4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <asm/namei.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int link_path_walk(const char *name, struct nameidata *nd);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
185 umode_t mode = inode->i_mode;
187 if (current->fsuid == inode->i_uid)
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
198 if (in_group_p(inode->i_gid))
203 * If the DACs are ok we don't need any capability check.
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
219 * Searching includes executable on directories, else just read.
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
228 int permission(struct inode *inode, int mask, struct nameidata *nd)
231 struct vfsmount *mnt = NULL;
236 if (mask & MAY_WRITE) {
237 umode_t mode = inode->i_mode;
240 * Nobody gets write access to a read-only fs.
242 if (IS_RDONLY(inode) &&
243 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
247 * Nobody gets write access to an immutable file.
249 if (IS_IMMUTABLE(inode))
253 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
255 * MAY_EXEC on regular files is denied if the fs is mounted
256 * with the "noexec" flag.
258 if (mnt && (mnt->mnt_flags & MNT_NOEXEC))
262 /* Ordinary permission routines do not understand MAY_APPEND. */
263 submask = mask & ~MAY_APPEND;
264 if (inode->i_op && inode->i_op->permission) {
265 retval = inode->i_op->permission(inode, submask, nd);
268 * Exec permission on a regular file is denied if none
269 * of the execute bits are set.
271 * This check should be done by the ->permission()
274 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
275 !(inode->i_mode & S_IXUGO))
279 retval = generic_permission(inode, submask, NULL);
284 return security_inode_permission(inode, mask, nd);
288 * vfs_permission - check for access rights to a given path
289 * @nd: lookup result that describes the path
290 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
292 * Used to check for read/write/execute permissions on a path.
293 * We use "fsuid" for this, letting us set arbitrary permissions
294 * for filesystem access without changing the "normal" uids which
295 * are used for other things.
297 int vfs_permission(struct nameidata *nd, int mask)
299 return permission(nd->path.dentry->d_inode, mask, nd);
303 * file_permission - check for additional access rights to a given file
304 * @file: file to check access rights for
305 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
307 * Used to check for read/write/execute permissions on an already opened
311 * Do not use this function in new code. All access checks should
312 * be done using vfs_permission().
314 int file_permission(struct file *file, int mask)
316 return permission(file->f_path.dentry->d_inode, mask, NULL);
320 * get_write_access() gets write permission for a file.
321 * put_write_access() releases this write permission.
322 * This is used for regular files.
323 * We cannot support write (and maybe mmap read-write shared) accesses and
324 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
325 * can have the following values:
326 * 0: no writers, no VM_DENYWRITE mappings
327 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
328 * > 0: (i_writecount) users are writing to the file.
330 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
331 * except for the cases where we don't hold i_writecount yet. Then we need to
332 * use {get,deny}_write_access() - these functions check the sign and refuse
333 * to do the change if sign is wrong. Exclusion between them is provided by
334 * the inode->i_lock spinlock.
337 int get_write_access(struct inode * inode)
339 spin_lock(&inode->i_lock);
340 if (atomic_read(&inode->i_writecount) < 0) {
341 spin_unlock(&inode->i_lock);
344 atomic_inc(&inode->i_writecount);
345 spin_unlock(&inode->i_lock);
350 int deny_write_access(struct file * file)
352 struct inode *inode = file->f_path.dentry->d_inode;
354 spin_lock(&inode->i_lock);
355 if (atomic_read(&inode->i_writecount) > 0) {
356 spin_unlock(&inode->i_lock);
359 atomic_dec(&inode->i_writecount);
360 spin_unlock(&inode->i_lock);
365 void path_release(struct nameidata *nd)
367 dput(nd->path.dentry);
368 mntput(nd->path.mnt);
372 * release_open_intent - free up open intent resources
373 * @nd: pointer to nameidata
375 void release_open_intent(struct nameidata *nd)
377 if (nd->intent.open.file->f_path.dentry == NULL)
378 put_filp(nd->intent.open.file);
380 fput(nd->intent.open.file);
383 static inline struct dentry *
384 do_revalidate(struct dentry *dentry, struct nameidata *nd)
386 int status = dentry->d_op->d_revalidate(dentry, nd);
387 if (unlikely(status <= 0)) {
389 * The dentry failed validation.
390 * If d_revalidate returned 0 attempt to invalidate
391 * the dentry otherwise d_revalidate is asking us
392 * to return a fail status.
395 if (!d_invalidate(dentry)) {
401 dentry = ERR_PTR(status);
408 * Internal lookup() using the new generic dcache.
411 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
413 struct dentry * dentry = __d_lookup(parent, name);
415 /* lockess __d_lookup may fail due to concurrent d_move()
416 * in some unrelated directory, so try with d_lookup
419 dentry = d_lookup(parent, name);
421 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
422 dentry = do_revalidate(dentry, nd);
428 * Short-cut version of permission(), for calling by
429 * path_walk(), when dcache lock is held. Combines parts
430 * of permission() and generic_permission(), and tests ONLY for
431 * MAY_EXEC permission.
433 * If appropriate, check DAC only. If not appropriate, or
434 * short-cut DAC fails, then call permission() to do more
435 * complete permission check.
437 static int exec_permission_lite(struct inode *inode,
438 struct nameidata *nd)
440 umode_t mode = inode->i_mode;
442 if (inode->i_op && inode->i_op->permission)
445 if (current->fsuid == inode->i_uid)
447 else if (in_group_p(inode->i_gid))
453 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
456 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
459 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
464 return security_inode_permission(inode, MAY_EXEC, nd);
468 * This is called when everything else fails, and we actually have
469 * to go to the low-level filesystem to find out what we should do..
471 * We get the directory semaphore, and after getting that we also
472 * make sure that nobody added the entry to the dcache in the meantime..
475 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
477 struct dentry * result;
478 struct inode *dir = parent->d_inode;
480 mutex_lock(&dir->i_mutex);
482 * First re-do the cached lookup just in case it was created
483 * while we waited for the directory semaphore..
485 * FIXME! This could use version numbering or similar to
486 * avoid unnecessary cache lookups.
488 * The "dcache_lock" is purely to protect the RCU list walker
489 * from concurrent renames at this point (we mustn't get false
490 * negatives from the RCU list walk here, unlike the optimistic
493 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
495 result = d_lookup(parent, name);
497 struct dentry * dentry = d_alloc(parent, name);
498 result = ERR_PTR(-ENOMEM);
500 result = dir->i_op->lookup(dir, dentry, nd);
506 mutex_unlock(&dir->i_mutex);
511 * Uhhuh! Nasty case: the cache was re-populated while
512 * we waited on the semaphore. Need to revalidate.
