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))
230 * inode_permission - check for access rights to a given inode
231 * @inode: inode to check permission on
232 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
234 * Used to check for read/write/execute permissions on an inode.
235 * We use "fsuid" for this, letting us set arbitrary permissions
236 * for filesystem access without changing the "normal" uids which
237 * are used for other things.
239 int inode_permission(struct inode *inode, int mask)
243 if (mask & MAY_WRITE) {
244 umode_t mode = inode->i_mode;
247 * Nobody gets write access to a read-only fs.
249 if (IS_RDONLY(inode) &&
250 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
254 * Nobody gets write access to an immutable file.
256 if (IS_IMMUTABLE(inode))
260 if (inode->i_op->permission)
261 retval = inode->i_op->permission(inode, mask);
263 retval = generic_permission(inode, mask, NULL);
268 retval = devcgroup_inode_permission(inode, mask);
272 return security_inode_permission(inode,
273 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
277 * file_permission - check for additional access rights to a given file
278 * @file: file to check access rights for
279 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
281 * Used to check for read/write/execute permissions on an already opened
285 * Do not use this function in new code. All access checks should
286 * be done using inode_permission().
288 int file_permission(struct file *file, int mask)
290 return inode_permission(file->f_path.dentry->d_inode, mask);
294 * get_write_access() gets write permission for a file.
295 * put_write_access() releases this write permission.
296 * This is used for regular files.
297 * We cannot support write (and maybe mmap read-write shared) accesses and
298 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
299 * can have the following values:
300 * 0: no writers, no VM_DENYWRITE mappings
301 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
302 * > 0: (i_writecount) users are writing to the file.
304 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
305 * except for the cases where we don't hold i_writecount yet. Then we need to
306 * use {get,deny}_write_access() - these functions check the sign and refuse
307 * to do the change if sign is wrong. Exclusion between them is provided by
308 * the inode->i_lock spinlock.
311 int get_write_access(struct inode * inode)
313 spin_lock(&inode->i_lock);
314 if (atomic_read(&inode->i_writecount) < 0) {
315 spin_unlock(&inode->i_lock);
318 atomic_inc(&inode->i_writecount);
319 spin_unlock(&inode->i_lock);
324 int deny_write_access(struct file * file)
326 struct inode *inode = file->f_path.dentry->d_inode;
328 spin_lock(&inode->i_lock);
329 if (atomic_read(&inode->i_writecount) > 0) {
330 spin_unlock(&inode->i_lock);
333 atomic_dec(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
340 * path_get - get a reference to a path
341 * @path: path to get the reference to
343 * Given a path increment the reference count to the dentry and the vfsmount.
345 void path_get(struct path *path)
350 EXPORT_SYMBOL(path_get);
353 * path_put - put a reference to a path
354 * @path: path to put the reference to
356 * Given a path decrement the reference count to the dentry and the vfsmount.
358 void path_put(struct path *path)
363 EXPORT_SYMBOL(path_put);
366 * release_open_intent - free up open intent resources
367 * @nd: pointer to nameidata
369 void release_open_intent(struct nameidata *nd)
371 if (nd->intent.open.file->f_path.dentry == NULL)
372 put_filp(nd->intent.open.file);
374 fput(nd->intent.open.file);
377 static inline struct dentry *
378 do_revalidate(struct dentry *dentry, struct nameidata *nd)
380 int status = dentry->d_op->d_revalidate(dentry, nd);
381 if (unlikely(status <= 0)) {
383 * The dentry failed validation.
384 * If d_revalidate returned 0 attempt to invalidate
385 * the dentry otherwise d_revalidate is asking us
386 * to return a fail status.
389 if (!d_invalidate(dentry)) {
395 dentry = ERR_PTR(status);
402 * Internal lookup() using the new generic dcache.
405 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
407 struct dentry * dentry = __d_lookup(parent, name);
409 /* lockess __d_lookup may fail due to concurrent d_move()
410 * in some unrelated directory, so try with d_lookup
413 dentry = d_lookup(parent, name);
415 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
416 dentry = do_revalidate(dentry, nd);
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
431 static int exec_permission_lite(struct inode *inode)
433 umode_t mode = inode->i_mode;
435 if (inode->i_op->permission)
438 if (current_fsuid() == inode->i_uid)
440 else if (in_group_p(inode->i_gid))
446 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
449 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
452 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
457 return security_inode_permission(inode, MAY_EXEC);
461 * This is called when everything else fails, and we actually have
462 * to go to the low-level filesystem to find out what we should do..
464 * We get the directory semaphore, and after getting that we also
465 * make sure that nobody added the entry to the dcache in the meantime..
468 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
470 struct dentry * result;
471 struct inode *dir = parent->d_inode;
473 mutex_lock(&dir->i_mutex);
475 * First re-do the cached lookup just in case it was created
476 * while we waited for the directory semaphore..
