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/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <asm/namei.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int link_path_walk(const char *name
, struct nameidata
*nd
);
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
);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode
*inode
, int mask
,
183 int (*check_acl
)(struct inode
*inode
, int mask
))
185 umode_t mode
= inode
->i_mode
;
187 if (current
->fsuid
== inode
->i_uid
)
190 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
191 int error
= check_acl(inode
, mask
);
192 if (error
== -EACCES
)
193 goto check_capabilities
;
194 else if (error
!= -EAGAIN
)
198 if (in_group_p(inode
->i_gid
))
203 * If the DACs are ok we don't need any capability check.
205 if (((mode
& mask
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
)) == mask
))
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
213 if (!(mask
& MAY_EXEC
) ||
214 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
215 if (capable(CAP_DAC_OVERRIDE
))
219 * Searching includes executable on directories, else just read.
221 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
222 if (capable(CAP_DAC_READ_SEARCH
))
228 int permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
231 struct vfsmount
*mnt
= NULL
;
236 if (mask
& MAY_WRITE
) {
237 umode_t mode
= inode
->i_mode
;
240 * Nobody gets write access to a read-only fs.
242 if (IS_RDONLY(inode
) &&
243 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
247 * Nobody gets write access to an immutable file.
249 if (IS_IMMUTABLE(inode
))
253 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
)) {
255 * MAY_EXEC on regular files is denied if the fs is mounted
256 * with the "noexec" flag.
258 if (mnt
&& (mnt
->mnt_flags
& MNT_NOEXEC
))
262 /* Ordinary permission routines do not understand MAY_APPEND. */
263 submask
= mask
& ~MAY_APPEND
;
264 if (inode
->i_op
&& inode
->i_op
->permission
) {
265 retval
= inode
->i_op
->permission(inode
, submask
, nd
);
268 * Exec permission on a regular file is denied if none
269 * of the execute bits are set.
271 * This check should be done by the ->permission()
274 if ((mask
& MAY_EXEC
) && S_ISREG(inode
->i_mode
) &&
275 !(inode
->i_mode
& S_IXUGO
))
279 retval
= generic_permission(inode
, submask
, NULL
);
284 return security_inode_permission(inode
, mask
, nd
);
288 * vfs_permission - check for access rights to a given path
289 * @nd: lookup result that describes the path
290 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
292 * Used to check for read/write/execute permissions on a path.
293 * We use "fsuid" for this, letting us set arbitrary permissions
294 * for filesystem access without changing the "normal" uids which
295 * are used for other things.
297 int vfs_permission(struct nameidata
*nd
, int mask
)
299 return permission(nd
->path
.dentry
->d_inode
, mask
, nd
);
303 * file_permission - check for additional access rights to a given file
304 * @file: file to check access rights for
305 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
307 * Used to check for read/write/execute permissions on an already opened
311 * Do not use this function in new code. All access checks should
312 * be done using vfs_permission().
314 int file_permission(struct file
*file
, int mask
)
316 return permission(file
->f_path
.dentry
->d_inode
, mask
, NULL
);
320 * get_write_access() gets write permission for a file.
321 * put_write_access() releases this write permission.
322 * This is used for regular files.
323 * We cannot support write (and maybe mmap read-write shared) accesses and
324 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
325 * can have the following values:
326 * 0: no writers, no VM_DENYWRITE mappings
327 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
328 * > 0: (i_writecount) users are writing to the file.
330 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
331 * except for the cases where we don't hold i_writecount yet. Then we need to
332 * use {get,deny}_write_access() - these functions check the sign and refuse
333 * to do the change if sign is wrong. Exclusion between them is provided by
334 * the inode->i_lock spinlock.
337 int get_write_access(struct inode
* inode
)
339 spin_lock(&inode
->i_lock
);
340 if (atomic_read(&inode
->i_writecount
) < 0) {
341 spin_unlock(&inode
->i_lock
);
344 atomic_inc(&inode
->i_writecount
);
345 spin_unlock(&inode
->i_lock
);
350 int deny_write_access(struct file
* file
)
352 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
354 spin_lock(&inode
->i_lock
);
355 if (atomic_read(&inode
->i_writecount
) > 0) {
356 spin_unlock(&inode
->i_lock
);
359 atomic_dec(&inode
->i_writecount
);
360 spin_unlock(&inode
->i_lock
);
366 * path_put - put a reference to a path
367 * @path: path to put the reference to
369 * Given a path decrement the reference count to the dentry and the vfsmount.
371 void path_put(struct path
*path
)
376 EXPORT_SYMBOL(path_put
);
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
382 void release_open_intent(struct nameidata
*nd
)
384 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
385 put_filp(nd
->intent
.open
.file
);
387 fput(nd
->intent
.open
.file
);
390 static inline struct dentry
*
391 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
393 int status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
394 if (unlikely(status
<= 0)) {
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
402 if (!d_invalidate(dentry
)) {
408 dentry
= ERR_PTR(status
);
415 * Internal lookup() using the new generic dcache.
418 static struct dentry
* cached_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
420 struct dentry
* dentry
= __d_lookup(parent
, name
);
422 /* lockess __d_lookup may fail due to concurrent d_move()
423 * in some unrelated directory, so try with d_lookup
426 dentry
= d_lookup(parent
, name
);
428 if (dentry
&& dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
429 dentry
= do_revalidate(dentry
, nd
);
435 * Short-cut version of permission(), for calling by
436 * path_walk(), when dcache lock is held. Combines parts
437 * of permission() and generic_permission(), and tests ONLY for
438 * MAY_EXEC permission.
440 * If appropriate, check DAC only. If not appropriate, or
441 * short-cut DAC fails, then call permission() to do more
442 * complete permission check.
444 static int exec_permission_lite(struct inode
*inode
,
445 struct nameidata
*nd
)
447 umode_t mode
= inode
->i_mode
;
449 if (inode
->i_op
&& inode
->i_op
->permission
)
452 if (current
->fsuid
== inode
->i_uid
)
454 else if (in_group_p(inode
->i_gid
))
460 if ((inode
->i_mode
& S_IXUGO
) && capable(CAP_DAC_OVERRIDE
))
463 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_OVERRIDE
))
466 if (S_ISDIR(inode
->i_mode
) && capable(CAP_DAC_READ_SEARCH
))
471 return security_inode_permission(inode
, MAY_EXEC
, nd
);
475 * This is called when everything else fails, and we actually have
476 * to go to the low-level filesystem to find out what we should do..
478 * We get the directory semaphore, and after getting that we also
479 * make sure that nobody added the entry to the dcache in the meantime..
482 static struct dentry
* real_lookup(struct dentry
* parent
, struct qstr
* name
, struct nameidata
*nd
)
484 struct dentry
* result
;
485 struct inode
*dir
= parent
->d_inode
;
487 mutex_lock(&dir
->i_mutex
);
489 * First re-do the cached lookup just in case it was created
490 * while we waited for the directory semaphore..
492 * FIXME! This could use version numbering or similar to
493 * avoid unnecessary cache lookups.
495 * The "dcache_lock" is purely to protect the RCU list walker
496 * from concurrent renames at this point (we mustn't get false
497 * negatives from the RCU list walk here, unlike the optimistic
500 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
502 result
= d_lookup(parent
, name
);
504 struct dentry
* dentry
= d_alloc(parent
, name
);
505 result
= ERR_PTR(-ENOMEM
);
507 result
= dir
->i_op
->lookup(dir
, dentry
, nd
);
513 mutex_unlock(&dir
->i_mutex
);
518 * Uhhuh! Nasty case: the cache was re-populated while
519 * we waited on the semaphore. Need to revalidate.