514 mutex_unlock(&dir->i_mutex);
515 if (result->d_op && result->d_op->d_revalidate) {
516 result = do_revalidate(result, nd);
518 result = ERR_PTR(-ENOENT);
523 static int __emul_lookup_dentry(const char *, struct nameidata *);
526 static __always_inline int
527 walk_init_root(const char *name, struct nameidata *nd)
529 struct fs_struct *fs = current->fs;
531 read_lock(&fs->lock);
532 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
533 nd->path.mnt = mntget(fs->altrootmnt);
534 nd->path.dentry = dget(fs->altroot);
535 read_unlock(&fs->lock);
536 if (__emul_lookup_dentry(name,nd))
538 read_lock(&fs->lock);
540 nd->path.mnt = mntget(fs->rootmnt);
541 nd->path.dentry = dget(fs->root);
542 read_unlock(&fs->lock);
546 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
555 if (!walk_init_root(link, nd))
556 /* weird __emul_prefix() stuff did it */
559 res = link_path_walk(link, nd);
561 if (nd->depth || res || nd->last_type!=LAST_NORM)
564 * If it is an iterative symlinks resolution in open_namei() we
565 * have to copy the last component. And all that crap because of
566 * bloody create() on broken symlinks. Furrfu...
569 if (unlikely(!name)) {
573 strcpy(name, nd->last.name);
574 nd->last.name = name;
578 return PTR_ERR(link);
581 static inline void dput_path(struct path *path, struct nameidata *nd)
584 if (path->mnt != nd->path.mnt)
588 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
590 dput(nd->path.dentry);
591 if (nd->path.mnt != path->mnt)
592 mntput(nd->path.mnt);
593 nd->path.mnt = path->mnt;
594 nd->path.dentry = path->dentry;
597 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
601 struct dentry *dentry = path->dentry;
603 touch_atime(path->mnt, dentry);
604 nd_set_link(nd, NULL);
606 if (path->mnt != nd->path.mnt) {
607 path_to_nameidata(path, nd);
611 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
612 error = PTR_ERR(cookie);
613 if (!IS_ERR(cookie)) {
614 char *s = nd_get_link(nd);
617 error = __vfs_follow_link(nd, s);
618 if (dentry->d_inode->i_op->put_link)
619 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
628 * This limits recursive symlink follows to 8, while
629 * limiting consecutive symlinks to 40.
631 * Without that kind of total limit, nasty chains of consecutive
632 * symlinks can cause almost arbitrarily long lookups.
634 static inline int do_follow_link(struct path *path, struct nameidata *nd)
637 if (current->link_count >= MAX_NESTED_LINKS)
639 if (current->total_link_count >= 40)
641 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
643 err = security_inode_follow_link(path->dentry, nd);
646 current->link_count++;
647 current->total_link_count++;
649 err = __do_follow_link(path, nd);
650 current->link_count--;
659 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
661 struct vfsmount *parent;
662 struct dentry *mountpoint;
663 spin_lock(&vfsmount_lock);
664 parent=(*mnt)->mnt_parent;
665 if (parent == *mnt) {
666 spin_unlock(&vfsmount_lock);
670 mountpoint=dget((*mnt)->mnt_mountpoint);
671 spin_unlock(&vfsmount_lock);
673 *dentry = mountpoint;
679 /* no need for dcache_lock, as serialization is taken care in
682 static int __follow_mount(struct path *path)
685 while (d_mountpoint(path->dentry)) {
686 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
693 path->dentry = dget(mounted->mnt_root);
699 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
701 while (d_mountpoint(*dentry)) {
702 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
708 *dentry = dget(mounted->mnt_root);
712 /* no need for dcache_lock, as serialization is taken care in
715 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
717 struct vfsmount *mounted;
719 mounted = lookup_mnt(*mnt, *dentry);
724 *dentry = dget(mounted->mnt_root);
730 static __always_inline void follow_dotdot(struct nameidata *nd)
732 struct fs_struct *fs = current->fs;
735 struct vfsmount *parent;
736 struct dentry *old = nd->path.dentry;
738 read_lock(&fs->lock);
739 if (nd->path.dentry == fs->root &&
740 nd->path.mnt == fs->rootmnt) {
741 read_unlock(&fs->lock);
744 read_unlock(&fs->lock);
745 spin_lock(&dcache_lock);
746 if (nd->path.dentry != nd->path.mnt->mnt_root) {
747 nd->path.dentry = dget(nd->path.dentry->d_parent);
748 spin_unlock(&dcache_lock);
752 spin_unlock(&dcache_lock);
753 spin_lock(&vfsmount_lock);
754 parent = nd->path.mnt->mnt_parent;
755 if (parent == nd->path.mnt) {
756 spin_unlock(&vfsmount_lock);
760 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
761 spin_unlock(&vfsmount_lock);
763 mntput(nd->path.mnt);
764 nd->path.mnt = parent;
766 follow_mount(&nd->path.mnt, &nd->path.dentry);
770 * It's more convoluted than I'd like it to be, but... it's still fairly
771 * small and for now I'd prefer to have fast path as straight as possible.
772 * It _is_ time-critical.
774 static int do_lookup(struct nameidata *nd, struct qstr *name,
777 struct vfsmount *mnt = nd->path.mnt;
778 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
782 if (dentry->d_op && dentry->d_op->d_revalidate)
783 goto need_revalidate;
786 path->dentry = dentry;
787 __follow_mount(path);
791 dentry = real_lookup(nd->path.dentry, name, nd);
797 dentry = do_revalidate(dentry, nd);
805 return PTR_ERR(dentry);
810 * This is the basic name resolution function, turning a pathname into
811 * the final dentry. We expect 'base' to be positive and a directory.
813 * Returns 0 and nd will have valid dentry and mnt on success.
814 * Returns error and drops reference to input namei data on failure.
816 static int __link_path_walk(const char *name, struct nameidata *nd)
821 unsigned int lookup_flags = nd->flags;
828 inode = nd->path.dentry->d_inode;
830 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
832 /* At this point we know we have a real path component. */
838 nd->flags |= LOOKUP_CONTINUE;
839 err = exec_permission_lite(inode, nd);
841 err = vfs_permission(nd, MAY_EXEC);
846 c = *(const unsigned char *)name;
848 hash = init_name_hash();
851 hash = partial_name_hash(c, hash);
852 c = *(const unsigned char *)name;
853 } while (c && (c != '/'));
854 this.len = name - (const char *) this.name;
855 this.hash = end_name_hash(hash);
857 /* remove trailing slashes? */
860 while (*++name == '/');
862 goto last_with_slashes;
865 * "." and ".." are special - ".." especially so because it has
866 * to be able to know about the current root directory and
867 * parent relationships.
