4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <asm/uaccess.h>
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user
*filename
, char *page
)
121 unsigned long len
= PATH_MAX
;
123 if (!segment_eq(get_fs(), KERNEL_DS
)) {
124 if ((unsigned long) filename
>= TASK_SIZE
)
126 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
127 len
= TASK_SIZE
- (unsigned long) filename
;
130 retval
= strncpy_from_user(page
, filename
, len
);
134 return -ENAMETOOLONG
;
140 char * getname(const char __user
* filename
)
144 result
= ERR_PTR(-ENOMEM
);
147 int retval
= do_getname(filename
, tmp
);
152 result
= ERR_PTR(retval
);
155 audit_getname(result
);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name
)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname
);
171 * This does basic POSIX ACL permission checking
173 static int acl_permission_check(struct inode
*inode
, int mask
,
174 int (*check_acl
)(struct inode
*inode
, int mask
))
176 umode_t mode
= inode
->i_mode
;
178 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
180 if (current_fsuid() == inode
->i_uid
)
183 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
184 int error
= check_acl(inode
, mask
);
185 if (error
!= -EAGAIN
)
189 if (in_group_p(inode
->i_gid
))
194 * If the DACs are ok we don't need any capability check.
196 if ((mask
& ~mode
) == 0)
202 * generic_permission - check for access rights on a Posix-like filesystem
203 * @inode: inode to check access rights for
204 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
205 * @check_acl: optional callback to check for Posix ACLs
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things..
212 int generic_permission(struct inode
*inode
, int mask
,
213 int (*check_acl
)(struct inode
*inode
, int mask
))
218 * Do the basic POSIX ACL permission checks.
220 ret
= acl_permission_check(inode
, mask
, check_acl
);
225 * Read/write DACs are always overridable.
226 * Executable DACs are overridable if at least one exec bit is set.
228 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
229 if (capable(CAP_DAC_OVERRIDE
))
233 * Searching includes executable on directories, else just read.
235 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
236 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
237 if (capable(CAP_DAC_READ_SEARCH
))
244 * inode_permission - check for access rights to a given inode
245 * @inode: inode to check permission on
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
248 * Used to check for read/write/execute permissions on an inode.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 int inode_permission(struct inode
*inode
, int mask
)
257 if (mask
& MAY_WRITE
) {
258 umode_t mode
= inode
->i_mode
;
261 * Nobody gets write access to a read-only fs.
263 if (IS_RDONLY(inode
) &&
264 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
268 * Nobody gets write access to an immutable file.
270 if (IS_IMMUTABLE(inode
))
274 if (inode
->i_op
->permission
)
275 retval
= inode
->i_op
->permission(inode
, mask
);
277 retval
= generic_permission(inode
, mask
, inode
->i_op
->check_acl
);
282 retval
= devcgroup_inode_permission(inode
, mask
);
286 return security_inode_permission(inode
,
287 mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
|MAY_APPEND
));
291 * file_permission - check for additional access rights to a given file
292 * @file: file to check access rights for
293 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
295 * Used to check for read/write/execute permissions on an already opened
299 * Do not use this function in new code. All access checks should
300 * be done using inode_permission().
302 int file_permission(struct file
*file
, int mask
)
304 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
308 * get_write_access() gets write permission for a file.
309 * put_write_access() releases this write permission.
310 * This is used for regular files.
311 * We cannot support write (and maybe mmap read-write shared) accesses and
312 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
313 * can have the following values:
314 * 0: no writers, no VM_DENYWRITE mappings
315 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
316 * > 0: (i_writecount) users are writing to the file.
318 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
319 * except for the cases where we don't hold i_writecount yet. Then we need to
320 * use {get,deny}_write_access() - these functions check the sign and refuse
321 * to do the change if sign is wrong. Exclusion between them is provided by
322 * the inode->i_lock spinlock.
325 int get_write_access(struct inode
* inode
)
327 spin_lock(&inode
->i_lock
);
328 if (atomic_read(&inode
->i_writecount
) < 0) {
329 spin_unlock(&inode
->i_lock
);
332 atomic_inc(&inode
->i_writecount
);
333 spin_unlock(&inode
->i_lock
);
338 int deny_write_access(struct file
* file
)
340 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
342 spin_lock(&inode
->i_lock
);
343 if (atomic_read(&inode
->i_writecount
) > 0) {
344 spin_unlock(&inode
->i_lock
);
347 atomic_dec(&inode
->i_writecount
);
348 spin_unlock(&inode
->i_lock
);
354 * path_get - get a reference to a path
355 * @path: path to get the reference to
357 * Given a path increment the reference count to the dentry and the vfsmount.
359 void path_get(struct path
*path
)
364 EXPORT_SYMBOL(path_get
);
367 * path_put - put a reference to a path
368 * @path: path to put the reference to
370 * Given a path decrement the reference count to the dentry and the vfsmount.
372 void path_put(struct path
*path
)
377 EXPORT_SYMBOL(path_put
);
380 * release_open_intent - free up open intent resources
381 * @nd: pointer to nameidata
383 void release_open_intent(struct nameidata
*nd
)
385 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
386 put_filp(nd
->intent
.open
.file
);
388 fput(nd
->intent
.open
.file
);
391 static inline struct dentry
*
392 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
394 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
395 if (unlikely(status
<= 0)) {
397 * The dentry failed validation.
398 * If d_revalidate returned 0 attempt to invalidate
399 * the dentry otherwise d_revalidate is asking us
400 * to return a fail status.
403 if (!d_invalidate(dentry
)) {
409 dentry
= ERR_PTR(status
);
416 * force_reval_path - force revalidation of a dentry
418 * In some situations the path walking code will trust dentries without
419 * revalidating them. This causes problems for filesystems that depend on
420 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
421 * (which indicates that it's possible for the dentry to go stale), force
422 * a d_revalidate call before proceeding.
424 * Returns 0 if the revalidation was successful. If the revalidation fails,
425 * either return the error returned by d_revalidate or -ESTALE if the
426 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
427 * invalidate the dentry. It's up to the caller to handle putting references
428 * to the path if necessary.
431 force_reval_path(struct path
*path
, struct nameidata
*nd
)
434 struct dentry
*dentry
= path
->dentry
;
437 * only check on filesystems where it's possible for the dentry to
438 * become stale. It's assumed that if this flag is set then the
439 * d_revalidate op will also be defined.
441 if (!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
))
444 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
449 d_invalidate(dentry
);
456 * Short-cut version of permission(), for calling on directories
457 * during pathname resolution. Combines parts of permission()
458 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
460 * If appropriate, check DAC only. If not appropriate, or
461 * short-cut DAC fails, then call ->permission() to do more
462 * complete permission check.
