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>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
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 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
281 retval = devcgroup_inode_permission(inode, mask);
285 return security_inode_permission(inode,
286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
294 * Used to check for read/write/execute permissions on an already opened
298 * Do not use this function in new code. All access checks should
299 * be done using inode_permission().
301 int file_permission(struct file *file, int mask)
303 return inode_permission(file->f_path.dentry->d_inode, mask);
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
324 int get_write_access(struct inode * inode)
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
337 int deny_write_access(struct file * file)
339 struct inode *inode = file->f_path.dentry->d_inode;
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
353 * path_get - get a reference to a path
354 * @path: path to get the reference to
356 * Given a path increment the reference count to the dentry and the vfsmount.
358 void path_get(struct path *path)
363 EXPORT_SYMBOL(path_get);
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 * Short-cut version of permission(), for calling by
416 * path_walk(), when dcache lock is held. Combines parts
417 * of permission() and generic_permission(), and tests ONLY for
418 * MAY_EXEC permission.
420 * If appropriate, check DAC only. If not appropriate, or
421 * short-cut DAC fails, then call permission() to do more
422 * complete permission check.
424 static int exec_permission_lite(struct inode *inode)
428 if (inode->i_op->permission) {
429 ret = inode->i_op->permission(inode, MAY_EXEC);
434 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
438 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
443 return security_inode_permission(inode, MAY_EXEC);
446 static __always_inline void set_root(struct nameidata *nd)
449 struct fs_struct *fs = current->fs;
450 read_lock(&fs->lock);
453 read_unlock(&fs->lock);
457 static int link_path_walk(const char *, struct nameidata *);
459 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
473 res = link_path_walk(link, nd);
474 if (nd->depth || res || nd->last_type!=LAST_NORM)
477 * If it is an iterative symlinks resolution in open_namei() we
478 * have to copy the last component. And all that crap because of
479 * bloody create() on broken symlinks. Furrfu...
482 if (unlikely(!name)) {
486 strcpy(name, nd->last.name);
487 nd->last.name = name;
491 return PTR_ERR(link);
494 static void path_put_conditional(struct path *path, struct nameidata *nd)
497 if (path->mnt != nd->path.mnt)
501 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
503 dput(nd->path.dentry);
504 if (nd->path.mnt != path->mnt)
505 mntput(nd->path.mnt);
506 nd->path.mnt = path->mnt;
507 nd->path.dentry = path->dentry;
510 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
514 struct dentry *dentry = path->dentry;
516 touch_atime(path->mnt, dentry);
517 nd_set_link(nd, NULL);
519 if (path->mnt != nd->path.mnt) {
520 path_to_nameidata(path, nd);
524 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
525 error = PTR_ERR(cookie);
526 if (!IS_ERR(cookie)) {
527 char *s = nd_get_link(nd);
530 error = __vfs_follow_link(nd, s);
531 if (dentry->d_inode->i_op->put_link)
532 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
538 * This limits recursive symlink follows to 8, while
539 * limiting consecutive symlinks to 40.
541 * Without that kind of total limit, nasty chains of consecutive
542 * symlinks can cause almost arbitrarily long lookups.
544 static inline int do_follow_link(struct path *path, struct nameidata *nd)
547 if (current->link_count >= MAX_NESTED_LINKS)
549 if (current->total_link_count >= 40)
551 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
553 err = security_inode_follow_link(path->dentry, nd);
556 current->link_count++;
557 current->total_link_count++;
559 err = __do_follow_link(path, nd);
561 current->link_count--;
565 path_put_conditional(path, nd);
570 int follow_up(struct path *path)
572 struct vfsmount *parent;
573 struct dentry *mountpoint;
574 spin_lock(&vfsmount_lock);
575 parent = path->mnt->mnt_parent;
576 if (parent == path->mnt) {
577 spin_unlock(&vfsmount_lock);
581 mountpoint = dget(path->mnt->mnt_mountpoint);
582 spin_unlock(&vfsmount_lock);
584 path->dentry = mountpoint;
590 /* no need for dcache_lock, as serialization is taken care in
593 static int __follow_mount(struct path *path)
596 while (d_mountpoint(path->dentry)) {
597 struct vfsmount *mounted = lookup_mnt(path);
604 path->dentry = dget(mounted->mnt_root);
610 static void follow_mount(struct path *path)
612 while (d_mountpoint(path->dentry)) {
613 struct vfsmount *mounted = lookup_mnt(path);
619 path->dentry = dget(mounted->mnt_root);
623 /* no need for dcache_lock, as serialization is taken care in
626 int follow_down(struct path *path)
628 struct vfsmount *mounted;
630 mounted = lookup_mnt(path);
635 path->dentry = dget(mounted->mnt_root);
641 static __always_inline void follow_dotdot(struct nameidata *nd)
646 struct vfsmount *parent;
647 struct dentry *old = nd->path.dentry;
649 if (nd->path.dentry == nd->root.dentry &&
650 nd->path.mnt == nd->root.mnt) {
653 spin_lock(&dcache_lock);
654 if (nd->path.dentry != nd->path.mnt->mnt_root) {
655 nd->path.dentry = dget(nd->path.dentry->d_parent);
656 spin_unlock(&dcache_lock);
660 spin_unlock(&dcache_lock);
661 spin_lock(&vfsmount_lock);
662 parent = nd->path.mnt->mnt_parent;
663 if (parent == nd->path.mnt) {
664 spin_unlock(&vfsmount_lock);
668 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
669 spin_unlock(&vfsmount_lock);
671 mntput(nd->path.mnt);
672 nd->path.mnt = parent;
674 follow_mount(&nd->path);
678 * It's more convoluted than I'd like it to be, but... it's still fairly
679 * small and for now I'd prefer to have fast path as straight as possible.
680 * It _is_ time-critical.
682 static int do_lookup(struct nameidata *nd, struct qstr *name,
685 struct vfsmount *mnt = nd->path.mnt;
686 struct dentry *dentry, *parent;
689 * See if the low-level filesystem might want
690 * to use its own hash..