478 * FIXME! This could use version numbering or similar to
479 * avoid unnecessary cache lookups.
481 * The "dcache_lock" is purely to protect the RCU list walker
482 * from concurrent renames at this point (we mustn't get false
483 * negatives from the RCU list walk here, unlike the optimistic
486 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
488 result = d_lookup(parent, name);
490 struct dentry *dentry;
492 /* Don't create child dentry for a dead directory. */
493 result = ERR_PTR(-ENOENT);
497 dentry = d_alloc(parent, name);
498 result = ERR_PTR(-ENOMEM);
500 result = dir->i_op->lookup(dir, dentry, nd);
507 mutex_unlock(&dir->i_mutex);
512 * Uhhuh! Nasty case: the cache was re-populated while
513 * we waited on the semaphore. Need to revalidate.
515 mutex_unlock(&dir->i_mutex);
516 if (result->d_op && result->d_op->d_revalidate) {
517 result = do_revalidate(result, nd);
519 result = ERR_PTR(-ENOENT);
525 * Wrapper to retry pathname resolution whenever the underlying
526 * file system returns an ESTALE.
528 * Retry the whole path once, forcing real lookup requests
529 * instead of relying on the dcache.
531 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
533 struct path save = nd->path;
536 /* make sure the stuff we saved doesn't go away */
539 result = __link_path_walk(name, nd);
540 if (result == -ESTALE) {
541 /* nd->path had been dropped */
544 nd->flags |= LOOKUP_REVAL;
545 result = __link_path_walk(name, nd);
553 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
561 struct fs_struct *fs = current->fs;
565 read_lock(&fs->lock);
568 read_unlock(&fs->lock);
571 res = link_path_walk(link, nd);
572 if (nd->depth || res || nd->last_type!=LAST_NORM)
575 * If it is an iterative symlinks resolution in open_namei() we
576 * have to copy the last component. And all that crap because of
577 * bloody create() on broken symlinks. Furrfu...
580 if (unlikely(!name)) {
584 strcpy(name, nd->last.name);
585 nd->last.name = name;
589 return PTR_ERR(link);
592 static void path_put_conditional(struct path *path, struct nameidata *nd)
595 if (path->mnt != nd->path.mnt)
599 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
601 dput(nd->path.dentry);
602 if (nd->path.mnt != path->mnt)
603 mntput(nd->path.mnt);
604 nd->path.mnt = path->mnt;
605 nd->path.dentry = path->dentry;
608 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
612 struct dentry *dentry = path->dentry;
614 touch_atime(path->mnt, dentry);
615 nd_set_link(nd, NULL);
617 if (path->mnt != nd->path.mnt) {
618 path_to_nameidata(path, nd);
622 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
623 error = PTR_ERR(cookie);
624 if (!IS_ERR(cookie)) {
625 char *s = nd_get_link(nd);
628 error = __vfs_follow_link(nd, s);
629 if (dentry->d_inode->i_op->put_link)
630 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
638 * This limits recursive symlink follows to 8, while
639 * limiting consecutive symlinks to 40.
641 * Without that kind of total limit, nasty chains of consecutive
642 * symlinks can cause almost arbitrarily long lookups.
644 static inline int do_follow_link(struct path *path, struct nameidata *nd)
647 if (current->link_count >= MAX_NESTED_LINKS)
649 if (current->total_link_count >= 40)
651 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
653 err = security_inode_follow_link(path->dentry, nd);
656 current->link_count++;
657 current->total_link_count++;
659 err = __do_follow_link(path, nd);
660 current->link_count--;
664 path_put_conditional(path, nd);
669 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
671 struct vfsmount *parent;
672 struct dentry *mountpoint;
673 spin_lock(&vfsmount_lock);
674 parent=(*mnt)->mnt_parent;
675 if (parent == *mnt) {
676 spin_unlock(&vfsmount_lock);
680 mountpoint=dget((*mnt)->mnt_mountpoint);
681 spin_unlock(&vfsmount_lock);
683 *dentry = mountpoint;
689 /* no need for dcache_lock, as serialization is taken care in
692 static int __follow_mount(struct path *path)
695 while (d_mountpoint(path->dentry)) {
696 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
703 path->dentry = dget(mounted->mnt_root);
709 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
711 while (d_mountpoint(*dentry)) {
712 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
718 *dentry = dget(mounted->mnt_root);
722 /* no need for dcache_lock, as serialization is taken care in
725 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
727 struct vfsmount *mounted;
729 mounted = lookup_mnt(*mnt, *dentry);
734 *dentry = dget(mounted->mnt_root);
740 static __always_inline void follow_dotdot(struct nameidata *nd)
742 struct fs_struct *fs = current->fs;
745 struct vfsmount *parent;
746 struct dentry *old = nd->path.dentry;
748 read_lock(&fs->lock);
749 if (nd->path.dentry == fs->root.dentry &&
750 nd->path.mnt == fs->root.mnt) {
751 read_unlock(&fs->lock);
754 read_unlock(&fs->lock);
755 spin_lock(&dcache_lock);
756 if (nd->path.dentry != nd->path.mnt->mnt_root) {
757 nd->path.dentry = dget(nd->path.dentry->d_parent);
758 spin_unlock(&dcache_lock);
762 spin_unlock(&dcache_lock);
763 spin_lock(&vfsmount_lock);
764 parent = nd->path.mnt->mnt_parent;
765 if (parent == nd->path.mnt) {
766 spin_unlock(&vfsmount_lock);
770 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
771 spin_unlock(&vfsmount_lock);
773 mntput(nd->path.mnt);
774 nd->path.mnt = parent;
776 follow_mount(&nd->path.mnt, &nd->path.dentry);
780 * It's more convoluted than I'd like it to be, but... it's still fairly
781 * small and for now I'd prefer to have fast path as straight as possible.
