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) ||
216 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
217 if (capable(CAP_DAC_OVERRIDE))
221 * Searching includes executable on directories, else just read.
223 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
224 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
225 if (capable(CAP_DAC_READ_SEARCH))
231 int inode_permission(struct inode *inode, int mask)
235 if (mask & MAY_WRITE) {
236 umode_t mode = inode->i_mode;
239 * Nobody gets write access to a read-only fs.
241 if (IS_RDONLY(inode) &&
242 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
246 * Nobody gets write access to an immutable file.
248 if (IS_IMMUTABLE(inode))
252 /* Ordinary permission routines do not understand MAY_APPEND. */
253 if (inode->i_op && inode->i_op->permission) {
254 retval = inode->i_op->permission(inode, mask);
257 * Exec permission on a regular file is denied if none
258 * of the execute bits are set.
260 * This check should be done by the ->permission()
263 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
264 !(inode->i_mode & S_IXUGO))
268 retval = generic_permission(inode, mask, NULL);
273 retval = devcgroup_inode_permission(inode, mask);
277 return security_inode_permission(inode,
278 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
282 * vfs_permission - check for access rights to a given path
283 * @nd: lookup result that describes the path
284 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
286 * Used to check for read/write/execute permissions on a path.
287 * We use "fsuid" for this, letting us set arbitrary permissions
288 * for filesystem access without changing the "normal" uids which
289 * are used for other things.
291 int vfs_permission(struct nameidata *nd, int mask)
293 return inode_permission(nd->path.dentry->d_inode, mask);
297 * file_permission - check for additional access rights to a given file
298 * @file: file to check access rights for
299 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
301 * Used to check for read/write/execute permissions on an already opened
305 * Do not use this function in new code. All access checks should
306 * be done using vfs_permission().
308 int file_permission(struct file *file, int mask)
310 return inode_permission(file->f_path.dentry->d_inode, mask);
314 * get_write_access() gets write permission for a file.
315 * put_write_access() releases this write permission.
316 * This is used for regular files.
317 * We cannot support write (and maybe mmap read-write shared) accesses and
318 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
319 * can have the following values:
320 * 0: no writers, no VM_DENYWRITE mappings
321 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
322 * > 0: (i_writecount) users are writing to the file.
324 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
325 * except for the cases where we don't hold i_writecount yet. Then we need to
326 * use {get,deny}_write_access() - these functions check the sign and refuse
327 * to do the change if sign is wrong. Exclusion between them is provided by
328 * the inode->i_lock spinlock.
331 int get_write_access(struct inode * inode)
333 spin_lock(&inode->i_lock);
334 if (atomic_read(&inode->i_writecount) < 0) {
335 spin_unlock(&inode->i_lock);
338 atomic_inc(&inode->i_writecount);
339 spin_unlock(&inode->i_lock);
344 int deny_write_access(struct file * file)
346 struct inode *inode = file->f_path.dentry->d_inode;
348 spin_lock(&inode->i_lock);
349 if (atomic_read(&inode->i_writecount) > 0) {
350 spin_unlock(&inode->i_lock);
353 atomic_dec(&inode->i_writecount);
354 spin_unlock(&inode->i_lock);
360 * path_get - get a reference to a path
361 * @path: path to get the reference to
363 * Given a path increment the reference count to the dentry and the vfsmount.
365 void path_get(struct path *path)
370 EXPORT_SYMBOL(path_get);
373 * path_put - put a reference to a path
374 * @path: path to put the reference to
376 * Given a path decrement the reference count to the dentry and the vfsmount.
378 void path_put(struct path *path)
383 EXPORT_SYMBOL(path_put);
386 * release_open_intent - free up open intent resources
387 * @nd: pointer to nameidata
389 void release_open_intent(struct nameidata *nd)
391 if (nd->intent.open.file->f_path.dentry == NULL)
392 put_filp(nd->intent.open.file);
394 fput(nd->intent.open.file);
397 static inline struct dentry *
398 do_revalidate(struct dentry *dentry, struct nameidata *nd)
400 int status = dentry->d_op->d_revalidate(dentry, nd);
401 if (unlikely(status <= 0)) {
403 * The dentry failed validation.
404 * If d_revalidate returned 0 attempt to invalidate
405 * the dentry otherwise d_revalidate is asking us
406 * to return a fail status.
409 if (!d_invalidate(dentry)) {
415 dentry = ERR_PTR(status);
422 * Internal lookup() using the new generic dcache.
425 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
427 struct dentry * dentry = __d_lookup(parent, name);
429 /* lockess __d_lookup may fail due to concurrent d_move()
430 * in some unrelated directory, so try with d_lookup
433 dentry = d_lookup(parent, name);
435 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
436 dentry = do_revalidate(dentry, nd);
442 * Short-cut version of permission(), for calling by
443 * path_walk(), when dcache lock is held. Combines parts
444 * of permission() and generic_permission(), and tests ONLY for
445 * MAY_EXEC permission.
447 * If appropriate, check DAC only. If not appropriate, or
448 * short-cut DAC fails, then call permission() to do more
449 * complete permission check.
451 static int exec_permission_lite(struct inode *inode)
453 umode_t mode = inode->i_mode;
455 if (inode->i_op && inode->i_op->permission)
458 if (current->fsuid == inode->i_uid)
460 else if (in_group_p(inode->i_gid))
466 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
469 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
472 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
477 return security_inode_permission(inode, MAY_EXEC);
481 * This is called when everything else fails, and we actually have
482 * to go to the low-level filesystem to find out what we should do..
484 * We get the directory semaphore, and after getting that we also
485 * make sure that nobody added the entry to the dcache in the meantime..
488 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
490 struct dentry * result;
491 struct inode *dir = parent->d_inode;
493 mutex_lock(&dir->i_mutex);
495 * First re-do the cached lookup just in case it was created
496 * while we waited for the directory semaphore..
498 * FIXME! This could use version numbering or similar to
499 * avoid unnecessary cache lookups.
501 * The "dcache_lock" is purely to protect the RCU list walker
502 * from concurrent renames at this point (we mustn't get false
503 * negatives from the RCU list walk here, unlike the optimistic
506 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
508 result = d_lookup(parent, name);
510 struct dentry *dentry;
512 /* Don't create child dentry for a dead directory. */
513 result = ERR_PTR(-ENOENT);
517 dentry = d_alloc(parent, name);
518 result = ERR_PTR(-ENOMEM);
520 result = dir->i_op->lookup(dir, dentry, nd);
527 mutex_unlock(&dir->i_mutex);
532 * Uhhuh! Nasty case: the cache was re-populated while
533 * we waited on the semaphore. Need to revalidate.