692 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
693 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
698 dentry = __d_lookup(nd->path.dentry, name);
701 if (dentry->d_op && dentry->d_op->d_revalidate)
702 goto need_revalidate;
705 path->dentry = dentry;
706 __follow_mount(path);
710 parent = nd->path.dentry;
711 dir = parent->d_inode;
713 mutex_lock(&dir->i_mutex);
715 * First re-do the cached lookup just in case it was created
716 * while we waited for the directory semaphore..
718 * FIXME! This could use version numbering or similar to
719 * avoid unnecessary cache lookups.
721 * The "dcache_lock" is purely to protect the RCU list walker
722 * from concurrent renames at this point (we mustn't get false
723 * negatives from the RCU list walk here, unlike the optimistic
726 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
728 dentry = d_lookup(parent, name);
732 /* Don't create child dentry for a dead directory. */
733 dentry = ERR_PTR(-ENOENT);
737 new = d_alloc(parent, name);
738 dentry = ERR_PTR(-ENOMEM);
740 dentry = dir->i_op->lookup(dir, new, nd);
747 mutex_unlock(&dir->i_mutex);
754 * Uhhuh! Nasty case: the cache was re-populated while
755 * we waited on the semaphore. Need to revalidate.
757 mutex_unlock(&dir->i_mutex);
758 if (dentry->d_op && dentry->d_op->d_revalidate) {
759 dentry = do_revalidate(dentry, nd);
761 dentry = ERR_PTR(-ENOENT);
768 dentry = do_revalidate(dentry, nd);
776 return PTR_ERR(dentry);
781 * This is the basic name resolution function, turning a pathname into
782 * the final dentry. We expect 'base' to be positive and a directory.
784 * Returns 0 and nd will have valid dentry and mnt on success.
785 * Returns error and drops reference to input namei data on failure.
787 static int link_path_walk(const char *name, struct nameidata *nd)
792 unsigned int lookup_flags = nd->flags;
799 inode = nd->path.dentry->d_inode;
801 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
803 /* At this point we know we have a real path component. */
809 nd->flags |= LOOKUP_CONTINUE;
810 err = exec_permission_lite(inode);
815 c = *(const unsigned char *)name;
817 hash = init_name_hash();
820 hash = partial_name_hash(c, hash);
821 c = *(const unsigned char *)name;
822 } while (c && (c != '/'));
823 this.len = name - (const char *) this.name;
824 this.hash = end_name_hash(hash);
826 /* remove trailing slashes? */
829 while (*++name == '/');
831 goto last_with_slashes;
834 * "." and ".." are special - ".." especially so because it has
835 * to be able to know about the current root directory and
836 * parent relationships.
838 if (this.name[0] == '.') switch (this.len) {
842 if (this.name[1] != '.')
845 inode = nd->path.dentry->d_inode;
850 /* This does the actual lookups.. */
851 err = do_lookup(nd, &this, &next);
856 inode = next.dentry->d_inode;
860 if (inode->i_op->follow_link) {
861 err = do_follow_link(&next, nd);
865 inode = nd->path.dentry->d_inode;
869 path_to_nameidata(&next, nd);
871 if (!inode->i_op->lookup)
874 /* here ends the main loop */
877 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
879 /* Clear LOOKUP_CONTINUE iff it was previously unset */
880 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
881 if (lookup_flags & LOOKUP_PARENT)
883 if (this.name[0] == '.') switch (this.len) {
887 if (this.name[1] != '.')
890 inode = nd->path.dentry->d_inode;
895 err = do_lookup(nd, &this, &next);
898 inode = next.dentry->d_inode;
899 if ((lookup_flags & LOOKUP_FOLLOW)
900 && inode && inode->i_op->follow_link) {
901 err = do_follow_link(&next, nd);
904 inode = nd->path.dentry->d_inode;
906 path_to_nameidata(&next, nd);
910 if (lookup_flags & LOOKUP_DIRECTORY) {
912 if (!inode->i_op->lookup)
918 nd->last_type = LAST_NORM;
919 if (this.name[0] != '.')
922 nd->last_type = LAST_DOT;
923 else if (this.len == 2 && this.name[1] == '.')
924 nd->last_type = LAST_DOTDOT;
929 * We bypassed the ordinary revalidation routines.
930 * We may need to check the cached dentry for staleness.
932 if (nd->path.dentry && nd->path.dentry->d_sb &&
933 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
935 /* Note: we do not d_invalidate() */
936 if (!nd->path.dentry->d_op->d_revalidate(
937 nd->path.dentry, nd))
943 path_put_conditional(&next, nd);
951 static int path_walk(const char *name, struct nameidata *nd)
953 struct path save = nd->path;
956 current->total_link_count = 0;
958 /* make sure the stuff we saved doesn't go away */
961 result = link_path_walk(name, nd);
962 if (result == -ESTALE) {
963 /* nd->path had been dropped */
964 current->total_link_count = 0;
967 nd->flags |= LOOKUP_REVAL;
968 result = link_path_walk(name, nd);
976 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
982 nd->last_type = LAST_ROOT; /* if there are only slashes... */
991 } else if (dfd == AT_FDCWD) {
992 struct fs_struct *fs = current->fs;
993 read_lock(&fs->lock);
996 read_unlock(&fs->lock);
998 struct dentry *dentry;
1000 file = fget_light(dfd, &fput_needed);
1005 dentry = file->f_path.dentry;
1008 if (!S_ISDIR(dentry->d_inode->i_mode))
1011 retval = file_permission(file, MAY_EXEC);
1015 nd->path = file->f_path;
1016 path_get(&file->f_path);
1018 fput_light(file, fput_needed);
1023 fput_light(file, fput_needed);
1028 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1029 static int do_path_lookup(int dfd, const char *name,
1030 unsigned int flags, struct nameidata *nd)
1032 int retval = path_init(dfd, name, flags, nd);
1034 retval = path_walk(name, nd);
1035 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1036 nd->path.dentry->d_inode))
1037 audit_inode(name, nd->path.dentry);
1039 path_put(&nd->root);
1040 nd->root.mnt = NULL;
1045 int path_lookup(const char *name, unsigned int flags,
1046 struct nameidata *nd)
1048 return do_path_lookup(AT_FDCWD, name, flags, nd);
1051 int kern_path(const char *name, unsigned int flags, struct path *path)
1053 struct nameidata nd;
1054 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1061 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1062 * @dentry: pointer to dentry of the base directory
1063 * @mnt: pointer to vfs mount of the base directory
1064 * @name: pointer to file name
1065 * @flags: lookup flags
1066 * @nd: pointer to nameidata
1068 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1069 const char *name, unsigned int flags,
1070 struct nameidata *nd)
1074 /* same as do_path_lookup */
1075 nd->last_type = LAST_ROOT;
1079 nd->path.dentry = dentry;
1081 path_get(&nd->path);
1082 nd->root = nd->path;
1083 path_get(&nd->root);
1085 retval = path_walk(name, nd);
1086 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1087 nd->path.dentry->d_inode))
1088 audit_inode(name, nd->path.dentry);
1090 path_put(&nd->root);
1091 nd->root.mnt = NULL;
1096 static struct dentry *__lookup_hash(struct qstr *name,
1097 struct dentry *base, struct nameidata *nd)
1099 struct dentry *dentry;
1100 struct inode *inode;
1103 inode = base->d_inode;
1106 * See if the low-level filesystem might want
1107 * to use its own hash..