464 static int exec_permission(struct inode
*inode
)
468 if (inode
->i_op
->permission
) {
469 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
);
474 ret
= acl_permission_check(inode
, MAY_EXEC
, inode
->i_op
->check_acl
);
478 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
483 return security_inode_permission(inode
, MAY_EXEC
);
486 static __always_inline
void set_root(struct nameidata
*nd
)
489 struct fs_struct
*fs
= current
->fs
;
490 read_lock(&fs
->lock
);
493 read_unlock(&fs
->lock
);
497 static int link_path_walk(const char *, struct nameidata
*);
499 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
513 res
= link_path_walk(link
, nd
);
514 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
517 * If it is an iterative symlinks resolution in open_namei() we
518 * have to copy the last component. And all that crap because of
519 * bloody create() on broken symlinks. Furrfu...
522 if (unlikely(!name
)) {
526 strcpy(name
, nd
->last
.name
);
527 nd
->last
.name
= name
;
531 return PTR_ERR(link
);
534 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
537 if (path
->mnt
!= nd
->path
.mnt
)
541 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
543 dput(nd
->path
.dentry
);
544 if (nd
->path
.mnt
!= path
->mnt
)
545 mntput(nd
->path
.mnt
);
546 nd
->path
.mnt
= path
->mnt
;
547 nd
->path
.dentry
= path
->dentry
;
550 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
554 struct dentry
*dentry
= path
->dentry
;
556 touch_atime(path
->mnt
, dentry
);
557 nd_set_link(nd
, NULL
);
559 if (path
->mnt
!= nd
->path
.mnt
) {
560 path_to_nameidata(path
, nd
);
564 nd
->last_type
= LAST_BIND
;
565 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
566 error
= PTR_ERR(cookie
);
567 if (!IS_ERR(cookie
)) {
568 char *s
= nd_get_link(nd
);
571 error
= __vfs_follow_link(nd
, s
);
572 else if (nd
->last_type
== LAST_BIND
) {
573 error
= force_reval_path(&nd
->path
, nd
);
577 if (dentry
->d_inode
->i_op
->put_link
)
578 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
584 * This limits recursive symlink follows to 8, while
585 * limiting consecutive symlinks to 40.
587 * Without that kind of total limit, nasty chains of consecutive
588 * symlinks can cause almost arbitrarily long lookups.
590 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
593 if (current
->link_count
>= MAX_NESTED_LINKS
)
595 if (current
->total_link_count
>= 40)
597 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
599 err
= security_inode_follow_link(path
->dentry
, nd
);
602 current
->link_count
++;
603 current
->total_link_count
++;
605 err
= __do_follow_link(path
, nd
);
607 current
->link_count
--;
611 path_put_conditional(path
, nd
);
616 int follow_up(struct path
*path
)
618 struct vfsmount
*parent
;
619 struct dentry
*mountpoint
;
620 spin_lock(&vfsmount_lock
);
621 parent
= path
->mnt
->mnt_parent
;
622 if (parent
== path
->mnt
) {
623 spin_unlock(&vfsmount_lock
);
627 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
628 spin_unlock(&vfsmount_lock
);
630 path
->dentry
= mountpoint
;
636 /* no need for dcache_lock, as serialization is taken care in
639 static int __follow_mount(struct path
*path
)
642 while (d_mountpoint(path
->dentry
)) {
643 struct vfsmount
*mounted
= lookup_mnt(path
);
650 path
->dentry
= dget(mounted
->mnt_root
);
656 static void follow_mount(struct path
*path
)
658 while (d_mountpoint(path
->dentry
)) {
659 struct vfsmount
*mounted
= lookup_mnt(path
);
665 path
->dentry
= dget(mounted
->mnt_root
);
669 /* no need for dcache_lock, as serialization is taken care in
672 int follow_down(struct path
*path
)
674 struct vfsmount
*mounted
;
676 mounted
= lookup_mnt(path
);
681 path
->dentry
= dget(mounted
->mnt_root
);
687 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
692 struct dentry
*old
= nd
->path
.dentry
;
694 if (nd
->path
.dentry
== nd
->root
.dentry
&&
695 nd
->path
.mnt
== nd
->root
.mnt
) {
698 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
699 /* rare case of legitimate dget_parent()... */
700 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
704 if (!follow_up(&nd
->path
))
707 follow_mount(&nd
->path
);
711 * It's more convoluted than I'd like it to be, but... it's still fairly
712 * small and for now I'd prefer to have fast path as straight as possible.
713 * It _is_ time-critical.
715 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
718 struct vfsmount
*mnt
= nd
->path
.mnt
;
719 struct dentry
*dentry
, *parent
;
722 * See if the low-level filesystem might want
723 * to use its own hash..
725 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
726 int err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
, name
);
731 dentry
= __d_lookup(nd
->path
.dentry
, name
);
734 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
735 goto need_revalidate
;
738 path
->dentry
= dentry
;
739 __follow_mount(path
);
743 parent
= nd
->path
.dentry
;
744 dir
= parent
->d_inode
;
746 mutex_lock(&dir
->i_mutex
);
748 * First re-do the cached lookup just in case it was created
749 * while we waited for the directory semaphore..
751 * FIXME! This could use version numbering or similar to
752 * avoid unnecessary cache lookups.
754 * The "dcache_lock" is purely to protect the RCU list walker
755 * from concurrent renames at this point (we mustn't get false
756 * negatives from the RCU list walk here, unlike the optimistic
759 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
761 dentry
= d_lookup(parent
, name
);
765 /* Don't create child dentry for a dead directory. */
766 dentry
= ERR_PTR(-ENOENT
);
770 new = d_alloc(parent
, name
);
771 dentry
= ERR_PTR(-ENOMEM
);
773 dentry
= dir
->i_op
->lookup(dir
, new, nd
);
780 mutex_unlock(&dir
->i_mutex
);
787 * Uhhuh! Nasty case: the cache was re-populated while
788 * we waited on the semaphore. Need to revalidate.
790 mutex_unlock(&dir
->i_mutex
);
791 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
) {
792 dentry
= do_revalidate(dentry
, nd
);
794 dentry
= ERR_PTR(-ENOENT
);
801 dentry
= do_revalidate(dentry
, nd
);
809 return PTR_ERR(dentry
);
813 * This is a temporary kludge to deal with "automount" symlinks; proper
814 * solution is to trigger them on follow_mount(), so that do_lookup()
815 * would DTRT. To be killed before 2.6.34-final.
817 static inline int follow_on_final(struct inode
*inode
, unsigned lookup_flags
)
819 return inode
&& unlikely(inode
->i_op
->follow_link
) &&
820 ((lookup_flags
& LOOKUP_FOLLOW
) || S_ISDIR(inode
->i_mode
));
825 * This is the basic name resolution function, turning a pathname into
826 * the final dentry. We expect 'base' to be positive and a directory.
828 * Returns 0 and nd will have valid dentry and mnt on success.
829 * Returns error and drops reference to input namei data on failure.
