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>
38 int do_truncate(struct dentry
*dentry
, loff_t length
, unsigned int time_attrs
,
42 struct iattr newattrs
;
44 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
48 newattrs
.ia_size
= length
;
49 newattrs
.ia_valid
= ATTR_SIZE
| time_attrs
;
51 newattrs
.ia_file
= filp
;
52 newattrs
.ia_valid
|= ATTR_FILE
;
55 /* Remove suid, sgid, and file capabilities on truncate too */
56 ret
= dentry_needs_remove_privs(dentry
);
60 newattrs
.ia_valid
|= ret
| ATTR_FORCE
;
62 inode_lock(dentry
->d_inode
);
63 /* Note any delegations or leases have already been broken: */
64 ret
= notify_change(dentry
, &newattrs
, NULL
);
65 inode_unlock(dentry
->d_inode
);
68 EXPORT_SYMBOL_GPL(do_truncate
);
70 long vfs_truncate(const struct path
*path
, loff_t length
)
75 inode
= path
->dentry
->d_inode
;
77 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
78 if (S_ISDIR(inode
->i_mode
))
80 if (!S_ISREG(inode
->i_mode
))
83 error
= mnt_want_write(path
->mnt
);
87 error
= inode_permission(inode
, MAY_WRITE
);
89 goto mnt_drop_write_and_out
;
93 goto mnt_drop_write_and_out
;
95 error
= get_write_access(inode
);
97 goto mnt_drop_write_and_out
;
100 * Make sure that there are no leases. get_write_access() protects
101 * against the truncate racing with a lease-granting setlease().
103 error
= break_lease(inode
, O_WRONLY
);
105 goto put_write_and_out
;
107 error
= locks_verify_truncate(inode
, NULL
, length
);
109 error
= security_path_truncate(path
);
111 error
= do_truncate(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
)))
191 sb_start_write(inode
->i_sb
);
192 error
= locks_verify_truncate(inode
, f
.file
, length
);
194 error
= security_path_truncate(&f
.file
->f_path
);
196 error
= do_truncate(dentry
, length
, ATTR_MTIME
|ATTR_CTIME
, f
.file
);
197 sb_end_write(inode
->i_sb
);
204 SYSCALL_DEFINE2(ftruncate
, unsigned int, fd
, unsigned long, length
)
206 return do_sys_ftruncate(fd
, length
, 1);
210 COMPAT_SYSCALL_DEFINE2(ftruncate
, unsigned int, fd
, compat_ulong_t
, length
)
212 return do_sys_ftruncate(fd
, length
, 1);
216 /* LFS versions of truncate are only needed on 32 bit machines */
217 #if BITS_PER_LONG == 32
218 SYSCALL_DEFINE2(truncate64
, const char __user
*, path
, loff_t
, length
)
220 return do_sys_truncate(path
, length
);
223 SYSCALL_DEFINE2(ftruncate64
, unsigned int, fd
, loff_t
, length
)
225 return do_sys_ftruncate(fd
, length
, 0);
227 #endif /* BITS_PER_LONG == 32 */
230 int vfs_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
232 struct inode
*inode
= file_inode(file
);
235 if (offset
< 0 || len
<= 0)
238 /* Return error if mode is not supported */
239 if (mode
& ~FALLOC_FL_SUPPORTED_MASK
)
242 /* Punch hole and zero range are mutually exclusive */
243 if ((mode
& (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
)) ==
244 (FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_ZERO_RANGE
))
247 /* Punch hole must have keep size set */
248 if ((mode
& FALLOC_FL_PUNCH_HOLE
) &&
249 !(mode
& FALLOC_FL_KEEP_SIZE
))
252 /* Collapse range should only be used exclusively. */
253 if ((mode
& FALLOC_FL_COLLAPSE_RANGE
) &&
254 (mode
& ~FALLOC_FL_COLLAPSE_RANGE
))
257 /* Insert range should only be used exclusively. */
258 if ((mode
& FALLOC_FL_INSERT_RANGE
) &&
259 (mode
& ~FALLOC_FL_INSERT_RANGE
))
262 /* Unshare range should only be used with allocate mode. */
263 if ((mode
& FALLOC_FL_UNSHARE_RANGE
) &&
264 (mode
& ~(FALLOC_FL_UNSHARE_RANGE
| FALLOC_FL_KEEP_SIZE
)))
267 if (!(file
->f_mode
& FMODE_WRITE
))
271 * We can only allow pure fallocate on append only files
273 if ((mode
& ~FALLOC_FL_KEEP_SIZE
) && IS_APPEND(inode
))
276 if (IS_IMMUTABLE(inode
))
280 * We cannot allow any fallocate operation on an active swapfile
282 if (IS_SWAPFILE(inode
))
286 * Revalidate the write permissions, in case security policy has
287 * changed since the files were opened.