782 * It _is_ time-critical.
784 static int do_lookup(struct nameidata *nd, struct qstr *name,
787 struct vfsmount *mnt = nd->path.mnt;
788 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
792 if (dentry->d_op && dentry->d_op->d_revalidate)
793 goto need_revalidate;
796 path->dentry = dentry;
797 __follow_mount(path);
801 dentry = real_lookup(nd->path.dentry, name, nd);
807 dentry = do_revalidate(dentry, nd);
815 return PTR_ERR(dentry);
820 * This is the basic name resolution function, turning a pathname into
821 * the final dentry. We expect 'base' to be positive and a directory.
823 * Returns 0 and nd will have valid dentry and mnt on success.
824 * Returns error and drops reference to input namei data on failure.
826 static int __link_path_walk(const char *name, struct nameidata *nd)
831 unsigned int lookup_flags = nd->flags;
838 inode = nd->path.dentry->d_inode;
840 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
842 /* At this point we know we have a real path component. */
848 nd->flags |= LOOKUP_CONTINUE;
849 err = exec_permission_lite(inode);
851 err = inode_permission(nd->path.dentry->d_inode,
857 c = *(const unsigned char *)name;
859 hash = init_name_hash();
862 hash = partial_name_hash(c, hash);
863 c = *(const unsigned char *)name;
864 } while (c && (c != '/'));
865 this.len = name - (const char *) this.name;
866 this.hash = end_name_hash(hash);
868 /* remove trailing slashes? */
871 while (*++name == '/');
873 goto last_with_slashes;
876 * "." and ".." are special - ".." especially so because it has
877 * to be able to know about the current root directory and
878 * parent relationships.
880 if (this.name[0] == '.') switch (this.len) {
884 if (this.name[1] != '.')
887 inode = nd->path.dentry->d_inode;
893 * See if the low-level filesystem might want
894 * to use its own hash..
896 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
897 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
902 /* This does the actual lookups.. */
903 err = do_lookup(nd, &this, &next);
908 inode = next.dentry->d_inode;
912 if (inode->i_op->follow_link) {
913 err = do_follow_link(&next, nd);
917 inode = nd->path.dentry->d_inode;
921 path_to_nameidata(&next, nd);
923 if (!inode->i_op->lookup)
926 /* here ends the main loop */
929 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
931 /* Clear LOOKUP_CONTINUE iff it was previously unset */
932 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
933 if (lookup_flags & LOOKUP_PARENT)
935 if (this.name[0] == '.') switch (this.len) {
939 if (this.name[1] != '.')
942 inode = nd->path.dentry->d_inode;
947 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
948 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
953 err = do_lookup(nd, &this, &next);
956 inode = next.dentry->d_inode;
957 if ((lookup_flags & LOOKUP_FOLLOW)
958 && inode && inode->i_op->follow_link) {
959 err = do_follow_link(&next, nd);
962 inode = nd->path.dentry->d_inode;
964 path_to_nameidata(&next, nd);
968 if (lookup_flags & LOOKUP_DIRECTORY) {
970 if (!inode->i_op->lookup)
976 nd->last_type = LAST_NORM;
977 if (this.name[0] != '.')
980 nd->last_type = LAST_DOT;
981 else if (this.len == 2 && this.name[1] == '.')
982 nd->last_type = LAST_DOTDOT;
987 * We bypassed the ordinary revalidation routines.
988 * We may need to check the cached dentry for staleness.
