_change(idmap
, dentry
, &newattrs
, NULL
);
66 inode_unlock(dentry
->d_inode
);
70 long vfs_truncate(const struct path
*path
, loff_t length
)
72 struct mnt_idmap
*idmap
;
76 inode
= path
->dentry
->d_inode
;
78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 if (S_ISDIR(inode
->i_mode
))
81 if (!S_ISREG(inode
->i_mode
))
84 error
= mnt_want_write(path
->mnt
);
88 idmap
= mnt_idmap(path
->mnt
);
89 error
= inode_permission(idmap
, inode
, MAY_WRITE
);
91 goto mnt_drop_write_and_out
;
95 goto mnt_drop_write_and_out
;
97 error
= get_write_access(inode
);
99 goto mnt_drop_write_and_out
;
102 * Make sure that there are no leases. get_write_access() protects
103 * against the truncate racing with a lease-granting setlease().
105 error
= break_lease(inode
, O_WRONLY
);
107 goto put_write_and_out
;
109 error
= security_path_truncate(path
);
111 error
= do_truncate(idmap
, path
->dentry
, length
, 0, NULL
);
114 put_write_access(inode
);
115 mnt_drop_write_and_out
:
116 mnt_drop_write(path
->mnt
);
120 EXPORT_SYMBOL_GPL(vfs_truncate
);
122 long do_sys_truncate(const char __user
*pathname
, loff_t length
)
124 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
128 if (length
< 0) /* sorry, but loff_t says... */
132 error
= user_path_at(AT_FDCWD
, pathname
, lookup_flags
, &path
);
134 error
= vfs_truncate(&path
, length
);
137 if (retry_estale(error
, lookup_flags
)) {
138 lookup_flags
|= LOOKUP_REVAL
;
144 SYSCALL_DEFINE2(truncate
, const char __user
*, path
, long, length
)
146 return do_sys_truncate(path
, length
);
150 COMPAT_SYSCALL_DEFINE2(truncate
, const char __user
*, path
, compat_off_t
, length
)
152 return do_sys_truncate(path
, length
);
156 long do_ftruncate(struct file
*file
, loff_t length
, int small
)
159 struct dentry
*dentry
;
162 /* explicitly opened as large or we are on 64-bit box */
163 if (file
->f_flags
& O_LARGEFILE
)
166 dentry
= file
->f_path
.dentry
;
167 inode
= dentry
->d_inode
;
168 if (!S_ISREG(inode
->i_mode
) || !(file
->f_mode
& FMODE_WRITE
))
171 /* Cannot ftruncate over 2^31 bytes without large file support */
172 if (small
&& length
> MAX_NON_LFS
)
175 /* Check IS_APPEND on real upper inode */
176 if (IS_APPEND(file_inode(file
)))
178 sb_start_write(inode
->i_sb
);
179 error
= security_file_truncate(file
);
181 error
= do_truncate(file_mnt_idmap(file
), dentry
, length
,
182 ATTR_MTIME
| ATTR_CTIME
, file
);
183 sb_end_write(inode
->i_sb
);
188 long do_sys_ftruncate(unsigned int fd
, loff_t length
, int small
)
199 error
= do_ftruncate(f
.file
, length
, small
);
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 */
230 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
231 COMPAT_SYSCALL_DEFINE3(truncate64
, const char __user
*, pathname
,
232 compat_arg_u64_dual(length
))
234 return ksys_truncate(pathname
, compat_arg_u64_glue(length
));
238 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
239 COMPAT_SYSCALL_DEFINE3(ftruncate64
, unsigned int, fd
,
240 compat_arg_u64_dual(length
))
242 return ksys_ftruncate(fd
, compat_arg_u64_glue(length
));
246 int vfs_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
248 struct inode
*inode
= file_inode(file
);
251 if (offset
< 0 || len
<= 0)
254 /* Return error if mode is not supported */
255 if (mode
& ~FALLOC_FL_SUPPORTED_MASK
)
258 /* Punch hole and zero range are mutually exclusive */
259 if ((mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
)) ==
260 (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
))
263 /* Punch hole must have keep size set */
264 if ((mode
& FALLOC_FL_PUNCH_HOLE
) &&
265 !(mode
& FALLOC_FL_KEEP_SIZE
))
268 /* Collapse range should only be used exclusively. */
269 if ((mode
& FALLOC_FL_COLLAPSE_RANGE
) &&
270 (mode
& ~FALLOC_FL_COLLAPSE_RANGE
))
273 /* Insert range should only be used exclusively. */
274 if ((mode
& FALLOC_FL_INSERT_RANGE
) &&
275 (mode
& ~FALLOC_FL_INSERT_RANGE
))
278 /* Unshare range should only be used with allocate mode. */
279 if ((mode
& FALLOC_FL_UNSHARE_RANGE
) &&
280 (mode
& ~(FALLOC_FL_UNSHARE_RANGE
| FALLOC_FL_KEEP_SIZE
)))
283 if (!(file
->f_mode
& FMODE_WRITE
))
287 * We can only allow pure fallocate on append only files
289 if ((mode
& ~FALLOC_FL_KEEP_SIZE
) && IS_APPEND(inode
))
292 if (IS_IMMUTABLE(inode
))
296 * We cannot allow any fallocate operation on an active swapfile
298 if (IS_SWAPFILE(inode
))
302 * Revalidate the write permissions, in case security policy has
303 * changed since the files were opened.
