4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name, struct nameidata *nd);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
185 umode_t mode = inode->i_mode;
187 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
189 if (current_fsuid() == inode->i_uid)
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
200 if (in_group_p(inode->i_gid))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask & ~mode) == 0)
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask & MAY_EXEC) || execute_ok(inode))
216 if (capable(CAP_DAC_OVERRIDE))
220 * Searching includes executable on directories, else just read.
222 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
223 if (capable(CAP_DAC_READ_SEARCH))
229 int inode_permission(struct inode *inode, int mask)
233 if (mask & MAY_WRITE) {
234 umode_t mode = inode->i_mode;
237 * Nobody gets write access to a read-only fs.
239 if (IS_RDONLY(inode) &&
240 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
244 * Nobody gets write access to an immutable file.
246 if (IS_IMMUTABLE(inode))
250 if (inode->i_op && inode->i_op->permission)
251 retval = inode->i_op->permission(inode, mask);
253 retval = generic_permission(inode, mask, NULL);
258 retval = devcgroup_inode_permission(inode, mask);
262 return security_inode_permission(inode,
263 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
267 * vfs_permission - check for access rights to a given path
268 * @nd: lookup result that describes the path
269 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271 * Used to check for read/write/execute permissions on a path.
272 * We use "fsuid" for this, letting us set arbitrary permissions
273 * for filesystem access without changing the "normal" uids which
274 * are used for other things.
276 int vfs_permission(struct nameidata *nd, int mask)
278 return inode_permission(nd->path.dentry->d_inode, mask);
282 * file_permission - check for additional access rights to a given file
283 * @file: file to check access rights for
284 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
286 * Used to check for read/write/execute permissions on an already opened
290 * Do not use this function in new code. All access checks should
291 * be done using vfs_permission().
293 int file_permission(struct file *file, int mask)
295 return inode_permission(file->f_path.dentry->d_inode, mask);
299 * get_write_access() gets write permission for a file.
300 * put_write_access() releases this write permission.
301 * This is used for regular files.
302 * We cannot support write (and maybe mmap read-write shared) accesses and
303 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
304 * can have the following values:
305 * 0: no writers, no VM_DENYWRITE mappings
306 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
307 * > 0: (i_writecount) users are writing to the file.
309 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
310 * except for the cases where we don't hold i_writecount yet. Then we need to
311 * use {get,deny}_write_access() - these functions check the sign and refuse
312 * to do the change if sign is wrong. Exclusion between them is provided by
313 * the inode->i_lock spinlock.
316 int get_write_access(struct inode * inode)
318 spin_lock(&inode->i_lock);
319 if (atomic_read(&inode->i_writecount) < 0) {
320 spin_unlock(&inode->i_lock);
323 atomic_inc(&inode->i_writecount);
324 spin_unlock(&inode->i_lock);
329 int deny_write_access(struct file * file)
331 struct inode *inode = file->f_path.dentry->d_inode;
333 spin_lock(&inode->i_lock);
334 if (atomic_read(&inode->i_writecount) > 0) {
335 spin_unlock(&inode->i_lock);
338 atomic_dec(&inode->i_writecount);
339 spin_unlock(&inode->i_lock);
345 * path_get - get a reference to a path
346 * @path: path to get the reference to
348 * Given a path increment the reference count to the dentry and the vfsmount.
350 void path_get(struct path *path)
355 EXPORT_SYMBOL(path_get);
358 * path_put - put a reference to a path
359 * @path: path to put the reference to
361 * Given a path decrement the reference count to the dentry and the vfsmount.
363 void path_put(struct path *path)
368 EXPORT_SYMBOL(path_put);
371 * release_open_intent - free up open intent resources
372 * @nd: pointer to nameidata
374 void release_open_intent(struct nameidata *nd)
376 if (nd->intent.open.file->f_path.dentry == NULL)
377 put_filp(nd->intent.open.file);
379 fput(nd->intent.open.file);
382 static inline struct dentry *
383 do_revalidate(struct dentry *dentry, struct nameidata *nd)
385 int status = dentry->d_op->d_revalidate(dentry, nd);
386 if (unlikely(status <= 0)) {
388 * The dentry failed validation.
389 * If d_revalidate returned 0 attempt to invalidate
390 * the dentry otherwise d_revalidate is asking us
391 * to return a fail status.
394 if (!d_invalidate(dentry)) {
400 dentry = ERR_PTR(status);
407 * Internal lookup() using the new generic dcache.
410 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
412 struct dentry * dentry = __d_lookup(parent, name);
414 /* lockess __d_lookup may fail due to concurrent d_move()
415 * in some unrelated directory, so try with d_lookup
418 dentry = d_lookup(parent, name);
420 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
421 dentry = do_revalidate(dentry, nd);
427 * Short-cut version of permission(), for calling by
428 * path_walk(), when dcache lock is held. Combines parts
429 * of permission() and generic_permission(), and tests ONLY for
430 * MAY_EXEC permission.
432 * If appropriate, check DAC only. If not appropriate, or
433 * short-cut DAC fails, then call permission() to do more
434 * complete permission check.
436 static int exec_permission_lite(struct inode *inode)
438 umode_t mode = inode->i_mode;
440 if (inode->i_op && inode->i_op->permission)
443 if (current_fsuid() == inode->i_uid)
445 else if (in_group_p(inode->i_gid))
451 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
454 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
457 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
462 return security_inode_permission(inode, MAY_EXEC);
466 * This is called when everything else fails, and we actually have
467 * to go to the low-level filesystem to find out what we should do..
469 * We get the directory semaphore, and after getting that we also
470 * make sure that nobody added the entry to the dcache in the meantime..
473 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
475 struct dentry * result;
476 struct inode *dir = parent->d_inode;
478 mutex_lock(&dir->i_mutex);
480 * First re-do the cached lookup just in case it was created
481 * while we waited for the directory semaphore..
483 * FIXME! This could use version numbering or similar to
484 * avoid unnecessary cache lookups.
486 * The "dcache_lock" is purely to protect the RCU list walker
487 * from concurrent renames at this point (we mustn't get false
488 * negatives from the RCU list walk here, unlike the optimistic
491 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
493 result = d_lookup(parent, name);
495 struct dentry *dentry;
497 /* Don't create child dentry for a dead directory. */
498 result = ERR_PTR(-ENOENT);
502 dentry = d_alloc(parent, name);
503 result = ERR_PTR(-ENOMEM);
505 result = dir->i_op->lookup(dir, dentry, nd);
512 mutex_unlock(&dir->i_mutex);
517 * Uhhuh! Nasty case: the cache was re-populated while
518 * we waited on the semaphore. Need to revalidate.