521 mutex_unlock(&dir
->i_mutex
);
522 if (result
->d_op
&& result
->d_op
->d_revalidate
) {
523 result
= do_revalidate(result
, nd
);
525 result
= ERR_PTR(-ENOENT
);
530 static int __emul_lookup_dentry(const char *, struct nameidata
*);
533 static __always_inline
int
534 walk_init_root(const char *name
, struct nameidata
*nd
)
536 struct fs_struct
*fs
= current
->fs
;
538 read_lock(&fs
->lock
);
539 if (fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
540 nd
->path
.mnt
= mntget(fs
->altrootmnt
);
541 nd
->path
.dentry
= dget(fs
->altroot
);
542 read_unlock(&fs
->lock
);
543 if (__emul_lookup_dentry(name
,nd
))
545 read_lock(&fs
->lock
);
547 nd
->path
.mnt
= mntget(fs
->rootmnt
);
548 nd
->path
.dentry
= dget(fs
->root
);
549 read_unlock(&fs
->lock
);
553 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
562 if (!walk_init_root(link
, nd
))
563 /* weird __emul_prefix() stuff did it */
566 res
= link_path_walk(link
, nd
);
568 if (nd
->depth
|| res
|| nd
->last_type
!=LAST_NORM
)
571 * If it is an iterative symlinks resolution in open_namei() we
572 * have to copy the last component. And all that crap because of
573 * bloody create() on broken symlinks. Furrfu...
576 if (unlikely(!name
)) {
580 strcpy(name
, nd
->last
.name
);
581 nd
->last
.name
= name
;
585 return PTR_ERR(link
);
588 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
591 if (path
->mnt
!= nd
->path
.mnt
)
595 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
597 dput(nd
->path
.dentry
);
598 if (nd
->path
.mnt
!= path
->mnt
)
599 mntput(nd
->path
.mnt
);
600 nd
->path
.mnt
= path
->mnt
;
601 nd
->path
.dentry
= path
->dentry
;
604 static __always_inline
int __do_follow_link(struct path
*path
, struct nameidata
*nd
)
608 struct dentry
*dentry
= path
->dentry
;
610 touch_atime(path
->mnt
, dentry
);
611 nd_set_link(nd
, NULL
);
613 if (path
->mnt
!= nd
->path
.mnt
) {
614 path_to_nameidata(path
, nd
);
618 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
619 error
= PTR_ERR(cookie
);
620 if (!IS_ERR(cookie
)) {
621 char *s
= nd_get_link(nd
);
624 error
= __vfs_follow_link(nd
, s
);
625 if (dentry
->d_inode
->i_op
->put_link
)
626 dentry
->d_inode
->i_op
->put_link(dentry
, nd
, cookie
);
635 * This limits recursive symlink follows to 8, while
636 * limiting consecutive symlinks to 40.
638 * Without that kind of total limit, nasty chains of consecutive
639 * symlinks can cause almost arbitrarily long lookups.
641 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
644 if (current
->link_count
>= MAX_NESTED_LINKS
)
646 if (current
->total_link_count
>= 40)
648 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
650 err
= security_inode_follow_link(path
->dentry
, nd
);
653 current
->link_count
++;
654 current
->total_link_count
++;
656 err
= __do_follow_link(path
, nd
);
657 current
->link_count
--;
661 path_put_conditional(path
, nd
);
666 int follow_up(struct vfsmount
**mnt
, struct dentry
**dentry
)
668 struct vfsmount
*parent
;
669 struct dentry
*mountpoint
;
670 spin_lock(&vfsmount_lock
);
671 parent
=(*mnt
)->mnt_parent
;
672 if (parent
== *mnt
) {
673 spin_unlock(&vfsmount_lock
);
677 mountpoint
=dget((*mnt
)->mnt_mountpoint
);
678 spin_unlock(&vfsmount_lock
);
680 *dentry
= mountpoint
;
686 /* no need for dcache_lock, as serialization is taken care in
689 static int __follow_mount(struct path
*path
)
692 while (d_mountpoint(path
->dentry
)) {
693 struct vfsmount
*mounted
= lookup_mnt(path
->mnt
, path
->dentry
);
700 path
->dentry
= dget(mounted
->mnt_root
);
706 static void follow_mount(struct vfsmount
**mnt
, struct dentry
**dentry
)
708 while (d_mountpoint(*dentry
)) {
709 struct vfsmount
*mounted
= lookup_mnt(*mnt
, *dentry
);
715 *dentry
= dget(mounted
->mnt_root
);
719 /* no need for dcache_lock, as serialization is taken care in
722 int follow_down(struct vfsmount
**mnt
, struct dentry
**dentry
)
724 struct vfsmount
*mounted
;
726 mounted
= lookup_mnt(*mnt
, *dentry
);
731 *dentry
= dget(mounted
->mnt_root
);
737 static __always_inline
void follow_dotdot(struct nameidata
*nd
)
739 struct fs_struct
*fs
= current
->fs
;
742 struct vfsmount
*parent
;
743 struct dentry
*old
= nd
->path
.dentry
;
745 read_lock(&fs
->lock
);
746 if (nd
->path
.dentry
== fs
->root
&&
747 nd
->path
.mnt
== fs
->rootmnt
) {
748 read_unlock(&fs
->lock
);
751 read_unlock(&fs
->lock
);
752 spin_lock(&dcache_lock
);
753 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
754 nd
->path
.dentry
= dget(nd
->path
.dentry
->d_parent
);
755 spin_unlock(&dcache_lock
);
759 spin_unlock(&dcache_lock
);
760 spin_lock(&vfsmount_lock
);
761 parent
= nd
->path
.mnt
->mnt_parent
;
762 if (parent
== nd
->path
.mnt
) {
763 spin_unlock(&vfsmount_lock
);
767 nd
->path
.dentry
= dget(nd
->path
.mnt
->mnt_mountpoint
);
768 spin_unlock(&vfsmount_lock
);
770 mntput(nd
->path
.mnt
);
771 nd
->path
.mnt
= parent
;
773 follow_mount(&nd
->path
.mnt
, &nd
->path
.dentry
);
777 * It's more convoluted than I'd like it to be, but... it's still fairly
778 * small and for now I'd prefer to have fast path as straight as possible.
779 * It _is_ time-critical.
781 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
784 struct vfsmount
*mnt
= nd
->path
.mnt
;
785 struct dentry
*dentry
= __d_lookup(nd
->path
.dentry
, name
);
789 if (dentry
->d_op
&& dentry
->d_op
->d_revalidate
)
790 goto need_revalidate
;
793 path
->dentry
= dentry
;
794 __follow_mount(path
);
798 dentry
= real_lookup(nd
->path
.dentry
, name
, nd
);
804 dentry
= do_revalidate(dentry
, nd
);
812 return PTR_ERR(dentry
);
817 * This is the basic name resolution function, turning a pathname into
818 * the final dentry. We expect 'base' to be positive and a directory.
820 * Returns 0 and nd will have valid dentry and mnt on success.
821 * Returns error and drops reference to input namei data on failure.
823 static int __link_path_walk(const char *name
, struct nameidata
*nd
)
828 unsigned int lookup_flags
= nd
->flags
;
835 inode
= nd
->path
.dentry
->d_inode
;
837 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
839 /* At this point we know we have a real path component. */
845 nd
->flags
|= LOOKUP_CONTINUE
;
846 err
= exec_permission_lite(inode
, nd
);
848 err
= vfs_permission(nd
, MAY_EXEC
);
853 c
= *(const unsigned char *)name
;
855 hash
= init_name_hash();
858 hash
= partial_name_hash(c
, hash
);
859 c
= *(const unsigned char *)name
;
860 } while (c
&& (c
!= '/'));
861 this.len
= name
- (const char *) this.name
;
862 this.hash
= end_name_hash(hash
);
864 /* remove trailing slashes? */
867 while (*++name
== '/');
869 goto last_with_slashes
;
872 * "." and ".." are special - ".." especially so because it has
873 * to be able to know about the current root directory and
874 * parent relationships.