305 ret
= security_file_permission(file
, MAY_WRITE
);
309 ret
= fsnotify_file_area_perm(file
, MAY_WRITE
, &offset
, len
);
313 if (S_ISFIFO(inode
->i_mode
))
316 if (S_ISDIR(inode
->i_mode
))
319 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
322 /* Check for wrap through zero too */
323 if (((offset
+ len
) > inode
->i_sb
->s_maxbytes
) || ((offset
+ len
) < 0))
326 if (!file
->f_op
->fallocate
)
329 file_start_write(file
);
330 ret
= file
->f_op
->fallocate(file
, mode
, offset
, len
);
333 * Create inotify and fanotify events.
335 * To keep the logic simple always create events if fallocate succeeds.
336 * This implies that events are even created if the file size remains
337 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
340 fsnotify_modify(file
);
342 file_end_write(file
);
345 EXPORT_SYMBOL_GPL(vfs_fallocate
);
347 int ksys_fallocate(int fd
, int mode
, loff_t offset
, loff_t len
)
349 struct fd f
= fdget(fd
);
353 error
= vfs_fallocate(f
.file
, mode
, offset
, len
);
359 SYSCALL_DEFINE4(fallocate
, int, fd
, int, mode
, loff_t
, offset
, loff_t
, len
)
361 return ksys_fallocate(fd
, mode
, offset
, len
);
364 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
365 COMPAT_SYSCALL_DEFINE6(fallocate
, int, fd
, int, mode
, compat_arg_u64_dual(offset
),
366 compat_arg_u64_dual(len
))
368 return ksys_fallocate(fd
, mode
, compat_arg_u64_glue(offset
),
369 compat_arg_u64_glue(len
));
374 * access() needs to use the real uid/gid, not the effective uid/gid.
375 * We do this by temporarily clearing all FS-related capabilities and
376 * switching the fsuid/fsgid around to the real ones.
378 * Creating new credentials is expensive, so we try to skip doing it,
379 * which we can if the result would match what we already got.
381 static bool access_need_override_creds(int flags
)
383 const struct cred
*cred
;
385 if (flags
& AT_EACCESS
)
388 cred
= current_cred();
389 if (!uid_eq(cred
->fsuid
, cred
->uid
) ||
390 !gid_eq(cred
->fsgid
, cred
->gid
))
393 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
394 kuid_t root_uid
= make_kuid(cred
->user_ns
, 0);
395 if (!uid_eq(cred
->uid
, root_uid
)) {
396 if (!cap_isclear(cred
->cap_effective
))
399 if (!cap_isidentical(cred
->cap_effective
,
400 cred
->cap_permitted
))
408 static const struct cred
*access_override_creds(void)
410 const struct cred
*old_cred
;
411 struct cred
*override_cred
;
413 override_cred
= prepare_creds();
418 * XXX access_need_override_creds performs checks in hopes of skipping
419 * this work. Make sure it stays in sync if making any changes in this
423 override_cred
->fsuid
= override_cred
->uid
;
424 override_cred
->fsgid
= override_cred
->gid
;
426 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
427 /* Clear the capabilities if we switch to a non-root user */
428 kuid_t root_uid
= make_kuid(override_cred
->user_ns
, 0);
429 if (!uid_eq(override_cred
->uid
, root_uid
))
430 cap_clear(override_cred
->cap_effective
);
432 override_cred
->cap_effective
=
433 override_cred
->cap_permitted
;
437 * The new set of credentials can *only* be used in
438 * task-synchronous circumstances, and does not need
439 * RCU freeing, unless somebody then takes a separate
442 * NOTE! This is _only_ true because this credential
443 * is used purely for override_creds() that installs
444 * it as the subjective cred. Other threads will be
445 * accessing ->real_cred, not the subjective cred.
447 * If somebody _does_ make a copy of this (using the
448 * 'get_current_cred()' function), that will clear the
449 * non_rcu field, because now that other user may be
450 * expecting RCU freeing. But normal thread-synchronous
451 * cred accesses will keep things non-racy to avoid RCU
454 override_cred
->non_rcu
= 1;
456 old_cred
= override_creds(override_cred
);
458 /* override_cred() gets its own ref */
459 put_cred(override_cred
);
464 static long do_faccessat(int dfd
, const char __user
*filename
, int mode
, int flags
)
469 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
470 const struct cred
*old_cred
= NULL
;
472 if (mode
& ~S_IRWXO
) /* where's F_OK, X_OK, W_OK, R_OK? */
475 if (flags
& ~(AT_EACCESS
| AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
))
478 if (flags
& AT_SYMLINK_NOFOLLOW
)
479 lookup_flags
&= ~LOOKUP_FOLLOW
;
480 if (flags
& AT_EMPTY_PATH
)
481 lookup_flags
|= LOOKUP_EMPTY
;
483 if (access_need_override_creds(flags
)) {
484 old_cred
= access_override_creds();
490 res
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
494 inode
= d_backing_inode(path
.dentry
);
496 if ((mode
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
498 * MAY_EXEC on regular files is denied if the fs is mounted
499 * with the "noexec" flag.