520 mutex_unlock(&dir->i_mutex);
521 if (result->d_op && result->d_op->d_revalidate) {
522 result = do_revalidate(result, nd);
524 result = ERR_PTR(-ENOENT);
530 * Wrapper to retry pathname resolution whenever the underlying
531 * file system returns an ESTALE.
533 * Retry the whole path once, forcing real lookup requests
534 * instead of relying on the dcache.
536 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
538 struct path save = nd->path;
541 /* make sure the stuff we saved doesn't go away */
544 result = __link_path_walk(name, nd);
545 if (result == -ESTALE) {
546 /* nd->path had been dropped */
549 nd->flags |= LOOKUP_REVAL;
550 result = __link_path_walk(name, nd);
558 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
566 struct fs_struct *fs = current->fs;
570 read_lock(&fs->lock);
573 read_unlock(&fs->lock);
576 res = link_path_walk(link, nd);
577 if (nd->depth || res || nd->last_type!=LAST_NORM)
580 * If it is an iterative symlinks resolution in open_namei() we
581 * have to copy the last component. And all that crap because of
582 * bloody create() on broken symlinks. Furrfu...
585 if (unlikely(!name)) {
589 strcpy(name, nd->last.name);
590 nd->last.name = name;
594 return PTR_ERR(link);
597 static void path_put_conditional(struct path *path, struct nameidata *nd)
600 if (path->mnt != nd->path.mnt)
604 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
606 dput(nd->path.dentry);
607 if (nd->path.mnt != path->mnt)
608 mntput(nd->path.mnt);
609 nd->path.mnt = path->mnt;
610 nd->path.dentry = path->dentry;
613 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
617 struct dentry *dentry = path->dentry;
619 touch_atime(path->mnt, dentry);
620 nd_set_link(nd, NULL);
622 if (path->mnt != nd->path.mnt) {
623 path_to_nameidata(path, nd);
627 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
628 error = PTR_ERR(cookie);
629 if (!IS_ERR(cookie)) {
630 char *s = nd_get_link(nd);
633 error = __vfs_follow_link(nd, s);
634 if (dentry->d_inode->i_op->put_link)
635 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
643 * This limits recursive symlink follows to 8, while
644 * limiting consecutive symlinks to 40.
646 * Without that kind of total limit, nasty chains of consecutive
647 * symlinks can cause almost arbitrarily long lookups.
649 static inline int do_follow_link(struct path *path, struct nameidata *nd)
652 if (current->link_count >= MAX_NESTED_LINKS)
654 if (current->total_link_count >= 40)
656 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
658 err = security_inode_follow_link(path->dentry, nd);
661 current->link_count++;
662 current->total_link_count++;
664 err = __do_follow_link(path, nd);
665 current->link_count--;
669 path_put_conditional(path, nd);
674 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
676 struct vfsmount *parent;
677 struct dentry *mountpoint;
678 spin_lock(&vfsmount_lock);
679 parent=(*mnt)->mnt_parent;
680 if (parent == *mnt) {
681 spin_unlock(&vfsmount_lock);
685 mountpoint=dget((*mnt)->mnt_mountpoint);
686 spin_unlock(&vfsmount_lock);
688 *dentry = mountpoint;
694 /* no need for dcache_lock, as serialization is taken care in
697 static int __follow_mount(struct path *path)
700 while (d_mountpoint(path->dentry)) {
701 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
708 path->dentry = dget(mounted->mnt_root);
714 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
716 while (d_mountpoint(*dentry)) {
717 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
723 *dentry = dget(mounted->mnt_root);
727 /* no need for dcache_lock, as serialization is taken care in
730 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
732 struct vfsmount *mounted;
734 mounted = lookup_mnt(*mnt, *dentry);
739 *dentry = dget(mounted->mnt_root);
745 static __always_inline void follow_dotdot(struct nameidata *nd)
747 struct fs_struct *fs = current->fs;
750 struct vfsmount *parent;
751 struct dentry *old = nd->path.dentry;
753 read_lock(&fs->lock);
754 if (nd->path.dentry == fs->root.dentry &&
755 nd->path.mnt == fs->root.mnt) {
756 read_unlock(&fs->lock);
759 read_unlock(&fs->lock);
760 spin_lock(&dcache_lock);
761 if (nd->path.dentry != nd->path.mnt->mnt_root) {
762 nd->path.dentry = dget(nd->path.dentry->d_parent);
763 spin_unlock(&dcache_lock);
767 spin_unlock(&dcache_lock);
768 spin_lock(&vfsmount_lock);
769 parent = nd->path.mnt->mnt_parent;
770 if (parent == nd->path.mnt) {
771 spin_unlock(&vfsmount_lock);
775 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
776 spin_unlock(&vfsmount_lock);
778 mntput(nd->path.mnt);
779 nd->path.mnt = parent;
781 follow_mount(&nd->path.mnt, &nd->path.dentry);
785 * It's more convoluted than I'd like it to be, but... it's still fairly
786 * small and for now I'd prefer to have fast path as straight as possible.
787 * It _is_ time-critical.
789 static int do_lookup(struct nameidata *nd, struct qstr *name,
792 struct vfsmount *mnt = nd->path.mnt;
793 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
797 if (dentry->d_op && dentry->d_op->d_revalidate)
798 goto need_revalidate;
801 path->dentry = dentry;
802 __follow_mount(path);
806 dentry = real_lookup(nd->path.dentry, name, nd);
812 dentry = do_revalidate(dentry, nd);
820 return PTR_ERR(dentry);
825 * This is the basic name resolution function, turning a pathname into
826 * the final dentry. We expect 'base' to be positive and a directory.
828 * Returns 0 and nd will have valid dentry and mnt on success.
829 * Returns error and drops reference to input namei data on failure.
831 static int __link_path_walk(const char *name, struct nameidata *nd)
836 unsigned int lookup_flags = nd->flags;
843 inode = nd->path.dentry->d_inode;
845 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
847 /* At this point we know we have a real path component. */
853 nd->flags |= LOOKUP_CONTINUE;
854 err = exec_permission_lite(inode);
856 err = vfs_permission(nd, MAY_EXEC);
861 c = *(const unsigned char *)name;
863 hash = init_name_hash();
866 hash = partial_name_hash(c, hash);
867 c = *(const unsigned char *)name;
868 } while (c && (c != '/'));
869 this.len = name - (const char *) this.name;
870 this.hash = end_name_hash(hash);
872 /* remove trailing slashes? */
875 while (*++name == '/');
877 goto last_with_slashes;
880 * "." and ".." are special - ".." especially so because it has
881 * to be able to know about the current root directory and
882 * parent relationships.