535 mutex_unlock(&dir->i_mutex);
536 if (result->d_op && result->d_op->d_revalidate) {
537 result = do_revalidate(result, nd);
539 result = ERR_PTR(-ENOENT);
545 static __always_inline void
546 walk_init_root(const char *name, struct nameidata *nd)
548 struct fs_struct *fs = current->fs;
550 read_lock(&fs->lock);
553 read_unlock(&fs->lock);
557 * Wrapper to retry pathname resolution whenever the underlying
558 * file system returns an ESTALE.
560 * Retry the whole path once, forcing real lookup requests
561 * instead of relying on the dcache.
563 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
565 struct path save = nd->path;
568 /* make sure the stuff we saved doesn't go away */
571 result = __link_path_walk(name, nd);
572 if (result == -ESTALE) {
573 /* nd->path had been dropped */
576 nd->flags |= LOOKUP_REVAL;
577 result = __link_path_walk(name, nd);
585 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
594 walk_init_root(link, nd);
596 res = link_path_walk(link, nd);
597 if (nd->depth || res || nd->last_type!=LAST_NORM)
600 * If it is an iterative symlinks resolution in open_namei() we
601 * have to copy the last component. And all that crap because of
602 * bloody create() on broken symlinks. Furrfu...
605 if (unlikely(!name)) {
609 strcpy(name, nd->last.name);
610 nd->last.name = name;
614 return PTR_ERR(link);
617 static void path_put_conditional(struct path *path, struct nameidata *nd)
620 if (path->mnt != nd->path.mnt)
624 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
626 dput(nd->path.dentry);
627 if (nd->path.mnt != path->mnt)
628 mntput(nd->path.mnt);
629 nd->path.mnt = path->mnt;
630 nd->path.dentry = path->dentry;
633 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
637 struct dentry *dentry = path->dentry;
639 touch_atime(path->mnt, dentry);
640 nd_set_link(nd, NULL);
642 if (path->mnt != nd->path.mnt) {
643 path_to_nameidata(path, nd);
647 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
648 error = PTR_ERR(cookie);
649 if (!IS_ERR(cookie)) {
650 char *s = nd_get_link(nd);
653 error = __vfs_follow_link(nd, s);
654 if (dentry->d_inode->i_op->put_link)
655 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
663 * This limits recursive symlink follows to 8, while
664 * limiting consecutive symlinks to 40.
666 * Without that kind of total limit, nasty chains of consecutive
667 * symlinks can cause almost arbitrarily long lookups.
669 static inline int do_follow_link(struct path *path, struct nameidata *nd)
672 if (current->link_count >= MAX_NESTED_LINKS)
674 if (current->total_link_count >= 40)
676 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
678 err = security_inode_follow_link(path->dentry, nd);
681 current->link_count++;
682 current->total_link_count++;
684 err = __do_follow_link(path, nd);
685 current->link_count--;
689 path_put_conditional(path, nd);
694 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
696 struct vfsmount *parent;
697 struct dentry *mountpoint;
698 spin_lock(&vfsmount_lock);
699 parent=(*mnt)->mnt_parent;
700 if (parent == *mnt) {
701 spin_unlock(&vfsmount_lock);
705 mountpoint=dget((*mnt)->mnt_mountpoint);
706 spin_unlock(&vfsmount_lock);
708 *dentry = mountpoint;
714 /* no need for dcache_lock, as serialization is taken care in
717 static int __follow_mount(struct path *path)
720 while (d_mountpoint(path->dentry)) {
721 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
728 path->dentry = dget(mounted->mnt_root);
734 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
736 while (d_mountpoint(*dentry)) {
737 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
743 *dentry = dget(mounted->mnt_root);
747 /* no need for dcache_lock, as serialization is taken care in
750 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
752 struct vfsmount *mounted;
754 mounted = lookup_mnt(*mnt, *dentry);
759 *dentry = dget(mounted->mnt_root);
765 static __always_inline void follow_dotdot(struct nameidata *nd)
767 struct fs_struct *fs = current->fs;
770 struct vfsmount *parent;
771 struct dentry *old = nd->path.dentry;
773 read_lock(&fs->lock);
774 if (nd->path.dentry == fs->root.dentry &&
775 nd->path.mnt == fs->root.mnt) {
776 read_unlock(&fs->lock);
779 read_unlock(&fs->lock);
780 spin_lock(&dcache_lock);
781 if (nd->path.dentry != nd->path.mnt->mnt_root) {
782 nd->path.dentry = dget(nd->path.dentry->d_parent);
783 spin_unlock(&dcache_lock);
787 spin_unlock(&dcache_lock);
788 spin_lock(&vfsmount_lock);
789 parent = nd->path.mnt->mnt_parent;
790 if (parent == nd->path.mnt) {
791 spin_unlock(&vfsmount_lock);
795 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
796 spin_unlock(&vfsmount_lock);
798 mntput(nd->path.mnt);
799 nd->path.mnt = parent;
801 follow_mount(&nd->path.mnt, &nd->path.dentry);
805 * It's more convoluted than I'd like it to be, but... it's still fairly
806 * small and for now I'd prefer to have fast path as straight as possible.
807 * It _is_ time-critical.
809 static int do_lookup(struct nameidata *nd, struct qstr *name,
812 struct vfsmount *mnt = nd->path.mnt;
813 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
817 if (dentry->d_op && dentry->d_op->d_revalidate)
818 goto need_revalidate;
821 path->dentry = dentry;
822 __follow_mount(path);
826 dentry = real_lookup(nd->path.dentry, name, nd);
832 dentry = do_revalidate(dentry, nd);
840 return PTR_ERR(dentry);
844 * This is a temporary kludge to deal with "automount" symlinks; proper
845 * solution is to trigger them on follow_mount(), so that do_lookup()
846 * would DTRT. To be killed before 2.6.34-final.
848 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
850 return inode && unlikely(inode->i_op->follow_link) &&
851 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
856 * This is the basic name resolution function, turning a pathname into
857 * the final dentry. We expect 'base' to be positive and a directory.
859 * Returns 0 and nd will have valid dentry and mnt on success.
860 * Returns error and drops reference to input namei data on failure.