1109 if (base->d_op && base->d_op->d_hash) {
1110 err = base->d_op->d_hash(base, name);
1111 dentry = ERR_PTR(err);
1116 dentry = __d_lookup(base, name);
1118 /* lockess __d_lookup may fail due to concurrent d_move()
1119 * in some unrelated directory, so try with d_lookup
1122 dentry = d_lookup(base, name);
1124 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1125 dentry = do_revalidate(dentry, nd);
1130 /* Don't create child dentry for a dead directory. */
1131 dentry = ERR_PTR(-ENOENT);
1132 if (IS_DEADDIR(inode))
1135 new = d_alloc(base, name);
1136 dentry = ERR_PTR(-ENOMEM);
1139 dentry = inode->i_op->lookup(inode, new, nd);
1150 * Restricted form of lookup. Doesn't follow links, single-component only,
1151 * needs parent already locked. Doesn't follow mounts.
1154 static struct dentry *lookup_hash(struct nameidata *nd)
1158 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1160 return ERR_PTR(err);
1161 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1164 static int __lookup_one_len(const char *name, struct qstr *this,
1165 struct dentry *base, int len)
1175 hash = init_name_hash();
1177 c = *(const unsigned char *)name++;
1178 if (c == '/' || c == '\0')
1180 hash = partial_name_hash(c, hash);
1182 this->hash = end_name_hash(hash);
1187 * lookup_one_len - filesystem helper to lookup single pathname component
1188 * @name: pathname component to lookup
1189 * @base: base directory to lookup from
1190 * @len: maximum length @len should be interpreted to
1192 * Note that this routine is purely a helper for filesystem usage and should
1193 * not be called by generic code. Also note that by using this function the
1194 * nameidata argument is passed to the filesystem methods and a filesystem
1195 * using this helper needs to be prepared for that.
1197 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1202 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1204 err = __lookup_one_len(name, &this, base, len);
1206 return ERR_PTR(err);
1208 err = inode_permission(base->d_inode, MAY_EXEC);
1210 return ERR_PTR(err);
1211 return __lookup_hash(&this, base, NULL);
1214 int user_path_at(int dfd, const char __user *name, unsigned flags,
1217 struct nameidata nd;
1218 char *tmp = getname(name);
1219 int err = PTR_ERR(tmp);
1222 BUG_ON(flags & LOOKUP_PARENT);
1224 err = do_path_lookup(dfd, tmp, flags, &nd);
1232 static int user_path_parent(int dfd, const char __user *path,
1233 struct nameidata *nd, char **name)
1235 char *s = getname(path);
1241 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1251 * It's inline, so penalty for filesystems that don't use sticky bit is
1254 static inline int check_sticky(struct inode *dir, struct inode *inode)
1256 uid_t fsuid = current_fsuid();
1258 if (!(dir->i_mode & S_ISVTX))
1260 if (inode->i_uid == fsuid)
1262 if (dir->i_uid == fsuid)
1264 return !capable(CAP_FOWNER);
1268 * Check whether we can remove a link victim from directory dir, check
1269 * whether the type of victim is right.
1270 * 1. We can't do it if dir is read-only (done in permission())
1271 * 2. We should have write and exec permissions on dir
1272 * 3. We can't remove anything from append-only dir
1273 * 4. We can't do anything with immutable dir (done in permission())
1274 * 5. If the sticky bit on dir is set we should either
1275 * a. be owner of dir, or
1276 * b. be owner of victim, or
1277 * c. have CAP_FOWNER capability
1278 * 6. If the victim is append-only or immutable we can't do antyhing with
1279 * links pointing to it.
1280 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1281 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1282 * 9. We can't remove a root or mountpoint.
1283 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1284 * nfs_async_unlink().
1286 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1290 if (!victim->d_inode)
1293 BUG_ON(victim->d_parent->d_inode != dir);
1294 audit_inode_child(victim->d_name.name, victim, dir);
1296 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1301 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1302 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1305 if (!S_ISDIR(victim->d_inode->i_mode))
1307 if (IS_ROOT(victim))
1309 } else if (S_ISDIR(victim->d_inode->i_mode))
1311 if (IS_DEADDIR(dir))
1313 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1318 /* Check whether we can create an object with dentry child in directory
1320 * 1. We can't do it if child already exists (open has special treatment for
1321 * this case, but since we are inlined it's OK)
1322 * 2. We can't do it if dir is read-only (done in permission())
1323 * 3. We should have write and exec permissions on dir
1324 * 4. We can't do it if dir is immutable (done in permission())
1326 static inline int may_create(struct inode *dir, struct dentry *child)
1330 if (IS_DEADDIR(dir))
1332 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1336 * O_DIRECTORY translates into forcing a directory lookup.