502 if (path_noexec(&path
))
503 goto out_path_release
;
506 res
= inode_permission(mnt_idmap(path
.mnt
), inode
, mode
| MAY_ACCESS
);
507 /* SuS v2 requires we report a read only fs too */
508 if (res
|| !(mode
& S_IWOTH
) || special_file(inode
->i_mode
))
509 goto out_path_release
;
511 * This is a rare case where using __mnt_is_readonly()
512 * is OK without a mnt_want/drop_write() pair. Since
513 * no actual write to the fs is performed here, we do
514 * not need to telegraph to that to anyone.
516 * By doing this, we accept that this access is
517 * inherently racy and know that the fs may change
518 * state before we even see this result.
520 if (__mnt_is_readonly(path
.mnt
))
525 if (retry_estale(res
, lookup_flags
)) {
526 lookup_flags
|= LOOKUP_REVAL
;
531 revert_creds(old_cred
);
536 SYSCALL_DEFINE3(faccessat
, int, dfd
, const char __user
*, filename
, int, mode
)
538 return do_faccessat(dfd
, filename
, mode
, 0);
541 SYSCALL_DEFINE4(faccessat2
, int, dfd
, const char __user
*, filename
, int, mode
,
544 return do_faccessat(dfd
, filename
, mode
, flags
);
547 SYSCALL_DEFINE2(access
, const char __user
*, filename
, int, mode
)
549 return do_faccessat(AT_FDCWD
, filename
, mode
, 0);
552 SYSCALL_DEFINE1(chdir
, const char __user
*, filename
)
556 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
558 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
562 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
566 set_fs_pwd(current
->fs
, &path
);
570 if (retry_estale(error
, lookup_flags
)) {
571 lookup_flags
|= LOOKUP_REVAL
;
578 SYSCALL_DEFINE1(fchdir
, unsigned int, fd
)
580 struct fd f
= fdget_raw(fd
);
588 if (!d_can_lookup(f
.file
->f_path
.dentry
))
591 error
= file_permission(f
.file
, MAY_EXEC
| MAY_CHDIR
);
593 set_fs_pwd(current
->fs
, &f
.file
->f_path
);
600 SYSCALL_DEFINE1(chroot
, const char __user
*, filename
)
604 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
606 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
610 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
615 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT
))
617 error
= security_path_chroot(&path
);
621 set_fs_root(current
->fs
, &path
);
625 if (retry_estale(error
, lookup_flags
)) {
626 lookup_flags
|= LOOKUP_REVAL
;
633 int chmod_common(const struct path
*path
, umode_t mode
)
635 struct inode
*inode
= path
->dentry
->d_inode
;
636 struct inode
*delegated_inode
= NULL
;
637 struct iattr newattrs
;
640 error
= mnt_want_write(path
->mnt
);
645 error
= security_path_chmod(path
, mode
);
648 newattrs
.ia_mode
= (mode
& S_IALLUGO
) | (inode
->i_mode
& ~S_IALLUGO
);
649 newattrs
.ia_valid
= ATTR_MODE
| ATTR_CTIME
;
650 error
= notify_change(mnt_idmap(path
->mnt
), path
->dentry
,
651 &newattrs
, &delegated_inode
);
654 if (delegated_inode
) {
655 error
= break_deleg_wait(&delegated_inode
);
659 mnt_drop_write(path
->mnt
);
663 int vfs_fchmod(struct file
*file
, umode_t mode
)
666 return chmod_common(&file
->f_path
, mode
);
669 SYSCALL_DEFINE2(fchmod
, unsigned int, fd
, umode_t
, mode
)
671 struct fd f
= fdget(fd
);
675 err
= vfs_fchmod(f
.file
, mode
);
681 static int do_fchmodat(int dfd
, const char __user
*filename
, umode_t mode
,
686 unsigned int lookup_flags
;
688 if (unlikely(flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)))
691 lookup_flags
= (flags
& AT_SYMLINK_NOFOLLOW
) ? 0 : LOOKUP_FOLLOW
;
692 if (flags
& AT_EMPTY_PATH
)
693 lookup_flags
|= LOOKUP_EMPTY
;
696 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
698 error
= chmod_common(&path
, mode
);
700 if (retry_estale(error
, lookup_flags
)) {
701 lookup_flags
|= LOOKUP_REVAL
;
708 SYSCALL_DEFINE4(fchmodat2
, int, dfd
, const char __user
*, filename
,
709 umode_t
, mode
, unsigned int, flags
)
711 return do_fchmodat(dfd
, filename
, mode
, flags
);
714 SYSCALL_DEFINE3(fchmodat
, int, dfd
, const char __user
*, filename
,
717 return do_fchmodat(dfd
, filename
, mode
, 0);
720 SYSCALL_DEFINE2(chmod
, const char __user
*, filename
, umode_t
, mode
)
722 return do_fchmodat(AT_FDCWD
, filename
, mode
, 0);
726 * Check whether @kuid is valid and if so generate and set vfsuid_t in
729 * Return: true if @kuid is valid, false if not.