884 if (this.name[0] == '.') switch (this.len) {
888 if (this.name[1] != '.')
891 inode = nd->path.dentry->d_inode;
897 * See if the low-level filesystem might want
898 * to use its own hash..
900 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
901 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
906 /* This does the actual lookups.. */
907 err = do_lookup(nd, &this, &next);
912 inode = next.dentry->d_inode;
919 if (inode->i_op->follow_link) {
920 err = do_follow_link(&next, nd);
924 inode = nd->path.dentry->d_inode;
931 path_to_nameidata(&next, nd);
933 if (!inode->i_op->lookup)
936 /* here ends the main loop */
939 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
941 /* Clear LOOKUP_CONTINUE iff it was previously unset */
942 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
943 if (lookup_flags & LOOKUP_PARENT)
945 if (this.name[0] == '.') switch (this.len) {
949 if (this.name[1] != '.')
952 inode = nd->path.dentry->d_inode;
957 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
958 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
963 err = do_lookup(nd, &this, &next);
966 inode = next.dentry->d_inode;
967 if ((lookup_flags & LOOKUP_FOLLOW)
968 && inode && inode->i_op && inode->i_op->follow_link) {
969 err = do_follow_link(&next, nd);
972 inode = nd->path.dentry->d_inode;
974 path_to_nameidata(&next, nd);
978 if (lookup_flags & LOOKUP_DIRECTORY) {
980 if (!inode->i_op || !inode->i_op->lookup)
986 nd->last_type = LAST_NORM;
987 if (this.name[0] != '.')
990 nd->last_type = LAST_DOT;
991 else if (this.len == 2 && this.name[1] == '.')
992 nd->last_type = LAST_DOTDOT;
997 * We bypassed the ordinary revalidation routines.
998 * We may need to check the cached dentry for staleness.
1000 if (nd->path.dentry && nd->path.dentry->d_sb &&
1001 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1003 /* Note: we do not d_invalidate() */
1004 if (!nd->path.dentry->d_op->d_revalidate(
1005 nd->path.dentry, nd))
1011 path_put_conditional(&next, nd);
1014 path_put(&nd->path);
1019 static int path_walk(const char *name, struct nameidata *nd)
1021 current->total_link_count = 0;
1022 return link_path_walk(name, nd);
1025 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1026 static int do_path_lookup(int dfd, const char *name,
1027 unsigned int flags, struct nameidata *nd)
1032 struct fs_struct *fs = current->fs;
1034 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1039 read_lock(&fs->lock);
1040 nd->path = fs->root;
1041 path_get(&fs->root);
1042 read_unlock(&fs->lock);
1043 } else if (dfd == AT_FDCWD) {
1044 read_lock(&fs->lock);
1047 read_unlock(&fs->lock);
1049 struct dentry *dentry;
1051 file = fget_light(dfd, &fput_needed);
1056 dentry = file->f_path.dentry;
1059 if (!S_ISDIR(dentry->d_inode->i_mode))
1062 retval = file_permission(file, MAY_EXEC);
1066 nd->path = file->f_path;
1067 path_get(&file->f_path);
1069 fput_light(file, fput_needed);
1072 retval = path_walk(name, nd);
1073 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1074 nd->path.dentry->d_inode))
1075 audit_inode(name, nd->path.dentry);
1080 fput_light(file, fput_needed);
1084 int path_lookup(const char *name, unsigned int flags,
1085 struct nameidata *nd)
1087 return do_path_lookup(AT_FDCWD, name, flags, nd);
1090 int kern_path(const char *name, unsigned int flags, struct path *path)
1092 struct nameidata nd;
1093 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1100 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1101 * @dentry: pointer to dentry of the base directory
1102 * @mnt: pointer to vfs mount of the base directory
1103 * @name: pointer to file name
1104 * @flags: lookup flags
1105 * @nd: pointer to nameidata
1107 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1108 const char *name, unsigned int flags,
1109 struct nameidata *nd)
1113 /* same as do_path_lookup */
1114 nd->last_type = LAST_ROOT;
1118 nd->path.dentry = dentry;
1120 path_get(&nd->path);
1122 retval = path_walk(name, nd);
1123 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1124 nd->path.dentry->d_inode))
1125 audit_inode(name, nd->path.dentry);
1132 * path_lookup_open - lookup a file path with open intent
1133 * @dfd: the directory to use as base, or AT_FDCWD
1134 * @name: pointer to file name
1135 * @lookup_flags: lookup intent flags
1136 * @nd: pointer to nameidata
1137 * @open_flags: open intent flags
1139 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1140 struct nameidata *nd, int open_flags)
1142 struct file *filp = get_empty_filp();
1147 nd->intent.open.file = filp;
1148 nd->intent.open.flags = open_flags;
1149 nd->intent.open.create_mode = 0;
1150 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1151 if (IS_ERR(nd->intent.open.file)) {
1153 err = PTR_ERR(nd->intent.open.file);
1154 path_put(&nd->path);
1156 } else if (err != 0)
1157 release_open_intent(nd);
1161 static struct dentry *__lookup_hash(struct qstr *name,
1162 struct dentry *base, struct nameidata *nd)
1164 struct dentry *dentry;
1165 struct inode *inode;
1168 inode = base->d_inode;
1171 * See if the low-level filesystem might want
1172 * to use its own hash..
1174 if (base->d_op && base->d_op->d_hash) {
1175 err = base->d_op->d_hash(base, name);
1176 dentry = ERR_PTR(err);
1181 dentry = cached_lookup(base, name, nd);
1185 /* Don't create child dentry for a dead directory. */
1186 dentry = ERR_PTR(-ENOENT);
1187 if (IS_DEADDIR(inode))
1190 new = d_alloc(base, name);
1191 dentry = ERR_PTR(-ENOMEM);
1194 dentry = inode->i_op->lookup(inode, new, nd);
1205 * Restricted form of lookup. Doesn't follow links, single-component only,
1206 * needs parent already locked. Doesn't follow mounts.
1209 static struct dentry *lookup_hash(struct nameidata *nd)
1213 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1215 return ERR_PTR(err);
1216 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1219 static int __lookup_one_len(const char *name, struct qstr *this,
1220 struct dentry *base, int len)
1230 hash = init_name_hash();
1232 c = *(const unsigned char *)name++;
1233 if (c == '/' || c == '\0')
1235 hash = partial_name_hash(c, hash);
1237 this->hash = end_name_hash(hash);
1242 * lookup_one_len - filesystem helper to lookup single pathname component
1243 * @name: pathname component to lookup
1244 * @base: base directory to lookup from
1245 * @len: maximum length @len should be interpreted to
1247 * Note that this routine is purely a helper for filesystem usage and should
1248 * not be called by generic code. Also note that by using this function the
1249 * nameidata argument is passed to the filesystem methods and a filesystem
1250 * using this helper needs to be prepared for that.