1338 static inline int lookup_flags(unsigned int f)
1340 unsigned long retval = LOOKUP_FOLLOW;
1343 retval &= ~LOOKUP_FOLLOW;
1345 if (f & O_DIRECTORY)
1346 retval |= LOOKUP_DIRECTORY;
1352 * p1 and p2 should be directories on the same fs.
1354 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1359 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1363 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1365 p = d_ancestor(p2, p1);
1367 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1368 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1372 p = d_ancestor(p1, p2);
1374 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1375 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1379 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1380 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1384 void unlock_rename(struct dentry *p1, struct dentry *p2)
1386 mutex_unlock(&p1->d_inode->i_mutex);
1388 mutex_unlock(&p2->d_inode->i_mutex);
1389 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1393 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1394 struct nameidata *nd)
1396 int error = may_create(dir, dentry);
1401 if (!dir->i_op->create)
1402 return -EACCES; /* shouldn't it be ENOSYS? */
1405 error = security_inode_create(dir, dentry, mode);
1409 error = dir->i_op->create(dir, dentry, mode, nd);
1411 fsnotify_create(dir, dentry);
1415 int may_open(struct path *path, int acc_mode, int flag)
1417 struct dentry *dentry = path->dentry;
1418 struct inode *inode = dentry->d_inode;
1424 switch (inode->i_mode & S_IFMT) {
1428 if (acc_mode & MAY_WRITE)
1433 if (path->mnt->mnt_flags & MNT_NODEV)
1442 error = inode_permission(inode, acc_mode);
1446 error = ima_path_check(path, acc_mode ?
1447 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1448 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1454 * An append-only file must be opened in append mode for writing.
1456 if (IS_APPEND(inode)) {
1458 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1464 /* O_NOATIME can only be set by the owner or superuser */
1465 if (flag & O_NOATIME)
1466 if (!is_owner_or_cap(inode)) {
1472 * Ensure there are no outstanding leases on the file.
1474 error = break_lease(inode, flag);
1478 if (flag & O_TRUNC) {
1479 error = get_write_access(inode);
1484 * Refuse to truncate files with mandatory locks held on them.
1486 error = locks_verify_locked(inode);
1488 error = security_path_truncate(path, 0,
1489 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1493 error = do_truncate(dentry, 0,
1494 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1497 put_write_access(inode);
1501 if (flag & FMODE_WRITE)
1506 ima_counts_put(path, acc_mode ?
1507 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1508 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1513 * Be careful about ever adding any more callers of this
1514 * function. Its flags must be in the namei format, not
1515 * what get passed to sys_open().
1517 static int __open_namei_create(struct nameidata *nd, struct path *path,
1521 struct dentry *dir = nd->path.dentry;
1523 if (!IS_POSIXACL(dir->d_inode))
1524 mode &= ~current_umask();
1525 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1528 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1530 mutex_unlock(&dir->d_inode->i_mutex);
1531 dput(nd->path.dentry);
1532 nd->path.dentry = path->dentry;
1535 /* Don't check for write permission, don't truncate */
1536 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1540 * Note that while the flag value (low two bits) for sys_open means:
1545 * it is changed into
1546 * 00 - no permissions needed
1547 * 01 - read-permission
1548 * 10 - write-permission
1550 * for the internal routines (ie open_namei()/follow_link() etc)
1551 * This is more logical, and also allows the 00 "no perm needed"
1552 * to be used for symlinks (where the permissions are checked
1556 static inline int open_to_namei_flags(int flag)
1558 if ((flag+1) & O_ACCMODE)
1563 static int open_will_write_to_fs(int flag, struct inode *inode)
1566 * We'll never write to the fs underlying
1569 if (special_file(inode->i_mode))
1571 return (flag & O_TRUNC);
1575 * Note that the low bits of the passed in "open_flag"
1576 * are not the same as in the local variable "flag". See
1577 * open_to_namei_flags() for more details.
1579 struct file *do_filp_open(int dfd, const char *pathname,
1580 int open_flag, int mode, int acc_mode)
1583 struct nameidata nd;
1585 struct path path, save;
1589 int flag = open_to_namei_flags(open_flag);
1592 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1593 * check for O_DSYNC if the need any syncing at all we enforce it's
1594 * always set instead of having to deal with possibly weird behaviour
1595 * for malicious applications setting only __O_SYNC.
1597 if (open_flag & __O_SYNC)
1598 open_flag |= O_DSYNC;
1601 acc_mode = MAY_OPEN | ACC_MODE(flag);
1603 /* O_TRUNC implies we need access checks for write permissions */
1605 acc_mode |= MAY_WRITE;
1607 /* Allow the LSM permission hook to distinguish append
1608 access from general write access. */
1609 if (flag & O_APPEND)
1610 acc_mode |= MAY_APPEND;
1613 * The simplest case - just a plain lookup.
1615 if (!(flag & O_CREAT)) {
1616 filp = get_empty_filp();
1619 return ERR_PTR(-ENFILE);
1620 nd.intent.open.file = filp;
1621 nd.intent.open.flags = flag;
1622 nd.intent.open.create_mode = 0;
1623 error = do_path_lookup(dfd, pathname,
1624 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1625 if (IS_ERR(nd.intent.open.file)) {
1627 error = PTR_ERR(nd.intent.open.file);
1631 release_open_intent(&nd);
1633 return ERR_PTR(error);
1638 * Create - we need to know the parent.