731 static inline bool setattr_vfsuid(struct iattr
*attr
, kuid_t kuid
)
733 if (!uid_valid(kuid
))
735 attr
->ia_valid
|= ATTR_UID
;
736 attr
->ia_vfsuid
= VFSUIDT_INIT(kuid
);
741 * Check whether @kgid is valid and if so generate and set vfsgid_t in
744 * Return: true if @kgid is valid, false if not.
746 static inline bool setattr_vfsgid(struct iattr
*attr
, kgid_t kgid
)
748 if (!gid_valid(kgid
))
750 attr
->ia_valid
|= ATTR_GID
;
751 attr
->ia_vfsgid
= VFSGIDT_INIT(kgid
);
755 int chown_common(const struct path
*path
, uid_t user
, gid_t group
)
757 struct mnt_idmap
*idmap
;
758 struct user_namespace
*fs_userns
;
759 struct inode
*inode
= path
->dentry
->d_inode
;
760 struct inode
*delegated_inode
= NULL
;
762 struct iattr newattrs
;
766 uid
= make_kuid(current_user_ns(), user
);
767 gid
= make_kgid(current_user_ns(), group
);
769 idmap
= mnt_idmap(path
->mnt
);
770 fs_userns
= i_user_ns(inode
);
773 newattrs
.ia_vfsuid
= INVALID_VFSUID
;
774 newattrs
.ia_vfsgid
= INVALID_VFSGID
;
775 newattrs
.ia_valid
= ATTR_CTIME
;
776 if ((user
!= (uid_t
)-1) && !setattr_vfsuid(&newattrs
, uid
))
778 if ((group
!= (gid_t
)-1) && !setattr_vfsgid(&newattrs
, gid
))
781 if (!S_ISDIR(inode
->i_mode
))
782 newattrs
.ia_valid
|= ATTR_KILL_SUID
| ATTR_KILL_PRIV
|
783 setattr_should_drop_sgid(idmap
, inode
);
784 /* Continue to send actual fs values, not the mount values. */
785 error
= security_path_chown(
787 from_vfsuid(idmap
, fs_userns
, newattrs
.ia_vfsuid
),
788 from_vfsgid(idmap
, fs_userns
, newattrs
.ia_vfsgid
));
790 error
= notify_change(idmap
, path
->dentry
, &newattrs
,
793 if (delegated_inode
) {
794 error
= break_deleg_wait(&delegated_inode
);
801 int do_fchownat(int dfd
, const char __user
*filename
, uid_t user
, gid_t group
,
808 if ((flag
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
811 lookup_flags
= (flag
& AT_SYMLINK_NOFOLLOW
) ? 0 : LOOKUP_FOLLOW
;
812 if (flag
& AT_EMPTY_PATH
)
813 lookup_flags
|= LOOKUP_EMPTY
;
815 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
818 error
= mnt_want_write(path
.mnt
);
821 error
= chown_common(&path
, user
, group
);
822 mnt_drop_write(path
.mnt
);
825 if (retry_estale(error
, lookup_flags
)) {
826 lookup_flags
|= LOOKUP_REVAL
;
833 SYSCALL_DEFINE5(fchownat
, int, dfd
, const char __user
*, filename
, uid_t
, user
,
834 gid_t
, group
, int, flag
)
836 return do_fchownat(dfd
, filename
, user
, group
, flag
);
839 SYSCALL_DEFINE3(chown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
841 return do_fchownat(AT_FDCWD
, filename
, user
, group
, 0);
844 SYSCALL_DEFINE3(lchown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
846 return do_fchownat(AT_FDCWD
, filename
, user
, group
,
847 AT_SYMLINK_NOFOLLOW
);
850 int vfs_fchown(struct file
*file
, uid_t user
, gid_t group
)
854 error
= mnt_want_write_file(file
);
858 error
= chown_common(&file
->f_path
, user
, group
);
859 mnt_drop_write_file(file
);
863 int ksys_fchown(unsigned int fd
, uid_t user
, gid_t group
)
865 struct fd f
= fdget(fd
);
869 error
= vfs_fchown(f
.file
, user
, group
);
875 SYSCALL_DEFINE3(fchown
, unsigned int, fd
, uid_t
, user
, gid_t
, group
)
877 return ksys_fchown(fd
, user
, group
);
880 static inline int file_get_write_access(struct file
*f
)
884 error
= get_write_access(f
->f_inode
);
887 error
= mnt_get_write_access(f
->f_path
.mnt
);
890 if (unlikely(f
->f_mode
& FMODE_BACKING
)) {
891 error
= mnt_get_write_access(backing_file_user_path(f
)->mnt
);
898 mnt_put_write_access(f
->f_path
.mnt
);
900 put_write_access(f
->f_inode
);
904 static int do_dentry_open(struct file
*f
,
906 int (*open
)(struct inode
*, struct file
*))
908 static const struct file_operations empty_fops
= {};
911 path_get(&f
->f_path
);
913 f
->f_mapping
= inode
->i_mapping
;
914 f
->f_wb_err
= filemap_sample_wb_err(f
->f_mapping
);
915 f
->f_sb_err
= file_sample_sb_err(f
);
917 if (unlikely(f
->f_flags
& O_PATH
)) {
918 f
->f_mode
= FMODE_PATH
| FMODE_OPENED
;
919 f
->f_op
= &empty_fops
;
923 if ((f
