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 <linux/device_cgroup.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 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
189 if (current_fsuid() == inode->i_uid)
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
200 if (in_group_p(inode->i_gid))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask & ~mode) == 0)
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask & MAY_EXEC) || execute_ok(inode))
216 if (capable(CAP_DAC_OVERRIDE))
220 * Searching includes executable on directories, else just read.
222 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
223 if (capable(CAP_DAC_READ_SEARCH))
229 int inode_permission(struct inode *inode, int mask)
233 if (mask & MAY_WRITE) {
234 umode_t mode = inode->i_mode;
237 * Nobody gets write access to a read-only fs.
239 if (IS_RDONLY(inode) &&
240 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
244 * Nobody gets write access to an immutable file.
246 if (IS_IMMUTABLE(inode))
250 if (inode->i_op && inode->i_op->permission)
251 retval = inode->i_op->permission(inode, mask);
253 retval = generic_permission(inode, mask, NULL);
258 retval = devcgroup_inode_permission(inode, mask);
262 return security_inode_permission(inode,
263 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
267 * vfs_permission - check for access rights to a given path
268 * @nd: lookup result that describes the path
269 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271 * Used to check for read/write/execute permissions on a path.
272 * We use "fsuid" for this, letting us set arbitrary permissions
273 * for filesystem access without changing the "normal" uids which
274 * are used for other things.
276 int vfs_permission(struct nameidata *nd, int mask)
278 return inode_permission(nd->path.dentry->d_inode, mask);
282 * file_permission - check for additional access rights to a given file
283 * @file: file to check access rights for
284 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
286 * Used to check for read/write/execute permissions on an already opened
290 * Do not use this function in new code. All access checks should
291 * be done using vfs_permission().
293 int file_permission(struct file *file, int mask)
295 return inode_permission(file->f_path.dentry->d_inode, mask);
299 * get_write_access() gets write permission for a file.
300 * put_write_access() releases this write permission.
301 * This is used for regular files.
302 * We cannot support write (and maybe mmap read-write shared) accesses and
303 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
304 * can have the following values:
305 * 0: no writers, no VM_DENYWRITE mappings
306 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
307 * > 0: (i_writecount) users are writing to the file.
309 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
310 * except for the cases where we don't hold i_writecount yet. Then we need to
311 * use {get,deny}_write_access() - these functions check the sign and refuse
312 * to do the change if sign is wrong. Exclusion between them is provided by
313 * the inode->i_lock spinlock.
316 int get_write_access(struct inode * inode)
318 spin_lock(&inode->i_lock);
319 if (atomic_read(&inode->i_writecount) < 0) {
320 spin_unlock(&inode->i_lock);
323 atomic_inc(&inode->i_writecount);
324 spin_unlock(&inode->i_lock);
329 int deny_write_access(struct file * file)
331 struct inode *inode = file->f_path.dentry->d_inode;
333 spin_lock(&inode->i_lock);
334 if (atomic_read(&inode->i_writecount) > 0) {
335 spin_unlock(&inode->i_lock);
338 atomic_dec(&inode->i_writecount);
339 spin_unlock(&inode->i_lock);
345 * path_get - get a reference to a path
346 * @path: path to get the reference to
348 * Given a path increment the reference count to the dentry and the vfsmount.
350 void path_get(struct path *path)
355 EXPORT_SYMBOL(path_get);
358 * path_put - put a reference to a path
359 * @path: path to put the reference to
361 * Given a path decrement the reference count to the dentry and the vfsmount.
363 void path_put(struct path *path)
368 EXPORT_SYMBOL(path_put);
371 * release_open_intent - free up open intent resources
372 * @nd: pointer to nameidata
374 void release_open_intent(struct nameidata *nd)
376 if (nd->intent.open.file->f_path.dentry == NULL)
377 put_filp(nd->intent.open.file);
379 fput(nd->intent.open.file);
382 static inline struct dentry *
383 do_revalidate(struct dentry *dentry, struct nameidata *nd)
385 int status = dentry->d_op->d_revalidate(dentry, nd);
386 if (unlikely(status <= 0)) {
388 * The dentry failed validation.
389 * If d_revalidate returned 0 attempt to invalidate
390 * the dentry otherwise d_revalidate is asking us
391 * to return a fail status.
394 if (!d_invalidate(dentry)) {
400 dentry = ERR_PTR(status);
407 * Internal lookup() using the new generic dcache.
410 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
412 struct dentry * dentry = __d_lookup(parent, name);
414 /* lockess __d_lookup may fail due to concurrent d_move()
415 * in some unrelated directory, so try with d_lookup
418 dentry = d_lookup(parent, name);
420 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
421 dentry = do_revalidate(dentry, nd);
427 * Short-cut version of permission(), for calling by
428 * path_walk(), when dcache lock is held. Combines parts
429 * of permission() and generic_permission(), and tests ONLY for
430 * MAY_EXEC permission.
432 * If appropriate, check DAC only. If not appropriate, or
433 * short-cut DAC fails, then call permission() to do more
434 * complete permission check.
436 static int exec_permission_lite(struct inode *inode)
438 umode_t mode = inode->i_mode;
440 if (inode->i_op && inode->i_op->permission)
443 if (current_fsuid() == inode->i_uid)
445 else if (in_group_p(inode->i_gid))
451 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
454 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
457 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
462 return security_inode_permission(inode, MAY_EXEC);
466 * This is called when everything else fails, and we actually have
467 * to go to the low-level filesystem to find out what we should do..
469 * We get the directory semaphore, and after getting that we also
470 * make sure that nobody added the entry to the dcache in the meantime..
473 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
475 struct dentry * result;
476 struct inode *dir = parent->d_inode;
478 mutex_lock(&dir->i_mutex);
480 * First re-do the cached lookup just in case it was created
481 * while we waited for the directory semaphore..
483 * FIXME! This could use version numbering or similar to
484 * avoid unnecessary cache lookups.
486 * The "dcache_lock" is purely to protect the RCU list walker
487 * from concurrent renames at this point (we mustn't get false
488 * negatives from the RCU list walk here, unlike the optimistic
491 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
493 result = d_lookup(parent, name);
495 struct dentry *dentry;
497 /* Don't create child dentry for a dead directory. */
498 result = ERR_PTR(-ENOENT);
502 dentry = d_alloc(parent, name);
503 result = ERR_PTR(-ENOMEM);
505 result = dir->i_op->lookup(dir, dentry, nd);
512 mutex_unlock(&dir->i_mutex);
517 * Uhhuh! Nasty case: the cache was re-populated while
518 * we waited on the semaphore. Need to revalidate.