289 ret
= security_file_permission(file
, MAY_WRITE
);
293 if (S_ISFIFO(inode
->i_mode
))
296 if (S_ISDIR(inode
->i_mode
))
299 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
302 /* Check for wrap through zero too */
303 if (((offset
+ len
) > inode
->i_sb
->s_maxbytes
) || ((offset
+ len
) < 0))
306 if (!file
->f_op
->fallocate
)
309 file_start_write(file
);
310 ret
= file
->f_op
->fallocate(file
, mode
, offset
, len
);
313 * Create inotify and fanotify events.
315 * To keep the logic simple always create events if fallocate succeeds.
316 * This implies that events are even created if the file size remains
317 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
320 fsnotify_modify(file
);
322 file_end_write(file
);
325 EXPORT_SYMBOL_GPL(vfs_fallocate
);
327 int ksys_fallocate(int fd
, int mode
, loff_t offset
, loff_t len
)
329 struct fd f
= fdget(fd
);
333 error
= vfs_fallocate(f
.file
, mode
, offset
, len
);
339 SYSCALL_DEFINE4(fallocate
, int, fd
, int, mode
, loff_t
, offset
, loff_t
, len
)
341 return ksys_fallocate(fd
, mode
, offset
, len
);
345 * access() needs to use the real uid/gid, not the effective uid/gid.
346 * We do this by temporarily clearing all FS-related capabilities and
347 * switching the fsuid/fsgid around to the real ones.
349 static const struct cred
*access_override_creds(void)
351 const struct cred
*old_cred
;
352 struct cred
*override_cred
;
354 override_cred
= prepare_creds();
358 override_cred
->fsuid
= override_cred
->uid
;
359 override_cred
->fsgid
= override_cred
->gid
;
361 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
362 /* Clear the capabilities if we switch to a non-root user */
363 kuid_t root_uid
= make_kuid(override_cred
->user_ns
, 0);
364 if (!uid_eq(override_cred
->uid
, root_uid
))
365 cap_clear(override_cred
->cap_effective
);
367 override_cred
->cap_effective
=
368 override_cred
->cap_permitted
;
372 * The new set of credentials can *only* be used in
373 * task-synchronous circumstances, and does not need
374 * RCU freeing, unless somebody then takes a separate
377 * NOTE! This is _only_ true because this credential
378 * is used purely for override_creds() that installs
379 * it as the subjective cred. Other threads will be
380 * accessing ->real_cred, not the subjective cred.
382 * If somebody _does_ make a copy of this (using the
383 * 'get_current_cred()' function), that will clear the
384 * non_rcu field, because now that other user may be
385 * expecting RCU freeing. But normal thread-synchronous
386 * cred accesses will keep things non-RCY.
388 override_cred
->non_rcu
= 1;
390 old_cred
= override_creds(override_cred
);
392 /* override_cred() gets its own ref */
393 put_cred(override_cred
);
398 static long do_faccessat(int dfd
, const char __user
*filename
, int mode
, int flags
)
403 unsigned int lookup_flags
= LOOKUP_FOLLOW
;
404 const struct cred
*old_cred
= NULL
;
406 if (mode
& ~S_IRWXO
) /* where's F_OK, X_OK, W_OK, R_OK? */
409 if (flags
& ~(AT_EACCESS
| AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
))
412 if (flags
& AT_SYMLINK_NOFOLLOW
)
413 lookup_flags
&= ~LOOKUP_FOLLOW
;
414 if (flags
& AT_EMPTY_PATH
)
415 lookup_flags
|= LOOKUP_EMPTY
;
417 if (!(flags
& AT_EACCESS
)) {
418 old_cred
= access_override_creds();
424 res
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
428 inode
= d_backing_inode(path
.dentry
);
430 if ((mode
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
432 * MAY_EXEC on regular files is denied if the fs is mounted
433 * with the "noexec" flag.