869 if (this.name[0] == '.') switch (this.len) {
873 if (this.name[1] != '.')
876 inode = nd->path.dentry->d_inode;
882 * See if the low-level filesystem might want
883 * to use its own hash..
885 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
886 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
891 /* This does the actual lookups.. */
892 err = do_lookup(nd, &this, &next);
897 inode = next.dentry->d_inode;
904 if (inode->i_op->follow_link) {
905 err = do_follow_link(&next, nd);
909 inode = nd->path.dentry->d_inode;
916 path_to_nameidata(&next, nd);
918 if (!inode->i_op->lookup)
921 /* here ends the main loop */
924 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
926 /* Clear LOOKUP_CONTINUE iff it was previously unset */
927 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
928 if (lookup_flags & LOOKUP_PARENT)
930 if (this.name[0] == '.') switch (this.len) {
934 if (this.name[1] != '.')
937 inode = nd->path.dentry->d_inode;
942 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
943 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
948 err = do_lookup(nd, &this, &next);
951 inode = next.dentry->d_inode;
952 if ((lookup_flags & LOOKUP_FOLLOW)
953 && inode && inode->i_op && inode->i_op->follow_link) {
954 err = do_follow_link(&next, nd);
957 inode = nd->path.dentry->d_inode;
959 path_to_nameidata(&next, nd);
963 if (lookup_flags & LOOKUP_DIRECTORY) {
965 if (!inode->i_op || !inode->i_op->lookup)
971 nd->last_type = LAST_NORM;
972 if (this.name[0] != '.')
975 nd->last_type = LAST_DOT;
976 else if (this.len == 2 && this.name[1] == '.')
977 nd->last_type = LAST_DOTDOT;
982 * We bypassed the ordinary revalidation routines.
983 * We may need to check the cached dentry for staleness.
985 if (nd->path.dentry && nd->path.dentry->d_sb &&
986 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
988 /* Note: we do not d_invalidate() */
989 if (!nd->path.dentry->d_op->d_revalidate(
990 nd->path.dentry, nd))
996 dput_path(&next, nd);
1005 * Wrapper to retry pathname resolution whenever the underlying
1006 * file system returns an ESTALE.
1008 * Retry the whole path once, forcing real lookup requests
1009 * instead of relying on the dcache.
1011 static int link_path_walk(const char *name, struct nameidata *nd)
1013 struct nameidata save = *nd;
1016 /* make sure the stuff we saved doesn't go away */
1017 dget(save.path.dentry);
1018 mntget(save.path.mnt);
1020 result = __link_path_walk(name, nd);
1021 if (result == -ESTALE) {
1023 dget(nd->path.dentry);
1024 mntget(nd->path.mnt);
1025 nd->flags |= LOOKUP_REVAL;
1026 result = __link_path_walk(name, nd);
1029 dput(save.path.dentry);
1030 mntput(save.path.mnt);
1035 static int path_walk(const char *name, struct nameidata *nd)
1037 current->total_link_count = 0;
1038 return link_path_walk(name, nd);
1042 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1043 * everything is done. Returns 0 and drops input nd, if lookup failed;
1045 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1047 if (path_walk(name, nd))
1048 return 0; /* something went wrong... */
1050 if (!nd->path.dentry->d_inode ||
1051 S_ISDIR(nd->path.dentry->d_inode->i_mode)) {
1052 struct dentry *old_dentry = nd->path.dentry;
1053 struct vfsmount *old_mnt = nd->path.mnt;
1054 struct qstr last = nd->last;
1055 int last_type = nd->last_type;
1056 struct fs_struct *fs = current->fs;
1059 * NAME was not found in alternate root or it's a directory.
1060 * Try to find it in the normal root:
1062 nd->last_type = LAST_ROOT;
1063 read_lock(&fs->lock);
1064 nd->path.mnt = mntget(fs->rootmnt);
1065 nd->path.dentry = dget(fs->root);
1066 read_unlock(&fs->lock);
1067 if (path_walk(name, nd) == 0) {
1068 if (nd->path.dentry->d_inode) {
1075 nd->path.dentry = old_dentry;
1076 nd->path.mnt = old_mnt;
1078 nd->last_type = last_type;
1083 void set_fs_altroot(void)
1085 char *emul = __emul_prefix();
1086 struct nameidata nd;
1087 struct vfsmount *mnt = NULL, *oldmnt;
1088 struct dentry *dentry = NULL, *olddentry;
1090 struct fs_struct *fs = current->fs;
1094 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1097 dentry = nd.path.dentry;
1100 write_lock(&fs->lock);
1101 oldmnt = fs->altrootmnt;
1102 olddentry = fs->altroot;
1103 fs->altrootmnt = mnt;
1104 fs->altroot = dentry;
1105 write_unlock(&fs->lock);
1112 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1113 static int do_path_lookup(int dfd, const char *name,
1114 unsigned int flags, struct nameidata *nd)
1119 struct fs_struct *fs = current->fs;
1121 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1126 read_lock(&fs->lock);
1127 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1128 nd->path.mnt = mntget(fs->altrootmnt);
1129 nd->path.dentry = dget(fs->altroot);
1130 read_unlock(&fs->lock);
1131 if (__emul_lookup_dentry(name,nd))
1132 goto out; /* found in altroot */
1133 read_lock(&fs->lock);
1135 nd->path.mnt = mntget(fs->rootmnt);
1136 nd->path.dentry = dget(fs->root);
1137 read_unlock(&fs->lock);
1138 } else if (dfd == AT_FDCWD) {
1139 read_lock(&fs->lock);
1140 nd->path.mnt = mntget(fs->pwdmnt);
1141 nd->path.dentry = dget(fs->pwd);
1142 read_unlock(&fs->lock);
1144 struct dentry *dentry;
1146 file = fget_light(dfd, &fput_needed);
1151 dentry = file->f_path.dentry;
1154 if (!S_ISDIR(dentry->d_inode->i_mode))
1157 retval = file_permission(file, MAY_EXEC);
1161 nd->path.mnt = mntget(file->f_path.mnt);
1162 nd->path.dentry = dget(dentry);
1164 fput_light(file, fput_needed);
1167 retval = path_walk(name, nd);
1169 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1170 nd->path.dentry->d_inode))
1171 audit_inode(name, nd->path.dentry);
1176 fput_light(file, fput_needed);
1180 int path_lookup(const char *name, unsigned int flags,
1181 struct nameidata *nd)
1183 return do_path_lookup(AT_FDCWD, name, flags, nd);
1187 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1188 * @dentry: pointer to dentry of the base directory
1189 * @mnt: pointer to vfs mount of the base directory
1190 * @name: pointer to file name
1191 * @flags: lookup flags
1192 * @nd: pointer to nameidata
1194 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1195 const char *name, unsigned int flags,
