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
);
69 EXPORT_SYMBOL_GPL(do_truncate
);
71 long vfs_truncate(const struct path
*path
, loff_t length
)
73 struct user_namespace
*mnt_userns
;
77 inode
= path
->dentry
->d_inode
;
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 if (S_ISDIR(inode
->i_mode
))
82 if (!S_ISREG(inode
->i_mode
))
85 error
= mnt_want_write(path
->mnt
);
89 mnt_userns
= mnt_user_ns(path
->mnt
);
90 error
= inode_permission(mnt_userns
, inode
, MAY_WRITE
);
92 goto mnt_drop_write_and_out
;
96 goto mnt_drop_write_and_out
;
98 error
= get_write_access(inode
);
100 goto mnt_drop_write_and_out
;
103 * Make sure that there are no leases. get_write_access() protects
104 * against the truncate racing with a lease-granting setlease().
106 error
= break_lease(inode
, O_WRONLY
);
108 goto put_write_and_out
;
110 error
= security_path_truncate(path
);
112 error
= do_truncate(mnt_userns
, path
->dentry
, length
, 0, NULL
);
115 put_write_access(inode
);
116 mnt_drop_write_and_out
:
117 mnt_drop_write(path
->mnt
);
121 EXPORT_SYMBOL_GPL(vfs_truncate
);
123 long do_sys_truncate(const char __user
*pathname
, loff_t length
)
125 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
129 if (length
< 0) /* sorry, but loff_t says... */
133 error
= user_path_at(AT_FDCWD
, pathname
, lookup_flags
, &path
);
135 error
= vfs_truncate(&path
, length
);
138 if (retry_estale(error
, lookup_flags
)) {
139 lookup_flags
|= LOOKUP_REVAL
;
145 SYSCALL_DEFINE2(truncate
, const char __user
*, path
, long, length
)
147 return do_sys_truncate(path
, length
);
151 COMPAT_SYSCALL_DEFINE2(truncate
, const char __user
*, path
, compat_off_t
, length
)
153 return do_sys_truncate(path
, length
);
157 long do_sys_ftruncate(unsigned int fd
, loff_t length
, int small
)
160 struct dentry
*dentry
;
172 /* explicitly opened as large or we are on 64-bit box */
173 if (f
.file
->f_flags
& O_LARGEFILE
)
176 dentry
= f
.file
->f_path
.dentry
;
177 inode
= dentry
->d_inode
;
179 if (!S_ISREG(inode
->i_mode
) || !(f
.file
->f_mode
& FMODE_WRITE
))
183 /* Cannot ftruncate over 2^31 bytes without large file support */
184 if (small
&& length
> MAX_NON_LFS
)
188 /* Check IS_APPEND on real upper inode */
189 if (IS_APPEND(file_inode(f
.file
)))
191 sb_start_write(inode
->i_sb
);
192 error
= security_path_truncate(&f
.file
->f_path
);
194 error
= do_truncate(file_mnt_user_ns(f
.file
), dentry
, length
,
195 ATTR_MTIME
| ATTR_CTIME
, f
.file
);
196 sb_end_write(inode
->i_sb
);
203 SYSCALL_DEFINE2(ftruncate
, unsigned int, fd
, unsigned long, length
)
205 return do_sys_ftruncate(fd
, length
, 1);
209 COMPAT_SYSCALL_DEFINE2(ftruncate
, unsigned int, fd
, compat_ulong_t
, length
)
211 return do_sys_ftruncate(fd
, length
, 1);
215 /* LFS versions of truncate are only needed on 32 bit machines */
216 #if BITS_PER_LONG == 32
217 SYSCALL_DEFINE2(truncate64
, const char __user
*, path
, loff_t
, length
)
219 return do_sys_truncate(path
, length
);
222 SYSCALL_DEFINE2(ftruncate64
, unsigned int, fd
, loff_t
, length
)
224 return do_sys_ftruncate(fd
, length
, 0);
226 #endif /* BITS_PER_LONG == 32 */
229 int vfs_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
231 struct inode
*inode
= file_inode(file
);
234 if (offset
< 0 || len
<= 0)
237 /* Return error if mode is not supported */
238 if (mode
& ~FALLOC_FL_SUPPORTED_MASK
)
241 /* Punch hole and zero range are mutually exclusive */
242 if ((mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
)) ==
243 (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
))
246 /* Punch hole must have keep size set */
247 if ((mode
& FALLOC_FL_PUNCH_HOLE
) &&
248 !(mode
& FALLOC_FL_KEEP_SIZE
))
251 /* Collapse range should only be used exclusively. */
252 if ((mode
& FALLOC_FL_COLLAPSE_RANGE
) &&
253 (mode
& ~FALLOC_FL_COLLAPSE_RANGE
))
256 /* Insert range should only be used exclusively. */
257 if ((mode
& FALLOC_FL_INSERT_RANGE
) &&
258 (mode
& ~FALLOC_FL_INSERT_RANGE
))
261 /* Unshare range should only be used with allocate mode. */
262 if ((mode
& FALLOC_FL_UNSHARE_RANGE
) &&
263 (mode
& ~(FALLOC_FL_UNSHARE_RANGE
| FALLOC_FL_KEEP_SIZE
)))
266 if (!(file
->f_mode
& FMODE_WRITE
))
270 * We can only allow pure fallocate on append only files
272 if ((mode
& ~FALLOC_FL_KEEP_SIZE
) && IS_APPEND(inode
))
275 if (IS_IMMUTABLE(inode
))
279 * We cannot allow any fallocate operation on an active swapfile
281 if (IS_SWAPFILE(inode
))
285 * Revalidate the write permissions, in case security policy has
286 * changed since the files were opened.