1640 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1642 return ERR_PTR(error);
1643 error = path_walk(pathname, &nd);
1647 return ERR_PTR(error);
1649 if (unlikely(!audit_dummy_context()))
1650 audit_inode(pathname, nd.path.dentry);
1653 * We have the parent and last component. First of all, check
1654 * that we are not asked to creat(2) an obvious directory - that
1658 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1662 filp = get_empty_filp();
1665 nd.intent.open.file = filp;
1666 nd.intent.open.flags = flag;
1667 nd.intent.open.create_mode = mode;
1668 dir = nd.path.dentry;
1669 nd.flags &= ~LOOKUP_PARENT;
1670 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1672 nd.flags |= LOOKUP_EXCL;
1673 mutex_lock(&dir->d_inode->i_mutex);
1674 path.dentry = lookup_hash(&nd);
1675 path.mnt = nd.path.mnt;
1678 error = PTR_ERR(path.dentry);
1679 if (IS_ERR(path.dentry)) {
1680 mutex_unlock(&dir->d_inode->i_mutex);
1684 if (IS_ERR(nd.intent.open.file)) {
1685 error = PTR_ERR(nd.intent.open.file);
1686 goto exit_mutex_unlock;
1689 /* Negative dentry, just create the file */
1690 if (!path.dentry->d_inode) {
1692 * This write is needed to ensure that a
1693 * ro->rw transition does not occur between
1694 * the time when the file is created and when
1695 * a permanent write count is taken through
1696 * the 'struct file' in nameidata_to_filp().
1698 error = mnt_want_write(nd.path.mnt);
1700 goto exit_mutex_unlock;
1701 error = __open_namei_create(&nd, &path, flag, mode);
1703 mnt_drop_write(nd.path.mnt);
1706 filp = nameidata_to_filp(&nd, open_flag);
1708 ima_counts_put(&nd.path,
1709 acc_mode & (MAY_READ | MAY_WRITE |
1711 mnt_drop_write(nd.path.mnt);
1718 * It already exists.
1720 mutex_unlock(&dir->d_inode->i_mutex);
1721 audit_inode(pathname, path.dentry);
1727 if (__follow_mount(&path)) {
1729 if (flag & O_NOFOLLOW)
1734 if (!path.dentry->d_inode)
1736 if (path.dentry->d_inode->i_op->follow_link)
1739 path_to_nameidata(&path, &nd);
1741 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1746 * 1. may_open() truncates a file
1747 * 2. a rw->ro mount transition occurs
1748 * 3. nameidata_to_filp() fails due to
1750 * That would be inconsistent, and should
1751 * be avoided. Taking this mnt write here
1752 * ensures that (2) can not occur.
1754 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1756 error = mnt_want_write(nd.path.mnt);
1760 error = may_open(&nd.path, acc_mode, flag);
1763 mnt_drop_write(nd.path.mnt);
1766 filp = nameidata_to_filp(&nd, open_flag);
1768 ima_counts_put(&nd.path,
1769 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1771 * It is now safe to drop the mnt write
1772 * because the filp has had a write taken
1776 mnt_drop_write(nd.path.mnt);
1782 mutex_unlock(&dir->d_inode->i_mutex);
1784 path_put_conditional(&path, &nd);
1786 if (!IS_ERR(nd.intent.open.file))
1787 release_open_intent(&nd);
1792 return ERR_PTR(error);
1796 if (flag & O_NOFOLLOW)
1799 * This is subtle. Instead of calling do_follow_link() we do the
1800 * thing by hands. The reason is that this way we have zero link_count
1801 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1802 * After that we have the parent and last component, i.e.
1803 * we are in the same situation as after the first path_walk().
1804 * Well, almost - if the last component is normal we get its copy
1805 * stored in nd->last.name and we will have to putname() it when we
1806 * are done. Procfs-like symlinks just set LAST_BIND.
1808 nd.flags |= LOOKUP_PARENT;
1809 error = security_inode_follow_link(path.dentry, &nd);
1814 error = __do_follow_link(&path, &nd);
1815 if (error == -ESTALE) {
1816 /* nd.path had been dropped */
1819 nd.flags |= LOOKUP_REVAL;
1820 error = __do_follow_link(&path, &nd);
1825 /* Does someone understand code flow here? Or it is only
1826 * me so stupid? Anathema to whoever designed this non-sense
1827 * with "intent.open".
1829 release_open_intent(&nd);
1832 return ERR_PTR(error);
1834 nd.flags &= ~LOOKUP_PARENT;
1835 if (nd.last_type == LAST_BIND)
1838 if (nd.last_type != LAST_NORM)
1840 if (nd.last.name[nd.last.len]) {
1841 __putname(nd.last.name);
1846 __putname(nd.last.name);
1849 dir = nd.path.dentry;
1850 mutex_lock(&dir->d_inode->i_mutex);
1851 path.dentry = lookup_hash(&nd);
1852 path.mnt = nd.path.mnt;
1853 __putname(nd.last.name);
1858 * filp_open - open file and return file pointer
1860 * @filename: path to open
1861 * @flags: open flags as per the open(2) second argument
1862 * @mode: mode for the new file if O_CREAT is set, else ignored
1864 * This is the helper to open a file from kernelspace if you really
1865 * have to. But in generally you should not do this, so please move
1866 * along, nothing to see here..
1868 struct file *filp_open(const char *filename, int flags, int mode)
1870 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1872 EXPORT_SYMBOL(filp_open);
1875 * lookup_create - lookup a dentry, creating it if it doesn't exist
1876 * @nd: nameidata info
1877 * @is_dir: directory flag
1879 * Simple function to lookup and return a dentry and create it
1880 * if it doesn't exist. Is SMP-safe.
1882 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1884 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1886 struct dentry *dentry = ERR_PTR(-EEXIST);
1888 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1890 * Yucky last component or no last component at all?
1891 * (foo/., foo/.., /////)
1893 if (nd->last_type != LAST_NORM)
1895 nd->flags &= ~LOOKUP_PARENT;
1896 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1897 nd->intent.open.flags = O_EXCL;
1900 * Do the final lookup.
1902 dentry = lookup_hash(nd);
1906 if (dentry->d_inode)
1909 * Special case - lookup gave negative, but... we had foo/bar/
1910 * From the vfs_mknod() POV we just have a negative dentry -
1911 * all is fine. Let's be bastards - you had / on the end, you've
1912 * been asking for (non-existent) directory. -ENOENT for you.