1196 struct nameidata *nd)
1200 /* same as do_path_lookup */
1201 nd->last_type = LAST_ROOT;
1205 nd->path.mnt = mntget(mnt);
1206 nd->path.dentry = dget(dentry);
1208 retval = path_walk(name, nd);
1209 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1210 nd->path.dentry->d_inode))
1211 audit_inode(name, nd->path.dentry);
1217 static int __path_lookup_intent_open(int dfd, const char *name,
1218 unsigned int lookup_flags, struct nameidata *nd,
1219 int open_flags, int create_mode)
1221 struct file *filp = get_empty_filp();
1226 nd->intent.open.file = filp;
1227 nd->intent.open.flags = open_flags;
1228 nd->intent.open.create_mode = create_mode;
1229 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1230 if (IS_ERR(nd->intent.open.file)) {
1232 err = PTR_ERR(nd->intent.open.file);
1235 } else if (err != 0)
1236 release_open_intent(nd);
1241 * path_lookup_open - lookup a file path with open intent
1242 * @dfd: the directory to use as base, or AT_FDCWD
1243 * @name: pointer to file name
1244 * @lookup_flags: lookup intent flags
1245 * @nd: pointer to nameidata
1246 * @open_flags: open intent flags
1248 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1249 struct nameidata *nd, int open_flags)
1251 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1256 * path_lookup_create - lookup a file path with open + create intent
1257 * @dfd: the directory to use as base, or AT_FDCWD
1258 * @name: pointer to file name
1259 * @lookup_flags: lookup intent flags
1260 * @nd: pointer to nameidata
1261 * @open_flags: open intent flags
1262 * @create_mode: create intent flags
1264 static int path_lookup_create(int dfd, const char *name,
1265 unsigned int lookup_flags, struct nameidata *nd,
1266 int open_flags, int create_mode)
1268 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1269 nd, open_flags, create_mode);
1272 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1273 struct nameidata *nd, int open_flags)
1275 char *tmp = getname(name);
1276 int err = PTR_ERR(tmp);
1279 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1285 static struct dentry *__lookup_hash(struct qstr *name,
1286 struct dentry *base, struct nameidata *nd)
1288 struct dentry *dentry;
1289 struct inode *inode;
1292 inode = base->d_inode;
1295 * See if the low-level filesystem might want
1296 * to use its own hash..
1298 if (base->d_op && base->d_op->d_hash) {
1299 err = base->d_op->d_hash(base, name);
1300 dentry = ERR_PTR(err);
1305 dentry = cached_lookup(base, name, nd);
1307 struct dentry *new = d_alloc(base, name);
1308 dentry = ERR_PTR(-ENOMEM);
1311 dentry = inode->i_op->lookup(inode, new, nd);
1322 * Restricted form of lookup. Doesn't follow links, single-component only,
1323 * needs parent already locked. Doesn't follow mounts.
1326 static struct dentry *lookup_hash(struct nameidata *nd)
1330 err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
1332 return ERR_PTR(err);
1333 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1336 static int __lookup_one_len(const char *name, struct qstr *this,
1337 struct dentry *base, int len)
1347 hash = init_name_hash();
1349 c = *(const unsigned char *)name++;
1350 if (c == '/' || c == '\0')
1352 hash = partial_name_hash(c, hash);
1354 this->hash = end_name_hash(hash);
1359 * lookup_one_len: filesystem helper to lookup single pathname component
1360 * @name: pathname component to lookup
1361 * @base: base directory to lookup from
1362 * @len: maximum length @len should be interpreted to
1364 * Note that this routine is purely a helper for filesystem useage and should
1365 * not be called by generic code. Also note that by using this function to
1366 * nameidata argument is passed to the filesystem methods and a filesystem
1367 * using this helper needs to be prepared for that.
1369 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1374 err = __lookup_one_len(name, &this, base, len);
1376 return ERR_PTR(err);
1378 err = permission(base->d_inode, MAY_EXEC, NULL);
1380 return ERR_PTR(err);
1381 return __lookup_hash(&this, base, NULL);
1385 * lookup_one_noperm - bad hack for sysfs
1386 * @name: pathname component to lookup
1387 * @base: base directory to lookup from
1389 * This is a variant of lookup_one_len that doesn't perform any permission
1390 * checks. It's a horrible hack to work around the braindead sysfs
1391 * architecture and should not be used anywhere else.
1393 * DON'T USE THIS FUNCTION EVER, thanks.
1395 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1400 err = __lookup_one_len(name, &this, base, strlen(name));
1402 return ERR_PTR(err);
1403 return __lookup_hash(&this, base, NULL);
1406 int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1407 struct nameidata *nd)
1409 char *tmp = getname(name);
1410 int err = PTR_ERR(tmp);
1413 err = do_path_lookup(dfd, tmp, flags, nd);
1419 int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1421 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1425 * It's inline, so penalty for filesystems that don't use sticky bit is
1428 static inline int check_sticky(struct inode *dir, struct inode *inode)
1430 if (!(dir->i_mode & S_ISVTX))
1432 if (inode->i_uid == current->fsuid)
1434 if (dir->i_uid == current->fsuid)
1436 return !capable(CAP_FOWNER);
1440 * Check whether we can remove a link victim from directory dir, check
1441 * whether the type of victim is right.
1442 * 1. We can't do it if dir is read-only (done in permission())
1443 * 2. We should have write and exec permissions on dir
1444 * 3. We can't remove anything from append-only dir
1445 * 4. We can't do anything with immutable dir (done in permission())
1446 * 5. If the sticky bit on dir is set we should either
1447 * a. be owner of dir, or
1448 * b. be owner of victim, or
1449 * c. have CAP_FOWNER capability
1450 * 6. If the victim is append-only or immutable we can't do antyhing with
1451 * links pointing to it.
1452 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1453 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1454 * 9. We can't remove a root or mountpoint.
1455 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1456 * nfs_async_unlink().