288 ret
= security_file_permission(file
, MAY_WRITE
);
292 if (S_ISFIFO(inode
->i_mode
))
295 if (S_ISDIR(inode
->i_mode
))
298 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
301 /* Check for wrap through zero too */
302 if (((offset
+ len
) > inode
->i_sb
->s_maxbytes
) || ((offset
+ len
) < 0))
305 if (!file
->f_op
->fallocate
)
308 file_start_write(file
);
309 ret
= file
->f_op
->fallocate(file
, mode
, offset
, len
);
312 * Create inotify and fanotify events.
314 * To keep the logic simple always create events if fallocate succeeds.
315 * This implies that events are even created if the file size remains
316 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
319 fsnotify_modify(file
);
321 file_end_write(file
);
324 EXPORT_SYMBOL_GPL(vfs_fallocate
);
326 int ksys_fallocate(int fd
, int mode
, loff_t offset
, loff_t len
)
328 struct fd f
= fdget(fd
);
332 error
= vfs_fallocate(f
.file
, mode
, offset
, len
);
338 SYSCALL_DEFINE4(fallocate
, int, fd
, int, mode
, loff_t
, offset
, loff_t
, len
)
340 return ksys_fallocate(fd
, mode
, offset
, len
);
344 * access() needs to use the real uid/gid, not the effective uid/gid.
345 * We do this by temporarily clearing all FS-related capabilities and
346 * switching the fsuid/fsgid around to the real ones.
348 static const struct cred
*access_override_creds(void)
350 const struct cred
*old_cred
;
351 struct cred
*override_cred
;
353 override_cred
= prepare_creds();
357 override_cred
->fsuid
= override_cred
->uid
;
358 override_cred
->fsgid
= override_cred
->gid
;
360 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
361 /* Clear the capabilities if we switch to a non-root user */
362 kuid_t root_uid
= make_kuid(override_cred
->user_ns
, 0);
363 if (!uid_eq(override_cred
->uid
, root_uid
))
364 cap_clear(override_cred
->cap_effective
);
366 override_cred
->cap_effective
=
367 override_cred
->cap_permitted
;
371 * The new set of credentials can *only* be used in
372 * task-synchronous circumstances, and does not need
373 * RCU freeing, unless somebody then takes a separate
376 * NOTE! This is _only_ true because this credential
377 * is used purely for override_creds() that installs
378 * it as the subjective cred. Other threads will be
379 * accessing ->real_cred, not the subjective cred.
381 * If somebody _does_ make a copy of this (using the
382 * 'get_current_cred()' function), that will clear the
383 * non_rcu field, because now that other user may be
384 * expecting RCU freeing. But normal thread-synchronous
385 * cred accesses will keep things non-RCY.
387 override_cred
->non_rcu
= 1;
389 old_cred
= override_creds(override_cred
);
391 /* override_cred() gets its own ref */
392 put_cred(override_cred
);
397 static long do_faccessat(int dfd
, const char __user
*filename
, int mode
, int flags
)
402 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
403 const struct cred
*old_cred
= NULL
;
405 if (mode
& ~S_IRWXO
) /* where's F_OK, X_OK, W_OK, R_OK? */
408 if (flags
& ~(AT_EACCESS
| AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
))
411 if (flags
& AT_SYMLINK_NOFOLLOW
)
412 lookup_flags
&= ~LOOKUP_FOLLOW
;
413 if (flags
& AT_EMPTY_PATH
)
414 lookup_flags
|= LOOKUP_EMPTY
;
416 if (!(flags
& AT_EACCESS
)) {
417 old_cred
= access_override_creds();
423 res
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
427 inode
= d_backing_inode(path
.dentry
);
429 if ((mode
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
431 * MAY_EXEC on regular files is denied if the fs is mounted
432 * with the "noexec" flag.