831 static int link_path_walk(const char *name
, struct nameidata
*nd
)
836 unsigned int lookup_flags
= nd
->flags
;
843 inode
= nd
->path
.dentry
->d_inode
;
845 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
847 /* At this point we know we have a real path component. */
853 nd
->flags
|= LOOKUP_CONTINUE
;
854 err
= exec_permission(inode
);
859 c
= *(const unsigned char *)name
;
861 hash
= init_name_hash();
864 hash
= partial_name_hash(c
, hash
);
865 c
= *(const unsigned char *)name
;
866 } while (c
&& (c
!= '/'));
867 this.len
= name
- (const char *) this.name
;
868 this.hash
= end_name_hash(hash
);
870 /* remove trailing slashes? */
873 while (*++name
== '/');
875 goto last_with_slashes
;
878 * "." and ".." are special - ".." especially so because it has
879 * to be able to know about the current root directory and
880 * parent relationships.
882 if (this.name
[0] == '.') switch (this.len
) {
886 if (this.name
[1] != '.')
889 inode
= nd
->path
.dentry
->d_inode
;
894 /* This does the actual lookups.. */
895 err
= do_lookup(nd
, &this, &next
);
900 inode
= next
.dentry
->d_inode
;
904 if (inode
->i_op
->follow_link
) {
905 err
= do_follow_link(&next
, nd
);
909 inode
= nd
->path
.dentry
->d_inode
;
913 path_to_nameidata(&next
, nd
);
915 if (!inode
->i_op
->lookup
)
918 /* here ends the main loop */
921 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
923 /* Clear LOOKUP_CONTINUE iff it was previously unset */
924 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
925 if (lookup_flags
& LOOKUP_PARENT
)
927 if (this.name
[0] == '.') switch (this.len
) {
931 if (this.name
[1] != '.')
934 inode
= nd
->path
.dentry
->d_inode
;
939 err
= do_lookup(nd
, &this, &next
);
942 inode
= next
.dentry
->d_inode
;
943 if (follow_on_final(inode
, lookup_flags
)) {
944 err
= do_follow_link(&next
, nd
);
947 inode
= nd
->path
.dentry
->d_inode
;
949 path_to_nameidata(&next
, nd
);
953 if (lookup_flags
& LOOKUP_DIRECTORY
) {
955 if (!inode
->i_op
->lookup
)
961 nd
->last_type
= LAST_NORM
;
962 if (this.name
[0] != '.')
965 nd
->last_type
= LAST_DOT
;
966 else if (this.len
== 2 && this.name
[1] == '.')
967 nd
->last_type
= LAST_DOTDOT
;
972 * We bypassed the ordinary revalidation routines.
973 * We may need to check the cached dentry for staleness.
975 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
976 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
978 /* Note: we do not d_invalidate() */
979 if (!nd
->path
.dentry
->d_op
->d_revalidate(
980 nd
->path
.dentry
, nd
))
986 path_put_conditional(&next
, nd
);
994 static int path_walk(const char *name
, struct nameidata
*nd
)
996 struct path save
= nd
->path
;
999 current
->total_link_count
= 0;
1001 /* make sure the stuff we saved doesn't go away */
1004 result
= link_path_walk(name
, nd
);
1005 if (result
== -ESTALE
) {
1006 /* nd->path had been dropped */
1007 current
->total_link_count
= 0;
1009 path_get(&nd
->path
);
1010 nd
->flags
|= LOOKUP_REVAL
;
1011 result
= link_path_walk(name
, nd
);
1019 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1025 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1028 nd
->root
.mnt
= NULL
;
1032 nd
->path
= nd
->root
;
1033 path_get(&nd
->root
);
1034 } else if (dfd
== AT_FDCWD
) {
1035 struct fs_struct
*fs
= current
->fs
;
1036 read_lock(&fs
->lock
);
1039 read_unlock(&fs
->lock
);
1041 struct dentry
*dentry
;
1043 file
= fget_light(dfd
, &fput_needed
);
1048 dentry
= file
->f_path
.dentry
;
1051 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1054 retval
= file_permission(file
, MAY_EXEC
);
1058 nd
->path
= file
->f_path
;
1059 path_get(&file
->f_path
);
1061 fput_light(file
, fput_needed
);
1066 fput_light(file
, fput_needed
);
1071 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1072 static int do_path_lookup(int dfd
, const char *name
,
1073 unsigned int flags
, struct nameidata
*nd
)
1075 int retval
= path_init(dfd
, name
, flags
, nd
);
1077 retval
= path_walk(name
, nd
);
1078 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1079 nd
->path
.dentry
->d_inode
))
1080 audit_inode(name
, nd
->path
.dentry
);
1082 path_put(&nd
->root
);
1083 nd
->root
.mnt
= NULL
;
1088 int path_lookup(const char *name
, unsigned int flags
,
1089 struct nameidata
*nd
)
1091 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1094 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1096 struct nameidata nd
;
1097 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1104 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1105 * @dentry: pointer to dentry of the base directory
1106 * @mnt: pointer to vfs mount of the base directory
1107 * @name: pointer to file name
1108 * @flags: lookup flags
1109 * @nd: pointer to nameidata
1111 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1112 const char *name
, unsigned int flags
,
1113 struct nameidata
*nd
)
1117 /* same as do_path_lookup */
1118 nd
->last_type
= LAST_ROOT
;
1122 nd
->path
.dentry
= dentry
;
1124 path_get(&nd
->path
);
1125 nd
->root
= nd
->path
;
1126 path_get(&nd
->root
);
1128 retval
= path_walk(name
, nd
);
1129 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1130 nd
->path
.dentry
->d_inode
))
1131 audit_inode(name
, nd
->path
.dentry
);
1133 path_put(&nd
->root
);
1134 nd
->root
.mnt
= NULL
;
1139 static struct dentry
*__lookup_hash(struct qstr
*name
,
1140 struct dentry
*base
, struct nameidata
*nd
)
1142 struct dentry
*dentry
;
1143 struct inode
*inode
;
1146 inode
= base
->d_inode
;
1149 * See if the low-level filesystem might want
1150 * to use its own hash..
1152 if (base
->d_op
&& base
->d_op
->d_hash
) {
1153 err
= base
->d_op
->d_hash(base
, name
);
1154 dentry
= ERR_PTR(err
);
1159 dentry
= __d_lookup(base
, name
);
1161 /* lockess __d_lookup may fail due to concurrent d_move()
1162 * in some unrelated directory, so try with d_lookup
1165 dentry
= d_lookup(base
, name
);
1167 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
1168 dentry
= do_revalidate(dentry
, nd
);
1173 /* Don't create child dentry for a dead directory. */
1174 dentry
= ERR_PTR(-ENOENT
);
1175 if (IS_DEADDIR(inode
))
1178 new = d_alloc(base
, name
);
1179 dentry
= ERR_PTR(-ENOMEM
);
1182 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1193 * Restricted form of lookup. Doesn't follow links, single-component only,
1194 * needs parent already locked. Doesn't follow mounts.