876 if (this.name
[0] == '.') switch (this.len
) {
880 if (this.name
[1] != '.')
883 inode
= nd
->path
.dentry
->d_inode
;
889 * See if the low-level filesystem might want
890 * to use its own hash..
892 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
893 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
898 /* This does the actual lookups.. */
899 err
= do_lookup(nd
, &this, &next
);
904 inode
= next
.dentry
->d_inode
;
911 if (inode
->i_op
->follow_link
) {
912 err
= do_follow_link(&next
, nd
);
916 inode
= nd
->path
.dentry
->d_inode
;
923 path_to_nameidata(&next
, nd
);
925 if (!inode
->i_op
->lookup
)
928 /* here ends the main loop */
931 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
933 /* Clear LOOKUP_CONTINUE iff it was previously unset */
934 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
935 if (lookup_flags
& LOOKUP_PARENT
)
937 if (this.name
[0] == '.') switch (this.len
) {
941 if (this.name
[1] != '.')
944 inode
= nd
->path
.dentry
->d_inode
;
949 if (nd
->path
.dentry
->d_op
&& nd
->path
.dentry
->d_op
->d_hash
) {
950 err
= nd
->path
.dentry
->d_op
->d_hash(nd
->path
.dentry
,
955 err
= do_lookup(nd
, &this, &next
);
958 inode
= next
.dentry
->d_inode
;
959 if ((lookup_flags
& LOOKUP_FOLLOW
)
960 && inode
&& inode
->i_op
&& inode
->i_op
->follow_link
) {
961 err
= do_follow_link(&next
, nd
);
964 inode
= nd
->path
.dentry
->d_inode
;
966 path_to_nameidata(&next
, nd
);
970 if (lookup_flags
& LOOKUP_DIRECTORY
) {
972 if (!inode
->i_op
|| !inode
->i_op
->lookup
)
978 nd
->last_type
= LAST_NORM
;
979 if (this.name
[0] != '.')
982 nd
->last_type
= LAST_DOT
;
983 else if (this.len
== 2 && this.name
[1] == '.')
984 nd
->last_type
= LAST_DOTDOT
;
989 * We bypassed the ordinary revalidation routines.
990 * We may need to check the cached dentry for staleness.
992 if (nd
->path
.dentry
&& nd
->path
.dentry
->d_sb
&&
993 (nd
->path
.dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)) {
995 /* Note: we do not d_invalidate() */
996 if (!nd
->path
.dentry
->d_op
->d_revalidate(
997 nd
->path
.dentry
, nd
))
1003 path_put_conditional(&next
, nd
);
1006 path_put(&nd
->path
);
1012 * Wrapper to retry pathname resolution whenever the underlying
1013 * file system returns an ESTALE.
1015 * Retry the whole path once, forcing real lookup requests
1016 * instead of relying on the dcache.
1018 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1020 struct nameidata save
= *nd
;
1023 /* make sure the stuff we saved doesn't go away */
1024 dget(save
.path
.dentry
);
1025 mntget(save
.path
.mnt
);
1027 result
= __link_path_walk(name
, nd
);
1028 if (result
== -ESTALE
) {
1030 dget(nd
->path
.dentry
);
1031 mntget(nd
->path
.mnt
);
1032 nd
->flags
|= LOOKUP_REVAL
;
1033 result
= __link_path_walk(name
, nd
);
1036 dput(save
.path
.dentry
);
1037 mntput(save
.path
.mnt
);
1042 static int path_walk(const char *name
, struct nameidata
*nd
)
1044 current
->total_link_count
= 0;
1045 return link_path_walk(name
, nd
);
1049 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1050 * everything is done. Returns 0 and drops input nd, if lookup failed;
1052 static int __emul_lookup_dentry(const char *name
, struct nameidata
*nd
)
1054 if (path_walk(name
, nd
))
1055 return 0; /* something went wrong... */
1057 if (!nd
->path
.dentry
->d_inode
||
1058 S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
)) {
1059 struct dentry
*old_dentry
= nd
->path
.dentry
;
1060 struct vfsmount
*old_mnt
= nd
->path
.mnt
;
1061 struct qstr last
= nd
->last
;
1062 int last_type
= nd
->last_type
;
1063 struct fs_struct
*fs
= current
->fs
;
1066 * NAME was not found in alternate root or it's a directory.
1067 * Try to find it in the normal root:
1069 nd
->last_type
= LAST_ROOT
;
1070 read_lock(&fs
->lock
);
1071 nd
->path
.mnt
= mntget(fs
->rootmnt
);
1072 nd
->path
.dentry
= dget(fs
->root
);
1073 read_unlock(&fs
->lock
);
1074 if (path_walk(name
, nd
) == 0) {
1075 if (nd
->path
.dentry
->d_inode
) {
1080 path_put(&nd
->path
);
1082 nd
->path
.dentry
= old_dentry
;
1083 nd
->path
.mnt
= old_mnt
;
1085 nd
->last_type
= last_type
;
1090 void set_fs_altroot(void)
1092 char *emul
= __emul_prefix();
1093 struct nameidata nd
;
1094 struct vfsmount
*mnt
= NULL
, *oldmnt
;
1095 struct dentry
*dentry
= NULL
, *olddentry
;
1097 struct fs_struct
*fs
= current
->fs
;
1101 err
= path_lookup(emul
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
|LOOKUP_NOALT
, &nd
);
1104 dentry
= nd
.path
.dentry
;
1107 write_lock(&fs
->lock
);
1108 oldmnt
= fs
->altrootmnt
;
1109 olddentry
= fs
->altroot
;
1110 fs
->altrootmnt
= mnt
;
1111 fs
->altroot
= dentry
;
1112 write_unlock(&fs
->lock
);
1119 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1120 static int do_path_lookup(int dfd
, const char *name
,
1121 unsigned int flags
, struct nameidata
*nd
)
1126 struct fs_struct
*fs
= current
->fs
;
1128 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1133 read_lock(&fs
->lock
);
1134 if (fs
->altroot
&& !(nd
->flags
& LOOKUP_NOALT
)) {
1135 nd
->path
.mnt
= mntget(fs
->altrootmnt
);
1136 nd
->path
.dentry
= dget(fs
->altroot
);
1137 read_unlock(&fs
->lock
);
1138 if (__emul_lookup_dentry(name
,nd
))
1139 goto out
; /* found in altroot */
1140 read_lock(&fs
->lock
);
1142 nd
->path
.mnt
= mntget(fs
->rootmnt
);
1143 nd
->path
.dentry
= dget(fs
->root
);
1144 read_unlock(&fs
->lock
);
1145 } else if (dfd
== AT_FDCWD
) {
1146 read_lock(&fs
->lock
);
1147 nd
->path
.mnt
= mntget(fs
->pwdmnt
);
1148 nd
->path
.dentry
= dget(fs
->pwd
);
1149 read_unlock(&fs
->lock
);
1151 struct dentry
*dentry
;
1153 file
= fget_light(dfd
, &fput_needed
);
1158 dentry
= file
->f_path
.dentry
;
1161 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1164 retval
= file_permission(file
, MAY_EXEC
);
1168 nd
->path
.mnt
= mntget(file
->f_path
.mnt
);
1169 nd
->path
.dentry
= dget(dentry
);
1171 fput_light(file
, fput_needed
);
1174 retval
= path_walk(name
, nd
);
1176 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1177 nd
->path
.dentry
->d_inode
))
1178 audit_inode(name
, nd
->path
.dentry
);
1183 fput_light(file
, fput_needed
);
1187 int path_lookup(const char *name
, unsigned int flags