435 if (path_noexec(&path
))
436 goto out_path_release
;
439 res
= inode_permission(mnt_user_ns(path
.mnt
), inode
, mode
| MAY_ACCESS
);
440 /* SuS v2 requires we report a read only fs too */
441 if (res
|| !(mode
& S_IWOTH
) || special_file(inode
->i_mode
))
442 goto out_path_release
;
444 * This is a rare case where using __mnt_is_readonly()
445 * is OK without a mnt_want/drop_write() pair. Since
446 * no actual write to the fs is performed here, we do
447 * not need to telegraph to that to anyone.
449 * By doing this, we accept that this access is
450 * inherently racy and know that the fs may change
451 * state before we even see this result.
453 if (__mnt_is_readonly(path
.mnt
))
458 if (retry_estale(res
, lookup_flags
)) {
459 lookup_flags
|= LOOKUP_REVAL
;
464 revert_creds(old_cred
);
469 SYSCALL_DEFINE3(faccessat
, int, dfd
, const char __user
*, filename
, int, mode
)
471 return do_faccessat(dfd
, filename
, mode
, 0);
474 SYSCALL_DEFINE4(faccessat2
, int, dfd
, const char __user
*, filename
, int, mode
,
477 return do_faccessat(dfd
, filename
, mode
, flags
);
480 SYSCALL_DEFINE2(access
, const char __user
*, filename
, int, mode
)
482 return do_faccessat(AT_FDCWD
, filename
, mode
, 0);
485 SYSCALL_DEFINE1(chdir
, const char __user
*, filename
)
489 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
491 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
495 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
499 set_fs_pwd(current
->fs
, &path
);
503 if (retry_estale(error
, lookup_flags
)) {
504 lookup_flags
|= LOOKUP_REVAL
;
511 SYSCALL_DEFINE1(fchdir
, unsigned int, fd
)
513 struct fd f
= fdget_raw(fd
);
521 if (!d_can_lookup(f
.file
->f_path
.dentry
))
524 error
= file_permission(f
.file
, MAY_EXEC
| MAY_CHDIR
);
526 set_fs_pwd(current
->fs
, &f
.file
->f_path
);
533 SYSCALL_DEFINE1(chroot
, const char __user
*, filename
)
537 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
539 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
543 error
= path_permission(&path
, MAY_EXEC
| MAY_CHDIR
);
548 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT
))
550 error
= security_path_chroot(&path
);
554 set_fs_root(current
->fs
, &path
);
558 if (retry_estale(error
, lookup_flags
)) {
559 lookup_flags
|= LOOKUP_REVAL
;
566 int chmod_common(const struct path
*path
, umode_t mode
)
568 struct inode
*inode
= path
->dentry
->d_inode
;
569 struct inode
*delegated_inode
= NULL
;
570 struct iattr newattrs
;
573 error
= mnt_want_write(path
->mnt
);
578 error
= security_path_chmod(path
, mode
);
581 newattrs
.ia_mode
= (mode
& S_IALLUGO
) | (inode
->i_mode
& ~S_IALLUGO
);
582 newattrs
.ia_valid
= ATTR_MODE
| ATTR_CTIME
;
583 error
= notify_change(mnt_user_ns(path
->mnt
), path
->dentry
,
584 &newattrs
, &delegated_inode
);
587 if (delegated_inode
) {
588 error
= break_deleg_wait(&delegated_inode
);
592 mnt_drop_write(path
->mnt
);
596 int vfs_fchmod(struct file
*file
, umode_t mode
)
599 return chmod_common(&file
->f_path
, mode
);
602 SYSCALL_DEFINE2(fchmod
, unsigned int, fd
, umode_t
, mode
)
604 struct fd f
= fdget(fd
);
608 err
= vfs_fchmod(f
.file
, mode
);
614 static int do_fchmodat(int dfd
, const char __user
*filename
, umode_t mode
)
618 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
620 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
622 error
= chmod_common(&path
, mode
);
624 if (retry_estale(error
, lookup_flags
)) {
625 lookup_flags
|= LOOKUP_REVAL
;
632 SYSCALL_DEFINE3(fchmodat
, int, dfd
, const char __user
*, filename
,