436 if (path_noexec(&path
))
437 goto out_path_release
;
440 res
= inode_permission(inode
, mode
| MAY_ACCESS
);
441 /* SuS v2 requires we report a read only fs too */
442 if (res
|| !(mode
& S_IWOTH
) || special_file(inode
->i_mode
))
443 goto out_path_release
;
445 * This is a rare case where using __mnt_is_readonly()
446 * is OK without a mnt_want/drop_write() pair. Since
447 * no actual write to the fs is performed here, we do
448 * not need to telegraph to that to anyone.
450 * By doing this, we accept that this access is
451 * inherently racy and know that the fs may change
452 * state before we even see this result.
454 if (__mnt_is_readonly(path
.mnt
))
459 if (retry_estale(res
, lookup_flags
)) {
460 lookup_flags
|= LOOKUP_REVAL
;
465 revert_creds(old_cred
);
470 SYSCALL_DEFINE3(faccessat
, int, dfd
, const char __user
*, filename
, int, mode
)
472 return do_faccessat(dfd
, filename
, mode
, 0);
475 SYSCALL_DEFINE4(faccessat2
, int, dfd
, const char __user
*, filename
, int, mode
,
478 return do_faccessat(dfd
, filename
, mode
, flags
);
481 SYSCALL_DEFINE2(access
, const char __user
*, filename
, int, mode
)
483 return do_faccessat(AT_FDCWD
, filename
, mode
, 0);
486 SYSCALL_DEFINE1(chdir
, const char __user
*, filename
)
490 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
492 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
496 error
= inode_permission(path
.dentry
->d_inode
, MAY_EXEC
| MAY_CHDIR
);
500 set_fs_pwd(current
->fs
, &path
);
504 if (retry_estale(error
, lookup_flags
)) {
505 lookup_flags
|= LOOKUP_REVAL
;
512 SYSCALL_DEFINE1(fchdir
, unsigned int, fd
)
514 struct fd f
= fdget_raw(fd
);
522 if (!d_can_lookup(f
.file
->f_path
.dentry
))
525 error
= inode_permission(file_inode(f
.file
), MAY_EXEC
| MAY_CHDIR
);
527 set_fs_pwd(current
->fs
, &f
.file
->f_path
);
534 SYSCALL_DEFINE1(chroot
, const char __user
*, filename
)
538 unsigned int lookup_flags
= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
540 error
= user_path_at(AT_FDCWD
, filename
, lookup_flags
, &path
);
544 error
= inode_permission(path
.dentry
->d_inode
, MAY_EXEC
| MAY_CHDIR
);
549 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT
))
551 error
= security_path_chroot(&path
);
555 set_fs_root(current
->fs
, &path
);
559 if (retry_estale(error
, lookup_flags
)) {
560 lookup_flags
|= LOOKUP_REVAL
;
567 int chmod_common(const struct path
*path
, umode_t mode
)
569 struct inode
*inode
= path
->dentry
->d_inode
;
570 struct inode
*delegated_inode
= NULL
;
571 struct iattr newattrs
;
574 error
= mnt_want_write(path
->mnt
);
579 error
= security_path_chmod(path
, mode
);
582 newattrs
.ia_mode
= (mode
& S_IALLUGO
) | (inode
->i_mode
& ~S_IALLUGO
);
583 newattrs
.ia_valid
= ATTR_MODE
| ATTR_CTIME
;
584 error
= notify_change(path
->dentry
, &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 inode
*inode
= path
->dentry
->d_inode
;
646 struct inode
*delegated_inode
= NULL
;
648 struct iattr newattrs
;
652 uid
= make_kuid(current_user_ns(), user
);
653 gid
= make_kgid(current_user_ns(), group
);
656 newattrs
.ia_valid
= ATTR_CTIME
;
657 if (user
!= (uid_t
) -1) {
660 newattrs
.ia_valid
|= ATTR_UID
;
661 newattrs
.ia_uid
= uid
;
663 if (group
!= (gid_t
) -1) {
666 newattrs
.ia_valid
|= ATTR_GID
;
667 newattrs
.ia_gid
= gid
;
669 if (!S_ISDIR(inode
->i_mode
))
671 ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
673 error
= security_path_chown(path
, uid
, gid
);
675 error
= notify_change(path
->dentry
, &newattrs
, &delegated_inode
);
677 if (delegated_inode
) {
678 error
= break_deleg_wait(&delegated_inode
);