1252 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1257 err = __lookup_one_len(name, &this, base, len);
1259 return ERR_PTR(err);
1261 err = inode_permission(base->d_inode, MAY_EXEC);
1263 return ERR_PTR(err);
1264 return __lookup_hash(&this, base, NULL);
1268 * lookup_one_noperm - bad hack for sysfs
1269 * @name: pathname component to lookup
1270 * @base: base directory to lookup from
1272 * This is a variant of lookup_one_len that doesn't perform any permission
1273 * checks. It's a horrible hack to work around the braindead sysfs
1274 * architecture and should not be used anywhere else.
1276 * DON'T USE THIS FUNCTION EVER, thanks.
1278 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1283 err = __lookup_one_len(name, &this, base, strlen(name));
1285 return ERR_PTR(err);
1286 return __lookup_hash(&this, base, NULL);
1289 int user_path_at(int dfd, const char __user *name, unsigned flags,
1292 struct nameidata nd;
1293 char *tmp = getname(name);
1294 int err = PTR_ERR(tmp);
1297 BUG_ON(flags & LOOKUP_PARENT);
1299 err = do_path_lookup(dfd, tmp, flags, &nd);
1307 static int user_path_parent(int dfd, const char __user *path,
1308 struct nameidata *nd, char **name)
1310 char *s = getname(path);
1316 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1326 * It's inline, so penalty for filesystems that don't use sticky bit is
1329 static inline int check_sticky(struct inode *dir, struct inode *inode)
1331 uid_t fsuid = current_fsuid();
1333 if (!(dir->i_mode & S_ISVTX))
1335 if (inode->i_uid == fsuid)
1337 if (dir->i_uid == fsuid)
1339 return !capable(CAP_FOWNER);
1343 * Check whether we can remove a link victim from directory dir, check
1344 * whether the type of victim is right.
1345 * 1. We can't do it if dir is read-only (done in permission())
1346 * 2. We should have write and exec permissions on dir
1347 * 3. We can't remove anything from append-only dir
1348 * 4. We can't do anything with immutable dir (done in permission())
1349 * 5. If the sticky bit on dir is set we should either
1350 * a. be owner of dir, or
1351 * b. be owner of victim, or
1352 * c. have CAP_FOWNER capability
1353 * 6. If the victim is append-only or immutable we can't do antyhing with
1354 * links pointing to it.
1355 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1356 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1357 * 9. We can't remove a root or mountpoint.
1358 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1359 * nfs_async_unlink().
1361 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1365 if (!victim->d_inode)
1368 BUG_ON(victim->d_parent->d_inode != dir);
1369 audit_inode_child(victim->d_name.name, victim, dir);
1371 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1376 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1377 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1380 if (!S_ISDIR(victim->d_inode->i_mode))
1382 if (IS_ROOT(victim))
1384 } else if (S_ISDIR(victim->d_inode->i_mode))
1386 if (IS_DEADDIR(dir))
1388 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1393 /* Check whether we can create an object with dentry child in directory
1395 * 1. We can't do it if child already exists (open has special treatment for
1396 * this case, but since we are inlined it's OK)
1397 * 2. We can't do it if dir is read-only (done in permission())
1398 * 3. We should have write and exec permissions on dir
1399 * 4. We can't do it if dir is immutable (done in permission())
1401 static inline int may_create(struct inode *dir, struct dentry *child)
1405 if (IS_DEADDIR(dir))
1407 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1411 * O_DIRECTORY translates into forcing a directory lookup.
1413 static inline int lookup_flags(unsigned int f)
1415 unsigned long retval = LOOKUP_FOLLOW;
1418 retval &= ~LOOKUP_FOLLOW;
1420 if (f & O_DIRECTORY)
1421 retval |= LOOKUP_DIRECTORY;
1427 * p1 and p2 should be directories on the same fs.
1429 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1434 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1438 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1440 p = d_ancestor(p2, p1);
1442 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1443 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1447 p = d_ancestor(p1, p2);
1449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1450 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1454 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1455 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1459 void unlock_rename(struct dentry *p1, struct dentry *p2)
1461 mutex_unlock(&p1->d_inode->i_mutex);
1463 mutex_unlock(&p2->d_inode->i_mutex);
1464 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1468 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1469 struct nameidata *nd)
1471 int error = may_create(dir, dentry);
1476 if (!dir->i_op || !dir->i_op->create)
1477 return -EACCES; /* shouldn't it be ENOSYS? */
1480 error = security_inode_create(dir, dentry, mode);
1484 error = dir->i_op->create(dir, dentry, mode, nd);
1486 fsnotify_create(dir, dentry);
1490 int may_open(struct nameidata *nd, int acc_mode, int flag)
1492 struct dentry *dentry = nd->path.dentry;
1493 struct inode *inode = dentry->d_inode;
1499 if (S_ISLNK(inode->i_mode))
1502 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1506 * FIFO's, sockets and device files are special: they don't
1507 * actually live on the filesystem itself, and as such you
1508 * can write to them even if the filesystem is read-only.
1510 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1512 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1513 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1519 error = vfs_permission(nd, acc_mode);
1523 * An append-only file must be opened in append mode for writing.
1525 if (IS_APPEND(inode)) {
1526 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1532 /* O_NOATIME can only be set by the owner or superuser */
1533 if (flag & O_NOATIME)
1534 if (!is_owner_or_cap(inode))
1538 * Ensure there are no outstanding leases on the file.
1540 error = break_lease(inode, flag);
1544 if (flag & O_TRUNC) {
1545 error = get_write_access(inode);
1550 * Refuse to truncate files with mandatory locks held on them.
1552 error = locks_verify_locked(inode);
1554 error = security_path_truncate(&nd->path, 0,
1555 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1559 error = do_truncate(dentry, 0,
1560 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1563 put_write_access(inode);
1567 if (flag & FMODE_WRITE)
1574 * Be careful about ever adding any more callers of this
1575 * function. Its flags must be in the namei format, not
1576 * what get passed to sys_open().