635 return do_fchmodat(dfd
, filename
, mode
);
638 SYSCALL_DEFINE2(chmod
, const char __user
*, filename
, umode_t
, mode
)
640 return do_fchmodat(AT_FDCWD
, filename
, mode
);
643 int chown_common(const struct path
*path
, uid_t user
, gid_t group
)
645 struct user_namespace
*mnt_userns
, *fs_userns
;
646 struct inode
*inode
= path
->dentry
->d_inode
;
647 struct inode
*delegated_inode
= NULL
;
649 struct iattr newattrs
;
653 uid
= make_kuid(current_user_ns(), user
);
654 gid
= make_kgid(current_user_ns(), group
);
656 mnt_userns
= mnt_user_ns(path
->mnt
);
657 fs_userns
= i_user_ns(inode
);
658 uid
= mapped_kuid_user(mnt_userns
, fs_userns
, uid
);
659 gid
= mapped_kgid_user(mnt_userns
, fs_userns
, gid
);
662 newattrs
.ia_valid
= ATTR_CTIME
;
663 if (user
!= (uid_t
) -1) {
666 newattrs
.ia_valid
|= ATTR_UID
;
667 newattrs
.ia_uid
= uid
;
669 if (group
!= (gid_t
) -1) {
672 newattrs
.ia_valid
|= ATTR_GID
;
673 newattrs
.ia_gid
= gid
;
675 if (!S_ISDIR(inode
->i_mode
))
677 ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
679 error
= security_path_chown(path
, uid
, gid
);
681 error
= notify_change(mnt_userns
, path
->dentry
, &newattrs
,
684 if (delegated_inode
) {
685 error
= break_deleg_wait(&delegated_inode
);
692 int do_fchownat(int dfd
, const char __user
*filename
, uid_t user
, gid_t group
,
699 if ((flag
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
702 lookup_flags
= (flag
& AT_SYMLINK_NOFOLLOW
) ? 0 : LOOKUP_FOLLOW
;
703 if (flag
& AT_EMPTY_PATH
)
704 lookup_flags
|= LOOKUP_EMPTY
;
706 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
709 error
= mnt_want_write(path
.mnt
);
712 error
= chown_common(&path
, user
, group
);
713 mnt_drop_write(path
.mnt
);
716 if (retry_estale(error
, lookup_flags
)) {
717 lookup_flags
|= LOOKUP_REVAL
;
724 SYSCALL_DEFINE5(fchownat
, int, dfd
, const char __user
*, filename
, uid_t
, user
,
725 gid_t
, group
, int, flag
)
727 return do_fchownat(dfd
, filename
, user
, group
, flag
);
730 SYSCALL_DEFINE3(chown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
732 return do_fchownat(AT_FDCWD
, filename
, user
, group
, 0);
735 SYSCALL_DEFINE3(lchown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
737 return do_fchownat(AT_FDCWD
, filename
, user
, group
,
738 AT_SYMLINK_NOFOLLOW
);
741 int vfs_fchown(struct file
*file
, uid_t user
, gid_t group
)
745 error
= mnt_want_write_file(file
);
749 error
= chown_common(&file
->f_path
, user
, group
);
750 mnt_drop_write_file(file
);
754 int ksys_fchown(unsigned int fd
, uid_t user
, gid_t group
)
756 struct fd f
= fdget(fd
);
760 error
= vfs_fchown(f
.file
, user
, group
);
766 SYSCALL_DEFINE3(fchown
, unsigned int, fd
, uid_t
, user
, gid_t
, group
)
768 return ksys_fchown(fd
, user
, group
);
771 static int do_dentry_open(struct file
*f
,
773 int (*open
)(struct inode
*, struct file
*))
775 static const struct file_operations empty_fops
= {};
778 path_get(&f
->f_path
);
780 f
->f_mapping
= inode
->i_mapping
;
781 f
->f_wb_err
= filemap_sample_wb_err(f
->f_mapping
);
782 f
->f_sb_err
= file_sample_sb_err(f
);
784 if (unlikely(f
->f_flags
& O_PATH
)) {
785 f
->f_mode
= FMODE_PATH
| FMODE_OPENED
;
786 f
->f_op
= &empty_fops
;
790 if ((f
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) == FMODE_READ
) {
791 i_readcount_inc(inode
);
792 } else if (f
->f_mode
& FMODE_WRITE
&& !special_file(inode
->i_mode
)) {
793 error
= get_write_access(inode
);
796 error
= __mnt_want_write(f
->f_path
.mnt
);
797 if (unlikely(error
)) {
798 put_write_access(inode
);
801 f
->f_mode