520 mutex_unlock(&dir->i_mutex);
521 if (result->d_op && result->d_op->d_revalidate) {
522 result = do_revalidate(result, nd);
524 result = ERR_PTR(-ENOENT);
530 static __always_inline void
531 walk_init_root(const char *name, struct nameidata *nd)
533 struct fs_struct *fs = current->fs;
535 read_lock(&fs->lock);
538 read_unlock(&fs->lock);
542 * Wrapper to retry pathname resolution whenever the underlying
543 * file system returns an ESTALE.
545 * Retry the whole path once, forcing real lookup requests
546 * instead of relying on the dcache.
548 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
550 struct path save = nd->path;
553 /* make sure the stuff we saved doesn't go away */
556 result = __link_path_walk(name, nd);
557 if (result == -ESTALE) {
558 /* nd->path had been dropped */
561 nd->flags |= LOOKUP_REVAL;
562 result = __link_path_walk(name, nd);
570 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
579 walk_init_root(link, nd);
581 res = link_path_walk(link, nd);
582 if (nd->depth || res || nd->last_type!=LAST_NORM)
585 * If it is an iterative symlinks resolution in open_namei() we
586 * have to copy the last component. And all that crap because of
587 * bloody create() on broken symlinks. Furrfu...
590 if (unlikely(!name)) {
594 strcpy(name, nd->last.name);
595 nd->last.name = name;
599 return PTR_ERR(link);
602 static void path_put_conditional(struct path *path, struct nameidata *nd)
605 if (path->mnt != nd->path.mnt)
609 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
611 dput(nd->path.dentry);
612 if (nd->path.mnt != path->mnt)
613 mntput(nd->path.mnt);
614 nd->path.mnt = path->mnt;
615 nd->path.dentry = path->dentry;
618 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
622 struct dentry *dentry = path->dentry;
624 touch_atime(path->mnt, dentry);
625 nd_set_link(nd, NULL);
627 if (path->mnt != nd->path.mnt) {
628 path_to_nameidata(path, nd);
632 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
633 error = PTR_ERR(cookie);
634 if (!IS_ERR(cookie)) {
635 char *s = nd_get_link(nd);
638 error = __vfs_follow_link(nd, s);
639 if (dentry->d_inode->i_op->put_link)
640 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
648 * This limits recursive symlink follows to 8, while
649 * limiting consecutive symlinks to 40.
651 * Without that kind of total limit, nasty chains of consecutive
652 * symlinks can cause almost arbitrarily long lookups.
654 static inline int do_follow_link(struct path *path, struct nameidata *nd)
657 if (current->link_count >= MAX_NESTED_LINKS)
659 if (current->total_link_count >= 40)
661 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
663 err = security_inode_follow_link(path->dentry, nd);
666 current->link_count++;
667 current->total_link_count++;
669 err = __do_follow_link(path, nd);
670 current->link_count--;
674 path_put_conditional(path, nd);
679 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
681 struct vfsmount *parent;
682 struct dentry *mountpoint;
683 spin_lock(&vfsmount_lock);
684 parent=(*mnt)->mnt_parent;
685 if (parent == *mnt) {
686 spin_unlock(&vfsmount_lock);
690 mountpoint=dget((*mnt)->mnt_mountpoint);
691 spin_unlock(&vfsmount_lock);
693 *dentry = mountpoint;
699 /* no need for dcache_lock, as serialization is taken care in
702 static int __follow_mount(struct path *path)
705 while (d_mountpoint(path->dentry)) {
706 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
713 path->dentry = dget(mounted->mnt_root);
719 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
721 while (d_mountpoint(*dentry)) {
722 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
728 *dentry = dget(mounted->mnt_root);
732 /* no need for dcache_lock, as serialization is taken care in
735 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
737 struct vfsmount *mounted;
739 mounted = lookup_mnt(*mnt, *dentry);
744 *dentry = dget(mounted->mnt_root);
750 static __always_inline void follow_dotdot(struct nameidata *nd)
752 struct fs_struct *fs = current->fs;
755 struct vfsmount *parent;
756 struct dentry *old = nd->path.dentry;
758 read_lock(&fs->lock);
759 if (nd->path.dentry == fs->root.dentry &&
760 nd->path.mnt == fs->root.mnt) {
761 read_unlock(&fs->lock);
764 read_unlock(&fs->lock);
765 spin_lock(&dcache_lock);
766 if (nd->path.dentry != nd->path.mnt->mnt_root) {
767 nd->path.dentry = dget(nd->path.dentry->d_parent);
768 spin_unlock(&dcache_lock);
772 spin_unlock(&dcache_lock);
773 spin_lock(&vfsmount_lock);
774 parent = nd->path.mnt->mnt_parent;
775 if (parent == nd->path.mnt) {
776 spin_unlock(&vfsmount_lock);
780 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
781 spin_unlock(&vfsmount_lock);
783 mntput(nd->path.mnt);
784 nd->path.mnt = parent;
786 follow_mount(&nd->path.mnt, &nd->path.dentry);
790 * It's more convoluted than I'd like it to be, but... it's still fairly
791 * small and for now I'd prefer to have fast path as straight as possible.
792 * It _is_ time-critical.
794 static int do_lookup(struct nameidata *nd, struct qstr *name,
797 struct vfsmount *mnt = nd->path.mnt;
798 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
802 if (dentry->d_op && dentry->d_op->d_revalidate)
803 goto need_revalidate;
806 path->dentry = dentry;
807 __follow_mount(path);
811 dentry = real_lookup(nd->path.dentry, name, nd);
817 dentry = do_revalidate(dentry, nd);
825 return PTR_ERR(dentry);
830 * This is the basic name resolution function, turning a pathname into
831 * the final dentry. We expect 'base' to be positive and a directory.
833 * Returns 0 and nd will have valid dentry and mnt on success.
834 * Returns error and drops reference to input namei data on failure.
836 static int __link_path_walk(const char *name, struct nameidata *nd)
841 unsigned int lookup_flags = nd->flags;
848 inode = nd->path.dentry->d_inode;
850 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
852 /* At this point we know we have a real path component. */
858 nd->flags |= LOOKUP_CONTINUE;
859 err = exec_permission_lite(inode);
861 err = vfs_permission(nd, MAY_EXEC);
866 c = *(const unsigned char *)name;
868 hash = init_name_hash();
871 hash = partial_name_hash(c, hash);
872 c = *(const unsigned char *)name;
873 } while (c && (c != '/'));
874 this.len = name - (const char *) this.name;
875 this.hash = end_name_hash(hash);
877 /* remove trailing slashes? */
880 while (*++name == '/');
882 goto last_with_slashes;
885 * "." and ".." are special - ".." especially so because it has
886 * to be able to know about the current root directory and
887 * parent relationships.