862 static int __link_path_walk(const char *name, struct nameidata *nd)
867 unsigned int lookup_flags = nd->flags;
874 inode = nd->path.dentry->d_inode;
876 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
878 /* At this point we know we have a real path component. */
884 nd->flags |= LOOKUP_CONTINUE;
885 err = exec_permission_lite(inode);
887 err = vfs_permission(nd, MAY_EXEC);
892 c = *(const unsigned char *)name;
894 hash = init_name_hash();
897 hash = partial_name_hash(c, hash);
898 c = *(const unsigned char *)name;
899 } while (c && (c != '/'));
900 this.len = name - (const char *) this.name;
901 this.hash = end_name_hash(hash);
903 /* remove trailing slashes? */
906 while (*++name == '/');
908 goto last_with_slashes;
911 * "." and ".." are special - ".." especially so because it has
912 * to be able to know about the current root directory and
913 * parent relationships.
915 if (this.name[0] == '.') switch (this.len) {
919 if (this.name[1] != '.')
922 inode = nd->path.dentry->d_inode;
928 * See if the low-level filesystem might want
929 * to use its own hash..
931 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
932 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
937 /* This does the actual lookups.. */
938 err = do_lookup(nd, &this, &next);
943 inode = next.dentry->d_inode;
950 if (inode->i_op->follow_link) {
951 err = do_follow_link(&next, nd);
955 inode = nd->path.dentry->d_inode;
962 path_to_nameidata(&next, nd);
964 if (!inode->i_op->lookup)
967 /* here ends the main loop */
970 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
972 /* Clear LOOKUP_CONTINUE iff it was previously unset */
973 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
974 if (lookup_flags & LOOKUP_PARENT)
976 if (this.name[0] == '.') switch (this.len) {
980 if (this.name[1] != '.')
983 inode = nd->path.dentry->d_inode;
988 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
989 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
994 err = do_lookup(nd, &this, &next);
997 inode = next.dentry->d_inode;
998 if (follow_on_final(inode, lookup_flags)) {
999 err = do_follow_link(&next, nd);
1002 inode = nd->path.dentry->d_inode;
1004 path_to_nameidata(&next, nd);
1008 if (lookup_flags & LOOKUP_DIRECTORY) {
1010 if (!inode->i_op || !inode->i_op->lookup)
1016 nd->last_type = LAST_NORM;
1017 if (this.name[0] != '.')
1020 nd->last_type = LAST_DOT;
1021 else if (this.len == 2 && this.name[1] == '.')
1022 nd->last_type = LAST_DOTDOT;
1027 * We bypassed the ordinary revalidation routines.
1028 * We may need to check the cached dentry for staleness.
1030 if (nd->path.dentry && nd->path.dentry->d_sb &&
1031 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1033 /* Note: we do not d_invalidate() */
1034 if (!nd->path.dentry->d_op->d_revalidate(
1035 nd->path.dentry, nd))
1041 path_put_conditional(&next, nd);
1044 path_put(&nd->path);
1049 static int path_walk(const char *name, struct nameidata *nd)
1051 current->total_link_count = 0;
1052 return link_path_walk(name, nd);
1055 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1056 static int do_path_lookup(int dfd, const char *name,
1057 unsigned int flags, struct nameidata *nd)
1062 struct fs_struct *fs = current->fs;
1064 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1069 read_lock(&fs->lock);
1070 nd->path = fs->root;
1071 path_get(&fs->root);
1072 read_unlock(&fs->lock);
1073 } else if (dfd == AT_FDCWD) {
1074 read_lock(&fs->lock);
1077 read_unlock(&fs->lock);
1079 struct dentry *dentry;
1081 file = fget_light(dfd, &fput_needed);
1086 dentry = file->f_path.dentry;
1089 if (!S_ISDIR(dentry->d_inode->i_mode))
1092 retval = file_permission(file, MAY_EXEC);
1096 nd->path = file->f_path;
1097 path_get(&file->f_path);
1099 fput_light(file, fput_needed);
1102 retval = path_walk(name, nd);
1103 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1104 nd->path.dentry->d_inode))
1105 audit_inode(name, nd->path.dentry);
1110 fput_light(file, fput_needed);
1114 int path_lookup(const char *name, unsigned int flags,
1115 struct nameidata *nd)
1117 return do_path_lookup(AT_FDCWD, name, flags, nd);
1121 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1122 * @dentry: pointer to dentry of the base directory
1123 * @mnt: pointer to vfs mount of the base directory
1124 * @name: pointer to file name
1125 * @flags: lookup flags
1126 * @nd: pointer to nameidata
1128 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1129 const char *name, unsigned int flags,
1130 struct nameidata *nd)
1134 /* same as do_path_lookup */
1135 nd->last_type = LAST_ROOT;
1139 nd->path.dentry = dentry;
1141 path_get(&nd->path);
1143 retval = path_walk(name, nd);
1144 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1145 nd->path.dentry->d_inode))
1146 audit_inode(name, nd->path.dentry);
1152 static int __path_lookup_intent_open(int dfd, const char *name,
1153 unsigned int lookup_flags, struct nameidata *nd,
1154 int open_flags, int create_mode)
1156 struct file *filp = get_empty_filp();
1161 nd->intent.open.file = filp;
1162 nd->intent.open.flags = open_flags;
1163 nd->intent.open.create_mode = create_mode;
1164 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1165 if (IS_ERR(nd->intent.open.file)) {
1167 err = PTR_ERR(nd->intent.open.file);
1168 path_put(&nd->path);
1170 } else if (err != 0)
1171 release_open_intent(nd);
1176 * path_lookup_open - lookup a file path with open intent
1177 * @dfd: the directory to use as base, or AT_FDCWD
1178 * @name: pointer to file name
1179 * @lookup_flags: lookup intent flags
1180 * @nd: pointer to nameidata
1181 * @open_flags: open intent flags
1183 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1184 struct nameidata *nd, int open_flags)
1186 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1191 * path_lookup_create - lookup a file path with open + create intent
1192 * @dfd: the directory to use as base, or AT_FDCWD
1193 * @name: pointer to file name
1194 * @lookup_flags: lookup intent flags
1195 * @nd: pointer to nameidata
1196 * @open_flags: open intent flags
1197 * @create_mode: create intent flags
1199 static int path_lookup_create(int dfd, const char *name,
1200 unsigned int lookup_flags, struct nameidata *nd,
1201 int open_flags, int create_mode)
1203 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1204 nd, open_flags, create_mode);
1207 static struct dentry *__lookup_hash(struct qstr *name,
1208 struct dentry *base, struct nameidata *nd)
1210 struct dentry *dentry;
1211 struct inode *inode;
1214 inode = base->d_inode;
1217 * See if the low-level filesystem might want
1218 * to use its own hash..