|= FMODE_WRITER
;
804 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
805 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
))
806 f
->f_mode
|= FMODE_ATOMIC_POS
;
808 f
->f_op
= fops_get(inode
->i_fop
);
809 if (WARN_ON(!f
->f_op
)) {
814 error
= security_file_open(f
);
818 error
= break_lease(locks_inode(f
), f
->f_flags
);
822 /* normally all 3 are set; ->open() can clear them if needed */
823 f
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
825 open
= f
->f_op
->open
;
827 error
= open(inode
, f
);
831 f
->f_mode
|= FMODE_OPENED
;
832 if ((f
->f_mode
& FMODE_READ
) &&
833 likely(f
->f_op
->read
|| f
->f_op
->read_iter
))
834 f
->f_mode
|= FMODE_CAN_READ
;
835 if ((f
->f_mode
& FMODE_WRITE
) &&
836 likely(f
->f_op
->write
|| f
->f_op
->write_iter
))
837 f
->f_mode
|= FMODE_CAN_WRITE
;
839 f
->f_write_hint
= WRITE_LIFE_NOT_SET
;
840 f
->f_flags
&= ~(O_CREAT
| O_EXCL
| O_NOCTTY
| O_TRUNC
);
842 file_ra_state_init(&f
->f_ra
, f
->f_mapping
->host
->i_mapping
);
844 /* NB: we're sure to have correct a_ops only after f_op->open */
845 if (f
->f_flags
& O_DIRECT
) {
846 if (!f
->f_mapping
->a_ops
|| !f
->f_mapping
->a_ops
->direct_IO
)
851 * XXX: Huge page cache doesn't support writing yet. Drop all page
852 * cache for this file before processing writes.
854 if (f
->f_mode
& FMODE_WRITE
) {
856 * Paired with smp_mb() in collapse_file() to ensure nr_thps
857 * is up to date and the update to i_writecount by
858 * get_write_access() is visible. Ensures subsequent insertion
859 * of THPs into the page cache will fail.
862 if (filemap_nr_thps(inode
->i_mapping
)) {
863 struct address_space
*mapping
= inode
->i_mapping
;
865 filemap_invalidate_lock(inode
->i_mapping
);
867 * unmap_mapping_range just need to be called once
868 * here, because the private pages is not need to be
869 * unmapped mapping (e.g. data segment of dynamic
870 * shared libraries here).
872 unmap_mapping_range(mapping
, 0, 0, 0);
873 truncate_inode_pages(mapping
, 0);
874 filemap_invalidate_unlock(inode
->i_mapping
);
881 if (WARN_ON_ONCE(error
> 0))
886 path_put(&f
->f_path
);
887 f
->f_path
.mnt
= NULL
;
888 f
->f_path
.dentry
= NULL
;
894 * finish_open - finish opening a file
895 * @file: file pointer
896 * @dentry: pointer to dentry
897 * @open: open callback
898 * @opened: state of open
900 * This can be used to finish opening a file passed to i_op->atomic_open().
902 * If the open callback is set to NULL, then the standard f_op->open()
903 * filesystem callback is substituted.
905 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
906 * the return value of d_splice_alias(), then the caller needs to perform dput()
907 * on it after finish_open().
909 * Returns zero on success or -errno if the open failed.
911 int finish_open(struct file
*file
, struct dentry
*dentry
,
912 int (*open
)(struct inode
*, struct file
*))
914 BUG_ON(file
->f_mode
& FMODE_OPENED
); /* once it's opened, it's opened */
916 file
->f_path
.dentry
= dentry
;
917 return do_dentry_open(file
, d_backing_inode(dentry
), open
);
919 EXPORT_SYMBOL(finish_open
);
922 * finish_no_open - finish ->atomic_open() without opening the file
924 * @file: file pointer
925 * @dentry: dentry or NULL (as returned from ->lookup())
927 * This can be used to set the result of a successful lookup in ->atomic_open().
929 * NB: unlike finish_open() this function does consume the dentry reference and
930 * the caller need not dput() it.
932 * Returns "0" which must be the return value of ->atomic_open() after having
933 * called this function.