889 if (this.name[0] == '.') switch (this.len) {
893 if (this.name[1] != '.')
896 inode = nd->path.dentry->d_inode;
902 * See if the low-level filesystem might want
903 * to use its own hash..
905 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
906 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
911 /* This does the actual lookups.. */
912 err = do_lookup(nd, &this, &next);
917 inode = next.dentry->d_inode;
924 if (inode->i_op->follow_link) {
925 err = do_follow_link(&next, nd);
929 inode = nd->path.dentry->d_inode;
936 path_to_nameidata(&next, nd);
938 if (!inode->i_op->lookup)
941 /* here ends the main loop */
944 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
946 /* Clear LOOKUP_CONTINUE iff it was previously unset */
947 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
948 if (lookup_flags & LOOKUP_PARENT)
950 if (this.name[0] == '.') switch (this.len) {
954 if (this.name[1] != '.')
957 inode = nd->path.dentry->d_inode;
962 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
963 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
968 err = do_lookup(nd, &this, &next);
971 inode = next.dentry->d_inode;
972 if ((lookup_flags & LOOKUP_FOLLOW)
973 && inode && inode->i_op && inode->i_op->follow_link) {
974 err = do_follow_link(&next, nd);
977 inode = nd->path.dentry->d_inode;
979 path_to_nameidata(&next, nd);
983 if (lookup_flags & LOOKUP_DIRECTORY) {
985 if (!inode->i_op || !inode->i_op->lookup)
991 nd->last_type = LAST_NORM;
992 if (this.name[0] != '.')
995 nd->last_type = LAST_DOT;
996 else if (this.len == 2 && this.name[1] == '.')
997 nd->last_type = LAST_DOTDOT;
1002 * We bypassed the ordinary revalidation routines.
1003 * We may need to check the cached dentry for staleness.
1005 if (nd->path.dentry && nd->path.dentry->d_sb &&
1006 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1008 /* Note: we do not d_invalidate() */
1009 if (!nd->path.dentry->d_op->d_revalidate(
1010 nd->path.dentry, nd))
1016 path_put_conditional(&next, nd);
1019 path_put(&nd->path);
1024 static int path_walk(const char *name, struct nameidata *nd)
1026 current->total_link_count = 0;
1027 return link_path_walk(name, nd);
1030 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1031 static int do_path_lookup(int dfd, const char *name,
1032 unsigned int flags, struct nameidata *nd)
1037 struct fs_struct *fs = current->fs;
1039 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1044 read_lock(&fs->lock);
1045 nd->path = fs->root;
1046 path_get(&fs->root);
1047 read_unlock(&fs->lock);
1048 } else if (dfd == AT_FDCWD) {
1049 read_lock(&fs->lock);
1052 read_unlock(&fs->lock);
1054 struct dentry *dentry;
1056 file = fget_light(dfd, &fput_needed);
1061 dentry = file->f_path.dentry;
1064 if (!S_ISDIR(dentry->d_inode->i_mode))
1067 retval = file_permission(file, MAY_EXEC);
1071 nd->path = file->f_path;
1072 path_get(&file->f_path);
1074 fput_light(file, fput_needed);
1077 retval = path_walk(name, nd);
1078 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1079 nd->path.dentry->d_inode))
1080 audit_inode(name, nd->path.dentry);
1085 fput_light(file, fput_needed);
1089 int path_lookup(const char *name, unsigned int flags,
1090 struct nameidata *nd)
1092 return do_path_lookup(AT_FDCWD, name, flags, nd);
1095 int kern_path(const char *name, unsigned int flags, struct path *path)
1097 struct nameidata nd;
1098 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1105 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1106 * @dentry: pointer to dentry of the base directory
1107 * @mnt: pointer to vfs mount of the base directory
1108 * @name: pointer to file name
1109 * @flags: lookup flags
1110 * @nd: pointer to nameidata
1112 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1113 const char *name, unsigned int flags,
1114 struct nameidata *nd)
1118 /* same as do_path_lookup */
1119 nd->last_type = LAST_ROOT;
1123 nd->path.dentry = dentry;
1125 path_get(&nd->path);
1127 retval = path_walk(name, nd);
1128 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1129 nd->path.dentry->d_inode))
1130 audit_inode(name, nd->path.dentry);
1137 * path_lookup_open - lookup a file path with open intent
1138 * @dfd: the directory to use as base, or AT_FDCWD
1139 * @name: pointer to file name
1140 * @lookup_flags: lookup intent flags
1141 * @nd: pointer to nameidata
1142 * @open_flags: open intent flags
1144 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1145 struct nameidata *nd, int open_flags)
1147 struct file *filp = get_empty_filp();
1152 nd->intent.open.file = filp;
1153 nd->intent.open.flags = open_flags;
1154 nd->intent.open.create_mode = 0;
1155 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1156 if (IS_ERR(nd->intent.open.file)) {
1158 err = PTR_ERR(nd->intent.open.file);
1159 path_put(&nd->path);
1161 } else if (err != 0)
1162 release_open_intent(nd);
1166 static struct dentry *__lookup_hash(struct qstr *name,
1167 struct dentry *base, struct nameidata *nd)
1169 struct dentry *dentry;
1170 struct inode *inode;
1173 inode = base->d_inode;
1176 * See if the low-level filesystem might want
1177 * to use its own hash..
1179 if (base->d_op && base->d_op->d_hash) {
1180 err = base->d_op->d_hash(base, name);
1181 dentry = ERR_PTR(err);
1186 dentry = cached_lookup(base, name, nd);
1190 /* Don't create child dentry for a dead directory. */
1191 dentry = ERR_PTR(-ENOENT);
1192 if (IS_DEADDIR(inode))
1195 new = d_alloc(base, name);
1196 dentry = ERR_PTR(-ENOMEM);
1199 dentry = inode->i_op->lookup(inode, new, nd);
1210 * Restricted form of lookup. Doesn't follow links, single-component only,
1211 * needs parent already locked. Doesn't follow mounts.
1214 static struct dentry *lookup_hash(struct nameidata *nd)
1218 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1220 return ERR_PTR(err);
1221 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1224 static int __lookup_one_len(const char *name, struct qstr *this,
1225 struct dentry *base, int len)
1235 hash = init_name_hash();
1237 c = *(const unsigned char *)name++;
1238 if (c == '/' || c == '\0')
1240 hash = partial_name_hash(c, hash);
1242 this->hash = end_name_hash(hash);
1247 * lookup_one_len - filesystem helper to lookup single pathname component
1248 * @name: pathname component to lookup
1249 * @base: base directory to lookup from
1250 * @len: maximum length @len should be interpreted to
1252 * Note that this routine is purely a helper for filesystem usage and should
1253 * not be called by generic code. Also note that by using this function the
1254 * nameidata argument is passed to the filesystem methods and a filesystem
1255 * using this helper needs to be prepared for that.