1220 if (base->d_op && base->d_op->d_hash) {
1221 err = base->d_op->d_hash(base, name);
1222 dentry = ERR_PTR(err);
1227 dentry = cached_lookup(base, name, nd);
1231 /* Don't create child dentry for a dead directory. */
1232 dentry = ERR_PTR(-ENOENT);
1233 if (IS_DEADDIR(inode))
1236 new = d_alloc(base, name);
1237 dentry = ERR_PTR(-ENOMEM);
1240 dentry = inode->i_op->lookup(inode, new, nd);
1251 * Restricted form of lookup. Doesn't follow links, single-component only,
1252 * needs parent already locked. Doesn't follow mounts.
1255 static struct dentry *lookup_hash(struct nameidata *nd)
1259 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1261 return ERR_PTR(err);
1262 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1265 static int __lookup_one_len(const char *name, struct qstr *this,
1266 struct dentry *base, int len)
1276 hash = init_name_hash();
1278 c = *(const unsigned char *)name++;
1279 if (c == '/' || c == '\0')
1281 hash = partial_name_hash(c, hash);
1283 this->hash = end_name_hash(hash);
1288 * lookup_one_len - filesystem helper to lookup single pathname component
1289 * @name: pathname component to lookup
1290 * @base: base directory to lookup from
1291 * @len: maximum length @len should be interpreted to
1293 * Note that this routine is purely a helper for filesystem usage and should
1294 * not be called by generic code. Also note that by using this function the
1295 * nameidata argument is passed to the filesystem methods and a filesystem
1296 * using this helper needs to be prepared for that.
1298 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1303 err = __lookup_one_len(name, &this, base, len);
1305 return ERR_PTR(err);
1307 err = inode_permission(base->d_inode, MAY_EXEC);
1309 return ERR_PTR(err);
1310 return __lookup_hash(&this, base, NULL);
1314 * lookup_one_noperm - bad hack for sysfs
1315 * @name: pathname component to lookup
1316 * @base: base directory to lookup from
1318 * This is a variant of lookup_one_len that doesn't perform any permission
1319 * checks. It's a horrible hack to work around the braindead sysfs
1320 * architecture and should not be used anywhere else.
1322 * DON'T USE THIS FUNCTION EVER, thanks.
1324 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1329 err = __lookup_one_len(name, &this, base, strlen(name));
1331 return ERR_PTR(err);
1332 return __lookup_hash(&this, base, NULL);
1335 int user_path_at(int dfd, const char __user *name, unsigned flags,
1338 struct nameidata nd;
1339 char *tmp = getname(name);
1340 int err = PTR_ERR(tmp);
1343 BUG_ON(flags & LOOKUP_PARENT);
1345 err = do_path_lookup(dfd, tmp, flags, &nd);
1353 static int user_path_parent(int dfd, const char __user *path,
1354 struct nameidata *nd, char **name)
1356 char *s = getname(path);
1362 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1372 * It's inline, so penalty for filesystems that don't use sticky bit is
1375 static inline int check_sticky(struct inode *dir, struct inode *inode)
1377 if (!(dir->i_mode & S_ISVTX))
1379 if (inode->i_uid == current->fsuid)
1381 if (dir->i_uid == current->fsuid)
1383 return !capable(CAP_FOWNER);
1387 * Check whether we can remove a link victim from directory dir, check
1388 * whether the type of victim is right.
1389 * 1. We can't do it if dir is read-only (done in permission())
1390 * 2. We should have write and exec permissions on dir
1391 * 3. We can't remove anything from append-only dir
1392 * 4. We can't do anything with immutable dir (done in permission())
1393 * 5. If the sticky bit on dir is set we should either
1394 * a. be owner of dir, or
1395 * b. be owner of victim, or
1396 * c. have CAP_FOWNER capability
1397 * 6. If the victim is append-only or immutable we can't do antyhing with
1398 * links pointing to it.
1399 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1400 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1401 * 9. We can't remove a root or mountpoint.
1402 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1403 * nfs_async_unlink().
1405 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1409 if (!victim->d_inode)
1412 BUG_ON(victim->d_parent->d_inode != dir);
1413 audit_inode_child(victim->d_name.name, victim, dir);
1415 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1420 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1421 IS_IMMUTABLE(victim->d_inode))
1424 if (!S_ISDIR(victim->d_inode->i_mode))
1426 if (IS_ROOT(victim))
1428 } else if (S_ISDIR(victim->d_inode->i_mode))
1430 if (IS_DEADDIR(dir))
1432 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1437 /* Check whether we can create an object with dentry child in directory
1439 * 1. We can't do it if child already exists (open has special treatment for
1440 * this case, but since we are inlined it's OK)
1441 * 2. We can't do it if dir is read-only (done in permission())
1442 * 3. We should have write and exec permissions on dir
1443 * 4. We can't do it if dir is immutable (done in permission())
1445 static inline int may_create(struct inode *dir, struct dentry *child)
1449 if (IS_DEADDIR(dir))
1451 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1455 * O_DIRECTORY translates into forcing a directory lookup.
1457 static inline int lookup_flags(unsigned int f)
1459 unsigned long retval = LOOKUP_FOLLOW;
1462 retval &= ~LOOKUP_FOLLOW;
1464 if (f & O_DIRECTORY)
1465 retval |= LOOKUP_DIRECTORY;
1471 * p1 and p2 should be directories on the same fs.
1473 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1478 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1482 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1484 for (p = p1; p->d_parent != p; p = p->d_parent) {
1485 if (p->d_parent == p2) {
1486 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1487 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1492 for (p = p2; p->d_parent != p; p = p->d_parent) {
1493 if (p->d_parent == p1) {
1494 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1495 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1500 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1501 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1505 void unlock_rename(struct dentry *p1, struct dentry *p2)
1507 mutex_unlock(&p1->d_inode->i_mutex);
1509 mutex_unlock(&p2->d_inode->i_mutex);
1510 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1514 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1515 struct nameidata *nd)
1517 int error = may_create(dir, dentry);
1522 if (!dir->i_op || !dir->i_op->create)
1523 return -EACCES; /* shouldn't it be ENOSYS? */
1526 error = security_inode_create(dir, dentry, mode);
1530 error = dir->i_op->create(dir, dentry, mode, nd);
1532 fsnotify_create(dir, dentry);
1536 int may_open(struct nameidata *nd, int acc_mode, int flag)
1538 struct dentry *dentry = nd->path.dentry;
1539 struct inode *inode = dentry->d_inode;
1545 if (S_ISLNK(inode->i_mode))
1548 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1552 * FIFO's, sockets and device files are special: they don't
1553 * actually live on the filesystem itself, and as such you
1554 * can write to them even if the filesystem is read-only.