1197 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1201 err
= exec_permission(nd
->path
.dentry
->d_inode
);
1203 return ERR_PTR(err
);
1204 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1207 static int __lookup_one_len(const char *name
, struct qstr
*this,
1208 struct dentry
*base
, int len
)
1218 hash
= init_name_hash();
1220 c
= *(const unsigned char *)name
++;
1221 if (c
== '/' || c
== '\0')
1223 hash
= partial_name_hash(c
, hash
);
1225 this->hash
= end_name_hash(hash
);
1230 * lookup_one_len - filesystem helper to lookup single pathname component
1231 * @name: pathname component to lookup
1232 * @base: base directory to lookup from
1233 * @len: maximum length @len should be interpreted to
1235 * Note that this routine is purely a helper for filesystem usage and should
1236 * not be called by generic code. Also note that by using this function the
1237 * nameidata argument is passed to the filesystem methods and a filesystem
1238 * using this helper needs to be prepared for that.
1240 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1245 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1247 err
= __lookup_one_len(name
, &this, base
, len
);
1249 return ERR_PTR(err
);
1251 err
= exec_permission(base
->d_inode
);
1253 return ERR_PTR(err
);
1254 return __lookup_hash(&this, base
, NULL
);
1257 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1260 struct nameidata nd
;
1261 char *tmp
= getname(name
);
1262 int err
= PTR_ERR(tmp
);
1265 BUG_ON(flags
& LOOKUP_PARENT
);
1267 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1275 static int user_path_parent(int dfd
, const char __user
*path
,
1276 struct nameidata
*nd
, char **name
)
1278 char *s
= getname(path
);
1284 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1294 * It's inline, so penalty for filesystems that don't use sticky bit is
1297 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1299 uid_t fsuid
= current_fsuid();
1301 if (!(dir
->i_mode
& S_ISVTX
))
1303 if (inode
->i_uid
== fsuid
)
1305 if (dir
->i_uid
== fsuid
)
1307 return !capable(CAP_FOWNER
);
1311 * Check whether we can remove a link victim from directory dir, check
1312 * whether the type of victim is right.
1313 * 1. We can't do it if dir is read-only (done in permission())
1314 * 2. We should have write and exec permissions on dir
1315 * 3. We can't remove anything from append-only dir
1316 * 4. We can't do anything with immutable dir (done in permission())
1317 * 5. If the sticky bit on dir is set we should either
1318 * a. be owner of dir, or
1319 * b. be owner of victim, or
1320 * c. have CAP_FOWNER capability
1321 * 6. If the victim is append-only or immutable we can't do antyhing with
1322 * links pointing to it.
1323 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1324 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1325 * 9. We can't remove a root or mountpoint.
1326 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1327 * nfs_async_unlink().
1329 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1333 if (!victim
->d_inode
)
1336 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1337 audit_inode_child(victim
, dir
);
1339 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1344 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1345 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1348 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1350 if (IS_ROOT(victim
))
1352 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1354 if (IS_DEADDIR(dir
))
1356 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1361 /* Check whether we can create an object with dentry child in directory
1363 * 1. We can't do it if child already exists (open has special treatment for
1364 * this case, but since we are inlined it's OK)
1365 * 2. We can't do it if dir is read-only (done in permission())
1366 * 3. We should have write and exec permissions on dir
1367 * 4. We can't do it if dir is immutable (done in permission())
1369 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1373 if (IS_DEADDIR(dir
))
1375 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1379 * O_DIRECTORY translates into forcing a directory lookup.
1381 static inline int lookup_flags(unsigned int f
)
1383 unsigned long retval
= LOOKUP_FOLLOW
;
1386 retval
&= ~LOOKUP_FOLLOW
;
1388 if (f
& O_DIRECTORY
)
1389 retval
|= LOOKUP_DIRECTORY
;
1395 * p1 and p2 should be directories on the same fs.
1397 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1402 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1406 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1408 p
= d_ancestor(p2
, p1
);
1410 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1411 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1415 p
= d_ancestor(p1
, p2
);
1417 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1418 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1422 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1423 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1427 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1429 mutex_unlock(&p1
->d_inode
->i_mutex
);
1431 mutex_unlock(&p2
->d_inode
->i_mutex
);
1432 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1436 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1437 struct nameidata
*nd
)
1439 int error
= may_create(dir
, dentry
);
1444 if (!dir
->i_op
->create
)
1445 return -EACCES
; /* shouldn't it be ENOSYS? */
1448 error
= security_inode_create(dir
, dentry
, mode
);
1452 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1454 fsnotify_create(dir
, dentry
);
1458 int may_open(struct path
*path
, int acc_mode
, int flag
)
1460 struct dentry
*dentry
= path
->dentry
;
1461 struct inode
*inode
= dentry
->d_inode
;
1467 switch (inode
->i_mode
& S_IFMT
) {
1471 if (acc_mode
& MAY_WRITE
)
1476 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1485 error
= inode_permission(inode
, acc_mode
);
1490 * An append-only file must be opened in append mode for writing.
1492 if (IS_APPEND(inode
)) {
1493 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1499 /* O_NOATIME can only be set by the owner or superuser */
1500 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
1504 * Ensure there are no outstanding leases on the file.
1506 return break_lease(inode
, flag
);
1509 static int handle_truncate(struct path
*path
)
1511 struct inode
*inode
= path
->dentry
->d_inode
;
1512 int error
= get_write_access(inode
);
1516 * Refuse to truncate files with mandatory locks held on them.
1518 error
= locks_verify_locked(inode
);
1520 error
= security_path_truncate(path
, 0,
1521 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
);
1523 error
= do_truncate(path
->dentry
, 0,
1524 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1527 put_write_access(inode
);
1532 * Be careful about ever adding any more callers of this
1533 * function. Its flags must be in the namei format, not
1534 * what get passed to sys_open().