685 int do_fchownat(int dfd
, const char __user
*filename
, uid_t user
, gid_t group
,
692 if ((flag
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
695 lookup_flags
= (flag
& AT_SYMLINK_NOFOLLOW
) ? 0 : LOOKUP_FOLLOW
;
696 if (flag
& AT_EMPTY_PATH
)
697 lookup_flags
|= LOOKUP_EMPTY
;
699 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
702 error
= mnt_want_write(path
.mnt
);
705 error
= chown_common(&path
, user
, group
);
706 mnt_drop_write(path
.mnt
);
709 if (retry_estale(error
, lookup_flags
)) {
710 lookup_flags
|= LOOKUP_REVAL
;
717 SYSCALL_DEFINE5(fchownat
, int, dfd
, const char __user
*, filename
, uid_t
, user
,
718 gid_t
, group
, int, flag
)
720 return do_fchownat(dfd
, filename
, user
, group
, flag
);
723 SYSCALL_DEFINE3(chown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
725 return do_fchownat(AT_FDCWD
, filename
, user
, group
, 0);
728 SYSCALL_DEFINE3(lchown
, const char __user
*, filename
, uid_t
, user
, gid_t
, group
)
730 return do_fchownat(AT_FDCWD
, filename
, user
, group
,
731 AT_SYMLINK_NOFOLLOW
);
734 int vfs_fchown(struct file
*file
, uid_t user
, gid_t group
)
738 error
= mnt_want_write_file(file
);
742 error
= chown_common(&file
->f_path
, user
, group
);
743 mnt_drop_write_file(file
);
747 int ksys_fchown(unsigned int fd
, uid_t user
, gid_t group
)
749 struct fd f
= fdget(fd
);
753 error
= vfs_fchown(f
.file
, user
, group
);
759 SYSCALL_DEFINE3(fchown
, unsigned int, fd
, uid_t
, user
, gid_t
, group
)
761 return ksys_fchown(fd
, user
, group
);
764 static int do_dentry_open(struct file
*f
,
766 int (*open
)(struct inode
*, struct file
*))
768 static const struct file_operations empty_fops
= {};
771 path_get(&f
->f_path
);
773 f
->f_mapping
= inode
->i_mapping
;
774 f
->f_wb_err
= filemap_sample_wb_err(f
->f_mapping
);
775 f
->f_sb_err
= file_sample_sb_err(f
);
777 if (unlikely(f
->f_flags
& O_PATH
)) {
778 f
->f_mode
= FMODE_PATH
| FMODE_OPENED
;
779 f
->f_op
= &empty_fops
;
783 if (f
->f_mode
& FMODE_WRITE
&& !special_file(inode
->i_mode
)) {
784 error
= get_write_access(inode
);
787 error
= __mnt_want_write(f
->f_path
.mnt
);
788 if (unlikely(error
)) {
789 put_write_access(inode
);
792 f
->f_mode
|= FMODE_WRITER
;
795 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
796 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
))
797 f
->f_mode
|= FMODE_ATOMIC_POS
;
799 f
->f_op
= fops_get(inode
->i_fop
);
800 if (WARN_ON(!f
->f_op
)) {
805 error
= security_file_open(f
);
809 error
= break_lease(locks_inode(f
), f
->f_flags
);
813 /* normally all 3 are set; ->open() can clear them if needed */
814 f
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
816 open
= f
->f_op
->open
;
818 error
= open(inode
, f
);
822 f
->f_mode
|= FMODE_OPENED
;
823 if ((f
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) == FMODE_READ
)
824 i_readcount_inc(inode
);
825 if ((f
->f_mode
& FMODE_READ
) &&
826 likely(f
->f_op
->read
|| f
->f_op
->read_iter
))
827 f
->f_mode
|= FMODE_CAN_READ
;
828 if ((f
->f_mode
& FMODE_WRITE
) &&
829 likely(f
->f_op
->write
|| f
->f_op
->write_iter
))
830 f
->f_mode
|= FMODE_CAN_WRITE
;
832 f
->f_write_hint
= WRITE_LIFE_NOT_SET
;
833 f
->f_flags
&= ~(O_CREAT
| O_EXCL
| O_NOCTTY
| O_TRUNC
);
835 file_ra_state_init(&f
->f_ra
, f
->f_mapping
->host
->i_mapping
);
837 /* NB: we're sure to have correct a_ops only after f_op->open */
838 if (f
->f_flags
& O_DIRECT
) {
839 if (!f
->f_mapping
->a_ops
|| !f
->f_mapping
->a_ops
->direct_IO
)
844 * XXX: Huge page cache doesn't support writing yet. Drop all page
845 * cache for this file before processing writes.