,
1188 struct nameidata
*nd
)
1190 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1194 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1195 * @dentry: pointer to dentry of the base directory
1196 * @mnt: pointer to vfs mount of the base directory
1197 * @name: pointer to file name
1198 * @flags: lookup flags
1199 * @nd: pointer to nameidata
1201 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1202 const char *name
, unsigned int flags
,
1203 struct nameidata
*nd
)
1207 /* same as do_path_lookup */
1208 nd
->last_type
= LAST_ROOT
;
1212 nd
->path
.mnt
= mntget(mnt
);
1213 nd
->path
.dentry
= dget(dentry
);
1215 retval
= path_walk(name
, nd
);
1216 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1217 nd
->path
.dentry
->d_inode
))
1218 audit_inode(name
, nd
->path
.dentry
);
1224 static int __path_lookup_intent_open(int dfd
, const char *name
,
1225 unsigned int lookup_flags
, struct nameidata
*nd
,
1226 int open_flags
, int create_mode
)
1228 struct file
*filp
= get_empty_filp();
1233 nd
->intent
.open
.file
= filp
;
1234 nd
->intent
.open
.flags
= open_flags
;
1235 nd
->intent
.open
.create_mode
= create_mode
;
1236 err
= do_path_lookup(dfd
, name
, lookup_flags
|LOOKUP_OPEN
, nd
);
1237 if (IS_ERR(nd
->intent
.open
.file
)) {
1239 err
= PTR_ERR(nd
->intent
.open
.file
);
1240 path_put(&nd
->path
);
1242 } else if (err
!= 0)
1243 release_open_intent(nd
);
1248 * path_lookup_open - lookup a file path with open intent
1249 * @dfd: the directory to use as base, or AT_FDCWD
1250 * @name: pointer to file name
1251 * @lookup_flags: lookup intent flags
1252 * @nd: pointer to nameidata
1253 * @open_flags: open intent flags
1255 int path_lookup_open(int dfd
, const char *name
, unsigned int lookup_flags
,
1256 struct nameidata
*nd
, int open_flags
)
1258 return __path_lookup_intent_open(dfd
, name
, lookup_flags
, nd
,
1263 * path_lookup_create - lookup a file path with open + create intent
1264 * @dfd: the directory to use as base, or AT_FDCWD
1265 * @name: pointer to file name
1266 * @lookup_flags: lookup intent flags
1267 * @nd: pointer to nameidata
1268 * @open_flags: open intent flags
1269 * @create_mode: create intent flags
1271 static int path_lookup_create(int dfd
, const char *name
,
1272 unsigned int lookup_flags
, struct nameidata
*nd
,
1273 int open_flags
, int create_mode
)
1275 return __path_lookup_intent_open(dfd
, name
, lookup_flags
|LOOKUP_CREATE
,
1276 nd
, open_flags
, create_mode
);
1279 int __user_path_lookup_open(const char __user
*name
, unsigned int lookup_flags
,
1280 struct nameidata
*nd
, int open_flags
)
1282 char *tmp
= getname(name
);
1283 int err
= PTR_ERR(tmp
);
1286 err
= __path_lookup_intent_open(AT_FDCWD
, tmp
, lookup_flags
, nd
, open_flags
, 0);
1292 static struct dentry
*__lookup_hash(struct qstr
*name
,
1293 struct dentry
*base
, struct nameidata
*nd
)
1295 struct dentry
*dentry
;
1296 struct inode
*inode
;
1299 inode
= base
->d_inode
;
1302 * See if the low-level filesystem might want
1303 * to use its own hash..
1305 if (base
->d_op
&& base
->d_op
->d_hash
) {
1306 err
= base
->d_op
->d_hash(base
, name
);
1307 dentry
= ERR_PTR(err
);
1312 dentry
= cached_lookup(base
, name
, nd
);
1314 struct dentry
*new = d_alloc(base
, name
);
1315 dentry
= ERR_PTR(-ENOMEM
);
1318 dentry
= inode
->i_op
->lookup(inode
, new, nd
);
1329 * Restricted form of lookup. Doesn't follow links, single-component only,
1330 * needs parent already locked. Doesn't follow mounts.
1333 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1337 err
= permission(nd
->path
.dentry
->d_inode
, MAY_EXEC
, nd
);
1339 return ERR_PTR(err
);
1340 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1343 static int __lookup_one_len(const char *name
, struct qstr
*this,
1344 struct dentry
*base
, int len
)
1354 hash
= init_name_hash();
1356 c
= *(const unsigned char *)name
++;
1357 if (c
== '/' || c
== '\0')
1359 hash
= partial_name_hash(c
, hash
);
1361 this->hash
= end_name_hash(hash
);
1366 * lookup_one_len: filesystem helper to lookup single pathname component
1367 * @name: pathname component to lookup
1368 * @base: base directory to lookup from
1369 * @len: maximum length @len should be interpreted to
1371 * Note that this routine is purely a helper for filesystem useage and should
1372 * not be called by generic code. Also note that by using this function to
1373 * nameidata argument is passed to the filesystem methods and a filesystem
1374 * using this helper needs to be prepared for that.
1376 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1381 err
= __lookup_one_len(name
, &this, base
, len
);
1383 return ERR_PTR(err
);
1385 err
= permission(base
->d_inode
, MAY_EXEC
, NULL
);
1387 return ERR_PTR(err
);
1388 return __lookup_hash(&this, base
, NULL
);
1392 * lookup_one_noperm - bad hack for sysfs
1393 * @name: pathname component to lookup
1394 * @base: base directory to lookup from
1396 * This is a variant of lookup_one_len that doesn't perform any permission
1397 * checks. It's a horrible hack to work around the braindead sysfs
1398 * architecture and should not be used anywhere else.
1400 * DON'T USE THIS FUNCTION EVER, thanks.
1402 struct dentry
*lookup_one_noperm(const char *name
, struct dentry
*base
)
1407 err
= __lookup_one_len(name
, &this, base
, strlen(name
));
1409 return ERR_PTR(err
);
1410 return __lookup_hash(&this, base
, NULL
);
1413 int __user_walk_fd(int dfd
, const char __user
*name
, unsigned flags
,
1414 struct nameidata
*nd
)
1416 char *tmp
= getname(name
);
1417 int err
= PTR_ERR(tmp
);
1420 err
= do_path_lookup(dfd
, tmp
, flags
, nd
);
1426 int __user_walk(const char __user
*name
, unsigned flags
, struct nameidata
*nd
)
1428 return __user_walk_fd(AT_FDCWD
, name
, flags
, nd
);
1432 * It's inline, so penalty for filesystems that don't use sticky bit is
1435 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1437 if (!(dir
->i_mode
& S_ISVTX
))
1439 if (inode
->i_uid
== current
->fsuid
)
1441 if (dir
->i_uid
== current
->fsuid
)
1443 return !capable(CAP_FOWNER
);
1447 * Check whether we can remove a link victim from directory dir, check
1448 * whether the type of victim is right.