1257 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1262 err = __lookup_one_len(name, &this, base, len);
1264 return ERR_PTR(err);
1266 err = inode_permission(base->d_inode, MAY_EXEC);
1268 return ERR_PTR(err);
1269 return __lookup_hash(&this, base, NULL);
1273 * lookup_one_noperm - bad hack for sysfs
1274 * @name: pathname component to lookup
1275 * @base: base directory to lookup from
1277 * This is a variant of lookup_one_len that doesn't perform any permission
1278 * checks. It's a horrible hack to work around the braindead sysfs
1279 * architecture and should not be used anywhere else.
1281 * DON'T USE THIS FUNCTION EVER, thanks.
1283 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1288 err = __lookup_one_len(name, &this, base, strlen(name));
1290 return ERR_PTR(err);
1291 return __lookup_hash(&this, base, NULL);
1294 int user_path_at(int dfd, const char __user *name, unsigned flags,
1297 struct nameidata nd;
1298 char *tmp = getname(name);
1299 int err = PTR_ERR(tmp);
1302 BUG_ON(flags & LOOKUP_PARENT);
1304 err = do_path_lookup(dfd, tmp, flags, &nd);
1312 static int user_path_parent(int dfd, const char __user *path,
1313 struct nameidata *nd, char **name)
1315 char *s = getname(path);
1321 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1331 * It's inline, so penalty for filesystems that don't use sticky bit is
1334 static inline int check_sticky(struct inode *dir, struct inode *inode)
1336 uid_t fsuid = current_fsuid();
1338 if (!(dir->i_mode & S_ISVTX))
1340 if (inode->i_uid == fsuid)
1342 if (dir->i_uid == fsuid)
1344 return !capable(CAP_FOWNER);
1348 * Check whether we can remove a link victim from directory dir, check
1349 * whether the type of victim is right.
1350 * 1. We can't do it if dir is read-only (done in permission())
1351 * 2. We should have write and exec permissions on dir
1352 * 3. We can't remove anything from append-only dir
1353 * 4. We can't do anything with immutable dir (done in permission())
1354 * 5. If the sticky bit on dir is set we should either
1355 * a. be owner of dir, or
1356 * b. be owner of victim, or
1357 * c. have CAP_FOWNER capability
1358 * 6. If the victim is append-only or immutable we can't do antyhing with
1359 * links pointing to it.
1360 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1361 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1362 * 9. We can't remove a root or mountpoint.
1363 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1364 * nfs_async_unlink().
1366 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1370 if (!victim->d_inode)
1373 BUG_ON(victim->d_parent->d_inode != dir);
1374 audit_inode_child(victim->d_name.name, victim, dir);
1376 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1381 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1382 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1385 if (!S_ISDIR(victim->d_inode->i_mode))
1387 if (IS_ROOT(victim))
1389 } else if (S_ISDIR(victim->d_inode->i_mode))
1391 if (IS_DEADDIR(dir))
1393 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1398 /* Check whether we can create an object with dentry child in directory
1400 * 1. We can't do it if child already exists (open has special treatment for
1401 * this case, but since we are inlined it's OK)
1402 * 2. We can't do it if dir is read-only (done in permission())
1403 * 3. We should have write and exec permissions on dir
1404 * 4. We can't do it if dir is immutable (done in permission())
1406 static inline int may_create(struct inode *dir, struct dentry *child)
1410 if (IS_DEADDIR(dir))
1412 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1416 * O_DIRECTORY translates into forcing a directory lookup.
1418 static inline int lookup_flags(unsigned int f)
1420 unsigned long retval = LOOKUP_FOLLOW;
1423 retval &= ~LOOKUP_FOLLOW;
1425 if (f & O_DIRECTORY)
1426 retval |= LOOKUP_DIRECTORY;
1432 * p1 and p2 should be directories on the same fs.
1434 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1439 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1443 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1445 p = d_ancestor(p2, p1);
1447 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1448 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1452 p = d_ancestor(p1, p2);
1454 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1455 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1459 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1460 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1464 void unlock_rename(struct dentry *p1, struct dentry *p2)
1466 mutex_unlock(&p1->d_inode->i_mutex);
1468 mutex_unlock(&p2->d_inode->i_mutex);
1469 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1473 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1474 struct nameidata *nd)
1476 int error = may_create(dir, dentry);
1481 if (!dir->i_op || !dir->i_op->create)
1482 return -EACCES; /* shouldn't it be ENOSYS? */
1485 error = security_inode_create(dir, dentry, mode);
1489 error = dir->i_op->create(dir, dentry, mode, nd);
1491 fsnotify_create(dir, dentry);
1495 int may_open(struct nameidata *nd, int acc_mode, int flag)
1497 struct dentry *dentry = nd->path.dentry;
1498 struct inode *inode = dentry->d_inode;
1504 if (S_ISLNK(inode->i_mode))
1507 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1511 * FIFO's, sockets and device files are special: they don't
1512 * actually live on the filesystem itself, and as such you
1513 * can write to them even if the filesystem is read-only.
1515 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1517 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1518 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1524 error = vfs_permission(nd, acc_mode);
1528 * An append-only file must be opened in append mode for writing.
1530 if (IS_APPEND(inode)) {
1531 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1537 /* O_NOATIME can only be set by the owner or superuser */
1538 if (flag & O_NOATIME)
1539 if (!is_owner_or_cap(inode))
1543 * Ensure there are no outstanding leases on the file.
1545 error = break_lease(inode, flag);
1549 if (flag & O_TRUNC) {
1550 error = get_write_access(inode);
1555 * Refuse to truncate files with mandatory locks held on them.
1557 error = locks_verify_locked(inode);
1559 error = security_path_truncate(&nd->path, 0,
1560 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1564 error = do_truncate(dentry, 0,
1565 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1568 put_write_access(inode);
1572 if (flag & FMODE_WRITE)
1579 * Be careful about ever adding any more callers of this
1580 * function. Its flags must be in the namei format, not
1581 * what get passed to sys_open().