1536 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1537 int open_flag
, int mode
)
1540 struct dentry
*dir
= nd
->path
.dentry
;
1542 if (!IS_POSIXACL(dir
->d_inode
))
1543 mode
&= ~current_umask();
1544 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
1547 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1549 mutex_unlock(&dir
->d_inode
->i_mutex
);
1550 dput(nd
->path
.dentry
);
1551 nd
->path
.dentry
= path
->dentry
;
1554 /* Don't check for write permission, don't truncate */
1555 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
1559 * Note that while the flag value (low two bits) for sys_open means:
1564 * it is changed into
1565 * 00 - no permissions needed
1566 * 01 - read-permission
1567 * 10 - write-permission
1569 * for the internal routines (ie open_namei()/follow_link() etc)
1570 * This is more logical, and also allows the 00 "no perm needed"
1571 * to be used for symlinks (where the permissions are checked
1575 static inline int open_to_namei_flags(int flag
)
1577 if ((flag
+1) & O_ACCMODE
)
1582 static int open_will_truncate(int flag
, struct inode
*inode
)
1585 * We'll never write to the fs underlying
1588 if (special_file(inode
->i_mode
))
1590 return (flag
& O_TRUNC
);
1593 static struct file
*finish_open(struct nameidata
*nd
,
1594 int open_flag
, int flag
, int acc_mode
)
1600 will_truncate
= open_will_truncate(flag
, nd
->path
.dentry
->d_inode
);
1601 if (will_truncate
) {
1602 error
= mnt_want_write(nd
->path
.mnt
);
1606 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
1609 mnt_drop_write(nd
->path
.mnt
);
1612 filp
= nameidata_to_filp(nd
);
1613 if (!IS_ERR(filp
)) {
1614 error
= ima_file_check(filp
, acc_mode
);
1617 filp
= ERR_PTR(error
);
1620 if (!IS_ERR(filp
)) {
1621 if (acc_mode
& MAY_WRITE
)
1622 vfs_dq_init(nd
->path
.dentry
->d_inode
);
1624 if (will_truncate
) {
1625 error
= handle_truncate(&nd
->path
);
1628 filp
= ERR_PTR(error
);
1633 * It is now safe to drop the mnt write
1634 * because the filp has had a write taken
1638 mnt_drop_write(nd
->path
.mnt
);
1642 if (!IS_ERR(nd
->intent
.open
.file
))
1643 release_open_intent(nd
);
1644 path_put(&nd
->path
);
1645 return ERR_PTR(error
);
1648 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
1649 int open_flag
, int flag
, int acc_mode
,
1650 int mode
, const char *pathname
,
1653 struct dentry
*dir
= nd
->path
.dentry
;
1659 mutex_lock(&dir
->d_inode
->i_mutex
);
1661 path
->dentry
= lookup_hash(nd
);
1662 path
->mnt
= nd
->path
.mnt
;
1664 error
= PTR_ERR(path
->dentry
);
1665 if (IS_ERR(path
->dentry
)) {
1666 mutex_unlock(&dir
->d_inode
->i_mutex
);
1670 if (IS_ERR(nd
->intent
.open
.file
)) {
1671 error
= PTR_ERR(nd
->intent
.open
.file
);
1672 goto exit_mutex_unlock
;
1675 /* Negative dentry, just create the file */
1676 if (!path
->dentry
->d_inode
) {
1678 * This write is needed to ensure that a
1679 * ro->rw transition does not occur between
1680 * the time when the file is created and when
1681 * a permanent write count is taken through
1682 * the 'struct file' in nameidata_to_filp().
1684 error
= mnt_want_write(nd
->path
.mnt
);
1686 goto exit_mutex_unlock
;
1687 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
1689 mnt_drop_write(nd
->path
.mnt
);
1692 filp
= nameidata_to_filp(nd
);
1693 mnt_drop_write(nd
->path
.mnt
);
1694 if (!IS_ERR(filp
)) {
1695 error
= ima_file_check(filp
, acc_mode
);
1698 filp
= ERR_PTR(error
);
1705 * It already exists.
1707 mutex_unlock(&dir
->d_inode
->i_mutex
);
1708 audit_inode(pathname
, path
->dentry
);
1714 if (__follow_mount(path
)) {
1716 if (flag
& O_NOFOLLOW
)
1721 if (!path
->dentry
->d_inode
)
1723 if (path
->dentry
->d_inode
->i_op
->follow_link
) {
1728 path_to_nameidata(path
, nd
);
1730 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
))
1732 filp
= finish_open(nd
, open_flag
, flag
, acc_mode
);
1736 mutex_unlock(&dir
->d_inode
->i_mutex
);
1738 path_put_conditional(path
, nd
);
1740 if (!IS_ERR(nd
->intent
.open
.file
))
1741 release_open_intent(nd
);
1742 path_put(&nd
->path
);
1743 return ERR_PTR(error
);
1747 * Note that the low bits of the passed in "open_flag"
1748 * are not the same as in the local variable "flag". See
1749 * open_to_namei_flags() for more details.
1751 struct file
*do_filp_open(int dfd
, const char *pathname
,
1752 int open_flag
, int mode
, int acc_mode
)
1755 struct nameidata nd
;
1759 int flag
= open_to_namei_flags(open_flag
);
1760 int force_reval
= 0;
1764 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1765 * check for O_DSYNC if the need any syncing at all we enforce it's
1766 * always set instead of having to deal with possibly weird behaviour
1767 * for malicious applications setting only __O_SYNC.
1769 if (open_flag
& __O_SYNC
)
1770 open_flag
|= O_DSYNC
;
1773 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
1775 /* O_TRUNC implies we need access checks for write permissions */
1777 acc_mode
|= MAY_WRITE
;
1779 /* Allow the LSM permission hook to distinguish append
1780 access from general write access. */
1781 if (flag
& O_APPEND
)
1782 acc_mode
|= MAY_APPEND
;
1785 * The simplest case - just a plain lookup.
1787 if (!(flag
& O_CREAT
)) {
1788 filp
= get_empty_filp();
1791 return ERR_PTR(-ENFILE
);
1792 nd
.intent
.open
.file
= filp
;
1793 filp
->f_flags
= open_flag
;
1794 nd
.intent
.open
.flags
= flag
;
1795 nd
.intent
.open
.create_mode
= 0;
1796 error
= do_path_lookup(dfd
, pathname
,
1797 lookup_flags(flag
)|LOOKUP_OPEN
, &nd
);
1798 if (IS_ERR(nd
.intent
.open
.file
)) {
1800 error
= PTR_ERR(nd
.intent
.open
.file
);
1804 release_open_intent(&nd
);
1806 return ERR_PTR(error
);
1811 * Create - we need to know the parent.
1814 error
= path_init(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
1816 return ERR_PTR(error
);
1818 nd
.flags
|= LOOKUP_REVAL
;
1819 error
= path_walk(pathname
, &nd
);
1823 return ERR_PTR(error
);
1825 if (unlikely(!audit_dummy_context()))
1826 audit_inode(pathname
, nd
.path
.dentry
);
1829 * We have the parent and last component. First of all, check
1830 * that we are not asked to creat(2) an obvious directory - that
1834 if (nd
.last_type
!= LAST_NORM
|| nd
.last
.name
[nd
.last
.len
])
1838 filp
= get_empty_filp();
1841 nd
.intent
.open
.file
= filp
;
1842 filp
->f_flags
= open_flag
;
1843 nd
.intent
.open
.flags
= flag
;
1844 nd
.intent
.open
.create_mode
= mode
;
1845 nd
.flags
&= ~LOOKUP_PARENT
;
1846 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_OPEN
;
1848 nd
.flags
|= LOOKUP_EXCL
;
1849 filp
= do_last(&nd
, &path
, open_flag
, flag
, acc_mode
, mode
,
1850 pathname
, &is_link
);
1858 filp
= finish_open(&nd
, open_flag
, flag
, acc_mode
);
1864 path_put_conditional(&path
, &nd
);
1866 if (!IS_ERR(nd
.intent
.open
.file
))
1867 release_open_intent(&nd
);
1872 return ERR_PTR(error
);
1876 if ((flag
& O_NOFOLLOW
) || count
++ == 32)
1879 * This is subtle. Instead of calling do_follow_link() we do the
1880 * thing by hands. The reason is that this way we have zero link_count
1881 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1882 * After that we have the parent and last component, i.e.