1578 static int __open_namei_create(struct nameidata *nd, struct path *path,
1582 struct dentry *dir = nd->path.dentry;
1584 if (!IS_POSIXACL(dir->d_inode))
1585 mode &= ~current->fs->umask;
1586 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1589 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1591 mutex_unlock(&dir->d_inode->i_mutex);
1592 dput(nd->path.dentry);
1593 nd->path.dentry = path->dentry;
1596 /* Don't check for write permission, don't truncate */
1597 return may_open(nd, 0, flag & ~O_TRUNC);
1601 * Note that while the flag value (low two bits) for sys_open means:
1606 * it is changed into
1607 * 00 - no permissions needed
1608 * 01 - read-permission
1609 * 10 - write-permission
1611 * for the internal routines (ie open_namei()/follow_link() etc)
1612 * This is more logical, and also allows the 00 "no perm needed"
1613 * to be used for symlinks (where the permissions are checked
1617 static inline int open_to_namei_flags(int flag)
1619 if ((flag+1) & O_ACCMODE)
1624 static int open_will_write_to_fs(int flag, struct inode *inode)
1627 * We'll never write to the fs underlying
1630 if (special_file(inode->i_mode))
1632 return (flag & O_TRUNC);
1636 * Note that the low bits of the passed in "open_flag"
1637 * are not the same as in the local variable "flag". See
1638 * open_to_namei_flags() for more details.
1640 struct file *do_filp_open(int dfd, const char *pathname,
1641 int open_flag, int mode)
1644 struct nameidata nd;
1645 int acc_mode, error;
1650 int flag = open_to_namei_flags(open_flag);
1652 acc_mode = MAY_OPEN | ACC_MODE(flag);
1654 /* O_TRUNC implies we need access checks for write permissions */
1656 acc_mode |= MAY_WRITE;
1658 /* Allow the LSM permission hook to distinguish append
1659 access from general write access. */
1660 if (flag & O_APPEND)
1661 acc_mode |= MAY_APPEND;
1664 * The simplest case - just a plain lookup.
1666 if (!(flag & O_CREAT)) {
1667 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1670 return ERR_PTR(error);
1675 * Create - we need to know the parent.
1677 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
1679 return ERR_PTR(error);
1682 * We have the parent and last component. First of all, check
1683 * that we are not asked to creat(2) an obvious directory - that
1687 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1691 filp = get_empty_filp();
1694 nd.intent.open.file = filp;
1695 nd.intent.open.flags = flag;
1696 nd.intent.open.create_mode = mode;
1697 dir = nd.path.dentry;
1698 nd.flags &= ~LOOKUP_PARENT;
1699 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1701 nd.flags |= LOOKUP_EXCL;
1702 mutex_lock(&dir->d_inode->i_mutex);
1703 path.dentry = lookup_hash(&nd);
1704 path.mnt = nd.path.mnt;
1707 error = PTR_ERR(path.dentry);
1708 if (IS_ERR(path.dentry)) {
1709 mutex_unlock(&dir->d_inode->i_mutex);
1713 if (IS_ERR(nd.intent.open.file)) {
1714 error = PTR_ERR(nd.intent.open.file);
1715 goto exit_mutex_unlock;
1718 /* Negative dentry, just create the file */
1719 if (!path.dentry->d_inode) {
1721 * This write is needed to ensure that a
1722 * ro->rw transition does not occur between
1723 * the time when the file is created and when
1724 * a permanent write count is taken through
1725 * the 'struct file' in nameidata_to_filp().
1727 error = mnt_want_write(nd.path.mnt);
1729 goto exit_mutex_unlock;
1730 error = __open_namei_create(&nd, &path, flag, mode);
1732 mnt_drop_write(nd.path.mnt);
1735 filp = nameidata_to_filp(&nd, open_flag);
1736 mnt_drop_write(nd.path.mnt);
1741 * It already exists.
1743 mutex_unlock(&dir->d_inode->i_mutex);
1744 audit_inode(pathname, path.dentry);
1750 if (__follow_mount(&path)) {
1752 if (flag & O_NOFOLLOW)
1757 if (!path.dentry->d_inode)
1759 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1762 path_to_nameidata(&path, &nd);
1764 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1769 * 1. may_open() truncates a file
1770 * 2. a rw->ro mount transition occurs
1771 * 3. nameidata_to_filp() fails due to
1773 * That would be inconsistent, and should
1774 * be avoided. Taking this mnt write here
1775 * ensures that (2) can not occur.
1777 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1779 error = mnt_want_write(nd.path.mnt);
1783 error = may_open(&nd, acc_mode, flag);
1786 mnt_drop_write(nd.path.mnt);
1789 filp = nameidata_to_filp(&nd, open_flag);
1791 * It is now safe to drop the mnt write
1792 * because the filp has had a write taken
1796 mnt_drop_write(nd.path.mnt);
1800 mutex_unlock(&dir->d_inode->i_mutex);
1802 path_put_conditional(&path, &nd);
1804 if (!IS_ERR(nd.intent.open.file))
1805 release_open_intent(&nd);
1808 return ERR_PTR(error);
1812 if (flag & O_NOFOLLOW)
1815 * This is subtle. Instead of calling do_follow_link() we do the
1816 * thing by hands. The reason is that this way we have zero link_count
1817 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1818 * After that we have the parent and last component, i.e.
1819 * we are in the same situation as after the first path_walk().
1820 * Well, almost - if the last component is normal we get its copy
1821 * stored in nd->last.name and we will have to putname() it when we
1822 * are done. Procfs-like symlinks just set LAST_BIND.
1824 nd.flags |= LOOKUP_PARENT;
1825 error = security_inode_follow_link(path.dentry, &nd);
1828 error = __do_follow_link(&path, &nd);
1830 /* Does someone understand code flow here? Or it is only
1831 * me so stupid? Anathema to whoever designed this non-sense
1832 * with "intent.open".
1834 release_open_intent(&nd);
1835 return ERR_PTR(error);
1837 nd.flags &= ~LOOKUP_PARENT;
1838 if (nd.last_type == LAST_BIND)
1841 if (nd.last_type != LAST_NORM)
1843 if (nd.last.name[nd.last.len]) {
1844 __putname(nd.last.name);
1849 __putname(nd.last.name);
1852 dir = nd.path.dentry;
1853 mutex_lock(&dir->d_inode->i_mutex);
1854 path.dentry = lookup_hash(&nd);
1855 path.mnt = nd.path.mnt;
1856 __putname(nd.last.name);
1861 * filp_open - open file and return file pointer
1863 * @filename: path to open
1864 * @flags: open flags as per the open(2) second argument
1865 * @mode: mode for the new file if O_CREAT is set, else ignored
1867 * This is the helper to open a file from kernelspace if you really
1868 * have to. But in generally you should not do this, so please move
1869 * along, nothing to see here..