1458 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1462 if (!victim->d_inode)
1465 BUG_ON(victim->d_parent->d_inode != dir);
1466 audit_inode_child(victim->d_name.name, victim, dir);
1468 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1473 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1474 IS_IMMUTABLE(victim->d_inode))
1477 if (!S_ISDIR(victim->d_inode->i_mode))
1479 if (IS_ROOT(victim))
1481 } else if (S_ISDIR(victim->d_inode->i_mode))
1483 if (IS_DEADDIR(dir))
1485 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1490 /* Check whether we can create an object with dentry child in directory
1492 * 1. We can't do it if child already exists (open has special treatment for
1493 * this case, but since we are inlined it's OK)
1494 * 2. We can't do it if dir is read-only (done in permission())
1495 * 3. We should have write and exec permissions on dir
1496 * 4. We can't do it if dir is immutable (done in permission())
1498 static inline int may_create(struct inode *dir, struct dentry *child,
1499 struct nameidata *nd)
1503 if (IS_DEADDIR(dir))
1505 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1509 * O_DIRECTORY translates into forcing a directory lookup.
1511 static inline int lookup_flags(unsigned int f)
1513 unsigned long retval = LOOKUP_FOLLOW;
1516 retval &= ~LOOKUP_FOLLOW;
1518 if (f & O_DIRECTORY)
1519 retval |= LOOKUP_DIRECTORY;
1525 * p1 and p2 should be directories on the same fs.
1527 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1532 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1536 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1538 for (p = p1; p->d_parent != p; p = p->d_parent) {
1539 if (p->d_parent == p2) {
1540 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1541 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1546 for (p = p2; p->d_parent != p; p = p->d_parent) {
1547 if (p->d_parent == p1) {
1548 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1549 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1554 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1555 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1559 void unlock_rename(struct dentry *p1, struct dentry *p2)
1561 mutex_unlock(&p1->d_inode->i_mutex);
1563 mutex_unlock(&p2->d_inode->i_mutex);
1564 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1568 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1569 struct nameidata *nd)
1571 int error = may_create(dir, dentry, nd);
1576 if (!dir->i_op || !dir->i_op->create)
1577 return -EACCES; /* shouldn't it be ENOSYS? */
1580 error = security_inode_create(dir, dentry, mode);
1584 error = dir->i_op->create(dir, dentry, mode, nd);
1586 fsnotify_create(dir, dentry);
1590 int may_open(struct nameidata *nd, int acc_mode, int flag)
1592 struct dentry *dentry = nd->path.dentry;
1593 struct inode *inode = dentry->d_inode;
1599 if (S_ISLNK(inode->i_mode))
1602 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1606 * FIFO's, sockets and device files are special: they don't
1607 * actually live on the filesystem itself, and as such you
1608 * can write to them even if the filesystem is read-only.
1610 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1612 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1613 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1617 } else if (IS_RDONLY(inode) && (acc_mode & MAY_WRITE))
1620 error = vfs_permission(nd, acc_mode);
1624 * An append-only file must be opened in append mode for writing.
1626 if (IS_APPEND(inode)) {
1627 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1633 /* O_NOATIME can only be set by the owner or superuser */
1634 if (flag & O_NOATIME)
1635 if (!is_owner_or_cap(inode))
1639 * Ensure there are no outstanding leases on the file.
1641 error = break_lease(inode, flag);
1645 if (flag & O_TRUNC) {
1646 error = get_write_access(inode);
1651 * Refuse to truncate files with mandatory locks held on them.
1653 error = locks_verify_locked(inode);
1657 error = do_truncate(dentry, 0,
1658 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1661 put_write_access(inode);
1665 if (flag & FMODE_WRITE)
1671 static int open_namei_create(struct nameidata *nd, struct path *path,
1675 struct dentry *dir = nd->path.dentry;
1677 if (!IS_POSIXACL(dir->d_inode))
1678 mode &= ~current->fs->umask;
1679 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1680 mutex_unlock(&dir->d_inode->i_mutex);
1681 dput(nd->path.dentry);
1682 nd->path.dentry = path->dentry;
1685 /* Don't check for write permission, don't truncate */
1686 return may_open(nd, 0, flag & ~O_TRUNC);
1692 * namei for open - this is in fact almost the whole open-routine.
1694 * Note that the low bits of "flag" aren't the same as in the open
1695 * system call - they are 00 - no permissions needed
1696 * 01 - read permission needed
1697 * 10 - write permission needed
1698 * 11 - read/write permissions needed
1699 * which is a lot more logical, and also allows the "no perm" needed
1700 * for symlinks (where the permissions are checked later).
1703 int open_namei(int dfd, const char *pathname, int flag,
1704 int mode, struct nameidata *nd)
1706 int acc_mode, error;
1711 acc_mode = ACC_MODE(flag);
1713 /* O_TRUNC implies we need access checks for write permissions */
1715 acc_mode |= MAY_WRITE;
1717 /* Allow the LSM permission hook to distinguish append
1718 access from general write access. */
1719 if (flag & O_APPEND)
1720 acc_mode |= MAY_APPEND;
1723 * The simplest case - just a plain lookup.
1725 if (!(flag & O_CREAT)) {
1726 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1734 * Create - we need to know the parent.
1736 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1741 * We have the parent and last component. First of all, check
1742 * that we are not asked to creat(2) an obvious directory - that
1746 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1749 dir = nd->path.dentry;
1750 nd->flags &= ~LOOKUP_PARENT;
1751 mutex_lock(&dir->d_inode->i_mutex);
1752 path.dentry = lookup_hash(nd);
1753 path.mnt = nd->path.mnt;
1756 error = PTR_ERR(path.dentry);
1757 if (IS_ERR(path.dentry)) {
1758 mutex_unlock(&dir->d_inode->i_mutex);
1762 if (IS_ERR(nd->intent.open.file)) {
1763 mutex_unlock(&dir->d_inode->i_mutex);
1764 error = PTR_ERR(nd->intent.open.file);
1768 /* Negative dentry, just create the file */
1769 if (!path.dentry->d_inode) {
1770 error = open_namei_create(nd, &path, flag, mode);
1777 * It already exists.
1779 mutex_unlock(&dir->d_inode->i_mutex);
1780 audit_inode(pathname, path.dentry);
1786 if (__follow_mount(&path)) {
1788 if (flag & O_NOFOLLOW)
1793 if (!path.dentry->d_inode)
1795 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1798 path_to_nameidata(&path, nd);
1800 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1803 error = may_open(nd, acc_mode, flag);
1809 dput_path(&path, nd);
1811 if (!IS_ERR(nd->intent.open.file))
1812 release_open_intent(nd);
1818 if (flag & O_NOFOLLOW)
1821 * This is subtle. Instead of calling do_follow_link() we do the
1822 * thing by hands. The reason is that this way we have zero link_count
1823 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1824 * After that we have the parent and last component, i.e.