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) == FMODE_READ
) {
924 i_readcount_inc(inode
);
925 } else if (f
->f_mode
& FMODE_WRITE
&& !special_file(inode
->i_mode
)) {
926 error
= file_get_write_access(f
);
929 f
->f_mode
|= FMODE_WRITER
;
932 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
933 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
))
934 f
->f_mode
|= FMODE_ATOMIC_POS
;
936 f
->f_op
= fops_get(inode
->i_fop
);
937 if (WARN_ON(!f
->f_op
)) {
942 error
= security_file_open(f
);
946 error
= break_lease(file_inode(f
), f
->f_flags
);
950 /* normally all 3 are set; ->open() can clear them if needed */
951 f
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
953 open
= f
->f_op
->open
;
955 error
= open(inode
, f
);
959 f
->f_mode
|= FMODE_OPENED
;
960 if ((f
->f_mode
& FMODE_READ
) &&
961 likely(f
->f_op
->read
|| f
->f_op
->read_iter
))
962 f
->f_mode
|= FMODE_CAN_READ
;
963 if ((f
->f_mode
& FMODE_WRITE
) &&
964 likely(f
->f_op
->write
|| f
->f_op
->write_iter
))
965 f
->f_mode
|= FMODE_CAN_WRITE
;
966 if ((f
->f_mode
& FMODE_LSEEK
) && !f
->f_op
->llseek
)
967 f
->f_mode
&= ~FMODE_LSEEK
;
968 if (f
->f_mapping
->a_ops
&& f
->f_mapping
->a_ops
->direct_IO
)
969 f
->f_mode
|= FMODE_CAN_ODIRECT
;
971 f
->f_flags
&= ~(O_CREAT
| O_EXCL
| O_NOCTTY
| O_TRUNC
);
972 f
->f_iocb_flags
= iocb_flags(f
);
974 file_ra_state_init(&f
->f_ra
, f
->f_mapping
->host
->i_mapping
);
976 if ((f
->f_flags
& O_DIRECT
) && !(f
->f_mode
& FMODE_CAN_ODIRECT
))
980 * XXX: Huge page cache doesn't support writing yet. Drop all page
981 * cache for this file before processing writes.
983 if (f
->f_mode
& FMODE_WRITE
) {
985 * Paired with smp_mb() in collapse_file() to ensure nr_thps
986 * is up to date and the update to i_writecount by
987 * get_write_access() is visible. Ensures subsequent insertion
988 * of THPs into the page cache will fail.
991 if (filemap_nr_thps(inode
->i_mapping
)) {
992 struct address_space
*mapping
= inode
->i_mapping
;
994 filemap_invalidate_lock(inode
->i_mapping
);
996 * unmap_mapping_range just need to be called once
997 * here, because the private pages is not need to be
998 * unmapped mapping (e.g. data segment of dynamic
999 * shared libraries here).
1001 unmap_mapping_range(mapping
, 0, 0, 0);
1002 truncate_inode_pages(mapping
, 0);
1003 filemap_invalidate_unlock(inode
->i_mapping
);
1008 * Once we return a file with FMODE_OPENED, __fput() will call
1009 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1015 if (WARN_ON_ONCE(error
> 0))
1020 path_put(&f
->f_path
);
1021 f
->f_path
.mnt
= NULL
;
1022 f
->f_path
.dentry
= NULL
;
1028 * finish_open - finish opening a file
1029 * @file: file pointer
1030 * @dentry: pointer to dentry
1031 * @open: open callback
1033 * This can be used to finish opening a file passed to i_op->atomic_open().
1035 * If the open callback is set to NULL, then the standard f_op->open()
1036 * filesystem callback is substituted.
1038 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1039 * the return value of d_splice_alias(), then the caller needs to perform dput()
1040 * on it after finish_open().
1042 * Returns zero on success or -errno if the open failed.
1044 int finish_open(struct file
*file
, struct dentry
*dentry
,
1045 int (*open
)(struct inode
*, struct file
*))
1047 BUG_ON(file
->f_mode
& FMODE_OPENED
); /* once it's opened, it's opened */
1049 file
->f_path
.dentry
= dentry
;
1050 return do_dentry_open(file
, d_backing_inode(dentry
), open
);
1052 EXPORT_SYMBOL(finish_open
);
1055 * finish_no_open - finish ->atomic_open() without opening the file
1057 * @file: file pointer
1058 * @dentry: dentry or NULL (as returned from ->lookup())
1060 * This can be used to set the result of a successful lookup in ->atomic_open().
1062 * NB: unlike finish_open() this function does consume the dentry reference and
1063 * the caller need not dput() it.
1065 * Returns "0" which must be the return value of ->atomic_open() after having
1066 * called this function.