1914 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1916 dentry = ERR_PTR(-ENOENT);
1921 dentry = ERR_PTR(-EEXIST);
1925 EXPORT_SYMBOL_GPL(lookup_create);
1927 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1929 int error = may_create(dir, dentry);
1934 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1937 if (!dir->i_op->mknod)
1940 error = devcgroup_inode_mknod(mode, dev);
1944 error = security_inode_mknod(dir, dentry, mode, dev);
1949 error = dir->i_op->mknod(dir, dentry, mode, dev);
1951 fsnotify_create(dir, dentry);
1955 static int may_mknod(mode_t mode)
1957 switch (mode & S_IFMT) {
1963 case 0: /* zero mode translates to S_IFREG */
1972 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1977 struct dentry *dentry;
1978 struct nameidata nd;
1983 error = user_path_parent(dfd, filename, &nd, &tmp);
1987 dentry = lookup_create(&nd, 0);
1988 if (IS_ERR(dentry)) {
1989 error = PTR_ERR(dentry);
1992 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1993 mode &= ~current_umask();
1994 error = may_mknod(mode);
1997 error = mnt_want_write(nd.path.mnt);
2000 error = security_path_mknod(&nd.path, dentry, mode, dev);
2002 goto out_drop_write;
2003 switch (mode & S_IFMT) {
2004 case 0: case S_IFREG:
2005 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2007 case S_IFCHR: case S_IFBLK:
2008 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2009 new_decode_dev(dev));
2011 case S_IFIFO: case S_IFSOCK:
2012 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2016 mnt_drop_write(nd.path.mnt);
2020 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2027 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2029 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2032 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2034 int error = may_create(dir, dentry);
2039 if (!dir->i_op->mkdir)
2042 mode &= (S_IRWXUGO|S_ISVTX);
2043 error = security_inode_mkdir(dir, dentry, mode);
2048 error = dir->i_op->mkdir(dir, dentry, mode);
2050 fsnotify_mkdir(dir, dentry);
2054 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2058 struct dentry *dentry;
2059 struct nameidata nd;
2061 error = user_path_parent(dfd, pathname, &nd, &tmp);
2065 dentry = lookup_create(&nd, 1);
2066 error = PTR_ERR(dentry);
2070 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2071 mode &= ~current_umask();
2072 error = mnt_want_write(nd.path.mnt);
2075 error = security_path_mkdir(&nd.path, dentry, mode);
2077 goto out_drop_write;
2078 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2080 mnt_drop_write(nd.path.mnt);
2084 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2091 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2093 return sys_mkdirat(AT_FDCWD, pathname, mode);
2097 * We try to drop the dentry early: we should have
2098 * a usage count of 2 if we're the only user of this
2099 * dentry, and if that is true (possibly after pruning
2100 * the dcache), then we drop the dentry now.
2102 * A low-level filesystem can, if it choses, legally
2105 * if (!d_unhashed(dentry))
2108 * if it cannot handle the case of removing a directory
2109 * that is still in use by something else..
2111 void dentry_unhash(struct dentry *dentry)
2114 shrink_dcache_parent(dentry);
2115 spin_lock(&dcache_lock);
2116 spin_lock(&dentry->d_lock);
2117 if (atomic_read(&dentry->d_count) == 2)
2119 spin_unlock(&dentry->d_lock);
2120 spin_unlock(&dcache_lock);
2123 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2125 int error = may_delete(dir, dentry, 1);
2130 if (!dir->i_op->rmdir)
2135 mutex_lock(&dentry->d_inode->i_mutex);
2136 dentry_unhash(dentry);
2137 if (d_mountpoint(dentry))
2140 error = security_inode_rmdir(dir, dentry);
2142 error = dir->i_op->rmdir(dir, dentry);
2144 dentry->d_inode->i_flags |= S_DEAD;
2147 mutex_unlock(&dentry->d_inode->i_mutex);
2156 static long do_rmdir(int dfd, const char __user *pathname)
2160 struct dentry *dentry;
2161 struct nameidata nd;
2163 error = user_path_parent(dfd, pathname, &nd, &name);
2167 switch(nd.last_type) {
2179 nd.flags &= ~LOOKUP_PARENT;
2181 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2182 dentry = lookup_hash(&nd);
2183 error = PTR_ERR(dentry);
2186 error = mnt_want_write(nd.path.mnt);
2189 error = security_path_rmdir(&nd.path, dentry);
2192 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2194 mnt_drop_write(nd.path.mnt);
2198 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2205 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2207 return do_rmdir(AT_FDCWD, pathname);
2210 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2212 int error = may_delete(dir, dentry, 0);
2217 if (!dir->i_op->unlink)
2222 mutex_lock(&dentry->d_inode->i_mutex);
2223 if (d_mountpoint(dentry))
2226 error = security_inode_unlink(dir, dentry);
2228 error = dir->i_op->unlink(dir, dentry);
2230 mutex_unlock(&dentry->d_inode->i_mutex);
2232 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2233 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2234 fsnotify_link_count(dentry->d_inode);
2242 * Make sure that the actual truncation of the file will occur outside its
2243 * directory's i_mutex. Truncate can take a long time if there is a lot of
2244 * writeout happening, and we don't want to prevent access to the directory
2245 * while waiting on the I/O.