1883 * we are in the same situation as after the first path_walk().
1884 * Well, almost - if the last component is normal we get its copy
1885 * stored in nd->last.name and we will have to putname() it when we
1886 * are done. Procfs-like symlinks just set LAST_BIND.
1888 nd
.flags
|= LOOKUP_PARENT
;
1889 error
= security_inode_follow_link(path
.dentry
, &nd
);
1892 error
= __do_follow_link(&path
, &nd
);
1895 /* Does someone understand code flow here? Or it is only
1896 * me so stupid? Anathema to whoever designed this non-sense
1897 * with "intent.open".
1899 release_open_intent(&nd
);
1902 if (error
== -ESTALE
&& !force_reval
) {
1906 return ERR_PTR(error
);
1908 nd
.flags
&= ~LOOKUP_PARENT
;
1909 if (nd
.last_type
== LAST_BIND
)
1912 if (nd
.last_type
!= LAST_NORM
)
1914 if (nd
.last
.name
[nd
.last
.len
]) {
1915 __putname(nd
.last
.name
);
1918 filp
= do_last(&nd
, &path
, open_flag
, flag
, acc_mode
, mode
,
1919 pathname
, &is_link
);
1920 __putname(nd
.last
.name
);
1929 * filp_open - open file and return file pointer
1931 * @filename: path to open
1932 * @flags: open flags as per the open(2) second argument
1933 * @mode: mode for the new file if O_CREAT is set, else ignored
1935 * This is the helper to open a file from kernelspace if you really
1936 * have to. But in generally you should not do this, so please move
1937 * along, nothing to see here..
1939 struct file
*filp_open(const char *filename
, int flags
, int mode
)
1941 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
1943 EXPORT_SYMBOL(filp_open
);
1946 * lookup_create - lookup a dentry, creating it if it doesn't exist
1947 * @nd: nameidata info
1948 * @is_dir: directory flag
1950 * Simple function to lookup and return a dentry and create it
1951 * if it doesn't exist. Is SMP-safe.
1953 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1955 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1957 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1959 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1961 * Yucky last component or no last component at all?
1962 * (foo/., foo/.., /////)
1964 if (nd
->last_type
!= LAST_NORM
)
1966 nd
->flags
&= ~LOOKUP_PARENT
;
1967 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
1968 nd
->intent
.open
.flags
= O_EXCL
;
1971 * Do the final lookup.
1973 dentry
= lookup_hash(nd
);
1977 if (dentry
->d_inode
)
1980 * Special case - lookup gave negative, but... we had foo/bar/
1981 * From the vfs_mknod() POV we just have a negative dentry -
1982 * all is fine. Let's be bastards - you had / on the end, you've
1983 * been asking for (non-existent) directory. -ENOENT for you.
1985 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
1987 dentry
= ERR_PTR(-ENOENT
);
1992 dentry
= ERR_PTR(-EEXIST
);
1996 EXPORT_SYMBOL_GPL(lookup_create
);
1998 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2000 int error
= may_create(dir
, dentry
);
2005 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2008 if (!dir
->i_op
->mknod
)
2011 error
= devcgroup_inode_mknod(mode
, dev
);
2015 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2020 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2022 fsnotify_create(dir
, dentry
);
2026 static int may_mknod(mode_t mode
)
2028 switch (mode
& S_IFMT
) {
2034 case 0: /* zero mode translates to S_IFREG */
2043 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2048 struct dentry
*dentry
;
2049 struct nameidata nd
;
2054 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2058 dentry
= lookup_create(&nd
, 0);
2059 if (IS_ERR(dentry
)) {
2060 error
= PTR_ERR(dentry
);
2063 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2064 mode
&= ~current_umask();
2065 error
= may_mknod(mode
);
2068 error
= mnt_want_write(nd
.path
.mnt
);
2071 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2073 goto out_drop_write
;
2074 switch (mode
& S_IFMT
) {
2075 case 0: case S_IFREG
:
2076 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2078 case S_IFCHR
: case S_IFBLK
:
2079 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2080 new_decode_dev(dev
));
2082 case S_IFIFO
: case S_IFSOCK
:
2083 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2087 mnt_drop_write(nd
.path
.mnt
);
2091 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2098 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2100 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2103 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2105 int error
= may_create(dir
, dentry
);
2110 if (!dir
->i_op
->mkdir
)
2113 mode
&= (S_IRWXUGO
|S_ISVTX
);
2114 error
= security_inode_mkdir(dir
, dentry
, mode
);
2119 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2121 fsnotify_mkdir(dir
, dentry
);
2125 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2129 struct dentry
*dentry
;
2130 struct nameidata nd
;
2132 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2136 dentry
= lookup_create(&nd
, 1);
2137 error
= PTR_ERR(dentry
);
2141 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2142 mode
&= ~current_umask();
2143 error
= mnt_want_write(nd
.path
.mnt
);
2146 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2148 goto out_drop_write
;
2149 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2151 mnt_drop_write(nd
.path
.mnt
);
2155 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2162 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2164 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2168 * We try to drop the dentry early: we should have
2169 * a usage count of 2 if we're the only user of this
2170 * dentry, and if that is true (possibly after pruning
2171 * the dcache), then we drop the dentry now.
2173 * A low-level filesystem can, if it choses, legally
2176 * if (!d_unhashed(dentry))
2179 * if it cannot handle the case of removing a directory
2180 * that is still in use by something else..
2182 void dentry_unhash(struct dentry
*dentry
)
2185 shrink_dcache_parent(dentry
);
2186 spin_lock(&dcache_lock
);
2187 spin_lock(&dentry
->d_lock
);
2188 if (atomic_read(&dentry
->d_count
) == 2)
2190 spin_unlock(&dentry
->d_lock
);
2191 spin_unlock(&dcache_lock
);
2194 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2196 int error
= may_delete(dir
, dentry
, 1);
2201 if (!dir
->i_op
->rmdir
)
2206 mutex_lock(&dentry
->d_inode
->i_mutex
);
2207 dentry_unhash(dentry
);
2208 if (d_mountpoint(dentry
))
2211 error
= security_inode_rmdir(dir
, dentry
);
2213 error
= dir
->i_op
->rmdir(dir
, dentry
);
2215 dentry
->d_inode
->i_flags
|= S_DEAD
;
2218 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2227 static long do_rmdir(int dfd
, const char __user
*pathname
)
2231 struct dentry
*dentry
;
2232 struct nameidata nd
;
2234 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2238 switch(nd
.last_type
) {
2250 nd
.flags
&= ~LOOKUP_PARENT
;
2252 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2253 dentry
= lookup_hash(&nd
);
2254 error
= PTR_ERR(dentry
);
2257 error
= mnt_want_write(nd
.path
.mnt
);
2260 error
= security_path_rmdir(&nd
.path
, dentry
);
2263 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2265 mnt_drop_write(nd
.path
.mnt
);
2269 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2276 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2278 return do_rmdir(AT_FDCWD
, pathname
);
2281 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2283 int error
= may_delete(dir
, dentry
, 0);
2288 if (!dir
->i_op
->unlink
)
2293 mutex_lock(&dentry
->d_inode
->i_mutex
);
2294 if (d_mountpoint(dentry
))
2297 error
= security_inode_unlink(dir
, dentry
);
2299 error
= dir
->i_op
->unlink(dir
, dentry
);
2301 dentry
->d_inode
->i_flags
|= S_DEAD
;
2304 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2306 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2307 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2308 fsnotify_link_count(dentry
->d_inode
);
2316 * Make sure that the actual truncation of the file will occur outside its
2317 * directory's i_mutex. Truncate can take a long time if there is a lot of
2318 * writeout happening, and we don't want to prevent access to the directory
2319 * while waiting on the I/O.