1068 int finish_no_open(struct file
*file
, struct dentry
*dentry
)
1070 file
->f_path
.dentry
= dentry
;
1073 EXPORT_SYMBOL(finish_no_open
);
1075 char *file_path(struct file
*filp
, char *buf
, int buflen
)
1077 return d_path(&filp
->f_path
, buf
, buflen
);
1079 EXPORT_SYMBOL(file_path
);
1082 * vfs_open - open the file at the given path
1083 * @path: path to open
1084 * @file: newly allocated file with f_flag initialized
1086 int vfs_open(const struct path
*path
, struct file
*file
)
1088 file
->f_path
= *path
;
1089 return do_dentry_open(file
, d_backing_inode(path
->dentry
), NULL
);
1092 struct file
*dentry_open(const struct path
*path
, int flags
,
1093 const struct cred
*cred
)
1098 /* We must always pass in a valid mount pointer. */
1101 f
= alloc_empty_file(flags
, cred
);
1103 error
= vfs_open(path
, f
);
1111 EXPORT_SYMBOL(dentry_open
);
1114 * dentry_create - Create and open a file
1115 * @path: path to create
1117 * @mode: mode bits for new file
1118 * @cred: credentials to use
1120 * Caller must hold the parent directory's lock, and have prepared
1121 * a negative dentry, placed in @path->dentry, for the new file.
1123 * Caller sets @path->mnt to the vfsmount of the filesystem where
1124 * the new file is to be created. The parent directory and the
1125 * negative dentry must reside on the same filesystem instance.
1127 * On success, returns a "struct file *". Otherwise a ERR_PTR
1130 struct file
*dentry_create(const struct path
*path
, int flags
, umode_t mode
,
1131 const struct cred
*cred
)
1136 f
= alloc_empty_file(flags
, cred
);
1140 error
= vfs_create(mnt_idmap(path
->mnt
),
1141 d_inode(path
->dentry
->d_parent
),
1142 path
->dentry
, mode
, true);
1144 error
= vfs_open(path
, f
);
1146 if (unlikely(error
)) {
1148 return ERR_PTR(error
);
1152 EXPORT_SYMBOL(dentry_create
);
1155 * kernel_file_open - open a file for kernel internal use
1156 * @path: path of the file to open
1157 * @flags: open flags
1159 * @cred: credentials for open
1161 * Open a file for use by in-kernel consumers. The file is not accounted
1162 * against nr_files and must not be installed into the file descriptor
1165 * Return: Opened file on success, an error pointer on failure.
1167 struct file
*kernel_file_open(const struct path
*path
, int flags
,
1168 struct inode
*inode
, const struct cred
*cred
)
1173 f
= alloc_empty_file_noaccount(flags
, cred
);
1178 error
= do_dentry_open(f
, inode
, NULL
);
1185 EXPORT_SYMBOL_GPL(kernel_file_open
);
1187 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1188 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1190 inline struct open_how
build_open_how(int flags
, umode_t mode
)
1192 struct open_how how
= {
1193 .flags
= flags
& VALID_OPEN_FLAGS
,
1194 .mode
= mode
& S_IALLUGO
,
1197 /* O_PATH beats everything else. */
1198 if (how
.flags
& O_PATH
)
1199 how
.flags
&= O_PATH_FLAGS
;
1200 /* Modes should only be set for create-like flags. */
1201 if (!WILL_CREATE(how
.flags
))
1206 inline int build_open_flags(const struct open_how
*how
, struct open_flags
*op
)
1208 u64 flags
= how
->flags
;
1209 u64 strip
= __FMODE_NONOTIFY
| O_CLOEXEC
;
1210 int lookup_flags
= 0;
1211 int acc_mode
= ACC_MODE(flags
);
1213 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS
),
1214 "struct open_flags doesn't yet handle flags > 32 bits");
1217 * Strip flags that either shouldn't be set by userspace like
1218 * FMODE_NONOTIFY or that aren't relevant in determining struct
1219 * open_flags like O_CLOEXEC.
1224 * Older syscalls implicitly clear all of the invalid flags or argument
1225 * values before calling build_open_flags(), but openat2(2) checks all
1228 if (flags
& ~VALID_OPEN_FLAGS
)
1230 if (how
->resolve
& ~VALID_RESOLVE_FLAGS
)
1233 /* Scoping flags are mutually exclusive. */
1234 if ((how
->resolve
& RESOLVE_BENEATH
) && (how
->resolve
& RESOLVE_IN_ROOT
))
1237 /* Deal with the mode. */
1238 if (WILL_CREATE(flags
)) {
1239 if (how
->mode
& ~S_IALLUGO
)
1241 op
->mode
= how
->mode
| S_IFREG
;
1249 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1250 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1251 * O_TMPFILE below which requires O_DIRECTORY being raised.
1253 if ((flags
& (O_DIRECTORY
| O_CREAT
)) == (O_DIRECTORY
| O_CREAT
))
1256 /* Now handle the creative implementation of O_TMPFILE. */
1257 if (flags
& __O_TMPFILE
) {
1259 * In order to ensure programs get explicit errors when trying
1260 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1261 * is raised alongside __O_TMPFILE.
1263 if (!(flags
& O_DIRECTORY
))
1265 if (!(acc_mode
& MAY_WRITE
))
1268 if (flags
& O_PATH
) {
1269 /* O_PATH only permits certain other flags to be set. */
1270 if (flags
& ~O_PATH_FLAGS
)
1276 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1277 * check for O_DSYNC if the need any syncing at all we enforce it's
1278 * always set instead of having to deal with possibly weird behaviour
1279 * for malicious applications setting only __O_SYNC.