2247 static long do_unlinkat(int dfd, const char __user *pathname)
2251 struct dentry *dentry;
2252 struct nameidata nd;
2253 struct inode *inode = NULL;
2255 error = user_path_parent(dfd, pathname, &nd, &name);
2260 if (nd.last_type != LAST_NORM)
2263 nd.flags &= ~LOOKUP_PARENT;
2265 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2266 dentry = lookup_hash(&nd);
2267 error = PTR_ERR(dentry);
2268 if (!IS_ERR(dentry)) {
2269 /* Why not before? Because we want correct error value */
2270 if (nd.last.name[nd.last.len])
2272 inode = dentry->d_inode;
2274 atomic_inc(&inode->i_count);
2275 error = mnt_want_write(nd.path.mnt);
2278 error = security_path_unlink(&nd.path, dentry);
2281 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2283 mnt_drop_write(nd.path.mnt);
2287 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2289 iput(inode); /* truncate the inode here */
2296 error = !dentry->d_inode ? -ENOENT :
2297 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2301 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2303 if ((flag & ~AT_REMOVEDIR) != 0)
2306 if (flag & AT_REMOVEDIR)
2307 return do_rmdir(dfd, pathname);
2309 return do_unlinkat(dfd, pathname);
2312 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2314 return do_unlinkat(AT_FDCWD, pathname);
2317 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2319 int error = may_create(dir, dentry);
2324 if (!dir->i_op->symlink)
2327 error = security_inode_symlink(dir, dentry, oldname);
2332 error = dir->i_op->symlink(dir, dentry, oldname);
2334 fsnotify_create(dir, dentry);
2338 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2339 int, newdfd, const char __user *, newname)
2344 struct dentry *dentry;
2345 struct nameidata nd;
2347 from = getname(oldname);
2349 return PTR_ERR(from);
2351 error = user_path_parent(newdfd, newname, &nd, &to);
2355 dentry = lookup_create(&nd, 0);
2356 error = PTR_ERR(dentry);
2360 error = mnt_want_write(nd.path.mnt);
2363 error = security_path_symlink(&nd.path, dentry, from);
2365 goto out_drop_write;
2366 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2368 mnt_drop_write(nd.path.mnt);
2372 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2380 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2382 return sys_symlinkat(oldname, AT_FDCWD, newname);
2385 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2387 struct inode *inode = old_dentry->d_inode;
2393 error = may_create(dir, new_dentry);
2397 if (dir->i_sb != inode->i_sb)
2401 * A link to an append-only or immutable file cannot be created.
2403 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2405 if (!dir->i_op->link)
2407 if (S_ISDIR(inode->i_mode))
2410 error = security_inode_link(old_dentry, dir, new_dentry);
2414 mutex_lock(&inode->i_mutex);
2416 error = dir->i_op->link(old_dentry, dir, new_dentry);
2417 mutex_unlock(&inode->i_mutex);
2419 fsnotify_link(dir, inode, new_dentry);
2424 * Hardlinks are often used in delicate situations. We avoid
2425 * security-related surprises by not following symlinks on the
2428 * We don't follow them on the oldname either to be compatible
2429 * with linux 2.0, and to avoid hard-linking to directories
2430 * and other special files. --ADM
2432 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2433 int, newdfd, const char __user *, newname, int, flags)
2435 struct dentry *new_dentry;
2436 struct nameidata nd;
2437 struct path old_path;
2441 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2444 error = user_path_at(olddfd, oldname,
2445 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2450 error = user_path_parent(newdfd, newname, &nd, &to);
2454 if (old_path.mnt != nd.path.mnt)
2456 new_dentry = lookup_create(&nd, 0);
2457 error = PTR_ERR(new_dentry);
2458 if (IS_ERR(new_dentry))
2460 error = mnt_want_write(nd.path.mnt);
2463 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2465 goto out_drop_write;
2466 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2468 mnt_drop_write(nd.path.mnt);
2472 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2477 path_put(&old_path);
2482 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2484 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2488 * The worst of all namespace operations - renaming directory. "Perverted"
2489 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2491 * a) we can get into loop creation. Check is done in is_subdir().
2492 * b) race potential - two innocent renames can create a loop together.
2493 * That's where 4.4 screws up. Current fix: serialization on
2494 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2496 * c) we have to lock _three_ objects - parents and victim (if it exists).
2497 * And that - after we got ->i_mutex on parents (until then we don't know
2498 * whether the target exists). Solution: try to be smart with locking
2499 * order for inodes. We rely on the fact that tree topology may change
2500 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2501 * move will be locked. Thus we can rank directories by the tree
2502 * (ancestors first) and rank all non-directories after them.
2503 * That works since everybody except rename does "lock parent, lookup,
2504 * lock child" and rename is under ->s_vfs_rename_mutex.
2505 * HOWEVER, it relies on the assumption that any object with ->lookup()
2506 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2507 * we'd better make sure that there's no link(2) for them.
2508 * d) some filesystems don't support opened-but-unlinked directories,
2509 * either because of layout or because they are not ready to deal with
2510 * all cases correctly. The latter will be fixed (taking this sort of
2511 * stuff into VFS), but the former is not going away. Solution: the same
2512 * trick as in rmdir().
2513 * e) conversion from fhandle to dentry may come in the wrong moment - when
2514 * we are removing the target. Solution: we will have to grab ->i_mutex
2515 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2516 * ->i_mutex on parents, which works but leads to some truely excessive
2519 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2520 struct inode *new_dir, struct dentry *new_dentry)
2523 struct inode *target;
2526 * If we are going to change the parent - check write permissions,
2527 * we'll need to flip '..'.