1583 static int __open_namei_create(struct nameidata *nd, struct path *path,
1587 struct dentry *dir = nd->path.dentry;
1589 if (!IS_POSIXACL(dir->d_inode))
1590 mode &= ~current->fs->umask;
1591 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1594 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1596 mutex_unlock(&dir->d_inode->i_mutex);
1597 dput(nd->path.dentry);
1598 nd->path.dentry = path->dentry;
1601 /* Don't check for write permission, don't truncate */
1602 return may_open(nd, 0, flag & ~O_TRUNC);
1606 * Note that while the flag value (low two bits) for sys_open means:
1611 * it is changed into
1612 * 00 - no permissions needed
1613 * 01 - read-permission
1614 * 10 - write-permission
1616 * for the internal routines (ie open_namei()/follow_link() etc)
1617 * This is more logical, and also allows the 00 "no perm needed"
1618 * to be used for symlinks (where the permissions are checked
1622 static inline int open_to_namei_flags(int flag)
1624 if ((flag+1) & O_ACCMODE)
1629 static int open_will_write_to_fs(int flag, struct inode *inode)
1632 * We'll never write to the fs underlying
1635 if (special_file(inode->i_mode))
1637 return (flag & O_TRUNC);
1641 * Note that the low bits of the passed in "open_flag"
1642 * are not the same as in the local variable "flag". See
1643 * open_to_namei_flags() for more details.
1645 struct file *do_filp_open(int dfd, const char *pathname,
1646 int open_flag, int mode)
1649 struct nameidata nd;
1650 int acc_mode, error;
1655 int flag = open_to_namei_flags(open_flag);
1657 acc_mode = MAY_OPEN | ACC_MODE(flag);
1659 /* O_TRUNC implies we need access checks for write permissions */
1661 acc_mode |= MAY_WRITE;
1663 /* Allow the LSM permission hook to distinguish append
1664 access from general write access. */
1665 if (flag & O_APPEND)
1666 acc_mode |= MAY_APPEND;
1669 * The simplest case - just a plain lookup.
1671 if (!(flag & O_CREAT)) {
1672 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1675 return ERR_PTR(error);
1680 * Create - we need to know the parent.
1682 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
1684 return ERR_PTR(error);
1687 * We have the parent and last component. First of all, check
1688 * that we are not asked to creat(2) an obvious directory - that
1692 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1696 filp = get_empty_filp();
1699 nd.intent.open.file = filp;
1700 nd.intent.open.flags = flag;
1701 nd.intent.open.create_mode = mode;
1702 dir = nd.path.dentry;
1703 nd.flags &= ~LOOKUP_PARENT;
1704 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1706 nd.flags |= LOOKUP_EXCL;
1707 mutex_lock(&dir->d_inode->i_mutex);
1708 path.dentry = lookup_hash(&nd);
1709 path.mnt = nd.path.mnt;
1712 error = PTR_ERR(path.dentry);
1713 if (IS_ERR(path.dentry)) {
1714 mutex_unlock(&dir->d_inode->i_mutex);
1718 if (IS_ERR(nd.intent.open.file)) {
1719 error = PTR_ERR(nd.intent.open.file);
1720 goto exit_mutex_unlock;
1723 /* Negative dentry, just create the file */
1724 if (!path.dentry->d_inode) {
1726 * This write is needed to ensure that a
1727 * ro->rw transition does not occur between
1728 * the time when the file is created and when
1729 * a permanent write count is taken through
1730 * the 'struct file' in nameidata_to_filp().
1732 error = mnt_want_write(nd.path.mnt);
1734 goto exit_mutex_unlock;
1735 error = __open_namei_create(&nd, &path, flag, mode);
1737 mnt_drop_write(nd.path.mnt);
1740 filp = nameidata_to_filp(&nd, open_flag);
1741 mnt_drop_write(nd.path.mnt);
1746 * It already exists.
1748 mutex_unlock(&dir->d_inode->i_mutex);
1749 audit_inode(pathname, path.dentry);
1755 if (__follow_mount(&path)) {
1757 if (flag & O_NOFOLLOW)
1762 if (!path.dentry->d_inode)
1764 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1767 path_to_nameidata(&path, &nd);
1769 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1774 * 1. may_open() truncates a file
1775 * 2. a rw->ro mount transition occurs
1776 * 3. nameidata_to_filp() fails due to
1778 * That would be inconsistent, and should
1779 * be avoided. Taking this mnt write here
1780 * ensures that (2) can not occur.
1782 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1784 error = mnt_want_write(nd.path.mnt);
1788 error = may_open(&nd, acc_mode, flag);
1791 mnt_drop_write(nd.path.mnt);
1794 filp = nameidata_to_filp(&nd, open_flag);
1796 * It is now safe to drop the mnt write
1797 * because the filp has had a write taken
1801 mnt_drop_write(nd.path.mnt);
1805 mutex_unlock(&dir->d_inode->i_mutex);
1807 path_put_conditional(&path, &nd);
1809 if (!IS_ERR(nd.intent.open.file))
1810 release_open_intent(&nd);
1813 return ERR_PTR(error);
1817 if (flag & O_NOFOLLOW)
1820 * This is subtle. Instead of calling do_follow_link() we do the
1821 * thing by hands. The reason is that this way we have zero link_count
1822 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1823 * After that we have the parent and last component, i.e.
1824 * we are in the same situation as after the first path_walk().
1825 * Well, almost - if the last component is normal we get its copy
1826 * stored in nd->last.name and we will have to putname() it when we
1827 * are done. Procfs-like symlinks just set LAST_BIND.
1829 nd.flags |= LOOKUP_PARENT;
1830 error = security_inode_follow_link(path.dentry, &nd);
1833 error = __do_follow_link(&path, &nd);
1835 /* Does someone understand code flow here? Or it is only
1836 * me so stupid? Anathema to whoever designed this non-sense
1837 * with "intent.open".
1839 release_open_intent(&nd);
1840 return ERR_PTR(error);
1842 nd.flags &= ~LOOKUP_PARENT;
1843 if (nd.last_type == LAST_BIND)
1846 if (nd.last_type != LAST_NORM)
1848 if (nd.last.name[nd.last.len]) {
1849 __putname(nd.last.name);
1854 __putname(nd.last.name);
1857 dir = nd.path.dentry;
1858 mutex_lock(&dir->d_inode->i_mutex);
1859 path.dentry = lookup_hash(&nd);
1860 path.mnt = nd.path.mnt;
1861 __putname(nd.last.name);
1866 * filp_open - open file and return file pointer
1868 * @filename: path to open
1869 * @flags: open flags as per the open(2) second argument
1870 * @mode: mode for the new file if O_CREAT is set, else ignored
1872 * This is the helper to open a file from kernelspace if you really
1873 * have to. But in generally you should not do this, so please move
1874 * along, nothing to see here..