1449 * 1. We can't do it if dir is read-only (done in permission())
1450 * 2. We should have write and exec permissions on dir
1451 * 3. We can't remove anything from append-only dir
1452 * 4. We can't do anything with immutable dir (done in permission())
1453 * 5. If the sticky bit on dir is set we should either
1454 * a. be owner of dir, or
1455 * b. be owner of victim, or
1456 * c. have CAP_FOWNER capability
1457 * 6. If the victim is append-only or immutable we can't do antyhing with
1458 * links pointing to it.
1459 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1460 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1461 * 9. We can't remove a root or mountpoint.
1462 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1463 * nfs_async_unlink().
1465 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1469 if (!victim
->d_inode
)
1472 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1473 audit_inode_child(victim
->d_name
.name
, victim
, dir
);
1475 error
= permission(dir
,MAY_WRITE
| MAY_EXEC
, NULL
);
1480 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1481 IS_IMMUTABLE(victim
->d_inode
))
1484 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1486 if (IS_ROOT(victim
))
1488 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1490 if (IS_DEADDIR(dir
))
1492 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1497 /* Check whether we can create an object with dentry child in directory
1499 * 1. We can't do it if child already exists (open has special treatment for
1500 * this case, but since we are inlined it's OK)
1501 * 2. We can't do it if dir is read-only (done in permission())
1502 * 3. We should have write and exec permissions on dir
1503 * 4. We can't do it if dir is immutable (done in permission())
1505 static inline int may_create(struct inode
*dir
, struct dentry
*child
,
1506 struct nameidata
*nd
)
1510 if (IS_DEADDIR(dir
))
1512 return permission(dir
,MAY_WRITE
| MAY_EXEC
, nd
);
1516 * O_DIRECTORY translates into forcing a directory lookup.
1518 static inline int lookup_flags(unsigned int f
)
1520 unsigned long retval
= LOOKUP_FOLLOW
;
1523 retval
&= ~LOOKUP_FOLLOW
;
1525 if (f
& O_DIRECTORY
)
1526 retval
|= LOOKUP_DIRECTORY
;
1532 * p1 and p2 should be directories on the same fs.
1534 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1539 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1543 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1545 for (p
= p1
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1546 if (p
->d_parent
== p2
) {
1547 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1548 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1553 for (p
= p2
; p
->d_parent
!= p
; p
= p
->d_parent
) {
1554 if (p
->d_parent
== p1
) {
1555 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1556 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1561 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1562 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1566 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1568 mutex_unlock(&p1
->d_inode
->i_mutex
);
1570 mutex_unlock(&p2
->d_inode
->i_mutex
);
1571 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1575 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1576 struct nameidata
*nd
)
1578 int error
= may_create(dir
, dentry
, nd
);
1583 if (!dir
->i_op
|| !dir
->i_op
->create
)
1584 return -EACCES
; /* shouldn't it be ENOSYS? */
1587 error
= security_inode_create(dir
, dentry
, mode
);
1591 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1593 fsnotify_create(dir
, dentry
);
1597 int may_open(struct nameidata
*nd
, int acc_mode
, int flag
)
1599 struct dentry
*dentry
= nd
->path
.dentry
;
1600 struct inode
*inode
= dentry
->d_inode
;
1606 if (S_ISLNK(inode
->i_mode
))
1609 if (S_ISDIR(inode
->i_mode
) && (acc_mode
& MAY_WRITE
))
1613 * FIFO's, sockets and device files are special: they don't
1614 * actually live on the filesystem itself, and as such you
1615 * can write to them even if the filesystem is read-only.
1617 if (S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
1619 } else if (S_ISBLK(inode
->i_mode
) || S_ISCHR(inode
->i_mode
)) {
1620 if (nd
->path
.mnt
->mnt_flags
& MNT_NODEV
)
1624 } else if (IS_RDONLY(inode
) && (acc_mode
& MAY_WRITE
))
1627 error
= vfs_permission(nd
, acc_mode
);
1631 * An append-only file must be opened in append mode for writing.
1633 if (IS_APPEND(inode
)) {
1634 if ((flag
& FMODE_WRITE
) && !(flag
& O_APPEND
))
1640 /* O_NOATIME can only be set by the owner or superuser */
1641 if (flag
& O_NOATIME
)
1642 if (!is_owner_or_cap(inode
))
1646 * Ensure there are no outstanding leases on the file.
1648 error
= break_lease(inode
, flag
);
1652 if (flag
& O_TRUNC
) {
1653 error
= get_write_access(inode
);
1658 * Refuse to truncate files with mandatory locks held on them.
1660 error
= locks_verify_locked(inode
);
1664 error
= do_truncate(dentry
, 0,
1665 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1668 put_write_access(inode
);
1672 if (flag
& FMODE_WRITE
)
1678 static int open_namei_create(struct nameidata
*nd
, struct path
*path
,
1682 struct dentry
*dir
= nd
->path
.dentry
;
1684 if (!IS_POSIXACL(dir
->d_inode
))
1685 mode
&= ~current
->fs
->umask
;
1686 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1687 mutex_unlock(&dir
->d_inode
->i_mutex
);
1688 dput(nd
->path
.dentry
);
1689 nd
->path
.dentry
= path
->dentry
;
1692 /* Don't check for write permission, don't truncate */
1693 return may_open(nd
, 0, flag
& ~O_TRUNC
);
1699 * namei for open - this is in fact almost the whole open-routine.
1701 * Note that the low bits of "flag" aren't the same as in the open
1702 * system call - they are 00 - no permissions needed
1703 * 01 - read permission needed
1704 * 10 - write permission needed
1705 * 11 - read/write permissions needed
1706 * which is a lot more logical, and also allows the "no perm" needed
1707 * for symlinks (where the permissions are checked later).
1710 int open_namei(int dfd
, const char *pathname
, int flag
,
1711 int mode
, struct nameidata
*nd
)
1713 int acc_mode
, error
;
1718 acc_mode
= ACC_MODE(flag
);
1720 /* O_TRUNC implies we need access checks for write permissions */
1722 acc_mode
|= MAY_WRITE
;
1724 /* Allow the LSM permission hook to distinguish append
1725 access from general write access. */
1726 if (flag
& O_APPEND
)
1727 acc_mode
|= MAY_APPEND
;
1730 * The simplest case - just a plain lookup.
1732 if (!(flag
& O_CREAT
)) {
1733 error
= path_lookup_open(dfd
, pathname
, lookup_flags(flag
),
1741 * Create - we need to know the parent.
1743 error
= path_lookup_create(dfd
,pathname
,LOOKUP_PARENT
,nd
,flag
,mode
);
1748 * We have the parent and last component. First of all, check
1749 * that we are not asked to creat(2) an obvious directory - that
1753 if (nd
->last_type
!= LAST_NORM
|| nd
->last
.name
[nd
->last
.len
])
1756 dir
= nd
->path
.dentry
;
1757 nd
->flags
&= ~LOOKUP_PARENT
;
1758 mutex_lock(&dir
->d_inode
->i_mutex
);
1759 path
.dentry
= lookup_hash(nd
);
1760 path
.mnt
= nd
->path
.mnt
;
1763 error
= PTR_ERR(path
.dentry
);
1764 if (IS_ERR(path
.dentry
)) {
1765 mutex_unlock(&dir
->d_inode
->i_mutex
);
1769 if (IS_ERR(nd
->intent
.open
.file
)) {
1770 mutex_unlock(&dir
->d_inode
->i_mutex
);
1771 error
= PTR_ERR(nd
->intent
.open
.file
);
1775 /* Negative dentry, just create the file */
1776 if (!path
.dentry
->d_inode
) {
1777 error
= open_namei_create(nd
, &path
, flag
, mode
);
1784 * It already exists.