2529 if (new_dir != old_dir) {
2530 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2535 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2539 target = new_dentry->d_inode;
2541 mutex_lock(&target->i_mutex);
2542 dentry_unhash(new_dentry);
2544 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2547 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2550 target->i_flags |= S_DEAD;
2551 mutex_unlock(&target->i_mutex);
2552 if (d_unhashed(new_dentry))
2553 d_rehash(new_dentry);
2557 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2558 d_move(old_dentry,new_dentry);
2562 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2563 struct inode *new_dir, struct dentry *new_dentry)
2565 struct inode *target;
2568 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2573 target = new_dentry->d_inode;
2575 mutex_lock(&target->i_mutex);
2576 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2579 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2581 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2582 d_move(old_dentry, new_dentry);
2585 mutex_unlock(&target->i_mutex);
2590 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2591 struct inode *new_dir, struct dentry *new_dentry)
2594 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2595 const char *old_name;
2597 if (old_dentry->d_inode == new_dentry->d_inode)
2600 error = may_delete(old_dir, old_dentry, is_dir);
2604 if (!new_dentry->d_inode)
2605 error = may_create(new_dir, new_dentry);
2607 error = may_delete(new_dir, new_dentry, is_dir);
2611 if (!old_dir->i_op->rename)
2614 vfs_dq_init(old_dir);
2615 vfs_dq_init(new_dir);
2617 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2620 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2622 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2624 const char *new_name = old_dentry->d_name.name;
2625 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2626 new_dentry->d_inode, old_dentry);
2628 fsnotify_oldname_free(old_name);
2633 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2634 int, newdfd, const char __user *, newname)
2636 struct dentry *old_dir, *new_dir;
2637 struct dentry *old_dentry, *new_dentry;
2638 struct dentry *trap;
2639 struct nameidata oldnd, newnd;
2644 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2648 error = user_path_parent(newdfd, newname, &newnd, &to);
2653 if (oldnd.path.mnt != newnd.path.mnt)
2656 old_dir = oldnd.path.dentry;
2658 if (oldnd.last_type != LAST_NORM)
2661 new_dir = newnd.path.dentry;
2662 if (newnd.last_type != LAST_NORM)
2665 oldnd.flags &= ~LOOKUP_PARENT;
2666 newnd.flags &= ~LOOKUP_PARENT;
2667 newnd.flags |= LOOKUP_RENAME_TARGET;
2669 trap = lock_rename(new_dir, old_dir);
2671 old_dentry = lookup_hash(&oldnd);
2672 error = PTR_ERR(old_dentry);
2673 if (IS_ERR(old_dentry))
2675 /* source must exist */
2677 if (!old_dentry->d_inode)
2679 /* unless the source is a directory trailing slashes give -ENOTDIR */
2680 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2682 if (oldnd.last.name[oldnd.last.len])
2684 if (newnd.last.name[newnd.last.len])
2687 /* source should not be ancestor of target */
2689 if (old_dentry == trap)
2691 new_dentry = lookup_hash(&newnd);
2692 error = PTR_ERR(new_dentry);
2693 if (IS_ERR(new_dentry))
2695 /* target should not be an ancestor of source */
2697 if (new_dentry == trap)
2700 error = mnt_want_write(oldnd.path.mnt);
2703 error = security_path_rename(&oldnd.path, old_dentry,
2704 &newnd.path, new_dentry);
2707 error = vfs_rename(old_dir->d_inode, old_dentry,
2708 new_dir->d_inode, new_dentry);
2710 mnt_drop_write(oldnd.path.mnt);
2716 unlock_rename(new_dir, old_dir);
2718 path_put(&newnd.path);
2721 path_put(&oldnd.path);
2727 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2729 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2732 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2736 len = PTR_ERR(link);
2741 if (len > (unsigned) buflen)
2743 if (copy_to_user(buffer, link, len))
2750 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2751 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2752 * using) it for any given inode is up to filesystem.
2754 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2756 struct nameidata nd;
2761 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2763 return PTR_ERR(cookie);
2765 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2766 if (dentry->d_inode->i_op->put_link)
2767 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2771 int vfs_follow_link(struct nameidata *nd, const char *link)
2773 return __vfs_follow_link(nd, link);
2776 /* get the link contents into pagecache */
2777 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2781 struct address_space *mapping = dentry->d_inode->i_mapping;
2782 page = read_mapping_page(mapping, 0, NULL);
2787 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2791 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2793 struct page *page = NULL;
2794 char *s = page_getlink(dentry, &page);
2795 int res = vfs_readlink(dentry,buffer,buflen,s);
2798 page_cache_release(page);
2803 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2805 struct page *page = NULL;
2806 nd_set_link(nd, page_getlink(dentry, &page));
2810 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2812 struct page *page = cookie;
2816 page_cache_release(page);
2821 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2823 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2825 struct address_space *mapping = inode->i_mapping;
2830 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2832 flags |= AOP_FLAG_NOFS;
2835 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2836 flags, &page, &fsdata);
2840 kaddr = kmap_atomic(page, KM_USER0);
2841 memcpy(kaddr, symname, len-1);
2842 kunmap_atomic(kaddr, KM_USER0);
2844 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2851 mark_inode_dirty(inode);
2857 int page_symlink(struct inode *inode, const char *symname, int len)
2859 return __page_symlink(inode, symname, len,
2860 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2863 const struct inode_operations page_symlink_inode_operations = {
2864 .readlink = generic_readlink,
2865 .follow_link = page_follow_link_light,
2866 .put_link = page_put_link,
2869 EXPORT_SYMBOL(user_path_at);
2870 EXPORT_SYMBOL(follow_down);
2871 EXPORT_SYMBOL(follow_up);
2872 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2873 EXPORT_SYMBOL(getname);
2874 EXPORT_SYMBOL(lock_rename);
2875 EXPORT_SYMBOL(lookup_one_len);
2876 EXPORT_SYMBOL(page_follow_link_light);
2877 EXPORT_SYMBOL(page_put_link);
2878 EXPORT_SYMBOL(page_readlink);
2879 EXPORT_SYMBOL(__page_symlink);
2880 EXPORT_SYMBOL(page_symlink);
2881 EXPORT_SYMBOL(page_symlink_inode_operations);
2882 EXPORT_SYMBOL(path_lookup);
2883 EXPORT_SYMBOL(kern_path);
2884 EXPORT_SYMBOL(vfs_path_lookup);
2885 EXPORT_SYMBOL(inode_permission);
2886 EXPORT_SYMBOL(file_permission);
2887 EXPORT_SYMBOL(unlock_rename);
2888 EXPORT_SYMBOL(vfs_create);
2889 EXPORT_SYMBOL(vfs_follow_link);
2890 EXPORT_SYMBOL(vfs_link);
2891 EXPORT_SYMBOL(vfs_mkdir);
2892 EXPORT_SYMBOL(vfs_mknod);
2893 EXPORT_SYMBOL(generic_permission);
2894 EXPORT_SYMBOL(vfs_readlink);
2895 EXPORT_SYMBOL(vfs_rename);
2896 EXPORT_SYMBOL(vfs_rmdir);
2897 EXPORT_SYMBOL(vfs_symlink);
2898 EXPORT_SYMBOL(vfs_unlink);
2899 EXPORT_SYMBOL(dentry_unhash);
2900 EXPORT_SYMBOL(generic_readlink);