935 int finish_no_open(struct file
*file
, struct dentry
*dentry
)
937 file
->f_path
.dentry
= dentry
;
940 EXPORT_SYMBOL(finish_no_open
);
942 char *file_path(struct file
*filp
, char *buf
, int buflen
)
944 return d_path(&filp
->f_path
, buf
, buflen
);
946 EXPORT_SYMBOL(file_path
);
949 * vfs_open - open the file at the given path
950 * @path: path to open
951 * @file: newly allocated file with f_flag initialized
952 * @cred: credentials to use
954 int vfs_open(const struct path
*path
, struct file
*file
)
956 file
->f_path
= *path
;
957 return do_dentry_open(file
, d_backing_inode(path
->dentry
), NULL
);
960 struct file
*dentry_open(const struct path
*path
, int flags
,
961 const struct cred
*cred
)
966 validate_creds(cred
);
968 /* We must always pass in a valid mount pointer. */
971 f
= alloc_empty_file(flags
, cred
);
973 error
= vfs_open(path
, f
);
981 EXPORT_SYMBOL(dentry_open
);
983 struct file
*open_with_fake_path(const struct path
*path
, int flags
,
984 struct inode
*inode
, const struct cred
*cred
)
986 struct file
*f
= alloc_empty_file_noaccount(flags
, cred
);
991 error
= do_dentry_open(f
, inode
, NULL
);
999 EXPORT_SYMBOL(open_with_fake_path
);
1001 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1002 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1004 inline struct open_how
build_open_how(int flags
, umode_t mode
)
1006 struct open_how how
= {
1007 .flags
= flags
& VALID_OPEN_FLAGS
,
1008 .mode
= mode
& S_IALLUGO
,
1011 /* O_PATH beats everything else. */
1012 if (how
.flags
& O_PATH
)
1013 how
.flags
&= O_PATH_FLAGS
;
1014 /* Modes should only be set for create-like flags. */
1015 if (!WILL_CREATE(how
.flags
))
1020 inline int build_open_flags(const struct open_how
*how
, struct open_flags
*op
)
1022 u64 flags
= how
->flags
;
1023 u64 strip
= FMODE_NONOTIFY
| O_CLOEXEC
;
1024 int lookup_flags
= 0;
1025 int acc_mode
= ACC_MODE(flags
);
1027 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS
),
1028 "struct open_flags doesn't yet handle flags > 32 bits");
1031 * Strip flags that either shouldn't be set by userspace like
1032 * FMODE_NONOTIFY or that aren't relevant in determining struct
1033 * open_flags like O_CLOEXEC.
1038 * Older syscalls implicitly clear all of the invalid flags or argument
1039 * values before calling build_open_flags(), but openat2(2) checks all
1042 if (flags
& ~VALID_OPEN_FLAGS
)
1044 if (how
->resolve
& ~VALID_RESOLVE_FLAGS
)
1047 /* Scoping flags are mutually exclusive. */
1048 if ((how
->resolve
& RESOLVE_BENEATH
) && (how
->resolve
& RESOLVE_IN_ROOT
))
1051 /* Deal with the mode. */
1052 if (WILL_CREATE(flags
)) {
1053 if (how
->mode
& ~S_IALLUGO
)
1055 op
->mode
= how
->mode
| S_IFREG
;
1063 * In order to ensure programs get explicit errors when trying to use
1064 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1065 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1066 * have to require userspace to explicitly set it.
1068 if (flags
& __O_TMPFILE
) {
1069 if ((flags
& O_TMPFILE_MASK
) != O_TMPFILE
)
1071 if (!(acc_mode
& MAY_WRITE
))
1074 if (flags
& O_PATH
) {
1075 /* O_PATH only permits certain other flags to be set. */
1076 if (flags
& ~O_PATH_FLAGS
)
1082 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1083 * check for O_DSYNC if the need any syncing at all we enforce it's
1084 * always set instead of having to deal with possibly weird behaviour
1085 * for malicious applications setting only __O_SYNC.