1786 mutex_unlock(&dir
->d_inode
->i_mutex
);
1787 audit_inode(pathname
, path
.dentry
);
1793 if (__follow_mount(&path
)) {
1795 if (flag
& O_NOFOLLOW
)
1800 if (!path
.dentry
->d_inode
)
1802 if (path
.dentry
->d_inode
->i_op
&& path
.dentry
->d_inode
->i_op
->follow_link
)
1805 path_to_nameidata(&path
, nd
);
1807 if (path
.dentry
->d_inode
&& S_ISDIR(path
.dentry
->d_inode
->i_mode
))
1810 error
= may_open(nd
, acc_mode
, flag
);
1816 path_put_conditional(&path
, nd
);
1818 if (!IS_ERR(nd
->intent
.open
.file
))
1819 release_open_intent(nd
);
1820 path_put(&nd
->path
);
1825 if (flag
& O_NOFOLLOW
)
1828 * This is subtle. Instead of calling do_follow_link() we do the
1829 * thing by hands. The reason is that this way we have zero link_count
1830 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1831 * After that we have the parent and last component, i.e.
1832 * we are in the same situation as after the first path_walk().
1833 * Well, almost - if the last component is normal we get its copy
1834 * stored in nd->last.name and we will have to putname() it when we
1835 * are done. Procfs-like symlinks just set LAST_BIND.
1837 nd
->flags
|= LOOKUP_PARENT
;
1838 error
= security_inode_follow_link(path
.dentry
, nd
);
1841 error
= __do_follow_link(&path
, nd
);
1843 /* Does someone understand code flow here? Or it is only
1844 * me so stupid? Anathema to whoever designed this non-sense
1845 * with "intent.open".
1847 release_open_intent(nd
);
1850 nd
->flags
&= ~LOOKUP_PARENT
;
1851 if (nd
->last_type
== LAST_BIND
)
1854 if (nd
->last_type
!= LAST_NORM
)
1856 if (nd
->last
.name
[nd
->last
.len
]) {
1857 __putname(nd
->last
.name
);
1862 __putname(nd
->last
.name
);
1865 dir
= nd
->path
.dentry
;
1866 mutex_lock(&dir
->d_inode
->i_mutex
);
1867 path
.dentry
= lookup_hash(nd
);
1868 path
.mnt
= nd
->path
.mnt
;
1869 __putname(nd
->last
.name
);
1874 * lookup_create - lookup a dentry, creating it if it doesn't exist
1875 * @nd: nameidata info
1876 * @is_dir: directory flag
1878 * Simple function to lookup and return a dentry and create it
1879 * if it doesn't exist. Is SMP-safe.
1881 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1883 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
1885 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
1887 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1889 * Yucky last component or no last component at all?
1890 * (foo/., foo/.., /////)
1892 if (nd
->last_type
!= LAST_NORM
)
1894 nd
->flags
&= ~LOOKUP_PARENT
;
1895 nd
->flags
|= LOOKUP_CREATE
;
1896 nd
->intent
.open
.flags
= O_EXCL
;
1899 * Do the final lookup.
1901 dentry
= lookup_hash(nd
);
1906 * Special case - lookup gave negative, but... we had foo/bar/
1907 * From the vfs_mknod() POV we just have a negative dentry -
1908 * all is fine. Let's be bastards - you had / on the end, you've
1909 * been asking for (non-existent) directory. -ENOENT for you.
1911 if (!is_dir
&& nd
->last
.name
[nd
->last
.len
] && !dentry
->d_inode
)
1916 dentry
= ERR_PTR(-ENOENT
);
1920 EXPORT_SYMBOL_GPL(lookup_create
);
1922 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1924 int error
= may_create(dir
, dentry
, NULL
);
1929 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
1932 if (!dir
->i_op
|| !dir
->i_op
->mknod
)
1935 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
1940 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
1942 fsnotify_create(dir
, dentry
);
1946 asmlinkage
long sys_mknodat(int dfd
, const char __user
*filename
, int mode
,
1951 struct dentry
* dentry
;
1952 struct nameidata nd
;
1956 tmp
= getname(filename
);
1958 return PTR_ERR(tmp
);
1960 error
= do_path_lookup(dfd
, tmp
, LOOKUP_PARENT
, &nd
);
1963 dentry
= lookup_create(&nd
, 0);
1964 error
= PTR_ERR(dentry
);
1966 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
1967 mode
&= ~current
->fs
->umask
;
1968 if (!IS_ERR(dentry
)) {
1969 switch (mode
& S_IFMT
) {
1970 case 0: case S_IFREG
:
1971 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
1973 case S_IFCHR
: case S_IFBLK
:
1974 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
1975 new_decode_dev(dev
));
1977 case S_IFIFO
: case S_IFSOCK
:
1978 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
1988 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
1996 asmlinkage
long sys_mknod(const char __user
*filename
, int mode
, unsigned dev
)
1998 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2001 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2003 int error
= may_create(dir
, dentry
, NULL
);
2008 if (!dir
->i_op
|| !dir
->i_op
->mkdir
)
2011 mode
&= (S_IRWXUGO
|S_ISVTX
);
2012 error
= security_inode_mkdir(dir
, dentry
, mode
);
2017 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2019 fsnotify_mkdir(dir
, dentry
);
2023 asmlinkage
long sys_mkdirat(int dfd
, const char __user
*pathname
, int mode
)
2027 struct dentry
*dentry
;
2028 struct nameidata nd
;
2030 tmp
= getname(pathname
);
2031 error
= PTR_ERR(tmp
);
2035 error
= do_path_lookup(dfd
, tmp
, LOOKUP_PARENT
, &nd
);
2038 dentry
= lookup_create(&nd
, 1);
2039 error
= PTR_ERR(dentry
);
2043 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2044 mode
&= ~current
->fs
->umask
;
2045 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2048 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2056 asmlinkage
long sys_mkdir(const char __user
*pathname
, int mode
)
2058 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2062 * We try to drop the dentry early: we should have
2063 * a usage count of 2 if we're the only user of this
2064 * dentry, and if that is true (possibly after pruning
2065 * the dcache), then we drop the dentry now.
2067 * A low-level filesystem can, if it choses, legally
2070 * if (!d_unhashed(dentry))
2073 * if it cannot handle the case of removing a directory
2074 * that is still in use by something else..
2076 void dentry_unhash(struct dentry
*dentry
)
2079 shrink_dcache_parent(dentry
);
2080 spin_lock(&dcache_lock
);
2081 spin_lock(&dentry
->d_lock
);
2082 if (atomic_read(&dentry
->d_count
) == 2)
2084 spin_unlock(&dentry
->d_lock
);
2085 spin_unlock(&dcache_lock
);
2088 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2090 int error
= may_delete(dir
, dentry
, 1);
2095 if (!dir
->i_op
|| !dir
->i_op
->rmdir
)
2100 mutex_lock(&dentry
->d_inode
->i_mutex
);
2101 dentry_unhash(dentry
);
2102 if (d_mountpoint(dentry
))
2105 error
= security_inode_rmdir(dir
, dentry
);
2107 error
= dir
->i_op
->rmdir(dir
, dentry
);
2109 dentry
->d_inode
->i_flags
|= S_DEAD
;
2112 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2121 static long do_rmdir(int dfd
, const char __user
*pathname
)
2125 struct dentry
*dentry
;
2126 struct nameidata nd
;
2128 name
= getname(pathname
);
2130 return PTR_ERR(name
);
2132 error
= do_path_lookup(dfd
, name
, LOOKUP_PARENT
, &nd
);
2136 switch(nd
.last_type
) {
2147 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2148 dentry
= lookup_hash(&nd
);
2149 error
= PTR_ERR(dentry
);
2152 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2155 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2163 asmlinkage
long sys_rmdir(const char __user
*pathname
)
2165 return do_rmdir(AT_FDCWD
, pathname
);
2168 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2170 int error
= may_delete(dir
, dentry
, 0);
2175 if (!dir
->i_op
|| !dir
->i_op
->unlink
)
2180 mutex_lock(&dentry
->d_inode
->i_mutex
);
2181 if (d_mountpoint(dentry
))
2184 error
= security_inode_unlink(dir
, dentry
);
2186 error
= dir
->i_op
->unlink(dir
, dentry
);
2188 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2190 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2191 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2192 fsnotify_link_count(dentry
->d_inode
);
2200 * Make sure that the actual truncation of the file will occur outside its
2201 * directory's i_mutex. Truncate can take a long time if there is a lot of
2202 * writeout happening, and we don't want to prevent access to the directory
2203 * while waiting on the I/O.