1825 * we are in the same situation as after the first path_walk().
1826 * Well, almost - if the last component is normal we get its copy
1827 * stored in nd->last.name and we will have to putname() it when we
1828 * are done. Procfs-like symlinks just set LAST_BIND.
1830 nd->flags |= LOOKUP_PARENT;
1831 error = security_inode_follow_link(path.dentry, nd);
1834 error = __do_follow_link(&path, nd);
1836 /* Does someone understand code flow here? Or it is only
1837 * me so stupid? Anathema to whoever designed this non-sense
1838 * with "intent.open".
1840 release_open_intent(nd);
1843 nd->flags &= ~LOOKUP_PARENT;
1844 if (nd->last_type == LAST_BIND)
1847 if (nd->last_type != LAST_NORM)
1849 if (nd->last.name[nd->last.len]) {
1850 __putname(nd->last.name);
1855 __putname(nd->last.name);
1858 dir = nd->path.dentry;
1859 mutex_lock(&dir->d_inode->i_mutex);
1860 path.dentry = lookup_hash(nd);
1861 path.mnt = nd->path.mnt;
1862 __putname(nd->last.name);
1867 * lookup_create - lookup a dentry, creating it if it doesn't exist
1868 * @nd: nameidata info
1869 * @is_dir: directory flag
1871 * Simple function to lookup and return a dentry and create it
1872 * if it doesn't exist. Is SMP-safe.
1874 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1876 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1878 struct dentry *dentry = ERR_PTR(-EEXIST);
1880 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1882 * Yucky last component or no last component at all?
1883 * (foo/., foo/.., /////)
1885 if (nd->last_type != LAST_NORM)
1887 nd->flags &= ~LOOKUP_PARENT;
1888 nd->flags |= LOOKUP_CREATE;
1889 nd->intent.open.flags = O_EXCL;
1892 * Do the final lookup.
1894 dentry = lookup_hash(nd);
1899 * Special case - lookup gave negative, but... we had foo/bar/
1900 * From the vfs_mknod() POV we just have a negative dentry -
1901 * all is fine. Let's be bastards - you had / on the end, you've
1902 * been asking for (non-existent) directory. -ENOENT for you.
1904 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1909 dentry = ERR_PTR(-ENOENT);
1913 EXPORT_SYMBOL_GPL(lookup_create);
1915 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1917 int error = may_create(dir, dentry, NULL);
1922 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1925 if (!dir->i_op || !dir->i_op->mknod)
1928 error = security_inode_mknod(dir, dentry, mode, dev);
1933 error = dir->i_op->mknod(dir, dentry, mode, dev);
1935 fsnotify_create(dir, dentry);
1939 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1944 struct dentry * dentry;
1945 struct nameidata nd;
1949 tmp = getname(filename);
1951 return PTR_ERR(tmp);
1953 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1956 dentry = lookup_create(&nd, 0);
1957 error = PTR_ERR(dentry);
1959 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1960 mode &= ~current->fs->umask;
1961 if (!IS_ERR(dentry)) {
1962 switch (mode & S_IFMT) {
1963 case 0: case S_IFREG:
1964 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1966 case S_IFCHR: case S_IFBLK:
1967 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1968 new_decode_dev(dev));
1970 case S_IFIFO: case S_IFSOCK:
1971 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1981 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1989 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1991 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1994 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1996 int error = may_create(dir, dentry, NULL);
2001 if (!dir->i_op || !dir->i_op->mkdir)
2004 mode &= (S_IRWXUGO|S_ISVTX);
2005 error = security_inode_mkdir(dir, dentry, mode);
2010 error = dir->i_op->mkdir(dir, dentry, mode);
2012 fsnotify_mkdir(dir, dentry);
2016 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2020 struct dentry *dentry;
2021 struct nameidata nd;
2023 tmp = getname(pathname);
2024 error = PTR_ERR(tmp);
2028 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2031 dentry = lookup_create(&nd, 1);
2032 error = PTR_ERR(dentry);
2036 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2037 mode &= ~current->fs->umask;
2038 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2041 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2049 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2051 return sys_mkdirat(AT_FDCWD, pathname, mode);
2055 * We try to drop the dentry early: we should have
2056 * a usage count of 2 if we're the only user of this
2057 * dentry, and if that is true (possibly after pruning
2058 * the dcache), then we drop the dentry now.
2060 * A low-level filesystem can, if it choses, legally
2063 * if (!d_unhashed(dentry))
2066 * if it cannot handle the case of removing a directory
2067 * that is still in use by something else..
2069 void dentry_unhash(struct dentry *dentry)
2072 shrink_dcache_parent(dentry);
2073 spin_lock(&dcache_lock);
2074 spin_lock(&dentry->d_lock);
2075 if (atomic_read(&dentry->d_count) == 2)
2077 spin_unlock(&dentry->d_lock);
2078 spin_unlock(&dcache_lock);
2081 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2083 int error = may_delete(dir, dentry, 1);
2088 if (!dir->i_op || !dir->i_op->rmdir)
2093 mutex_lock(&dentry->d_inode->i_mutex);
2094 dentry_unhash(dentry);
2095 if (d_mountpoint(dentry))
2098 error = security_inode_rmdir(dir, dentry);
2100 error = dir->i_op->rmdir(dir, dentry);
2102 dentry->d_inode->i_flags |= S_DEAD;
2105 mutex_unlock(&dentry->d_inode->i_mutex);
2114 static long do_rmdir(int dfd, const char __user *pathname)
2118 struct dentry *dentry;
2119 struct nameidata nd;
2121 name = getname(pathname);
2123 return PTR_ERR(name);
2125 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2129 switch(nd.last_type) {
2140 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2141 dentry = lookup_hash(&nd);
2142 error = PTR_ERR(dentry);
2145 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2148 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2156 asmlinkage long sys_rmdir(const char __user *pathname)
2158 return do_rmdir(AT_FDCWD, pathname);
2161 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2163 int error = may_delete(dir, dentry, 0);
2168 if (!dir->i_op || !dir->i_op->unlink)
2173 mutex_lock(&dentry->d_inode->i_mutex);
2174 if (d_mountpoint(dentry))
2177 error = security_inode_unlink(dir, dentry);
2179 error = dir->i_op->unlink(dir, dentry);
2181 mutex_unlock(&dentry->d_inode->i_mutex);
2183 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2184 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2185 fsnotify_link_count(dentry->d_inode);
2193 * Make sure that the actual truncation of the file will occur outside its
2194 * directory's i_mutex. Truncate can take a long time if there is a lot of
2195 * writeout happening, and we don't want to prevent access to the directory
2196 * while waiting on the I/O.