2321 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2325 struct dentry
*dentry
;
2326 struct nameidata nd
;
2327 struct inode
*inode
= NULL
;
2329 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2334 if (nd
.last_type
!= LAST_NORM
)
2337 nd
.flags
&= ~LOOKUP_PARENT
;
2339 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2340 dentry
= lookup_hash(&nd
);
2341 error
= PTR_ERR(dentry
);
2342 if (!IS_ERR(dentry
)) {
2343 /* Why not before? Because we want correct error value */
2344 if (nd
.last
.name
[nd
.last
.len
])
2346 inode
= dentry
->d_inode
;
2348 atomic_inc(&inode
->i_count
);
2349 error
= mnt_want_write(nd
.path
.mnt
);
2352 error
= security_path_unlink(&nd
.path
, dentry
);
2355 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2357 mnt_drop_write(nd
.path
.mnt
);
2361 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2363 iput(inode
); /* truncate the inode here */
2370 error
= !dentry
->d_inode
? -ENOENT
:
2371 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2375 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2377 if ((flag
& ~AT_REMOVEDIR
) != 0)
2380 if (flag
& AT_REMOVEDIR
)
2381 return do_rmdir(dfd
, pathname
);
2383 return do_unlinkat(dfd
, pathname
);
2386 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2388 return do_unlinkat(AT_FDCWD
, pathname
);
2391 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2393 int error
= may_create(dir
, dentry
);
2398 if (!dir
->i_op
->symlink
)
2401 error
= security_inode_symlink(dir
, dentry
, oldname
);
2406 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2408 fsnotify_create(dir
, dentry
);
2412 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2413 int, newdfd
, const char __user
*, newname
)
2418 struct dentry
*dentry
;
2419 struct nameidata nd
;
2421 from
= getname(oldname
);
2423 return PTR_ERR(from
);
2425 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2429 dentry
= lookup_create(&nd
, 0);
2430 error
= PTR_ERR(dentry
);
2434 error
= mnt_want_write(nd
.path
.mnt
);
2437 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2439 goto out_drop_write
;
2440 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2442 mnt_drop_write(nd
.path
.mnt
);
2446 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2454 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2456 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2459 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2461 struct inode
*inode
= old_dentry
->d_inode
;
2467 error
= may_create(dir
, new_dentry
);
2471 if (dir
->i_sb
!= inode
->i_sb
)
2475 * A link to an append-only or immutable file cannot be created.
2477 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2479 if (!dir
->i_op
->link
)
2481 if (S_ISDIR(inode
->i_mode
))
2484 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2488 mutex_lock(&inode
->i_mutex
);
2490 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2491 mutex_unlock(&inode
->i_mutex
);
2493 fsnotify_link(dir
, inode
, new_dentry
);
2498 * Hardlinks are often used in delicate situations. We avoid
2499 * security-related surprises by not following symlinks on the
2502 * We don't follow them on the oldname either to be compatible
2503 * with linux 2.0, and to avoid hard-linking to directories
2504 * and other special files. --ADM
2506 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2507 int, newdfd
, const char __user
*, newname
, int, flags
)
2509 struct dentry
*new_dentry
;
2510 struct nameidata nd
;
2511 struct path old_path
;
2515 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2518 error
= user_path_at(olddfd
, oldname
,
2519 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2524 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2528 if (old_path
.mnt
!= nd
.path
.mnt
)
2530 new_dentry
= lookup_create(&nd
, 0);
2531 error
= PTR_ERR(new_dentry
);
2532 if (IS_ERR(new_dentry
))
2534 error
= mnt_want_write(nd
.path
.mnt
);
2537 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2539 goto out_drop_write
;
2540 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2542 mnt_drop_write(nd
.path
.mnt
);
2546 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2551 path_put(&old_path
);
2556 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
2558 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2562 * The worst of all namespace operations - renaming directory. "Perverted"
2563 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2565 * a) we can get into loop creation. Check is done in is_subdir().
2566 * b) race potential - two innocent renames can create a loop together.
2567 * That's where 4.4 screws up. Current fix: serialization on
2568 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2570 * c) we have to lock _three_ objects - parents and victim (if it exists).
2571 * And that - after we got ->i_mutex on parents (until then we don't know
2572 * whether the target exists). Solution: try to be smart with locking
2573 * order for inodes. We rely on the fact that tree topology may change
2574 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2575 * move will be locked. Thus we can rank directories by the tree
2576 * (ancestors first) and rank all non-directories after them.
2577 * That works since everybody except rename does "lock parent, lookup,
2578 * lock child" and rename is under ->s_vfs_rename_mutex.
2579 * HOWEVER, it relies on the assumption that any object with ->lookup()
2580 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2581 * we'd better make sure that there's no link(2) for them.
2582 * d) some filesystems don't support opened-but-unlinked directories,
2583 * either because of layout or because they are not ready to deal with
2584 * all cases correctly. The latter will be fixed (taking this sort of
2585 * stuff into VFS), but the former is not going away. Solution: the same
2586 * trick as in rmdir().
2587 * e) conversion from fhandle to dentry may come in the wrong moment - when
2588 * we are removing the target. Solution: we will have to grab ->i_mutex
2589 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2590 * ->i_mutex on parents, which works but leads to some truely excessive
2593 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2594 struct inode
*new_dir
, struct dentry
*new_dentry
)
2597 struct inode
*target
;
2600 * If we are going to change the parent - check write permissions,
2601 * we'll need to flip '..'.