1087 if (flags
& __O_SYNC
)
1090 op
->open_flag
= flags
;
1092 /* O_TRUNC implies we need access checks for write permissions */
1093 if (flags
& O_TRUNC
)
1094 acc_mode
|= MAY_WRITE
;
1096 /* Allow the LSM permission hook to distinguish append
1097 access from general write access. */
1098 if (flags
& O_APPEND
)
1099 acc_mode
|= MAY_APPEND
;
1101 op
->acc_mode
= acc_mode
;
1103 op
->intent
= flags
& O_PATH
? 0 : LOOKUP_OPEN
;
1105 if (flags
& O_CREAT
) {
1106 op
->intent
|= LOOKUP_CREATE
;
1107 if (flags
& O_EXCL
) {
1108 op
->intent
|= LOOKUP_EXCL
;
1109 flags
|= O_NOFOLLOW
;
1113 if (flags
& O_DIRECTORY
)
1114 lookup_flags
|= LOOKUP_DIRECTORY
;
1115 if (!(flags
& O_NOFOLLOW
))
1116 lookup_flags
|= LOOKUP_FOLLOW
;
1118 if (how
->resolve
& RESOLVE_NO_XDEV
)
1119 lookup_flags
|= LOOKUP_NO_XDEV
;
1120 if (how
->resolve
& RESOLVE_NO_MAGICLINKS
)
1121 lookup_flags
|= LOOKUP_NO_MAGICLINKS
;
1122 if (how
->resolve
& RESOLVE_NO_SYMLINKS
)
1123 lookup_flags
|= LOOKUP_NO_SYMLINKS
;
1124 if (how
->resolve
& RESOLVE_BENEATH
)
1125 lookup_flags
|= LOOKUP_BENEATH
;
1126 if (how
->resolve
& RESOLVE_IN_ROOT
)
1127 lookup_flags
|= LOOKUP_IN_ROOT
;
1128 if (how
->resolve
& RESOLVE_CACHED
) {
1129 /* Don't bother even trying for create/truncate/tmpfile open */
1130 if (flags
& (O_TRUNC
| O_CREAT
| O_TMPFILE
))
1132 lookup_flags
|= LOOKUP_CACHED
;
1135 op
->lookup_flags
= lookup_flags
;
1140 * file_open_name - open file and return file pointer
1142 * @name: struct filename containing path to open
1143 * @flags: open flags as per the open(2) second argument
1144 * @mode: mode for the new file if O_CREAT is set, else ignored
1146 * This is the helper to open a file from kernelspace if you really
1147 * have to. But in generally you should not do this, so please move
1148 * along, nothing to see here..
1150 struct file
*file_open_name(struct filename
*name
, int flags
, umode_t mode
)
1152 struct open_flags op
;
1153 struct open_how how
= build_open_how(flags
, mode
);
1154 int err
= build_open_flags(&how
, &op
);
1156 return ERR_PTR(err
);
1157 return do_filp_open(AT_FDCWD
, name
, &op
);
1161 * filp_open - open file and return file pointer
1163 * @filename: path to open
1164 * @flags: open flags as per the open(2) second argument
1165 * @mode: mode for the new file if O_CREAT is set, else ignored
1167 * This is the helper to open a file from kernelspace if you really
1168 * have to. But in generally you should not do this, so please move
1169 * along, nothing to see here..
1171 struct file
*filp_open(const char *filename
, int flags
, umode_t mode
)
1173 struct filename
*name
= getname_kernel(filename
);
1174 struct file
*file
= ERR_CAST(name
);
1176 if (!IS_ERR(name
)) {
1177 file
= file_open_name(name
, flags
, mode
);
1182 EXPORT_SYMBOL(filp_open
);
1184 struct file
*file_open_root(const struct path
*root
,
1185 const char *filename
, int flags
, umode_t mode
)
1187 struct open_flags op
;
1188 struct open_how how
= build_open_how(flags
, mode
);
1189 int err
= build_open_flags(&how
, &op
);
1191 return ERR_PTR(err
);
1192 return do_file_open_root(root
, filename
, &op
);
1194 EXPORT_SYMBOL(file_open_root
);
1196 static long do_sys_openat2(int dfd
, const char __user
*filename
,
1197 struct open_how
*how
)
1199 struct open_flags op
;
1200 int fd
= build_open_flags(how
, &op
);
1201 struct filename
*tmp
;
1206 tmp
= getname(filename
);
1208 return PTR_ERR(tmp
);
1210 fd
= get_unused_fd_flags(how
->flags
);
1212 struct file
*f
= do_filp_open(dfd
, tmp
, &op
);
1225 long do_sys_open(int dfd
, const char __user
*filename
, int flags
, umode_t mode
)
1227 struct open_how how
= build_open_how(flags
, mode
);
1228 return do_sys_openat2(dfd
, filename
, &how
);
1232 SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1234 if (force_o_largefile())
1235 flags
|= O_LARGEFILE
;
1236 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1239 SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
,
1242 if (force_o_largefile())
1243 flags
|= O_LARGEFILE
;
1244 return do_sys_open(dfd
, filename
, flags
, mode
);
1247 SYSCALL_DEFINE4(openat2
, int, dfd
, const char __user
*, filename
,
1248 struct open_how __user
*, how
, size_t, usize
)
1251 struct open_how tmp
;
1253 BUILD_BUG_ON(sizeof(struct open_how
) < OPEN_HOW_SIZE_VER0
);
1254 BUILD_BUG_ON(sizeof(struct open_how
) != OPEN_HOW_SIZE_LATEST
);
1256 if (unlikely(usize
< OPEN_HOW_SIZE_VER0
))
1259 err
= copy_struct_from_user(&tmp
, sizeof(tmp
), how
, usize
);
1263 /* O_LARGEFILE is only allowed for non-O_PATH. */
1264 if (!(tmp
.flags
& O_PATH
) && force_o_largefile())
1265 tmp
.flags
|= O_LARGEFILE
;
1267 return do_sys_openat2(dfd
, filename
, &tmp
);
1270 #ifdef CONFIG_COMPAT
1272 * Exactly like sys_open(), except that it doesn't set the
1275 COMPAT_SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1277 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1281 * Exactly like sys_openat(), except that it doesn't set the
1284 COMPAT_SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1286 return do_sys_open(dfd
, filename
, flags
, mode
);
1293 * For backward compatibility? Maybe this should be moved
1294 * into arch/i386 instead?