1871 struct file *filp_open(const char *filename, int flags, int mode)
1873 return do_filp_open(AT_FDCWD, filename, flags, mode);
1875 EXPORT_SYMBOL(filp_open);
1878 * lookup_create - lookup a dentry, creating it if it doesn't exist
1879 * @nd: nameidata info
1880 * @is_dir: directory flag
1882 * Simple function to lookup and return a dentry and create it
1883 * if it doesn't exist. Is SMP-safe.
1885 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1887 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1889 struct dentry *dentry = ERR_PTR(-EEXIST);
1891 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1893 * Yucky last component or no last component at all?
1894 * (foo/., foo/.., /////)
1896 if (nd->last_type != LAST_NORM)
1898 nd->flags &= ~LOOKUP_PARENT;
1899 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1900 nd->intent.open.flags = O_EXCL;
1903 * Do the final lookup.
1905 dentry = lookup_hash(nd);
1909 if (dentry->d_inode)
1912 * Special case - lookup gave negative, but... we had foo/bar/
1913 * From the vfs_mknod() POV we just have a negative dentry -
1914 * all is fine. Let's be bastards - you had / on the end, you've
1915 * been asking for (non-existent) directory. -ENOENT for you.
1917 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1919 dentry = ERR_PTR(-ENOENT);
1924 dentry = ERR_PTR(-EEXIST);
1928 EXPORT_SYMBOL_GPL(lookup_create);
1930 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1932 int error = may_create(dir, dentry);
1937 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1940 if (!dir->i_op || !dir->i_op->mknod)
1943 error = devcgroup_inode_mknod(mode, dev);
1947 error = security_inode_mknod(dir, dentry, mode, dev);
1952 error = dir->i_op->mknod(dir, dentry, mode, dev);
1954 fsnotify_create(dir, dentry);
1958 static int may_mknod(mode_t mode)
1960 switch (mode & S_IFMT) {
1966 case 0: /* zero mode translates to S_IFREG */
1975 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1980 struct dentry *dentry;
1981 struct nameidata nd;
1986 error = user_path_parent(dfd, filename, &nd, &tmp);
1990 dentry = lookup_create(&nd, 0);
1991 if (IS_ERR(dentry)) {
1992 error = PTR_ERR(dentry);
1995 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1996 mode &= ~current->fs->umask;
1997 error = may_mknod(mode);
2000 error = mnt_want_write(nd.path.mnt);
2003 error = security_path_mknod(&nd.path, dentry, mode, dev);
2005 goto out_drop_write;
2006 switch (mode & S_IFMT) {
2007 case 0: case S_IFREG:
2008 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2010 case S_IFCHR: case S_IFBLK:
2011 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2012 new_decode_dev(dev));
2014 case S_IFIFO: case S_IFSOCK:
2015 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2019 mnt_drop_write(nd.path.mnt);
2023 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2030 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2032 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2035 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2037 int error = may_create(dir, dentry);
2042 if (!dir->i_op || !dir->i_op->mkdir)
2045 mode &= (S_IRWXUGO|S_ISVTX);
2046 error = security_inode_mkdir(dir, dentry, mode);
2051 error = dir->i_op->mkdir(dir, dentry, mode);
2053 fsnotify_mkdir(dir, dentry);
2057 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2061 struct dentry *dentry;
2062 struct nameidata nd;
2064 error = user_path_parent(dfd, pathname, &nd, &tmp);
2068 dentry = lookup_create(&nd, 1);
2069 error = PTR_ERR(dentry);
2073 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2074 mode &= ~current->fs->umask;
2075 error = mnt_want_write(nd.path.mnt);
2078 error = security_path_mkdir(&nd.path, dentry, mode);
2080 goto out_drop_write;
2081 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2083 mnt_drop_write(nd.path.mnt);
2087 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2094 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2096 return sys_mkdirat(AT_FDCWD, pathname, mode);
2100 * We try to drop the dentry early: we should have
2101 * a usage count of 2 if we're the only user of this
2102 * dentry, and if that is true (possibly after pruning
2103 * the dcache), then we drop the dentry now.
2105 * A low-level filesystem can, if it choses, legally
2108 * if (!d_unhashed(dentry))
2111 * if it cannot handle the case of removing a directory
2112 * that is still in use by something else..
2114 void dentry_unhash(struct dentry *dentry)
2117 shrink_dcache_parent(dentry);
2118 spin_lock(&dcache_lock);
2119 spin_lock(&dentry->d_lock);
2120 if (atomic_read(&dentry->d_count) == 2)
2122 spin_unlock(&dentry->d_lock);
2123 spin_unlock(&dcache_lock);
2126 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2128 int error = may_delete(dir, dentry, 1);
2133 if (!dir->i_op || !dir->i_op->rmdir)
2138 mutex_lock(&dentry->d_inode->i_mutex);
2139 dentry_unhash(dentry);
2140 if (d_mountpoint(dentry))
2143 error = security_inode_rmdir(dir, dentry);
2145 error = dir->i_op->rmdir(dir, dentry);
2147 dentry->d_inode->i_flags |= S_DEAD;
2150 mutex_unlock(&dentry->d_inode->i_mutex);
2159 static long do_rmdir(int dfd, const char __user *pathname)
2163 struct dentry *dentry;
2164 struct nameidata nd;
2166 error = user_path_parent(dfd, pathname, &nd, &name);
2170 switch(nd.last_type) {
2182 nd.flags &= ~LOOKUP_PARENT;
2184 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2185 dentry = lookup_hash(&nd);
2186 error = PTR_ERR(dentry);
2189 error = mnt_want_write(nd.path.mnt);
2192 error = security_path_rmdir(&nd.path, dentry);
2195 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2197 mnt_drop_write(nd.path.mnt);
2201 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2208 asmlinkage long sys_rmdir(const char __user *pathname)
2210 return do_rmdir(AT_FDCWD, pathname);
2213 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2215 int error = may_delete(dir, dentry, 0);
2220 if (!dir->i_op || !dir->i_op->unlink)
2225 mutex_lock(&dentry->d_inode->i_mutex);
2226 if (d_mountpoint(dentry))
2229 error = security_inode_unlink(dir, dentry);
2231 error = dir->i_op->unlink(dir, dentry);
2233 mutex_unlock(&dentry->d_inode->i_mutex);
2235 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2236 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2237 fsnotify_link_count(dentry->d_inode);
2245 * Make sure that the actual truncation of the file will occur outside its
2246 * directory's i_mutex. Truncate can take a long time if there is a lot of
2247 * writeout happening, and we don't want to prevent access to the directory
2248 * while waiting on the I/O.