2603 if (new_dir
!= old_dir
) {
2604 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
2609 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2613 target
= new_dentry
->d_inode
;
2615 mutex_lock(&target
->i_mutex
);
2616 dentry_unhash(new_dentry
);
2618 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2621 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2624 target
->i_flags
|= S_DEAD
;
2625 mutex_unlock(&target
->i_mutex
);
2626 if (d_unhashed(new_dentry
))
2627 d_rehash(new_dentry
);
2631 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2632 d_move(old_dentry
,new_dentry
);
2636 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2637 struct inode
*new_dir
, struct dentry
*new_dentry
)
2639 struct inode
*target
;
2642 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2647 target
= new_dentry
->d_inode
;
2649 mutex_lock(&target
->i_mutex
);
2650 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2653 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2656 target
->i_flags
|= S_DEAD
;
2657 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2658 d_move(old_dentry
, new_dentry
);
2661 mutex_unlock(&target
->i_mutex
);
2666 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2667 struct inode
*new_dir
, struct dentry
*new_dentry
)
2670 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2671 const char *old_name
;
2673 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2676 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2680 if (!new_dentry
->d_inode
)
2681 error
= may_create(new_dir
, new_dentry
);
2683 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2687 if (!old_dir
->i_op
->rename
)
2690 vfs_dq_init(old_dir
);
2691 vfs_dq_init(new_dir
);
2693 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2696 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2698 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2700 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
2701 new_dentry
->d_inode
, old_dentry
);
2702 fsnotify_oldname_free(old_name
);
2707 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
2708 int, newdfd
, const char __user
*, newname
)
2710 struct dentry
*old_dir
, *new_dir
;
2711 struct dentry
*old_dentry
, *new_dentry
;
2712 struct dentry
*trap
;
2713 struct nameidata oldnd
, newnd
;
2718 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
2722 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
2727 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2730 old_dir
= oldnd
.path
.dentry
;
2732 if (oldnd
.last_type
!= LAST_NORM
)
2735 new_dir
= newnd
.path
.dentry
;
2736 if (newnd
.last_type
!= LAST_NORM
)
2739 oldnd
.flags
&= ~LOOKUP_PARENT
;
2740 newnd
.flags
&= ~LOOKUP_PARENT
;
2741 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
2743 trap
= lock_rename(new_dir
, old_dir
);
2745 old_dentry
= lookup_hash(&oldnd
);
2746 error
= PTR_ERR(old_dentry
);
2747 if (IS_ERR(old_dentry
))
2749 /* source must exist */
2751 if (!old_dentry
->d_inode
)
2753 /* unless the source is a directory trailing slashes give -ENOTDIR */
2754 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2756 if (oldnd
.last
.name
[oldnd
.last
.len
])
2758 if (newnd
.last
.name
[newnd
.last
.len
])
2761 /* source should not be ancestor of target */
2763 if (old_dentry
== trap
)
2765 new_dentry
= lookup_hash(&newnd
);
2766 error
= PTR_ERR(new_dentry
);
2767 if (IS_ERR(new_dentry
))
2769 /* target should not be an ancestor of source */
2771 if (new_dentry
== trap
)
2774 error
= mnt_want_write(oldnd
.path
.mnt
);
2777 error
= security_path_rename(&oldnd
.path
, old_dentry
,
2778 &newnd
.path
, new_dentry
);
2781 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2782 new_dir
->d_inode
, new_dentry
);
2784 mnt_drop_write(oldnd
.path
.mnt
);
2790 unlock_rename(new_dir
, old_dir
);
2792 path_put(&newnd
.path
);
2795 path_put(&oldnd
.path
);
2801 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
2803 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2806 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2810 len
= PTR_ERR(link
);
2815 if (len
> (unsigned) buflen
)
2817 if (copy_to_user(buffer
, link
, len
))
2824 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2825 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2826 * using) it for any given inode is up to filesystem.
2828 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2830 struct nameidata nd
;
2835 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2837 return PTR_ERR(cookie
);
2839 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2840 if (dentry
->d_inode
->i_op
->put_link
)
2841 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2845 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2847 return __vfs_follow_link(nd
, link
);
2850 /* get the link contents into pagecache */
2851 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2855 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2856 page
= read_mapping_page(mapping
, 0, NULL
);
2861 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
2865 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2867 struct page
*page
= NULL
;
2868 char *s
= page_getlink(dentry
, &page
);
2869 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2872 page_cache_release(page
);
2877 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2879 struct page
*page
= NULL
;
2880 nd_set_link(nd
, page_getlink(dentry
, &page
));
2884 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2886 struct page
*page
= cookie
;
2890 page_cache_release(page
);
2895 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2897 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
2899 struct address_space
*mapping
= inode
->i_mapping
;
2904 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
2906 flags
|= AOP_FLAG_NOFS
;
2909 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2910 flags
, &page
, &fsdata
);
2914 kaddr
= kmap_atomic(page
, KM_USER0
);
2915 memcpy(kaddr
, symname
, len
-1);
2916 kunmap_atomic(kaddr
, KM_USER0
);
2918 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2925 mark_inode_dirty(inode
);
2931 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2933 return __page_symlink(inode
, symname
, len
,
2934 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
2937 const struct inode_operations page_symlink_inode_operations
= {
2938 .readlink
= generic_readlink
,
2939 .follow_link
= page_follow_link_light
,
2940 .put_link
= page_put_link
,
2943 EXPORT_SYMBOL(user_path_at
);
2944 EXPORT_SYMBOL(follow_down
);
2945 EXPORT_SYMBOL(follow_up
);
2946 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2947 EXPORT_SYMBOL(getname
);
2948 EXPORT_SYMBOL(lock_rename
);
2949 EXPORT_SYMBOL(lookup_one_len
);
2950 EXPORT_SYMBOL(page_follow_link_light
);
2951 EXPORT_SYMBOL(page_put_link
);
2952 EXPORT_SYMBOL(page_readlink
);
2953 EXPORT_SYMBOL(__page_symlink
);
2954 EXPORT_SYMBOL(page_symlink
);
2955 EXPORT_SYMBOL(page_symlink_inode_operations
);
2956 EXPORT_SYMBOL(path_lookup
);
2957 EXPORT_SYMBOL(kern_path
);
2958 EXPORT_SYMBOL(vfs_path_lookup
);
2959 EXPORT_SYMBOL(inode_permission
);
2960 EXPORT_SYMBOL(file_permission
);
2961 EXPORT_SYMBOL(unlock_rename
);
2962 EXPORT_SYMBOL(vfs_create
);
2963 EXPORT_SYMBOL(vfs_follow_link
);
2964 EXPORT_SYMBOL(vfs_link
);
2965 EXPORT_SYMBOL(vfs_mkdir
);
2966 EXPORT_SYMBOL(vfs_mknod
);
2967 EXPORT_SYMBOL(generic_permission
);
2968 EXPORT_SYMBOL(vfs_readlink
);
2969 EXPORT_SYMBOL(vfs_rename
);
2970 EXPORT_SYMBOL(vfs_rmdir
);
2971 EXPORT_SYMBOL(vfs_symlink
);
2972 EXPORT_SYMBOL(vfs_unlink
);
2973 EXPORT_SYMBOL(dentry_unhash
);
2974 EXPORT_SYMBOL(generic_readlink
);