1876 struct file *filp_open(const char *filename, int flags, int mode)
1878 return do_filp_open(AT_FDCWD, filename, flags, mode);
1880 EXPORT_SYMBOL(filp_open);
1883 * lookup_create - lookup a dentry, creating it if it doesn't exist
1884 * @nd: nameidata info
1885 * @is_dir: directory flag
1887 * Simple function to lookup and return a dentry and create it
1888 * if it doesn't exist. Is SMP-safe.
1890 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1892 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1894 struct dentry *dentry = ERR_PTR(-EEXIST);
1896 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1898 * Yucky last component or no last component at all?
1899 * (foo/., foo/.., /////)
1901 if (nd->last_type != LAST_NORM)
1903 nd->flags &= ~LOOKUP_PARENT;
1904 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1905 nd->intent.open.flags = O_EXCL;
1908 * Do the final lookup.
1910 dentry = lookup_hash(nd);
1914 if (dentry->d_inode)
1917 * Special case - lookup gave negative, but... we had foo/bar/
1918 * From the vfs_mknod() POV we just have a negative dentry -
1919 * all is fine. Let's be bastards - you had / on the end, you've
1920 * been asking for (non-existent) directory. -ENOENT for you.
1922 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1924 dentry = ERR_PTR(-ENOENT);
1929 dentry = ERR_PTR(-EEXIST);
1933 EXPORT_SYMBOL_GPL(lookup_create);
1935 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1937 int error = may_create(dir, dentry);
1942 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1945 if (!dir->i_op || !dir->i_op->mknod)
1948 error = devcgroup_inode_mknod(mode, dev);
1952 error = security_inode_mknod(dir, dentry, mode, dev);
1957 error = dir->i_op->mknod(dir, dentry, mode, dev);
1959 fsnotify_create(dir, dentry);
1963 static int may_mknod(mode_t mode)
1965 switch (mode & S_IFMT) {
1971 case 0: /* zero mode translates to S_IFREG */
1980 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1985 struct dentry *dentry;
1986 struct nameidata nd;
1991 error = user_path_parent(dfd, filename, &nd, &tmp);
1995 dentry = lookup_create(&nd, 0);
1996 if (IS_ERR(dentry)) {
1997 error = PTR_ERR(dentry);
2000 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2001 mode &= ~current->fs->umask;
2002 error = may_mknod(mode);
2005 error = mnt_want_write(nd.path.mnt);
2008 error = security_path_mknod(&nd.path, dentry, mode, dev);
2010 goto out_drop_write;
2011 switch (mode & S_IFMT) {
2012 case 0: case S_IFREG:
2013 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2015 case S_IFCHR: case S_IFBLK:
2016 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2017 new_decode_dev(dev));
2019 case S_IFIFO: case S_IFSOCK:
2020 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2024 mnt_drop_write(nd.path.mnt);
2028 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2035 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2037 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2040 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2042 int error = may_create(dir, dentry);
2047 if (!dir->i_op || !dir->i_op->mkdir)
2050 mode &= (S_IRWXUGO|S_ISVTX);
2051 error = security_inode_mkdir(dir, dentry, mode);
2056 error = dir->i_op->mkdir(dir, dentry, mode);
2058 fsnotify_mkdir(dir, dentry);
2062 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2066 struct dentry *dentry;
2067 struct nameidata nd;
2069 error = user_path_parent(dfd, pathname, &nd, &tmp);
2073 dentry = lookup_create(&nd, 1);
2074 error = PTR_ERR(dentry);
2078 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2079 mode &= ~current->fs->umask;
2080 error = mnt_want_write(nd.path.mnt);
2083 error = security_path_mkdir(&nd.path, dentry, mode);
2085 goto out_drop_write;
2086 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2088 mnt_drop_write(nd.path.mnt);
2092 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2099 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2101 return sys_mkdirat(AT_FDCWD, pathname, mode);
2105 * We try to drop the dentry early: we should have
2106 * a usage count of 2 if we're the only user of this
2107 * dentry, and if that is true (possibly after pruning
2108 * the dcache), then we drop the dentry now.
2110 * A low-level filesystem can, if it choses, legally
2113 * if (!d_unhashed(dentry))
2116 * if it cannot handle the case of removing a directory
2117 * that is still in use by something else..
2119 void dentry_unhash(struct dentry *dentry)
2122 shrink_dcache_parent(dentry);
2123 spin_lock(&dcache_lock);
2124 spin_lock(&dentry->d_lock);
2125 if (atomic_read(&dentry->d_count) == 2)
2127 spin_unlock(&dentry->d_lock);
2128 spin_unlock(&dcache_lock);
2131 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2133 int error = may_delete(dir, dentry, 1);
2138 if (!dir->i_op || !dir->i_op->rmdir)
2143 mutex_lock(&dentry->d_inode->i_mutex);
2144 dentry_unhash(dentry);
2145 if (d_mountpoint(dentry))
2148 error = security_inode_rmdir(dir, dentry);
2150 error = dir->i_op->rmdir(dir, dentry);
2152 dentry->d_inode->i_flags |= S_DEAD;
2155 mutex_unlock(&dentry->d_inode->i_mutex);
2164 static long do_rmdir(int dfd, const char __user *pathname)
2168 struct dentry *dentry;
2169 struct nameidata nd;
2171 error = user_path_parent(dfd, pathname, &nd, &name);
2175 switch(nd.last_type) {
2187 nd.flags &= ~LOOKUP_PARENT;
2189 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2190 dentry = lookup_hash(&nd);
2191 error = PTR_ERR(dentry);
2194 error = mnt_want_write(nd.path.mnt);
2197 error = security_path_rmdir(&nd.path, dentry);
2200 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2202 mnt_drop_write(nd.path.mnt);
2206 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2213 asmlinkage long sys_rmdir(const char __user *pathname)
2215 return do_rmdir(AT_FDCWD, pathname);
2218 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2220 int error = may_delete(dir, dentry, 0);
2225 if (!dir->i_op || !dir->i_op->unlink)
2230 mutex_lock(&dentry->d_inode->i_mutex);
2231 if (d_mountpoint(dentry))
2234 error = security_inode_unlink(dir, dentry);
2236 error = dir->i_op->unlink(dir, dentry);
2238 mutex_unlock(&dentry->d_inode->i_mutex);
2240 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2241 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2242 fsnotify_link_count(dentry->d_inode);
2250 * Make sure that the actual truncation of the file will occur outside its
2251 * directory's i_mutex. Truncate can take a long time if there is a lot of
2252 * writeout happening, and we don't want to prevent access to the directory
2253 * while waiting on the I/O.