2250 static long do_unlinkat(int dfd, const char __user *pathname)
2254 struct dentry *dentry;
2255 struct nameidata nd;
2256 struct inode *inode = NULL;
2258 error = user_path_parent(dfd, pathname, &nd, &name);
2263 if (nd.last_type != LAST_NORM)
2266 nd.flags &= ~LOOKUP_PARENT;
2268 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2269 dentry = lookup_hash(&nd);
2270 error = PTR_ERR(dentry);
2271 if (!IS_ERR(dentry)) {
2272 /* Why not before? Because we want correct error value */
2273 if (nd.last.name[nd.last.len])
2275 inode = dentry->d_inode;
2277 atomic_inc(&inode->i_count);
2278 error = mnt_want_write(nd.path.mnt);
2281 error = security_path_unlink(&nd.path, dentry);
2284 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2286 mnt_drop_write(nd.path.mnt);
2290 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2292 iput(inode); /* truncate the inode here */
2299 error = !dentry->d_inode ? -ENOENT :
2300 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2304 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2306 if ((flag & ~AT_REMOVEDIR) != 0)
2309 if (flag & AT_REMOVEDIR)
2310 return do_rmdir(dfd, pathname);
2312 return do_unlinkat(dfd, pathname);
2315 asmlinkage long sys_unlink(const char __user *pathname)
2317 return do_unlinkat(AT_FDCWD, pathname);
2320 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2322 int error = may_create(dir, dentry);
2327 if (!dir->i_op || !dir->i_op->symlink)
2330 error = security_inode_symlink(dir, dentry, oldname);
2335 error = dir->i_op->symlink(dir, dentry, oldname);
2337 fsnotify_create(dir, dentry);
2341 asmlinkage long sys_symlinkat(const char __user *oldname,
2342 int newdfd, const char __user *newname)
2347 struct dentry *dentry;
2348 struct nameidata nd;
2350 from = getname(oldname);
2352 return PTR_ERR(from);
2354 error = user_path_parent(newdfd, newname, &nd, &to);
2358 dentry = lookup_create(&nd, 0);
2359 error = PTR_ERR(dentry);
2363 error = mnt_want_write(nd.path.mnt);
2366 error = security_path_symlink(&nd.path, dentry, from);
2368 goto out_drop_write;
2369 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2371 mnt_drop_write(nd.path.mnt);
2375 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2383 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2385 return sys_symlinkat(oldname, AT_FDCWD, newname);
2388 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2390 struct inode *inode = old_dentry->d_inode;
2396 error = may_create(dir, new_dentry);
2400 if (dir->i_sb != inode->i_sb)
2404 * A link to an append-only or immutable file cannot be created.
2406 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2408 if (!dir->i_op || !dir->i_op->link)
2410 if (S_ISDIR(inode->i_mode))
2413 error = security_inode_link(old_dentry, dir, new_dentry);
2417 mutex_lock(&inode->i_mutex);
2419 error = dir->i_op->link(old_dentry, dir, new_dentry);
2420 mutex_unlock(&inode->i_mutex);
2422 fsnotify_link(dir, inode, new_dentry);
2427 * Hardlinks are often used in delicate situations. We avoid
2428 * security-related surprises by not following symlinks on the
2431 * We don't follow them on the oldname either to be compatible
2432 * with linux 2.0, and to avoid hard-linking to directories
2433 * and other special files. --ADM
2435 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2436 int newdfd, const char __user *newname,
2439 struct dentry *new_dentry;
2440 struct nameidata nd;
2441 struct path old_path;
2445 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2448 error = user_path_at(olddfd, oldname,
2449 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2454 error = user_path_parent(newdfd, newname, &nd, &to);
2458 if (old_path.mnt != nd.path.mnt)
2460 new_dentry = lookup_create(&nd, 0);
2461 error = PTR_ERR(new_dentry);
2462 if (IS_ERR(new_dentry))
2464 error = mnt_want_write(nd.path.mnt);
2467 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2469 goto out_drop_write;
2470 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2472 mnt_drop_write(nd.path.mnt);
2476 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2481 path_put(&old_path);
2486 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2488 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2492 * The worst of all namespace operations - renaming directory. "Perverted"
2493 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2495 * a) we can get into loop creation. Check is done in is_subdir().
2496 * b) race potential - two innocent renames can create a loop together.
2497 * That's where 4.4 screws up. Current fix: serialization on
2498 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2500 * c) we have to lock _three_ objects - parents and victim (if it exists).
2501 * And that - after we got ->i_mutex on parents (until then we don't know
2502 * whether the target exists). Solution: try to be smart with locking
2503 * order for inodes. We rely on the fact that tree topology may change
2504 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2505 * move will be locked. Thus we can rank directories by the tree
2506 * (ancestors first) and rank all non-directories after them.
2507 * That works since everybody except rename does "lock parent, lookup,
2508 * lock child" and rename is under ->s_vfs_rename_mutex.
2509 * HOWEVER, it relies on the assumption that any object with ->lookup()
2510 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2511 * we'd better make sure that there's no link(2) for them.
2512 * d) some filesystems don't support opened-but-unlinked directories,
2513 * either because of layout or because they are not ready to deal with
2514 * all cases correctly. The latter will be fixed (taking this sort of
2515 * stuff into VFS), but the former is not going away. Solution: the same
2516 * trick as in rmdir().
2517 * e) conversion from fhandle to dentry may come in the wrong moment - when
2518 * we are removing the target. Solution: we will have to grab ->i_mutex
2519 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2520 * ->i_mutex on parents, which works but leads to some truely excessive
2523 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2524 struct inode *new_dir, struct dentry *new_dentry)
2527 struct inode *target;
2530 * If we are going to change the parent - check write permissions,
2531 * we'll need to flip '..'.