847 if ((f
->f_mode
& FMODE_WRITE
) && filemap_nr_thps(inode
->i_mapping
))
848 truncate_pagecache(inode
, 0);
853 if (WARN_ON_ONCE(error
> 0))
856 if (f
->f_mode
& FMODE_WRITER
) {
857 put_write_access(inode
);
858 __mnt_drop_write(f
->f_path
.mnt
);
861 path_put(&f
->f_path
);
862 f
->f_path
.mnt
= NULL
;
863 f
->f_path
.dentry
= NULL
;
869 * finish_open - finish opening a file
870 * @file: file pointer
871 * @dentry: pointer to dentry
872 * @open: open callback
873 * @opened: state of open
875 * This can be used to finish opening a file passed to i_op->atomic_open().
877 * If the open callback is set to NULL, then the standard f_op->open()
878 * filesystem callback is substituted.
880 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
881 * the return value of d_splice_alias(), then the caller needs to perform dput()
882 * on it after finish_open().
884 * Returns zero on success or -errno if the open failed.
886 int finish_open(struct file
*file
, struct dentry
*dentry
,
887 int (*open
)(struct inode
*, struct file
*))
889 BUG_ON(file
->f_mode
& FMODE_OPENED
); /* once it's opened, it's opened */
891 file
->f_path
.dentry
= dentry
;
892 return do_dentry_open(file
, d_backing_inode(dentry
), open
);
894 EXPORT_SYMBOL(finish_open
);
897 * finish_no_open - finish ->atomic_open() without opening the file
899 * @file: file pointer
900 * @dentry: dentry or NULL (as returned from ->lookup())
902 * This can be used to set the result of a successful lookup in ->atomic_open().
904 * NB: unlike finish_open() this function does consume the dentry reference and
905 * the caller need not dput() it.
907 * Returns "0" which must be the return value of ->atomic_open() after having
908 * called this function.
910 int finish_no_open(struct file
*file
, struct dentry
*dentry
)
912 file
->f_path
.dentry
= dentry
;
915 EXPORT_SYMBOL(finish_no_open
);
917 char *file_path(struct file
*filp
, char *buf
, int buflen
)
919 return d_path(&filp
->f_path
, buf
, buflen
);
921 EXPORT_SYMBOL(file_path
);
924 * vfs_open - open the file at the given path
925 * @path: path to open
926 * @file: newly allocated file with f_flag initialized
927 * @cred: credentials to use
929 int vfs_open(const struct path
*path
, struct file
*file
)
931 file
->f_path
= *path
;
932 return do_dentry_open(file
, d_backing_inode(path
->dentry
), NULL
);
935 struct file
*dentry_open(const struct path
*path
, int flags
,
936 const struct cred
*cred
)
941 validate_creds(cred
);
943 /* We must always pass in a valid mount pointer. */
946 f
= alloc_empty_file(flags
, cred
);
948 error
= vfs_open(path
, f
);
956 EXPORT_SYMBOL(dentry_open
);
958 struct file
*open_with_fake_path(const struct path
*path
, int flags
,
959 struct inode
*inode
, const struct cred
*cred
)
961 struct file
*f
= alloc_empty_file_noaccount(flags
, cred
);
966 error
= do_dentry_open(f
, inode
, NULL
);
974 EXPORT_SYMBOL(open_with_fake_path
);
976 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
977 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
979 inline struct open_how
build_open_how(int flags
, umode_t mode
)
981 struct open_how how
= {
982 .flags
= flags
& VALID_OPEN_FLAGS
,
983 .mode
= mode
& S_IALLUGO
,
986 /* O_PATH beats everything else. */
987 if (how
.flags
& O_PATH