2255 static long do_unlinkat(int dfd, const char __user *pathname)
2259 struct dentry *dentry;
2260 struct nameidata nd;
2261 struct inode *inode = NULL;
2263 error = user_path_parent(dfd, pathname, &nd, &name);
2268 if (nd.last_type != LAST_NORM)
2271 nd.flags &= ~LOOKUP_PARENT;
2273 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2274 dentry = lookup_hash(&nd);
2275 error = PTR_ERR(dentry);
2276 if (!IS_ERR(dentry)) {
2277 /* Why not before? Because we want correct error value */
2278 if (nd.last.name[nd.last.len])
2280 inode = dentry->d_inode;
2282 atomic_inc(&inode->i_count);
2283 error = mnt_want_write(nd.path.mnt);
2286 error = security_path_unlink(&nd.path, dentry);
2289 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2291 mnt_drop_write(nd.path.mnt);
2295 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2297 iput(inode); /* truncate the inode here */
2304 error = !dentry->d_inode ? -ENOENT :
2305 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2309 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2311 if ((flag & ~AT_REMOVEDIR) != 0)
2314 if (flag & AT_REMOVEDIR)
2315 return do_rmdir(dfd, pathname);
2317 return do_unlinkat(dfd, pathname);
2320 asmlinkage long sys_unlink(const char __user *pathname)
2322 return do_unlinkat(AT_FDCWD, pathname);
2325 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2327 int error = may_create(dir, dentry);
2332 if (!dir->i_op || !dir->i_op->symlink)
2335 error = security_inode_symlink(dir, dentry, oldname);
2340 error = dir->i_op->symlink(dir, dentry, oldname);
2342 fsnotify_create(dir, dentry);
2346 asmlinkage long sys_symlinkat(const char __user *oldname,
2347 int newdfd, const char __user *newname)
2352 struct dentry *dentry;
2353 struct nameidata nd;
2355 from = getname(oldname);
2357 return PTR_ERR(from);
2359 error = user_path_parent(newdfd, newname, &nd, &to);
2363 dentry = lookup_create(&nd, 0);
2364 error = PTR_ERR(dentry);
2368 error = mnt_want_write(nd.path.mnt);
2371 error = security_path_symlink(&nd.path, dentry, from);
2373 goto out_drop_write;
2374 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2376 mnt_drop_write(nd.path.mnt);
2380 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2388 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2390 return sys_symlinkat(oldname, AT_FDCWD, newname);
2393 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2395 struct inode *inode = old_dentry->d_inode;
2401 error = may_create(dir, new_dentry);
2405 if (dir->i_sb != inode->i_sb)
2409 * A link to an append-only or immutable file cannot be created.
2411 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2413 if (!dir->i_op || !dir->i_op->link)
2415 if (S_ISDIR(inode->i_mode))
2418 error = security_inode_link(old_dentry, dir, new_dentry);
2422 mutex_lock(&inode->i_mutex);
2424 error = dir->i_op->link(old_dentry, dir, new_dentry);
2425 mutex_unlock(&inode->i_mutex);
2427 fsnotify_link(dir, inode, new_dentry);
2432 * Hardlinks are often used in delicate situations. We avoid
2433 * security-related surprises by not following symlinks on the
2436 * We don't follow them on the oldname either to be compatible
2437 * with linux 2.0, and to avoid hard-linking to directories
2438 * and other special files. --ADM
2440 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2441 int newdfd, const char __user *newname,
2444 struct dentry *new_dentry;
2445 struct nameidata nd;
2446 struct path old_path;
2450 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2453 error = user_path_at(olddfd, oldname,
2454 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2459 error = user_path_parent(newdfd, newname, &nd, &to);
2463 if (old_path.mnt != nd.path.mnt)
2465 new_dentry = lookup_create(&nd, 0);
2466 error = PTR_ERR(new_dentry);
2467 if (IS_ERR(new_dentry))
2469 error = mnt_want_write(nd.path.mnt);
2472 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2474 goto out_drop_write;
2475 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2477 mnt_drop_write(nd.path.mnt);
2481 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2486 path_put(&old_path);
2491 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2493 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2497 * The worst of all namespace operations - renaming directory. "Perverted"
2498 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2500 * a) we can get into loop creation. Check is done in is_subdir().
2501 * b) race potential - two innocent renames can create a loop together.
2502 * That's where 4.4 screws up. Current fix: serialization on
2503 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2505 * c) we have to lock _three_ objects - parents and victim (if it exists).
2506 * And that - after we got ->i_mutex on parents (until then we don't know
2507 * whether the target exists). Solution: try to be smart with locking
2508 * order for inodes. We rely on the fact that tree topology may change
2509 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2510 * move will be locked. Thus we can rank directories by the tree
2511 * (ancestors first) and rank all non-directories after them.
2512 * That works since everybody except rename does "lock parent, lookup,
2513 * lock child" and rename is under ->s_vfs_rename_mutex.
2514 * HOWEVER, it relies on the assumption that any object with ->lookup()
2515 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2516 * we'd better make sure that there's no link(2) for them.
2517 * d) some filesystems don't support opened-but-unlinked directories,
2518 * either because of layout or because they are not ready to deal with
2519 * all cases correctly. The latter will be fixed (taking this sort of
2520 * stuff into VFS), but the former is not going away. Solution: the same
2521 * trick as in rmdir().
2522 * e) conversion from fhandle to dentry may come in the wrong moment - when
2523 * we are removing the target. Solution: we will have to grab ->i_mutex
2524 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2525 * ->i_mutex on parents, which works but leads to some truely excessive
2528 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2529 struct inode *new_dir, struct dentry *new_dentry)
2532 struct inode *target;
2535 * If we are going to change the parent - check write permissions,
2536 * we'll need to flip '..'.