2533 if (new_dir != old_dir) {
2534 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2539 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2543 target = new_dentry->d_inode;
2545 mutex_lock(&target->i_mutex);
2546 dentry_unhash(new_dentry);
2548 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2551 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2554 target->i_flags |= S_DEAD;
2555 mutex_unlock(&target->i_mutex);
2556 if (d_unhashed(new_dentry))
2557 d_rehash(new_dentry);
2561 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2562 d_move(old_dentry,new_dentry);
2566 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2567 struct inode *new_dir, struct dentry *new_dentry)
2569 struct inode *target;
2572 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2577 target = new_dentry->d_inode;
2579 mutex_lock(&target->i_mutex);
2580 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2583 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2585 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2586 d_move(old_dentry, new_dentry);
2589 mutex_unlock(&target->i_mutex);
2594 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2595 struct inode *new_dir, struct dentry *new_dentry)
2598 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2599 const char *old_name;
2601 if (old_dentry->d_inode == new_dentry->d_inode)
2604 error = may_delete(old_dir, old_dentry, is_dir);
2608 if (!new_dentry->d_inode)
2609 error = may_create(new_dir, new_dentry);
2611 error = may_delete(new_dir, new_dentry, is_dir);
2615 if (!old_dir->i_op || !old_dir->i_op->rename)
2618 DQUOT_INIT(old_dir);
2619 DQUOT_INIT(new_dir);
2621 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2624 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2626 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2628 const char *new_name = old_dentry->d_name.name;
2629 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2630 new_dentry->d_inode, old_dentry);
2632 fsnotify_oldname_free(old_name);
2637 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2638 int newdfd, const char __user *newname)
2640 struct dentry *old_dir, *new_dir;
2641 struct dentry *old_dentry, *new_dentry;
2642 struct dentry *trap;
2643 struct nameidata oldnd, newnd;
2648 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2652 error = user_path_parent(newdfd, newname, &newnd, &to);
2657 if (oldnd.path.mnt != newnd.path.mnt)
2660 old_dir = oldnd.path.dentry;
2662 if (oldnd.last_type != LAST_NORM)
2665 new_dir = newnd.path.dentry;
2666 if (newnd.last_type != LAST_NORM)
2669 oldnd.flags &= ~LOOKUP_PARENT;
2670 newnd.flags &= ~LOOKUP_PARENT;
2671 newnd.flags |= LOOKUP_RENAME_TARGET;
2673 trap = lock_rename(new_dir, old_dir);
2675 old_dentry = lookup_hash(&oldnd);
2676 error = PTR_ERR(old_dentry);
2677 if (IS_ERR(old_dentry))
2679 /* source must exist */
2681 if (!old_dentry->d_inode)
2683 /* unless the source is a directory trailing slashes give -ENOTDIR */
2684 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2686 if (oldnd.last.name[oldnd.last.len])
2688 if (newnd.last.name[newnd.last.len])
2691 /* source should not be ancestor of target */
2693 if (old_dentry == trap)
2695 new_dentry = lookup_hash(&newnd);
2696 error = PTR_ERR(new_dentry);
2697 if (IS_ERR(new_dentry))
2699 /* target should not be an ancestor of source */
2701 if (new_dentry == trap)
2704 error = mnt_want_write(oldnd.path.mnt);
2707 error = security_path_rename(&oldnd.path, old_dentry,
2708 &newnd.path, new_dentry);
2711 error = vfs_rename(old_dir->d_inode, old_dentry,
2712 new_dir->d_inode, new_dentry);
2714 mnt_drop_write(oldnd.path.mnt);
2720 unlock_rename(new_dir, old_dir);
2722 path_put(&newnd.path);
2725 path_put(&oldnd.path);
2731 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2733 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2736 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2740 len = PTR_ERR(link);
2745 if (len > (unsigned) buflen)
2747 if (copy_to_user(buffer, link, len))
2754 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2755 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2756 * using) it for any given inode is up to filesystem.
2758 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2760 struct nameidata nd;
2765 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2767 return PTR_ERR(cookie);
2769 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2770 if (dentry->d_inode->i_op->put_link)
2771 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2775 int vfs_follow_link(struct nameidata *nd, const char *link)
2777 return __vfs_follow_link(nd, link);
2780 /* get the link contents into pagecache */
2781 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2785 struct address_space *mapping = dentry->d_inode->i_mapping;
2786 page = read_mapping_page(mapping, 0, NULL);
2791 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2795 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2797 struct page *page = NULL;
2798 char *s = page_getlink(dentry, &page);
2799 int res = vfs_readlink(dentry,buffer,buflen,s);
2802 page_cache_release(page);
2807 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2809 struct page *page = NULL;
2810 nd_set_link(nd, page_getlink(dentry, &page));
2814 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2816 struct page *page = cookie;
2820 page_cache_release(page);
2824 int __page_symlink(struct inode *inode, const char *symname, int len,
2827 struct address_space *mapping = inode->i_mapping;
2834 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2835 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2839 kaddr = kmap_atomic(page, KM_USER0);
2840 memcpy(kaddr, symname, len-1);
2841 kunmap_atomic(kaddr, KM_USER0);
2843 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2850 mark_inode_dirty(inode);
2856 int page_symlink(struct inode *inode, const char *symname, int len)
2858 return __page_symlink(inode, symname, len,
2859 mapping_gfp_mask(inode->i_mapping));
2862 const struct inode_operations page_symlink_inode_operations = {
2863 .readlink = generic_readlink,
2864 .follow_link = page_follow_link_light,
2865 .put_link = page_put_link,
2868 EXPORT_SYMBOL(user_path_at);
2869 EXPORT_SYMBOL(follow_down);
2870 EXPORT_SYMBOL(follow_up);
2871 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2872 EXPORT_SYMBOL(getname);
2873 EXPORT_SYMBOL(lock_rename);
2874 EXPORT_SYMBOL(lookup_one_len);
2875 EXPORT_SYMBOL(page_follow_link_light);
2876 EXPORT_SYMBOL(page_put_link);
2877 EXPORT_SYMBOL(page_readlink);
2878 EXPORT_SYMBOL(__page_symlink);
2879 EXPORT_SYMBOL(page_symlink);
2880 EXPORT_SYMBOL(page_symlink_inode_operations);
2881 EXPORT_SYMBOL(path_lookup);
2882 EXPORT_SYMBOL(kern_path);
2883 EXPORT_SYMBOL(vfs_path_lookup);
2884 EXPORT_SYMBOL(inode_permission);
2885 EXPORT_SYMBOL(vfs_permission);
2886 EXPORT_SYMBOL(file_permission);
2887 EXPORT_SYMBOL(unlock_rename);
2888 EXPORT_SYMBOL(vfs_create);
2889 EXPORT_SYMBOL(vfs_follow_link);
2890 EXPORT_SYMBOL(vfs_link);
2891 EXPORT_SYMBOL(vfs_mkdir);
2892 EXPORT_SYMBOL(vfs_mknod);
2893 EXPORT_SYMBOL(generic_permission);
2894 EXPORT_SYMBOL(vfs_readlink);
2895 EXPORT_SYMBOL(vfs_rename);
2896 EXPORT_SYMBOL(vfs_rmdir);
2897 EXPORT_SYMBOL(vfs_symlink);
2898 EXPORT_SYMBOL(vfs_unlink);
2899 EXPORT_SYMBOL(dentry_unhash);
2900 EXPORT_SYMBOL(generic_readlink);