1556 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1558 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1559 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1565 error = vfs_permission(nd, acc_mode);
1569 * An append-only file must be opened in append mode for writing.
1571 if (IS_APPEND(inode)) {
1572 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1578 /* O_NOATIME can only be set by the owner or superuser */
1579 if (flag & O_NOATIME)
1580 if (!is_owner_or_cap(inode))
1584 * Ensure there are no outstanding leases on the file.
1586 error = break_lease(inode, flag);
1590 if (flag & O_TRUNC) {
1591 error = get_write_access(inode);
1596 * Refuse to truncate files with mandatory locks held on them.
1598 error = locks_verify_locked(inode);
1602 error = do_truncate(dentry, 0,
1603 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1606 put_write_access(inode);
1610 if (flag & FMODE_WRITE)
1617 * Be careful about ever adding any more callers of this
1618 * function. Its flags must be in the namei format, not
1619 * what get passed to sys_open().
1621 static int __open_namei_create(struct nameidata *nd, struct path *path,
1625 struct dentry *dir = nd->path.dentry;
1627 if (!IS_POSIXACL(dir->d_inode))
1628 mode &= ~current->fs->umask;
1629 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1630 mutex_unlock(&dir->d_inode->i_mutex);
1631 dput(nd->path.dentry);
1632 nd->path.dentry = path->dentry;
1635 /* Don't check for write permission, don't truncate */
1636 return may_open(nd, 0, flag & ~O_TRUNC);
1640 * Note that while the flag value (low two bits) for sys_open means:
1645 * it is changed into
1646 * 00 - no permissions needed
1647 * 01 - read-permission
1648 * 10 - write-permission
1650 * for the internal routines (ie open_namei()/follow_link() etc)
1651 * This is more logical, and also allows the 00 "no perm needed"
1652 * to be used for symlinks (where the permissions are checked
1656 static inline int open_to_namei_flags(int flag)
1658 if ((flag+1) & O_ACCMODE)
1663 static int open_will_write_to_fs(int flag, struct inode *inode)
1666 * We'll never write to the fs underlying
1669 if (special_file(inode->i_mode))
1671 return (flag & O_TRUNC);
1675 * Note that the low bits of the passed in "open_flag"
1676 * are not the same as in the local variable "flag". See
1677 * open_to_namei_flags() for more details.
1679 struct file *do_filp_open(int dfd, const char *pathname,
1680 int open_flag, int mode)
1683 struct nameidata nd;
1684 int acc_mode, error;
1689 int flag = open_to_namei_flags(open_flag);
1691 acc_mode = MAY_OPEN | ACC_MODE(flag);
1693 /* O_TRUNC implies we need access checks for write permissions */
1695 acc_mode |= MAY_WRITE;
1697 /* Allow the LSM permission hook to distinguish append
1698 access from general write access. */
1699 if (flag & O_APPEND)
1700 acc_mode |= MAY_APPEND;
1703 * The simplest case - just a plain lookup.
1705 if (!(flag & O_CREAT)) {
1706 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1709 return ERR_PTR(error);
1714 * Create - we need to know the parent.
1716 error = path_lookup_create(dfd, pathname, LOOKUP_PARENT,
1719 return ERR_PTR(error);
1722 * We have the parent and last component. First of all, check
1723 * that we are not asked to creat(2) an obvious directory - that
1727 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1730 dir = nd.path.dentry;
1731 nd.flags &= ~LOOKUP_PARENT;
1732 mutex_lock(&dir->d_inode->i_mutex);
1733 path.dentry = lookup_hash(&nd);
1734 path.mnt = nd.path.mnt;
1737 error = PTR_ERR(path.dentry);
1738 if (IS_ERR(path.dentry)) {
1739 mutex_unlock(&dir->d_inode->i_mutex);
1743 if (IS_ERR(nd.intent.open.file)) {
1744 error = PTR_ERR(nd.intent.open.file);
1745 goto exit_mutex_unlock;
1748 /* Negative dentry, just create the file */
1749 if (!path.dentry->d_inode) {
1751 * This write is needed to ensure that a
1752 * ro->rw transition does not occur between
1753 * the time when the file is created and when
1754 * a permanent write count is taken through
1755 * the 'struct file' in nameidata_to_filp().
1757 error = mnt_want_write(nd.path.mnt);
1759 goto exit_mutex_unlock;
1760 error = __open_namei_create(&nd, &path, flag, mode);
1762 mnt_drop_write(nd.path.mnt);
1765 filp = nameidata_to_filp(&nd, open_flag);
1766 mnt_drop_write(nd.path.mnt);
1771 * It already exists.
1773 mutex_unlock(&dir->d_inode->i_mutex);
1774 audit_inode(pathname, path.dentry);
1780 if (__follow_mount(&path)) {
1782 if (flag & O_NOFOLLOW)
1787 if (!path.dentry->d_inode)
1789 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1792 path_to_nameidata(&path, &nd);
1794 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1799 * 1. may_open() truncates a file
1800 * 2. a rw->ro mount transition occurs
1801 * 3. nameidata_to_filp() fails due to
1803 * That would be inconsistent, and should
1804 * be avoided. Taking this mnt write here
1805 * ensures that (2) can not occur.
1807 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1809 error = mnt_want_write(nd.path.mnt);
1813 error = may_open(&nd, acc_mode, flag);
1816 mnt_drop_write(nd.path.mnt);
1819 filp = nameidata_to_filp(&nd, open_flag);
1821 * It is now safe to drop the mnt write
1822 * because the filp has had a write taken
1826 mnt_drop_write(nd.path.mnt);
1830 mutex_unlock(&dir->d_inode->i_mutex);
1832 path_put_conditional(&path, &nd);
1834 if (!IS_ERR(nd.intent.open.file))
1835 release_open_intent(&nd);
1837 return ERR_PTR(error);
1841 if (flag & O_NOFOLLOW)
1844 * This is subtle. Instead of calling do_follow_link() we do the
1845 * thing by hands. The reason is that this way we have zero link_count
1846 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1847 * After that we have the parent and last component, i.e.
1848 * we are in the same situation as after the first path_walk().
1849 * Well, almost - if the last component is normal we get its copy
1850 * stored in nd->last.name and we will have to putname() it when we
1851 * are done. Procfs-like symlinks just set LAST_BIND.
1853 nd.flags |= LOOKUP_PARENT;
1854 error = security_inode_follow_link(path.dentry, &nd);
1857 error = __do_follow_link(&path, &nd);
1859 /* Does someone understand code flow here? Or it is only
1860 * me so stupid? Anathema to whoever designed this non-sense
1861 * with "intent.open".