2198 static long do_unlinkat(int dfd, const char __user *pathname)
2202 struct dentry *dentry;
2203 struct nameidata nd;
2204 struct inode *inode = NULL;
2206 name = getname(pathname);
2208 return PTR_ERR(name);
2210 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2214 if (nd.last_type != LAST_NORM)
2216 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2217 dentry = lookup_hash(&nd);
2218 error = PTR_ERR(dentry);
2219 if (!IS_ERR(dentry)) {
2220 /* Why not before? Because we want correct error value */
2221 if (nd.last.name[nd.last.len])
2223 inode = dentry->d_inode;
2225 atomic_inc(&inode->i_count);
2226 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2230 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2232 iput(inode); /* truncate the inode here */
2240 error = !dentry->d_inode ? -ENOENT :
2241 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2245 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2247 if ((flag & ~AT_REMOVEDIR) != 0)
2250 if (flag & AT_REMOVEDIR)
2251 return do_rmdir(dfd, pathname);
2253 return do_unlinkat(dfd, pathname);
2256 asmlinkage long sys_unlink(const char __user *pathname)
2258 return do_unlinkat(AT_FDCWD, pathname);
2261 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2263 int error = may_create(dir, dentry, NULL);
2268 if (!dir->i_op || !dir->i_op->symlink)
2271 error = security_inode_symlink(dir, dentry, oldname);
2276 error = dir->i_op->symlink(dir, dentry, oldname);
2278 fsnotify_create(dir, dentry);
2282 asmlinkage long sys_symlinkat(const char __user *oldname,
2283 int newdfd, const char __user *newname)
2288 struct dentry *dentry;
2289 struct nameidata nd;
2291 from = getname(oldname);
2293 return PTR_ERR(from);
2294 to = getname(newname);
2295 error = PTR_ERR(to);
2299 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2302 dentry = lookup_create(&nd, 0);
2303 error = PTR_ERR(dentry);
2307 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from, S_IALLUGO);
2310 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2319 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2321 return sys_symlinkat(oldname, AT_FDCWD, newname);
2324 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2326 struct inode *inode = old_dentry->d_inode;
2332 error = may_create(dir, new_dentry, NULL);
2336 if (dir->i_sb != inode->i_sb)
2340 * A link to an append-only or immutable file cannot be created.
2342 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2344 if (!dir->i_op || !dir->i_op->link)
2346 if (S_ISDIR(old_dentry->d_inode->i_mode))
2349 error = security_inode_link(old_dentry, dir, new_dentry);
2353 mutex_lock(&old_dentry->d_inode->i_mutex);
2355 error = dir->i_op->link(old_dentry, dir, new_dentry);
2356 mutex_unlock(&old_dentry->d_inode->i_mutex);
2358 fsnotify_link(dir, old_dentry->d_inode, new_dentry);
2363 * Hardlinks are often used in delicate situations. We avoid
2364 * security-related surprises by not following symlinks on the
2367 * We don't follow them on the oldname either to be compatible
2368 * with linux 2.0, and to avoid hard-linking to directories
2369 * and other special files. --ADM
2371 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2372 int newdfd, const char __user *newname,
2375 struct dentry *new_dentry;
2376 struct nameidata nd, old_nd;
2380 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2383 to = getname(newname);
2387 error = __user_walk_fd(olddfd, oldname,
2388 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2392 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2396 if (old_nd.path.mnt != nd.path.mnt)
2398 new_dentry = lookup_create(&nd, 0);
2399 error = PTR_ERR(new_dentry);
2400 if (IS_ERR(new_dentry))
2402 error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
2405 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2409 path_release(&old_nd);
2416 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2418 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2422 * The worst of all namespace operations - renaming directory. "Perverted"
2423 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2425 * a) we can get into loop creation. Check is done in is_subdir().
2426 * b) race potential - two innocent renames can create a loop together.
2427 * That's where 4.4 screws up. Current fix: serialization on
2428 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2430 * c) we have to lock _three_ objects - parents and victim (if it exists).
2431 * And that - after we got ->i_mutex on parents (until then we don't know
2432 * whether the target exists). Solution: try to be smart with locking
2433 * order for inodes. We rely on the fact that tree topology may change
2434 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2435 * move will be locked. Thus we can rank directories by the tree
2436 * (ancestors first) and rank all non-directories after them.
2437 * That works since everybody except rename does "lock parent, lookup,
2438 * lock child" and rename is under ->s_vfs_rename_mutex.
2439 * HOWEVER, it relies on the assumption that any object with ->lookup()
2440 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2441 * we'd better make sure that there's no link(2) for them.
2442 * d) some filesystems don't support opened-but-unlinked directories,
2443 * either because of layout or because they are not ready to deal with
2444 * all cases correctly. The latter will be fixed (taking this sort of
2445 * stuff into VFS), but the former is not going away. Solution: the same
2446 * trick as in rmdir().
2447 * e) conversion from fhandle to dentry may come in the wrong moment - when
2448 * we are removing the target. Solution: we will have to grab ->i_mutex
2449 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2450 * ->i_mutex on parents, which works but leads to some truely excessive
2453 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2454 struct inode *new_dir, struct dentry *new_dentry)
2457 struct inode *target;
2460 * If we are going to change the parent - check write permissions,
2461 * we'll need to flip '..'.