)
988 how
.flags
&= O_PATH_FLAGS
;
989 /* Modes should only be set for create-like flags. */
990 if (!WILL_CREATE(how
.flags
))
995 inline int build_open_flags(const struct open_how
*how
, struct open_flags
*op
)
997 int flags
= how
->flags
;
998 int lookup_flags
= 0;
999 int acc_mode
= ACC_MODE(flags
);
1001 /* Must never be set by userspace */
1002 flags
&= ~(FMODE_NONOTIFY
| O_CLOEXEC
);
1005 * Older syscalls implicitly clear all of the invalid flags or argument
1006 * values before calling build_open_flags(), but openat2(2) checks all
1009 if (flags
& ~VALID_OPEN_FLAGS
)
1011 if (how
->resolve
& ~VALID_RESOLVE_FLAGS
)
1014 /* Scoping flags are mutually exclusive. */
1015 if ((how
->resolve
& RESOLVE_BENEATH
) && (how
->resolve
& RESOLVE_IN_ROOT
))
1018 /* Deal with the mode. */
1019 if (WILL_CREATE(flags
)) {
1020 if (how
->mode
& ~S_IALLUGO
)
1022 op
->mode
= how
->mode
| S_IFREG
;
1030 * In order to ensure programs get explicit errors when trying to use
1031 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1032 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1033 * have to require userspace to explicitly set it.
1035 if (flags
& __O_TMPFILE
) {
1036 if ((flags
& O_TMPFILE_MASK
) != O_TMPFILE
)
1038 if (!(acc_mode
& MAY_WRITE
))
1041 if (flags
& O_PATH
) {
1042 /* O_PATH only permits certain other flags to be set. */
1043 if (flags
& ~O_PATH_FLAGS
)
1049 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1050 * check for O_DSYNC if the need any syncing at all we enforce it's
1051 * always set instead of having to deal with possibly weird behaviour
1052 * for malicious applications setting only __O_SYNC.
1054 if (flags
& __O_SYNC
)
1057 op
->open_flag
= flags
;
1059 /* O_TRUNC implies we need access checks for write permissions */
1060 if (flags
& O_TRUNC
)
1061 acc_mode
|= MAY_WRITE
;
1063 /* Allow the LSM permission hook to distinguish append
1064 access from general write access. */
1065 if (flags
& O_APPEND
)
1066 acc_mode
|= MAY_APPEND
;
1068 op
->acc_mode
= acc_mode
;
1070 op
->intent
= flags
& O_PATH
? 0 : LOOKUP_OPEN
;
1072 if (flags
& O_CREAT
) {
1073 op
->intent
|= LOOKUP_CREATE
;
1074 if (flags
& O_EXCL
) {
1075 op
->intent
|= LOOKUP_EXCL
;
1076 flags
|= O_NOFOLLOW
;
1080 if (flags
& O_DIRECTORY
)
1081 lookup_flags
|= LOOKUP_DIRECTORY
;
1082 if (!(flags
& O_NOFOLLOW
))
1083 lookup_flags
|= LOOKUP_FOLLOW
;
1085 if (how
->resolve
& RESOLVE_NO_XDEV
)
1086 lookup_flags
|= LOOKUP_NO_XDEV
;
1087 if (how
->resolve
& RESOLVE_NO_MAGICLINKS
)
1088 lookup_flags
|= LOOKUP_NO_MAGICLINKS
;
1089 if (how
->resolve
& RESOLVE_NO_SYMLINKS
)
1090 lookup_flags
|= LOOKUP_NO_SYMLINKS
;
1091 if (how
->resolve
& RESOLVE_BENEATH
)
1092 lookup_flags
|= LOOKUP_BENEATH
;
1093 if (how
->resolve
& RESOLVE_IN_ROOT
)
1094 lookup_flags
|= LOOKUP_IN_ROOT
;
1096 op
->lookup_flags
= lookup_flags
;
1101 * file_open_name - open file and return file pointer
1103 * @name: struct filename containing path to open
1104 * @flags: open flags as per the open(2) second argument
1105 * @mode: mode for the new file if O_CREAT is set, else ignored
1107 * This is the helper to open a file from kernelspace if you really
1108 * have to. But in generally you should not do this, so please move
1109 * along, nothing to see here..