1863 release_open_intent(&nd);
1864 return ERR_PTR(error);
1866 nd.flags &= ~LOOKUP_PARENT;
1867 if (nd.last_type == LAST_BIND)
1870 if (nd.last_type != LAST_NORM)
1872 if (nd.last.name[nd.last.len]) {
1873 __putname(nd.last.name);
1878 __putname(nd.last.name);
1881 dir = nd.path.dentry;
1882 mutex_lock(&dir->d_inode->i_mutex);
1883 path.dentry = lookup_hash(&nd);
1884 path.mnt = nd.path.mnt;
1885 __putname(nd.last.name);
1890 * filp_open - open file and return file pointer
1892 * @filename: path to open
1893 * @flags: open flags as per the open(2) second argument
1894 * @mode: mode for the new file if O_CREAT is set, else ignored
1896 * This is the helper to open a file from kernelspace if you really
1897 * have to. But in generally you should not do this, so please move
1898 * along, nothing to see here..
1900 struct file *filp_open(const char *filename, int flags, int mode)
1902 return do_filp_open(AT_FDCWD, filename, flags, mode);
1904 EXPORT_SYMBOL(filp_open);
1907 * lookup_create - lookup a dentry, creating it if it doesn't exist
1908 * @nd: nameidata info
1909 * @is_dir: directory flag
1911 * Simple function to lookup and return a dentry and create it
1912 * if it doesn't exist. Is SMP-safe.
1914 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1916 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1918 struct dentry *dentry = ERR_PTR(-EEXIST);
1920 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1922 * Yucky last component or no last component at all?
1923 * (foo/., foo/.., /////)
1925 if (nd->last_type != LAST_NORM)
1927 nd->flags &= ~LOOKUP_PARENT;
1928 nd->flags |= LOOKUP_CREATE;
1929 nd->intent.open.flags = O_EXCL;
1932 * Do the final lookup.
1934 dentry = lookup_hash(nd);
1938 if (dentry->d_inode)
1941 * Special case - lookup gave negative, but... we had foo/bar/
1942 * From the vfs_mknod() POV we just have a negative dentry -
1943 * all is fine. Let's be bastards - you had / on the end, you've
1944 * been asking for (non-existent) directory. -ENOENT for you.
1946 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1948 dentry = ERR_PTR(-ENOENT);
1953 dentry = ERR_PTR(-EEXIST);
1957 EXPORT_SYMBOL_GPL(lookup_create);
1959 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1961 int error = may_create(dir, dentry);
1966 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1969 if (!dir->i_op || !dir->i_op->mknod)
1972 error = devcgroup_inode_mknod(mode, dev);
1976 error = security_inode_mknod(dir, dentry, mode, dev);
1981 error = dir->i_op->mknod(dir, dentry, mode, dev);
1983 fsnotify_create(dir, dentry);
1987 static int may_mknod(mode_t mode)
1989 switch (mode & S_IFMT) {
1995 case 0: /* zero mode translates to S_IFREG */
2004 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2009 struct dentry *dentry;
2010 struct nameidata nd;
2015 error = user_path_parent(dfd, filename, &nd, &tmp);
2019 dentry = lookup_create(&nd, 0);
2020 if (IS_ERR(dentry)) {
2021 error = PTR_ERR(dentry);
2024 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2025 mode &= ~current->fs->umask;
2026 error = may_mknod(mode);
2029 error = mnt_want_write(nd.path.mnt);
2032 switch (mode & S_IFMT) {
2033 case 0: case S_IFREG:
2034 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2036 case S_IFCHR: case S_IFBLK:
2037 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2038 new_decode_dev(dev));
2040 case S_IFIFO: case S_IFSOCK:
2041 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2044 mnt_drop_write(nd.path.mnt);
2048 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2055 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2057 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2060 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2062 int error = may_create(dir, dentry);
2067 if (!dir->i_op || !dir->i_op->mkdir)
2070 mode &= (S_IRWXUGO|S_ISVTX);
2071 error = security_inode_mkdir(dir, dentry, mode);
2076 error = dir->i_op->mkdir(dir, dentry, mode);
2078 fsnotify_mkdir(dir, dentry);
2082 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2086 struct dentry *dentry;
2087 struct nameidata nd;
2089 error = user_path_parent(dfd, pathname, &nd, &tmp);
2093 dentry = lookup_create(&nd, 1);
2094 error = PTR_ERR(dentry);
2098 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2099 mode &= ~current->fs->umask;
2100 error = mnt_want_write(nd.path.mnt);
2103 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2104 mnt_drop_write(nd.path.mnt);
2108 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2115 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2117 return sys_mkdirat(AT_FDCWD, pathname, mode);
2121 * We try to drop the dentry early: we should have
2122 * a usage count of 2 if we're the only user of this
2123 * dentry, and if that is true (possibly after pruning
2124 * the dcache), then we drop the dentry now.
2126 * A low-level filesystem can, if it choses, legally
2129 * if (!d_unhashed(dentry))
2132 * if it cannot handle the case of removing a directory
2133 * that is still in use by something else..
2135 void dentry_unhash(struct dentry *dentry)
2138 shrink_dcache_parent(dentry);
2139 spin_lock(&dcache_lock);
2140 spin_lock(&dentry->d_lock);
2141 if (atomic_read(&dentry->d_count) == 2)
2143 spin_unlock(&dentry->d_lock);
2144 spin_unlock(&dcache_lock);
2147 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2149 int error = may_delete(dir, dentry, 1);
2154 if (!dir->i_op || !dir->i_op->rmdir)
2159 mutex_lock(&dentry->d_inode->i_mutex);
2160 dentry_unhash(dentry);
2161 if (d_mountpoint(dentry))
2164 error = security_inode_rmdir(dir, dentry);
2166 error = dir->i_op->rmdir(dir, dentry);
2168 dentry->d_inode->i_flags |= S_DEAD;
2171 mutex_unlock(&dentry->d_inode->i_mutex);
2180 static long do_rmdir(int dfd, const char __user *pathname)
2184 struct dentry *dentry;
2185 struct nameidata nd;
2187 error = user_path_parent(dfd, pathname, &nd, &name);
2191 switch(nd.last_type) {
2202 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2203 dentry = lookup_hash(&nd);
2204 error = PTR_ERR(dentry);
2207 error = mnt_want_write(nd.path.mnt);
2210 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2211 mnt_drop_write(nd.path.mnt);
2215 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2222 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2224 return do_rmdir(AT_FDCWD, pathname);
2227 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2229 int error = may_delete(dir, dentry, 0);
2234 if (!dir->i_op || !dir->i_op->unlink)
2239 mutex_lock(&dentry->d_inode->i_mutex);
2240 if (d_mountpoint(dentry))
2243 error = security_inode_unlink(dir, dentry);
2245 error = dir->i_op->unlink(dir, dentry);
2247 mutex_unlock(&dentry->d_inode->i_mutex);
2249 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2250 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2251 fsnotify_link_count(dentry->d_inode);
2259 * Make sure that the actual truncation of the file will occur outside its
2260 * directory's i_mutex. Truncate can take a long time if there is a lot of
2261 * writeout happening, and we don't want to prevent access to the directory
2262 * while waiting on the I/O.