2463 if (new_dir != old_dir) {
2464 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2469 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2473 target = new_dentry->d_inode;
2475 mutex_lock(&target->i_mutex);
2476 dentry_unhash(new_dentry);
2478 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2481 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2484 target->i_flags |= S_DEAD;
2485 mutex_unlock(&target->i_mutex);
2486 if (d_unhashed(new_dentry))
2487 d_rehash(new_dentry);
2491 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2492 d_move(old_dentry,new_dentry);
2496 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2497 struct inode *new_dir, struct dentry *new_dentry)
2499 struct inode *target;
2502 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2507 target = new_dentry->d_inode;
2509 mutex_lock(&target->i_mutex);
2510 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2513 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2515 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2516 d_move(old_dentry, new_dentry);
2519 mutex_unlock(&target->i_mutex);
2524 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2525 struct inode *new_dir, struct dentry *new_dentry)
2528 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2529 const char *old_name;
2531 if (old_dentry->d_inode == new_dentry->d_inode)
2534 error = may_delete(old_dir, old_dentry, is_dir);
2538 if (!new_dentry->d_inode)
2539 error = may_create(new_dir, new_dentry, NULL);
2541 error = may_delete(new_dir, new_dentry, is_dir);
2545 if (!old_dir->i_op || !old_dir->i_op->rename)
2548 DQUOT_INIT(old_dir);
2549 DQUOT_INIT(new_dir);
2551 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2554 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2556 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2558 const char *new_name = old_dentry->d_name.name;
2559 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2560 new_dentry->d_inode, old_dentry);
2562 fsnotify_oldname_free(old_name);
2567 static int do_rename(int olddfd, const char *oldname,
2568 int newdfd, const char *newname)
2571 struct dentry * old_dir, * new_dir;
2572 struct dentry * old_dentry, *new_dentry;
2573 struct dentry * trap;
2574 struct nameidata oldnd, newnd;
2576 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2580 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2585 if (oldnd.path.mnt != newnd.path.mnt)
2588 old_dir = oldnd.path.dentry;
2590 if (oldnd.last_type != LAST_NORM)
2593 new_dir = newnd.path.dentry;
2594 if (newnd.last_type != LAST_NORM)
2597 trap = lock_rename(new_dir, old_dir);
2599 old_dentry = lookup_hash(&oldnd);
2600 error = PTR_ERR(old_dentry);
2601 if (IS_ERR(old_dentry))
2603 /* source must exist */
2605 if (!old_dentry->d_inode)
2607 /* unless the source is a directory trailing slashes give -ENOTDIR */
2608 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2610 if (oldnd.last.name[oldnd.last.len])
2612 if (newnd.last.name[newnd.last.len])
2615 /* source should not be ancestor of target */
2617 if (old_dentry == trap)
2619 new_dentry = lookup_hash(&newnd);
2620 error = PTR_ERR(new_dentry);
2621 if (IS_ERR(new_dentry))
2623 /* target should not be an ancestor of source */
2625 if (new_dentry == trap)
2628 error = vfs_rename(old_dir->d_inode, old_dentry,
2629 new_dir->d_inode, new_dentry);
2635 unlock_rename(new_dir, old_dir);
2637 path_release(&newnd);
2639 path_release(&oldnd);
2644 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2645 int newdfd, const char __user *newname)
2651 from = getname(oldname);
2653 return PTR_ERR(from);
2654 to = getname(newname);
2655 error = PTR_ERR(to);
2657 error = do_rename(olddfd, from, newdfd, to);
2664 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2666 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2669 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2673 len = PTR_ERR(link);
2678 if (len > (unsigned) buflen)
2680 if (copy_to_user(buffer, link, len))
2687 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2688 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2689 * using) it for any given inode is up to filesystem.
2691 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2693 struct nameidata nd;
2697 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2698 if (!IS_ERR(cookie)) {
2699 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2700 if (dentry->d_inode->i_op->put_link)
2701 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2702 cookie = ERR_PTR(res);
2704 return PTR_ERR(cookie);
2707 int vfs_follow_link(struct nameidata *nd, const char *link)
2709 return __vfs_follow_link(nd, link);
2712 /* get the link contents into pagecache */
2713 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2716 struct address_space *mapping = dentry->d_inode->i_mapping;
2717 page = read_mapping_page(mapping, 0, NULL);
2724 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2726 struct page *page = NULL;
2727 char *s = page_getlink(dentry, &page);
2728 int res = vfs_readlink(dentry,buffer,buflen,s);
2731 page_cache_release(page);
2736 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2738 struct page *page = NULL;
2739 nd_set_link(nd, page_getlink(dentry, &page));
2743 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2745 struct page *page = cookie;
2749 page_cache_release(page);
2753 int __page_symlink(struct inode *inode, const char *symname, int len,
2756 struct address_space *mapping = inode->i_mapping;
2763 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2764 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2768 kaddr = kmap_atomic(page, KM_USER0);
2769 memcpy(kaddr, symname, len-1);
2770 kunmap_atomic(kaddr, KM_USER0);
2772 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2779 mark_inode_dirty(inode);
2785 int page_symlink(struct inode *inode, const char *symname, int len)
2787 return __page_symlink(inode, symname, len,
2788 mapping_gfp_mask(inode->i_mapping));
2791 const struct inode_operations page_symlink_inode_operations = {
2792 .readlink = generic_readlink,
2793 .follow_link = page_follow_link_light,
2794 .put_link = page_put_link,
2797 EXPORT_SYMBOL(__user_walk);
2798 EXPORT_SYMBOL(__user_walk_fd);
2799 EXPORT_SYMBOL(follow_down);
2800 EXPORT_SYMBOL(follow_up);
2801 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2802 EXPORT_SYMBOL(getname);
2803 EXPORT_SYMBOL(lock_rename);
2804 EXPORT_SYMBOL(lookup_one_len);
2805 EXPORT_SYMBOL(page_follow_link_light);
2806 EXPORT_SYMBOL(page_put_link);
2807 EXPORT_SYMBOL(page_readlink);
2808 EXPORT_SYMBOL(__page_symlink);
2809 EXPORT_SYMBOL(page_symlink);
2810 EXPORT_SYMBOL(page_symlink_inode_operations);
2811 EXPORT_SYMBOL(path_lookup);
2812 EXPORT_SYMBOL(vfs_path_lookup);
2813 EXPORT_SYMBOL(path_release);
2814 EXPORT_SYMBOL(permission);
2815 EXPORT_SYMBOL(vfs_permission);
2816 EXPORT_SYMBOL(file_permission);
2817 EXPORT_SYMBOL(unlock_rename);
2818 EXPORT_SYMBOL(vfs_create);
2819 EXPORT_SYMBOL(vfs_follow_link);
2820 EXPORT_SYMBOL(vfs_link);
2821 EXPORT_SYMBOL(vfs_mkdir);
2822 EXPORT_SYMBOL(vfs_mknod);
2823 EXPORT_SYMBOL(generic_permission);
2824 EXPORT_SYMBOL(vfs_readlink);
2825 EXPORT_SYMBOL(vfs_rename);
2826 EXPORT_SYMBOL(vfs_rmdir);
2827 EXPORT_SYMBOL(vfs_symlink);
2828 EXPORT_SYMBOL(vfs_unlink);
2829 EXPORT_SYMBOL(dentry_unhash);
2830 EXPORT_SYMBOL(generic_readlink);