2538 if (new_dir != old_dir) {
2539 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2544 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2548 target = new_dentry->d_inode;
2550 mutex_lock(&target->i_mutex);
2551 dentry_unhash(new_dentry);
2553 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2556 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2559 target->i_flags |= S_DEAD;
2560 mutex_unlock(&target->i_mutex);
2561 if (d_unhashed(new_dentry))
2562 d_rehash(new_dentry);
2566 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2567 d_move(old_dentry,new_dentry);
2571 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2572 struct inode *new_dir, struct dentry *new_dentry)
2574 struct inode *target;
2577 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2582 target = new_dentry->d_inode;
2584 mutex_lock(&target->i_mutex);
2585 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2588 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2590 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2591 d_move(old_dentry, new_dentry);
2594 mutex_unlock(&target->i_mutex);
2599 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2600 struct inode *new_dir, struct dentry *new_dentry)
2603 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2604 const char *old_name;
2606 if (old_dentry->d_inode == new_dentry->d_inode)
2609 error = may_delete(old_dir, old_dentry, is_dir);
2613 if (!new_dentry->d_inode)
2614 error = may_create(new_dir, new_dentry);
2616 error = may_delete(new_dir, new_dentry, is_dir);
2620 if (!old_dir->i_op || !old_dir->i_op->rename)
2623 DQUOT_INIT(old_dir);
2624 DQUOT_INIT(new_dir);
2626 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2629 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2631 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2633 const char *new_name = old_dentry->d_name.name;
2634 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2635 new_dentry->d_inode, old_dentry);
2637 fsnotify_oldname_free(old_name);
2642 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2643 int newdfd, const char __user *newname)
2645 struct dentry *old_dir, *new_dir;
2646 struct dentry *old_dentry, *new_dentry;
2647 struct dentry *trap;
2648 struct nameidata oldnd, newnd;
2653 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2657 error = user_path_parent(newdfd, newname, &newnd, &to);
2662 if (oldnd.path.mnt != newnd.path.mnt)
2665 old_dir = oldnd.path.dentry;
2667 if (oldnd.last_type != LAST_NORM)
2670 new_dir = newnd.path.dentry;
2671 if (newnd.last_type != LAST_NORM)
2674 oldnd.flags &= ~LOOKUP_PARENT;
2675 newnd.flags &= ~LOOKUP_PARENT;
2676 newnd.flags |= LOOKUP_RENAME_TARGET;
2678 trap = lock_rename(new_dir, old_dir);
2680 old_dentry = lookup_hash(&oldnd);
2681 error = PTR_ERR(old_dentry);
2682 if (IS_ERR(old_dentry))
2684 /* source must exist */
2686 if (!old_dentry->d_inode)
2688 /* unless the source is a directory trailing slashes give -ENOTDIR */
2689 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2691 if (oldnd.last.name[oldnd.last.len])
2693 if (newnd.last.name[newnd.last.len])
2696 /* source should not be ancestor of target */
2698 if (old_dentry == trap)
2700 new_dentry = lookup_hash(&newnd);
2701 error = PTR_ERR(new_dentry);
2702 if (IS_ERR(new_dentry))
2704 /* target should not be an ancestor of source */
2706 if (new_dentry == trap)
2709 error = mnt_want_write(oldnd.path.mnt);
2712 error = security_path_rename(&oldnd.path, old_dentry,
2713 &newnd.path, new_dentry);
2716 error = vfs_rename(old_dir->d_inode, old_dentry,
2717 new_dir->d_inode, new_dentry);
2719 mnt_drop_write(oldnd.path.mnt);
2725 unlock_rename(new_dir, old_dir);
2727 path_put(&newnd.path);
2730 path_put(&oldnd.path);
2736 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2738 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2741 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2745 len = PTR_ERR(link);
2750 if (len > (unsigned) buflen)
2752 if (copy_to_user(buffer, link, len))
2759 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2760 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2761 * using) it for any given inode is up to filesystem.
2763 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2765 struct nameidata nd;
2770 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2772 return PTR_ERR(cookie);
2774 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2775 if (dentry->d_inode->i_op->put_link)
2776 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2780 int vfs_follow_link(struct nameidata *nd, const char *link)
2782 return __vfs_follow_link(nd, link);
2785 /* get the link contents into pagecache */
2786 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2790 struct address_space *mapping = dentry->d_inode->i_mapping;
2791 page = read_mapping_page(mapping, 0, NULL);
2796 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2800 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2802 struct page *page = NULL;
2803 char *s = page_getlink(dentry, &page);
2804 int res = vfs_readlink(dentry,buffer,buflen,s);
2807 page_cache_release(page);
2812 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2814 struct page *page = NULL;
2815 nd_set_link(nd, page_getlink(dentry, &page));
2819 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2821 struct page *page = cookie;
2825 page_cache_release(page);
2829 int __page_symlink(struct inode *inode, const char *symname, int len,
2832 struct address_space *mapping = inode->i_mapping;
2839 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2840 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2844 kaddr = kmap_atomic(page, KM_USER0);
2845 memcpy(kaddr, symname, len-1);
2846 kunmap_atomic(kaddr, KM_USER0);
2848 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2855 mark_inode_dirty(inode);
2861 int page_symlink(struct inode *inode, const char *symname, int len)
2863 return __page_symlink(inode, symname, len,
2864 mapping_gfp_mask(inode->i_mapping));
2867 const struct inode_operations page_symlink_inode_operations = {
2868 .readlink = generic_readlink,
2869 .follow_link = page_follow_link_light,
2870 .put_link = page_put_link,
2873 EXPORT_SYMBOL(user_path_at);
2874 EXPORT_SYMBOL(follow_down);
2875 EXPORT_SYMBOL(follow_up);
2876 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2877 EXPORT_SYMBOL(getname);
2878 EXPORT_SYMBOL(lock_rename);
2879 EXPORT_SYMBOL(lookup_one_len);
2880 EXPORT_SYMBOL(page_follow_link_light);
2881 EXPORT_SYMBOL(page_put_link);
2882 EXPORT_SYMBOL(page_readlink);
2883 EXPORT_SYMBOL(__page_symlink);
2884 EXPORT_SYMBOL(page_symlink);
2885 EXPORT_SYMBOL(page_symlink_inode_operations);
2886 EXPORT_SYMBOL(path_lookup);
2887 EXPORT_SYMBOL(kern_path);
2888 EXPORT_SYMBOL(vfs_path_lookup);
2889 EXPORT_SYMBOL(inode_permission);
2890 EXPORT_SYMBOL(vfs_permission);
2891 EXPORT_SYMBOL(file_permission);
2892 EXPORT_SYMBOL(unlock_rename);
2893 EXPORT_SYMBOL(vfs_create);
2894 EXPORT_SYMBOL(vfs_follow_link);
2895 EXPORT_SYMBOL(vfs_link);
2896 EXPORT_SYMBOL(vfs_mkdir);
2897 EXPORT_SYMBOL(vfs_mknod);
2898 EXPORT_SYMBOL(generic_permission);
2899 EXPORT_SYMBOL(vfs_readlink);
2900 EXPORT_SYMBOL(vfs_rename);
2901 EXPORT_SYMBOL(vfs_rmdir);
2902 EXPORT_SYMBOL(vfs_symlink);
2903 EXPORT_SYMBOL(vfs_unlink);
2904 EXPORT_SYMBOL(dentry_unhash);
2905 EXPORT_SYMBOL(generic_readlink);