2264 static long do_unlinkat(int dfd, const char __user *pathname)
2268 struct dentry *dentry;
2269 struct nameidata nd;
2270 struct inode *inode = NULL;
2272 error = user_path_parent(dfd, pathname, &nd, &name);
2277 if (nd.last_type != LAST_NORM)
2279 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2280 dentry = lookup_hash(&nd);
2281 error = PTR_ERR(dentry);
2282 if (!IS_ERR(dentry)) {
2283 /* Why not before? Because we want correct error value */
2284 if (nd.last.name[nd.last.len])
2286 inode = dentry->d_inode;
2288 atomic_inc(&inode->i_count);
2289 error = mnt_want_write(nd.path.mnt);
2292 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2293 mnt_drop_write(nd.path.mnt);
2297 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2299 iput(inode); /* truncate the inode here */
2306 error = !dentry->d_inode ? -ENOENT :
2307 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2311 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2313 if ((flag & ~AT_REMOVEDIR) != 0)
2316 if (flag & AT_REMOVEDIR)
2317 return do_rmdir(dfd, pathname);
2319 return do_unlinkat(dfd, pathname);
2322 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2324 return do_unlinkat(AT_FDCWD, pathname);
2327 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2329 int error = may_create(dir, dentry);
2334 if (!dir->i_op || !dir->i_op->symlink)
2337 error = security_inode_symlink(dir, dentry, oldname);
2342 error = dir->i_op->symlink(dir, dentry, oldname);
2344 fsnotify_create(dir, dentry);
2348 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2349 int, newdfd, const char __user *, newname)
2354 struct dentry *dentry;
2355 struct nameidata nd;
2357 from = getname(oldname);
2359 return PTR_ERR(from);
2361 error = user_path_parent(newdfd, newname, &nd, &to);
2365 dentry = lookup_create(&nd, 0);
2366 error = PTR_ERR(dentry);
2370 error = mnt_want_write(nd.path.mnt);
2373 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2374 mnt_drop_write(nd.path.mnt);
2378 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2386 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2388 return sys_symlinkat(oldname, AT_FDCWD, newname);
2391 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2393 struct inode *inode = old_dentry->d_inode;
2399 error = may_create(dir, new_dentry);
2403 if (dir->i_sb != inode->i_sb)
2407 * A link to an append-only or immutable file cannot be created.
2409 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2411 if (!dir->i_op || !dir->i_op->link)
2413 if (S_ISDIR(inode->i_mode))
2416 error = security_inode_link(old_dentry, dir, new_dentry);
2420 mutex_lock(&inode->i_mutex);
2422 error = dir->i_op->link(old_dentry, dir, new_dentry);
2423 mutex_unlock(&inode->i_mutex);
2425 fsnotify_link(dir, inode, new_dentry);
2430 * Hardlinks are often used in delicate situations. We avoid
2431 * security-related surprises by not following symlinks on the
2434 * We don't follow them on the oldname either to be compatible
2435 * with linux 2.0, and to avoid hard-linking to directories
2436 * and other special files. --ADM
2438 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2439 int, newdfd, const char __user *, newname, int, flags)
2441 struct dentry *new_dentry;
2442 struct nameidata nd;
2443 struct path old_path;
2447 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2450 error = user_path_at(olddfd, oldname,
2451 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2456 error = user_path_parent(newdfd, newname, &nd, &to);
2460 if (old_path.mnt != nd.path.mnt)
2462 new_dentry = lookup_create(&nd, 0);
2463 error = PTR_ERR(new_dentry);
2464 if (IS_ERR(new_dentry))
2466 error = mnt_want_write(nd.path.mnt);
2469 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2470 mnt_drop_write(nd.path.mnt);
2474 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2479 path_put(&old_path);
2484 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2486 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2490 * The worst of all namespace operations - renaming directory. "Perverted"
2491 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2493 * a) we can get into loop creation. Check is done in is_subdir().
2494 * b) race potential - two innocent renames can create a loop together.
2495 * That's where 4.4 screws up. Current fix: serialization on
2496 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2498 * c) we have to lock _three_ objects - parents and victim (if it exists).
2499 * And that - after we got ->i_mutex on parents (until then we don't know
2500 * whether the target exists). Solution: try to be smart with locking
2501 * order for inodes. We rely on the fact that tree topology may change
2502 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2503 * move will be locked. Thus we can rank directories by the tree
2504 * (ancestors first) and rank all non-directories after them.
2505 * That works since everybody except rename does "lock parent, lookup,
2506 * lock child" and rename is under ->s_vfs_rename_mutex.
2507 * HOWEVER, it relies on the assumption that any object with ->lookup()
2508 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2509 * we'd better make sure that there's no link(2) for them.
2510 * d) some filesystems don't support opened-but-unlinked directories,
2511 * either because of layout or because they are not ready to deal with
2512 * all cases correctly. The latter will be fixed (taking this sort of
2513 * stuff into VFS), but the former is not going away. Solution: the same
2514 * trick as in rmdir().
2515 * e) conversion from fhandle to dentry may come in the wrong moment - when
2516 * we are removing the target. Solution: we will have to grab ->i_mutex
2517 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2518 * ->i_mutex on parents, which works but leads to some truely excessive
2521 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2522 struct inode *new_dir, struct dentry *new_dentry)
2525 struct inode *target;
2528 * If we are going to change the parent - check write permissions,
2529 * we'll need to flip '..'.