2205 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2209 struct dentry
*dentry
;
2210 struct nameidata nd
;
2211 struct inode
*inode
= NULL
;
2213 name
= getname(pathname
);
2215 return PTR_ERR(name
);
2217 error
= do_path_lookup(dfd
, name
, LOOKUP_PARENT
, &nd
);
2221 if (nd
.last_type
!= LAST_NORM
)
2223 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2224 dentry
= lookup_hash(&nd
);
2225 error
= PTR_ERR(dentry
);
2226 if (!IS_ERR(dentry
)) {
2227 /* Why not before? Because we want correct error value */
2228 if (nd
.last
.name
[nd
.last
.len
])
2230 inode
= dentry
->d_inode
;
2232 atomic_inc(&inode
->i_count
);
2233 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2237 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2239 iput(inode
); /* truncate the inode here */
2247 error
= !dentry
->d_inode
? -ENOENT
:
2248 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2252 asmlinkage
long sys_unlinkat(int dfd
, const char __user
*pathname
, int flag
)
2254 if ((flag
& ~AT_REMOVEDIR
) != 0)
2257 if (flag
& AT_REMOVEDIR
)
2258 return do_rmdir(dfd
, pathname
);
2260 return do_unlinkat(dfd
, pathname
);
2263 asmlinkage
long sys_unlink(const char __user
*pathname
)
2265 return do_unlinkat(AT_FDCWD
, pathname
);
2268 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
, int mode
)
2270 int error
= may_create(dir
, dentry
, NULL
);
2275 if (!dir
->i_op
|| !dir
->i_op
->symlink
)
2278 error
= security_inode_symlink(dir
, dentry
, oldname
);
2283 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2285 fsnotify_create(dir
, dentry
);
2289 asmlinkage
long sys_symlinkat(const char __user
*oldname
,
2290 int newdfd
, const char __user
*newname
)
2295 struct dentry
*dentry
;
2296 struct nameidata nd
;
2298 from
= getname(oldname
);
2300 return PTR_ERR(from
);
2301 to
= getname(newname
);
2302 error
= PTR_ERR(to
);
2306 error
= do_path_lookup(newdfd
, to
, LOOKUP_PARENT
, &nd
);
2309 dentry
= lookup_create(&nd
, 0);
2310 error
= PTR_ERR(dentry
);
2314 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
, S_IALLUGO
);
2317 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2326 asmlinkage
long sys_symlink(const char __user
*oldname
, const char __user
*newname
)
2328 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2331 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2333 struct inode
*inode
= old_dentry
->d_inode
;
2339 error
= may_create(dir
, new_dentry
, NULL
);
2343 if (dir
->i_sb
!= inode
->i_sb
)
2347 * A link to an append-only or immutable file cannot be created.
2349 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2351 if (!dir
->i_op
|| !dir
->i_op
->link
)
2353 if (S_ISDIR(old_dentry
->d_inode
->i_mode
))
2356 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2360 mutex_lock(&old_dentry
->d_inode
->i_mutex
);
2362 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2363 mutex_unlock(&old_dentry
->d_inode
->i_mutex
);
2365 fsnotify_link(dir
, old_dentry
->d_inode
, new_dentry
);
2370 * Hardlinks are often used in delicate situations. We avoid
2371 * security-related surprises by not following symlinks on the
2374 * We don't follow them on the oldname either to be compatible
2375 * with linux 2.0, and to avoid hard-linking to directories
2376 * and other special files. --ADM
2378 asmlinkage
long sys_linkat(int olddfd
, const char __user
*oldname
,
2379 int newdfd
, const char __user
*newname
,
2382 struct dentry
*new_dentry
;
2383 struct nameidata nd
, old_nd
;
2387 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2390 to
= getname(newname
);
2394 error
= __user_walk_fd(olddfd
, oldname
,
2395 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2399 error
= do_path_lookup(newdfd
, to
, LOOKUP_PARENT
, &nd
);
2403 if (old_nd
.path
.mnt
!= nd
.path
.mnt
)
2405 new_dentry
= lookup_create(&nd
, 0);
2406 error
= PTR_ERR(new_dentry
);
2407 if (IS_ERR(new_dentry
))
2409 error
= vfs_link(old_nd
.path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2412 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2416 path_put(&old_nd
.path
);
2423 asmlinkage
long sys_link(const char __user
*oldname
, const char __user
*newname
)
2425 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
2429 * The worst of all namespace operations - renaming directory. "Perverted"
2430 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2432 * a) we can get into loop creation. Check is done in is_subdir().
2433 * b) race potential - two innocent renames can create a loop together.
2434 * That's where 4.4 screws up. Current fix: serialization on
2435 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2437 * c) we have to lock _three_ objects - parents and victim (if it exists).
2438 * And that - after we got ->i_mutex on parents (until then we don't know
2439 * whether the target exists). Solution: try to be smart with locking
2440 * order for inodes. We rely on the fact that tree topology may change
2441 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2442 * move will be locked. Thus we can rank directories by the tree
2443 * (ancestors first) and rank all non-directories after them.
2444 * That works since everybody except rename does "lock parent, lookup,
2445 * lock child" and rename is under ->s_vfs_rename_mutex.
2446 * HOWEVER, it relies on the assumption that any object with ->lookup()
2447 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2448 * we'd better make sure that there's no link(2) for them.
2449 * d) some filesystems don't support opened-but-unlinked directories,
2450 * either because of layout or because they are not ready to deal with
2451 * all cases correctly. The latter will be fixed (taking this sort of
2452 * stuff into VFS), but the former is not going away. Solution: the same
2453 * trick as in rmdir().
2454 * e) conversion from fhandle to dentry may come in the wrong moment - when
2455 * we are removing the target. Solution: we will have to grab ->i_mutex
2456 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2457 * ->i_mutex on parents, which works but leads to some truely excessive
2460 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
2461 struct inode
*new_dir
, struct dentry
*new_dentry
)
2464 struct inode
*target
;
2467 * If we are going to change the parent - check write permissions,
2468 * we'll need to flip '..'.