1296 SYSCALL_DEFINE2(creat
, const char __user
*, pathname
, umode_t
, mode
)
1298 int flags
= O_CREAT
| O_WRONLY
| O_TRUNC
;
1300 if (force_o_largefile())
1301 flags
|= O_LARGEFILE
;
1302 return do_sys_open(AT_FDCWD
, pathname
, flags
, mode
);
1307 * "id" is the POSIX thread ID. We use the
1308 * files pointer for this..
1310 int filp_close(struct file
*filp
, fl_owner_t id
)
1314 if (!file_count(filp
)) {
1315 printk(KERN_ERR
"VFS: Close: file count is 0\n");
1319 if (filp
->f_op
->flush
)
1320 retval
= filp
->f_op
->flush(filp
, id
);
1322 if (likely(!(filp
->f_mode
& FMODE_PATH
))) {
1323 dnotify_flush(filp
, id
);
1324 locks_remove_posix(filp
, id
);
1330 EXPORT_SYMBOL(filp_close
);
1333 * Careful here! We test whether the file pointer is NULL before
1334 * releasing the fd. This ensures that one clone task can't release
1335 * an fd while another clone is opening it.
1337 SYSCALL_DEFINE1(close
, unsigned int, fd
)
1339 int retval
= close_fd(fd
);
1341 /* can't restart close syscall because file table entry was cleared */
1342 if (unlikely(retval
== -ERESTARTSYS
||
1343 retval
== -ERESTARTNOINTR
||
1344 retval
== -ERESTARTNOHAND
||
1345 retval
== -ERESTART_RESTARTBLOCK
))
1352 * close_range() - Close all file descriptors in a given range.
1354 * @fd: starting file descriptor to close
1355 * @max_fd: last file descriptor to close
1356 * @flags: reserved for future extensions
1358 * This closes a range of file descriptors. All file descriptors
1359 * from @fd up to and including @max_fd are closed.
1360 * Currently, errors to close a given file descriptor are ignored.
1362 SYSCALL_DEFINE3(close_range
, unsigned int, fd
, unsigned int, max_fd
,
1363 unsigned int, flags
)
1365 return __close_range(fd
, max_fd
, flags
);
1369 * This routine simulates a hangup on the tty, to arrange that users
1370 * are given clean terminals at login time.
1372 SYSCALL_DEFINE0(vhangup
)
1374 if (capable(CAP_SYS_TTY_CONFIG
)) {
1382 * Called when an inode is about to be open.
1383 * We use this to disallow opening large files on 32bit systems if
1384 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1385 * on this flag in sys_open.
1387 int generic_file_open(struct inode
* inode
, struct file
* filp
)
1389 if (!(filp
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1394 EXPORT_SYMBOL(generic_file_open
);
1397 * This is used by subsystems that don't want seekable
1398 * file descriptors. The function is not supposed to ever fail, the only
1399 * reason it returns an 'int' and not 'void' is so that it can be plugged
1400 * directly into file_operations structure.
1402 int nonseekable_open(struct inode
*inode
, struct file
*filp
)
1404 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
);
1408 EXPORT_SYMBOL(nonseekable_open
);
1411 * stream_open is used by subsystems that want stream-like file descriptors.
1412 * Such file descriptors are not seekable and don't have notion of position
1413 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1414 * Contrary to file descriptors of other regular files, .read() and .write()
1415 * can run simultaneously.
1417 * stream_open never fails and is marked to return int so that it could be
1418 * directly used as file_operations.open .
1420 int stream_open(struct inode
*inode
, struct file
*filp
)
1422 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
| FMODE_ATOMIC_POS
);
1423 filp
->f_mode
|= FMODE_STREAM
;
1427 EXPORT_SYMBOL(stream_open
);