1111 struct file
*file_open_name(struct filename
*name
, int flags
, umode_t mode
)
1113 struct open_flags op
;
1114 struct open_how how
= build_open_how(flags
, mode
);
1115 int err
= build_open_flags(&how
, &op
);
1117 return ERR_PTR(err
);
1118 return do_filp_open(AT_FDCWD
, name
, &op
);
1122 * filp_open - open file and return file pointer
1124 * @filename: path to open
1125 * @flags: open flags as per the open(2) second argument
1126 * @mode: mode for the new file if O_CREAT is set, else ignored
1128 * This is the helper to open a file from kernelspace if you really
1129 * have to. But in generally you should not do this, so please move
1130 * along, nothing to see here..
1132 struct file
*filp_open(const char *filename
, int flags
, umode_t mode
)
1134 struct filename
*name
= getname_kernel(filename
);
1135 struct file
*file
= ERR_CAST(name
);
1137 if (!IS_ERR(name
)) {
1138 file
= file_open_name(name
, flags
, mode
);
1143 EXPORT_SYMBOL(filp_open
);
1145 struct file
*file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
1146 const char *filename
, int flags
, umode_t mode
)
1148 struct open_flags op
;
1149 struct open_how how
= build_open_how(flags
, mode
);
1150 int err
= build_open_flags(&how
, &op
);
1152 return ERR_PTR(err
);
1153 return do_file_open_root(dentry
, mnt
, filename
, &op
);
1155 EXPORT_SYMBOL(file_open_root
);
1157 static long do_sys_openat2(int dfd
, const char __user
*filename
,
1158 struct open_how
*how
)
1160 struct open_flags op
;
1161 int fd
= build_open_flags(how
, &op
);
1162 struct filename
*tmp
;
1167 tmp
= getname(filename
);
1169 return PTR_ERR(tmp
);
1171 fd
= get_unused_fd_flags(how
->flags
);
1173 struct file
*f
= do_filp_open(dfd
, tmp
, &op
);
1186 long do_sys_open(int dfd
, const char __user
*filename
, int flags
, umode_t mode
)
1188 struct open_how how
= build_open_how(flags
, mode
);
1189 return do_sys_openat2(dfd
, filename
, &how
);
1193 SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1195 if (force_o_largefile())
1196 flags
|= O_LARGEFILE
;
1197 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1200 SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
,
1203 if (force_o_largefile())
1204 flags
|= O_LARGEFILE
;
1205 return do_sys_open(dfd
, filename
, flags
, mode
);
1208 SYSCALL_DEFINE4(openat2
, int, dfd
, const char __user
*, filename
,
1209 struct open_how __user
*, how
, size_t, usize
)
1212 struct open_how tmp
;
1214 BUILD_BUG_ON(sizeof(struct open_how
) < OPEN_HOW_SIZE_VER0
);
1215 BUILD_BUG_ON(sizeof(struct open_how
) != OPEN_HOW_SIZE_LATEST
);
1217 if (unlikely(usize
< OPEN_HOW_SIZE_VER0
))
1220 err
= copy_struct_from_user(&tmp
, sizeof(tmp
), how
, usize
);
1224 /* O_LARGEFILE is only allowed for non-O_PATH. */
1225 if (!(tmp
.flags
& O_PATH
) && force_o_largefile())
1226 tmp
.flags
|= O_LARGEFILE
;
1228 return do_sys_openat2(dfd
, filename
, &tmp
);
1231 #ifdef CONFIG_COMPAT
1233 * Exactly like sys_open(), except that it doesn't set the
1236 COMPAT_SYSCALL_DEFINE3(open
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1238 return do_sys_open(AT_FDCWD
, filename
, flags
, mode
);
1242 * Exactly like sys_openat(), except that it doesn't set the
1245 COMPAT_SYSCALL_DEFINE4(openat
, int, dfd
, const char __user
*, filename
, int, flags
, umode_t
, mode
)
1247 return do_sys_open(dfd
, filename
, flags
, mode
);
1254 * For backward compatibility? Maybe this should be moved
1255 * into arch/i386 instead?