2531 if (new_dir != old_dir) {
2532 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2537 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2541 target = new_dentry->d_inode;
2543 mutex_lock(&target->i_mutex);
2544 dentry_unhash(new_dentry);
2546 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2549 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2552 target->i_flags |= S_DEAD;
2553 mutex_unlock(&target->i_mutex);
2554 if (d_unhashed(new_dentry))
2555 d_rehash(new_dentry);
2559 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2560 d_move(old_dentry,new_dentry);
2564 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2565 struct inode *new_dir, struct dentry *new_dentry)
2567 struct inode *target;
2570 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2575 target = new_dentry->d_inode;
2577 mutex_lock(&target->i_mutex);
2578 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2581 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2583 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2584 d_move(old_dentry, new_dentry);
2587 mutex_unlock(&target->i_mutex);
2592 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2593 struct inode *new_dir, struct dentry *new_dentry)
2596 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2597 const char *old_name;
2599 if (old_dentry->d_inode == new_dentry->d_inode)
2602 error = may_delete(old_dir, old_dentry, is_dir);
2606 if (!new_dentry->d_inode)
2607 error = may_create(new_dir, new_dentry);
2609 error = may_delete(new_dir, new_dentry, is_dir);
2613 if (!old_dir->i_op || !old_dir->i_op->rename)
2616 DQUOT_INIT(old_dir);
2617 DQUOT_INIT(new_dir);
2619 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2622 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2624 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2626 const char *new_name = old_dentry->d_name.name;
2627 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2628 new_dentry->d_inode, old_dentry);
2630 fsnotify_oldname_free(old_name);
2635 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2636 int, newdfd, const char __user *, newname)
2638 struct dentry *old_dir, *new_dir;
2639 struct dentry *old_dentry, *new_dentry;
2640 struct dentry *trap;
2641 struct nameidata oldnd, newnd;
2646 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2650 error = user_path_parent(newdfd, newname, &newnd, &to);
2655 if (oldnd.path.mnt != newnd.path.mnt)
2658 old_dir = oldnd.path.dentry;
2660 if (oldnd.last_type != LAST_NORM)
2663 new_dir = newnd.path.dentry;
2664 if (newnd.last_type != LAST_NORM)
2667 trap = lock_rename(new_dir, old_dir);
2669 old_dentry = lookup_hash(&oldnd);
2670 error = PTR_ERR(old_dentry);
2671 if (IS_ERR(old_dentry))
2673 /* source must exist */
2675 if (!old_dentry->d_inode)
2677 /* unless the source is a directory trailing slashes give -ENOTDIR */
2678 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2680 if (oldnd.last.name[oldnd.last.len])
2682 if (newnd.last.name[newnd.last.len])
2685 /* source should not be ancestor of target */
2687 if (old_dentry == trap)
2689 new_dentry = lookup_hash(&newnd);
2690 error = PTR_ERR(new_dentry);
2691 if (IS_ERR(new_dentry))
2693 /* target should not be an ancestor of source */
2695 if (new_dentry == trap)
2698 error = mnt_want_write(oldnd.path.mnt);
2701 error = vfs_rename(old_dir->d_inode, old_dentry,
2702 new_dir->d_inode, new_dentry);
2703 mnt_drop_write(oldnd.path.mnt);
2709 unlock_rename(new_dir, old_dir);
2711 path_put(&newnd.path);
2714 path_put(&oldnd.path);
2720 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2722 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2725 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2729 len = PTR_ERR(link);
2734 if (len > (unsigned) buflen)
2736 if (copy_to_user(buffer, link, len))
2743 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2744 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2745 * using) it for any given inode is up to filesystem.
2747 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2749 struct nameidata nd;
2754 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2756 return PTR_ERR(cookie);
2758 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2759 if (dentry->d_inode->i_op->put_link)
2760 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2764 int vfs_follow_link(struct nameidata *nd, const char *link)
2766 return __vfs_follow_link(nd, link);
2769 /* get the link contents into pagecache */
2770 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2773 struct address_space *mapping = dentry->d_inode->i_mapping;
2774 page = read_mapping_page(mapping, 0, NULL);
2781 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2783 struct page *page = NULL;
2784 char *s = page_getlink(dentry, &page);
2785 int res = vfs_readlink(dentry,buffer,buflen,s);
2788 page_cache_release(page);
2793 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2795 struct page *page = NULL;
2796 nd_set_link(nd, page_getlink(dentry, &page));
2800 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2802 struct page *page = cookie;
2806 page_cache_release(page);
2811 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2813 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2815 struct address_space *mapping = inode->i_mapping;
2820 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2822 flags |= AOP_FLAG_NOFS;
2825 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2826 flags, &page, &fsdata);
2830 kaddr = kmap_atomic(page, KM_USER0);
2831 memcpy(kaddr, symname, len-1);
2832 kunmap_atomic(kaddr, KM_USER0);
2834 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2841 mark_inode_dirty(inode);
2847 int page_symlink(struct inode *inode, const char *symname, int len)
2849 return __page_symlink(inode, symname, len,
2850 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2853 const struct inode_operations page_symlink_inode_operations = {
2854 .readlink = generic_readlink,
2855 .follow_link = page_follow_link_light,
2856 .put_link = page_put_link,
2859 EXPORT_SYMBOL(user_path_at);
2860 EXPORT_SYMBOL(follow_down);
2861 EXPORT_SYMBOL(follow_up);
2862 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2863 EXPORT_SYMBOL(getname);
2864 EXPORT_SYMBOL(lock_rename);
2865 EXPORT_SYMBOL(lookup_one_len);
2866 EXPORT_SYMBOL(page_follow_link_light);
2867 EXPORT_SYMBOL(page_put_link);
2868 EXPORT_SYMBOL(page_readlink);
2869 EXPORT_SYMBOL(__page_symlink);
2870 EXPORT_SYMBOL(page_symlink);
2871 EXPORT_SYMBOL(page_symlink_inode_operations);
2872 EXPORT_SYMBOL(path_lookup);
2873 EXPORT_SYMBOL(vfs_path_lookup);
2874 EXPORT_SYMBOL(inode_permission);
2875 EXPORT_SYMBOL(vfs_permission);
2876 EXPORT_SYMBOL(file_permission);
2877 EXPORT_SYMBOL(unlock_rename);
2878 EXPORT_SYMBOL(vfs_create);
2879 EXPORT_SYMBOL(vfs_follow_link);
2880 EXPORT_SYMBOL(vfs_link);
2881 EXPORT_SYMBOL(vfs_mkdir);
2882 EXPORT_SYMBOL(vfs_mknod);
2883 EXPORT_SYMBOL(generic_permission);
2884 EXPORT_SYMBOL(vfs_readlink);
2885 EXPORT_SYMBOL(vfs_rename);
2886 EXPORT_SYMBOL(vfs_rmdir);
2887 EXPORT_SYMBOL(vfs_symlink);
2888 EXPORT_SYMBOL(vfs_unlink);
2889 EXPORT_SYMBOL(dentry_unhash);
2890 EXPORT_SYMBOL(generic_readlink);