1281 if (flags
& __O_SYNC
)
1284 op
->open_flag
= flags
;
1286 /* O_TRUNC implies we need access checks for write permissions */
1287 if (flags
& O_TRUNC
)
1288 acc_mode
|= MAY_WRITE
;
1290 /* Allow the LSM permission hook to distinguish append
1291 access from general write access. */
1292 if (flags
& O_APPEND
)
1293 acc_mode
|= MAY_APPEND
;
1295 op
->acc_mode
= acc_mode
;
1297 op
->intent
= flags
& O_PATH
? 0 : LOOKUP_OPEN
;
1299 if (flags
& O_CREAT
) {
1300 op
->intent
|= LOOKUP_CREATE
;
1301 if (flags
& O_EXCL
) {
1302 op
->intent
|= LOOKUP_EXCL
;
1303 flags
|= O_NOFOLLOW
;
1307 if (flags
& O_DIRECTORY
)
1308 lookup_flags
|= LOOKUP_DIRECTORY
;
1309 if (!(flags
& O_NOFOLLOW
))
1310 lookup_flags
|= LOOKUP_FOLLOW
;
1312 if (how
->resolve
& RESOLVE_NO_XDEV
)
1313 lookup_flags
|= LOOKUP_NO_XDEV
;
1314 if (how
->resolve
& RESOLVE_NO_MAGICLINKS
)
1315 lookup_flags
|= LOOKUP_NO_MAGICLINKS
;
1316 if (how
->resolve
& RESOLVE_NO_SYMLINKS
)
1317 lookup_flags
|= LOOKUP_NO_SYMLINKS
;
1318 if (how
->resolve
& RESOLVE_BENEATH
)
1319 lookup_flags
|= LOOKUP_BENEATH
;
1320 if (how
->resolve
& RESOLVE_IN_ROOT
)
1321 lookup_flags
|= LOOKUP_IN_ROOT
;
1322 if (how
->resolve
& RESOLVE_CACHED
) {
1323 /* Don't bother even trying for create/truncate/tmpfile open */
1324 if (flags
& (O_TRUNC
| O_CREAT
| __O_TMPFILE
))
1326 lookup_flags
|= LOOKUP_CACHED
;
1329 op
->lookup_flags
= lookup_flags
;
1334 * file_open_name - open file and return file pointer
1336 * @name: struct filename containing path to open
1337 * @flags: open flags as per the open(2) second argument
1338 * @mode: mode for the new file if O_CREAT is set, else ignored
1340 * This is the helper to open a file from kernelspace if you really
1341 * have to. But in generally you should not do this, so please move
1342 * along, nothing to see here..
1344 struct file
*file_open_name(struct filename
*name
, int flags
, umode_t mode
)
1346 struct open_flags op
;
1347 struct open_how how
= build_open_how(flags
, mode
);
1348 int err
= build_open_flags(&how
, &op
);
1350 return ERR_PTR(err
);
1351 return do_filp_open(AT_FDCWD
, name
, &op
);
1355 * filp_open - open file and return file pointer
1357 * @filename: path to open
1358 * @flags: open flags as per the open(2) second argument
1359 * @mode: mode for the new file if O_CREAT is set, else ignored
1361 * This is the helper to open a file from kernelspace if you really
1362 * have to. But in generally you should not do this, so please move
1363 * along, nothing to see here..
1365 struct file
*filp_open(const char *filename
, int flags
, umode_t mode
)
1367 struct filename
*name
= getname_kernel(filename
);
1368 struct file
*file
= ERR_CAST(name
);
1370 if (!IS_ERR(name
)) {
1371 file
= file_open_name(name
, flags
, mode
);
1376 EXPORT_SYMBOL(filp_open
);
1378 struct file
*file_open_root(const struct path
*root
,
1379 const char *filename
, int flags
, umode_t mode
)
1381 struct open_flags op
;
1382 struct open_how how
= build_open_how(flags
, mode
);
1383 int err
= build_open_flags(&how
, &op
);
1385 return ERR_PTR(err
);
1386 return do_file_open_root(root
, filename
, &op
);
1388 EXPORT_SYMBOL(file_open_root
);
1390 static long do_sys_openat2(int dfd
, const char __user
*filename
,
1391 struct open_how
*how
)
1393 struct open_flags op
;
1394 int fd
= build_open_flags(how
, &op
);
1395 struct filename
*tmp
;
1400 tmp
= getname(filename
);
1402 return PTR_ERR(tmp
);
1404 fd
= get_unused_fd_flags(how
->flags
);
1406 struct file
*f
= do_filp_open(dfd
, tmp
, &op
);
1418 long do_sys_open(int dfd
, const char __user
*filename
, int flags
, umode_t mode
)
1420 struct open_how how
= build_open_how(flags
, mode
);
1421 return do_sys_openat2(dfd
, filename
, &how
);
1425 SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1427 if (force_o_largefile())
1428 flags
|= O_LARGEFILE
;
1429 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1432 SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
,
1435 if (force_o_largefile())
1436 flags
|= O_LARGEFILE
;
1437 return do_sys_open(dfd
, filename
, flags
, mode
);
1440 SYSCALL_DEFINE4(openat2
, int, dfd
, const char __user
*, filename
,
1441 struct open_how __user
*, how
, size_t, usize
)
1444 struct open_how tmp
;
1446 BUILD_BUG_ON(sizeof(struct open_how
) < OPEN_HOW_SIZE_VER0
);
1447 BUILD_BUG_ON(sizeof(struct open_how
) != OPEN_HOW_SIZE_LATEST
);
1449 if (unlikely(usize
< OPEN_HOW_SIZE_VER0
))
1452 err
= copy_struct_from_user(&tmp
, sizeof(tmp
), how
, usize
);
1456 audit_openat2_how(&tmp
);
1458 /* O_LARGEFILE is only allowed for non-O_PATH. */
1459 if (!(tmp
.flags
& O_PATH
) && force_o_largefile())
1460 tmp
.flags
|= O_LARGEFILE
;
1462 return do_sys_openat2(dfd
, filename
, &tmp
);
1465 #ifdef CONFIG_COMPAT
1467 * Exactly like sys_open(), except that it doesn't set the
1470 COMPAT_SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1472 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1476 * Exactly like sys_openat(), except that it doesn't set the
1479 COMPAT_SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1481 return do_sys_open(dfd
, filename
, flags
, mode
);
1488 * For backward compatibility? Maybe this should be moved
1489 * into arch/i386 instead?