1257 SYSCALL_DEFINE2(creat
, const char __user
*, pathname
, umode_t
, mode
)
1259 int flags
= O_CREAT
| O_WRONLY
| O_TRUNC
;
1261 if (force_o_largefile())
1262 flags
|= O_LARGEFILE
;
1263 return do_sys_open(AT_FDCWD
, pathname
, flags
, mode
);
1268 * "id" is the POSIX thread ID. We use the
1269 * files pointer for this..
1271 int filp_close(struct file
*filp
, fl_owner_t id
)
1275 if (!file_count(filp
)) {
1276 printk(KERN_ERR
"VFS: Close: file count is 0\n");
1280 if (filp
->f_op
->flush
)
1281 retval
= filp
->f_op
->flush(filp
, id
);
1283 if (likely(!(filp
->f_mode
& FMODE_PATH
))) {
1284 dnotify_flush(filp
, id
);
1285 locks_remove_posix(filp
, id
);
1291 EXPORT_SYMBOL(filp_close
);
1294 * Careful here! We test whether the file pointer is NULL before
1295 * releasing the fd. This ensures that one clone task can't release
1296 * an fd while another clone is opening it.
1298 SYSCALL_DEFINE1(close
, unsigned int, fd
)
1300 int retval
= close_fd(fd
);
1302 /* can't restart close syscall because file table entry was cleared */
1303 if (unlikely(retval
== -ERESTARTSYS
||
1304 retval
== -ERESTARTNOINTR
||
1305 retval
== -ERESTARTNOHAND
||
1306 retval
== -ERESTART_RESTARTBLOCK
))
1313 * close_range() - Close all file descriptors in a given range.
1315 * @fd: starting file descriptor to close
1316 * @max_fd: last file descriptor to close
1317 * @flags: reserved for future extensions
1319 * This closes a range of file descriptors. All file descriptors
1320 * from @fd up to and including @max_fd are closed.
1321 * Currently, errors to close a given file descriptor are ignored.
1323 SYSCALL_DEFINE3(close_range
, unsigned int, fd
, unsigned int, max_fd
,
1324 unsigned int, flags
)
1326 return __close_range(fd
, max_fd
, flags
);
1330 * This routine simulates a hangup on the tty, to arrange that users
1331 * are given clean terminals at login time.
1333 SYSCALL_DEFINE0(vhangup
)
1335 if (capable(CAP_SYS_TTY_CONFIG
)) {
1343 * Called when an inode is about to be open.
1344 * We use this to disallow opening large files on 32bit systems if
1345 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1346 * on this flag in sys_open.
1348 int generic_file_open(struct inode
* inode
, struct file
* filp
)
1350 if (!(filp
->f_flags
& O_LARGEFILE
) && i_size_read(inode
) > MAX_NON_LFS
)
1355 EXPORT_SYMBOL(generic_file_open
);
1358 * This is used by subsystems that don't want seekable
1359 * file descriptors. The function is not supposed to ever fail, the only
1360 * reason it returns an 'int' and not 'void' is so that it can be plugged
1361 * directly into file_operations structure.
1363 int nonseekable_open(struct inode
*inode
, struct file
*filp
)
1365 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
);
1369 EXPORT_SYMBOL(nonseekable_open
);
1372 * stream_open is used by subsystems that want stream-like file descriptors.
1373 * Such file descriptors are not seekable and don't have notion of position
1374 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1375 * Contrary to file descriptors of other regular files, .read() and .write()
1376 * can run simultaneously.
1378 * stream_open never fails and is marked to return int so that it could be
1379 * directly used as file_operations.open .
1381 int stream_open(struct inode
*inode
, struct file
*filp
)
1383 filp
->f_mode
&= ~(FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
| FMODE_ATOMIC_POS
);
1384 filp
->f_mode
|= FMODE_STREAM
;
1388 EXPORT_SYMBOL(stream_open
);