2470 if (new_dir
!= old_dir
) {
2471 error
= permission(old_dentry
->d_inode
, MAY_WRITE
, NULL
);
2476 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2480 target
= new_dentry
->d_inode
;
2482 mutex_lock(&target
->i_mutex
);
2483 dentry_unhash(new_dentry
);
2485 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2488 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2491 target
->i_flags
|= S_DEAD
;
2492 mutex_unlock(&target
->i_mutex
);
2493 if (d_unhashed(new_dentry
))
2494 d_rehash(new_dentry
);
2498 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2499 d_move(old_dentry
,new_dentry
);
2503 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
2504 struct inode
*new_dir
, struct dentry
*new_dentry
)
2506 struct inode
*target
;
2509 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2514 target
= new_dentry
->d_inode
;
2516 mutex_lock(&target
->i_mutex
);
2517 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
2520 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2522 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
2523 d_move(old_dentry
, new_dentry
);
2526 mutex_unlock(&target
->i_mutex
);
2531 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2532 struct inode
*new_dir
, struct dentry
*new_dentry
)
2535 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2536 const char *old_name
;
2538 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
2541 error
= may_delete(old_dir
, old_dentry
, is_dir
);
2545 if (!new_dentry
->d_inode
)
2546 error
= may_create(new_dir
, new_dentry
, NULL
);
2548 error
= may_delete(new_dir
, new_dentry
, is_dir
);
2552 if (!old_dir
->i_op
|| !old_dir
->i_op
->rename
)
2555 DQUOT_INIT(old_dir
);
2556 DQUOT_INIT(new_dir
);
2558 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
2561 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
2563 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
2565 const char *new_name
= old_dentry
->d_name
.name
;
2566 fsnotify_move(old_dir
, new_dir
, old_name
, new_name
, is_dir
,
2567 new_dentry
->d_inode
, old_dentry
);
2569 fsnotify_oldname_free(old_name
);
2574 static int do_rename(int olddfd
, const char *oldname
,
2575 int newdfd
, const char *newname
)
2578 struct dentry
* old_dir
, * new_dir
;
2579 struct dentry
* old_dentry
, *new_dentry
;
2580 struct dentry
* trap
;
2581 struct nameidata oldnd
, newnd
;
2583 error
= do_path_lookup(olddfd
, oldname
, LOOKUP_PARENT
, &oldnd
);
2587 error
= do_path_lookup(newdfd
, newname
, LOOKUP_PARENT
, &newnd
);
2592 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
2595 old_dir
= oldnd
.path
.dentry
;
2597 if (oldnd
.last_type
!= LAST_NORM
)
2600 new_dir
= newnd
.path
.dentry
;
2601 if (newnd
.last_type
!= LAST_NORM
)
2604 trap
= lock_rename(new_dir
, old_dir
);
2606 old_dentry
= lookup_hash(&oldnd
);
2607 error
= PTR_ERR(old_dentry
);
2608 if (IS_ERR(old_dentry
))
2610 /* source must exist */
2612 if (!old_dentry
->d_inode
)
2614 /* unless the source is a directory trailing slashes give -ENOTDIR */
2615 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
2617 if (oldnd
.last
.name
[oldnd
.last
.len
])
2619 if (newnd
.last
.name
[newnd
.last
.len
])
2622 /* source should not be ancestor of target */
2624 if (old_dentry
== trap
)
2626 new_dentry
= lookup_hash(&newnd
);
2627 error
= PTR_ERR(new_dentry
);
2628 if (IS_ERR(new_dentry
))
2630 /* target should not be an ancestor of source */
2632 if (new_dentry
== trap
)
2635 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
2636 new_dir
->d_inode
, new_dentry
);
2642 unlock_rename(new_dir
, old_dir
);
2644 path_put(&newnd
.path
);
2646 path_put(&oldnd
.path
);
2651 asmlinkage
long sys_renameat(int olddfd
, const char __user
*oldname
,
2652 int newdfd
, const char __user
*newname
)
2658 from
= getname(oldname
);
2660 return PTR_ERR(from
);
2661 to
= getname(newname
);
2662 error
= PTR_ERR(to
);
2664 error
= do_rename(olddfd
, from
, newdfd
, to
);
2671 asmlinkage
long sys_rename(const char __user
*oldname
, const char __user
*newname
)
2673 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
2676 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
2680 len
= PTR_ERR(link
);
2685 if (len
> (unsigned) buflen
)
2687 if (copy_to_user(buffer
, link
, len
))
2694 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2695 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2696 * using) it for any given inode is up to filesystem.
2698 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2700 struct nameidata nd
;
2704 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
2705 if (!IS_ERR(cookie
)) {
2706 int res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
2707 if (dentry
->d_inode
->i_op
->put_link
)
2708 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
2709 cookie
= ERR_PTR(res
);
2711 return PTR_ERR(cookie
);
2714 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
2716 return __vfs_follow_link(nd
, link
);
2719 /* get the link contents into pagecache */
2720 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
2723 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
2724 page
= read_mapping_page(mapping
, 0, NULL
);
2731 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
2733 struct page
*page
= NULL
;
2734 char *s
= page_getlink(dentry
, &page
);
2735 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
2738 page_cache_release(page
);
2743 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
2745 struct page
*page
= NULL
;
2746 nd_set_link(nd
, page_getlink(dentry
, &page
));
2750 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2752 struct page
*page
= cookie
;
2756 page_cache_release(page
);
2760 int __page_symlink(struct inode
*inode
, const char *symname
, int len
,
2763 struct address_space
*mapping
= inode
->i_mapping
;
2770 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
2771 AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
2775 kaddr
= kmap_atomic(page
, KM_USER0
);
2776 memcpy(kaddr
, symname
, len
-1);
2777 kunmap_atomic(kaddr
, KM_USER0
);
2779 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
2786 mark_inode_dirty(inode
);
2792 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
2794 return __page_symlink(inode
, symname
, len
,
2795 mapping_gfp_mask(inode
->i_mapping
));
2798 const struct inode_operations page_symlink_inode_operations
= {
2799 .readlink
= generic_readlink
,
2800 .follow_link
= page_follow_link_light
,
2801 .put_link
= page_put_link
,
2804 EXPORT_SYMBOL(__user_walk
);
2805 EXPORT_SYMBOL(__user_walk_fd
);
2806 EXPORT_SYMBOL(follow_down
);
2807 EXPORT_SYMBOL(follow_up
);
2808 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
2809 EXPORT_SYMBOL(getname
);
2810 EXPORT_SYMBOL(lock_rename
);
2811 EXPORT_SYMBOL(lookup_one_len
);
2812 EXPORT_SYMBOL(page_follow_link_light
);
2813 EXPORT_SYMBOL(page_put_link
);
2814 EXPORT_SYMBOL(page_readlink
);
2815 EXPORT_SYMBOL(__page_symlink
);
2816 EXPORT_SYMBOL(page_symlink
);
2817 EXPORT_SYMBOL(page_symlink_inode_operations
);
2818 EXPORT_SYMBOL(path_lookup
);
2819 EXPORT_SYMBOL(vfs_path_lookup
);
2820 EXPORT_SYMBOL(permission
);
2821 EXPORT_SYMBOL(vfs_permission
);
2822 EXPORT_SYMBOL(file_permission
);
2823 EXPORT_SYMBOL(unlock_rename
);
2824 EXPORT_SYMBOL(vfs_create
);
2825 EXPORT_SYMBOL(vfs_follow_link
);
2826 EXPORT_SYMBOL(vfs_link
);
2827 EXPORT_SYMBOL(vfs_mkdir
);
2828 EXPORT_SYMBOL(vfs_mknod
);
2829 EXPORT_SYMBOL(generic_permission
);
2830 EXPORT_SYMBOL(vfs_readlink
);
2831 EXPORT_SYMBOL(vfs_rename
);
2832 EXPORT_SYMBOL(vfs_rmdir
);
2833 EXPORT_SYMBOL(vfs_symlink
);
2834 EXPORT_SYMBOL(vfs_unlink
);
2835 EXPORT_SYMBOL(dentry_unhash
);
2836 EXPORT_SYMBOL(generic_readlink
);