1491 SYSCALL_DEFINE2(creat
, const char __user
*, pathname
, umode_t
, mode
)
1493 int flags
= O_CREAT
| O_WRONLY
| O_TRUNC
;
1495 if (force_o_largefile())
1496 flags
|= O_LARGEFILE
;
1497 return do_sys_open(AT_FDCWD
, pathname
, flags
, mode
);
1502 * "id" is the POSIX thread ID. We use the
1503 * files pointer for this..
1505 static int filp_flush(struct file
*filp
, fl_owner_t id
)
1509 if (CHECK_DATA_CORRUPTION(file_count(filp
) == 0,
1510 "VFS: Close: file count is 0 (f_op=%ps)",
1515 if (filp
->f_op
->flush
)
1516 retval
= filp
->f_op
->flush(filp
, id
);
1518 if (likely(!(filp
->f_mode
& FMODE_PATH
))) {
1519 dnotify_flush(filp
, id
);
1520 locks_remove_posix(filp
, id
);
1525 int filp_close(struct file
*filp
, fl_owner_t id
)
1529 retval
= filp_flush(filp
, id
);
1534 EXPORT_SYMBOL(filp_close
);
1537 * Careful here! We test whether the file pointer is NULL before
1538 * releasing the fd. This ensures that one clone task can't release
1539 * an fd while another clone is opening it.
1541 SYSCALL_DEFINE1(close
, unsigned int, fd
)
1546 file
= file_close_fd(fd
);
1550 retval
= filp_flush(file
, current
->files
);
1553 * We're returning to user space. Don't bother
1554 * with any delayed fput() cases.
1558 /* can't restart close syscall because file table entry was cleared */
1559 if (unlikely(retval
== -ERESTARTSYS
||
1560 retval
== -ERESTARTNOINTR
||
1561 retval
== -ERESTARTNOHAND
||
1562 retval
== -ERESTART_RESTARTBLOCK
))
1569 * sys_close_range() - Close all file descriptors in a given range.
1571 * @fd: starting file descriptor to close
1572 * @max_fd: last file descriptor to close
1573 * @flags: reserved for future extensions
1575 * This closes a range of file descriptors. All file descriptors
1576 * from @fd up to and including @max_fd are closed.
1577 * Currently, errors to close a given file descriptor are ignored.
1579 SYSCALL_DEFINE3(close_range
, unsigned int, fd
, unsigned int, max_fd
,
1580 unsigned int, flags
)
1582 return __close_range(fd
, max_fd
, flags
);
1586 * This routine simulates a hangup on the tty, to arrange that users
1587 * are given clean terminals at login time.
1589 SYSCALL_DEFINE0(vhangup
)
1591 if (capable(CAP_SYS_TTY_CONFIG
)) {
1599 * Called when an inode is about to be open.
1600 * We use this to disallow opening large files on 32bit systems if
1601 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1602 * on this flag in sys_open.
1604 int generic_file_open(struct inode
* inode
, struct file
* filp
)
1606 if (!(filp
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1611 EXPORT_SYMBOL(generic_file_open
);
1614 * This is used by subsystems that don't want seekable
1615 * file descriptors. The function is not supposed to ever fail, the only
1616 * reason it returns an 'int' and not 'void' is so that it can be plugged
1617 * directly into file_operations structure.
1619 int nonseekable_open(struct inode
*inode
, struct file
*filp
)
1621 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
);
1625 EXPORT_SYMBOL(nonseekable_open
);
1628 * stream_open is used by subsystems that want stream-like file descriptors.
1629 * Such file descriptors are not seekable and don't have notion of position
1630 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1631 * Contrary to file descriptors of other regular files, .read() and .write()
1632 * can run simultaneously.
1634 * stream_open never fails and is marked to return int so that it could be
1635 * directly used as file_operations.open .
1637 int stream_open(struct inode
*inode
, struct file
*filp
)
1639 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
| FMODE_ATOMIC_POS
);
1640 filp
->f_mode
|= FMODE_STREAM
;
1644 EXPORT_SYMBOL(stream_open
);