990 if (nd->path.dentry && nd->path.dentry->d_sb &&
991 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
993 /* Note: we do not d_invalidate() */
994 if (!nd->path.dentry->d_op->d_revalidate(
995 nd->path.dentry, nd))
1001 path_put_conditional(&next, nd);
1004 path_put(&nd->path);
1009 static int path_walk(const char *name, struct nameidata *nd)
1011 current->total_link_count = 0;
1012 return link_path_walk(name, nd);
1015 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1016 static int do_path_lookup(int dfd, const char *name,
1017 unsigned int flags, struct nameidata *nd)
1022 struct fs_struct *fs = current->fs;
1024 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1029 read_lock(&fs->lock);
1030 nd->path = fs->root;
1031 path_get(&fs->root);
1032 read_unlock(&fs->lock);
1033 } else if (dfd == AT_FDCWD) {
1034 read_lock(&fs->lock);
1037 read_unlock(&fs->lock);
1039 struct dentry *dentry;
1041 file = fget_light(dfd, &fput_needed);
1046 dentry = file->f_path.dentry;
1049 if (!S_ISDIR(dentry->d_inode->i_mode))
1052 retval = file_permission(file, MAY_EXEC);
1056 nd->path = file->f_path;
1057 path_get(&file->f_path);
1059 fput_light(file, fput_needed);
1062 retval = path_walk(name, nd);
1063 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1064 nd->path.dentry->d_inode))
1065 audit_inode(name, nd->path.dentry);
1070 fput_light(file, fput_needed);
1074 int path_lookup(const char *name, unsigned int flags,
1075 struct nameidata *nd)
1077 return do_path_lookup(AT_FDCWD, name, flags, nd);
1080 int kern_path(const char *name, unsigned int flags, struct path *path)
1082 struct nameidata nd;
1083 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1090 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1091 * @dentry: pointer to dentry of the base directory
1092 * @mnt: pointer to vfs mount of the base directory
1093 * @name: pointer to file name
1094 * @flags: lookup flags
1095 * @nd: pointer to nameidata
1097 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1098 const char *name, unsigned int flags,
1099 struct nameidata *nd)
1103 /* same as do_path_lookup */
1104 nd->last_type = LAST_ROOT;
1108 nd->path.dentry = dentry;
1110 path_get(&nd->path);
1112 retval = path_walk(name, nd);
1113 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1114 nd->path.dentry->d_inode))
1115 audit_inode(name, nd->path.dentry);
1122 * path_lookup_open - lookup a file path with open intent
1123 * @dfd: the directory to use as base, or AT_FDCWD
1124 * @name: pointer to file name
1125 * @lookup_flags: lookup intent flags
1126 * @nd: pointer to nameidata
1127 * @open_flags: open intent flags
1129 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1130 struct nameidata *nd, int open_flags)
1132 struct file *filp = get_empty_filp();
1137 nd->intent.open.file = filp;
1138 nd->intent.open.flags = open_flags;
1139 nd->intent.open.create_mode = 0;
1140 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1141 if (IS_ERR(nd->intent.open.file)) {
1143 err = PTR_ERR(nd->intent.open.file);
1144 path_put(&nd->path);
1146 } else if (err != 0)
1147 release_open_intent(nd);
1151 static struct dentry *__lookup_hash(struct qstr *name,
1152 struct dentry *base, struct nameidata *nd)
1154 struct dentry *dentry;
1155 struct inode *inode;
1158 inode = base->d_inode;
1161 * See if the low-level filesystem might want
1162 * to use its own hash..
1164 if (base->d_op && base->d_op->d_hash) {
1165 err = base->d_op->d_hash(base, name);
1166 dentry = ERR_PTR(err);
1171 dentry = cached_lookup(base, name, nd);
1175 /* Don't create child dentry for a dead directory. */
1176 dentry = ERR_PTR(-ENOENT);
1177 if (IS_DEADDIR(inode))
1180 new = d_alloc(base, name);
1181 dentry = ERR_PTR(-ENOMEM);
1184 dentry = inode->i_op->lookup(inode, new, nd);
1195 * Restricted form of lookup. Doesn't follow links, single-component only,
1196 * needs parent already locked. Doesn't follow mounts.
1199 static struct dentry *lookup_hash(struct nameidata *nd)
1203 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1205 return ERR_PTR(err);
1206 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1209 static int __lookup_one_len(const char *name, struct qstr *this,
1210 struct dentry *base, int len)
1220 hash = init_name_hash();
1222 c = *(const unsigned char *)name++;
1223 if (c == '/' || c == '\0')
1225 hash = partial_name_hash(c, hash);
1227 this->hash = end_name_hash(hash);
1232 * lookup_one_len - filesystem helper to lookup single pathname component
1233 * @name: pathname component to lookup
1234 * @base: base directory to lookup from
1235 * @len: maximum length @len should be interpreted to
1237 * Note that this routine is purely a helper for filesystem usage and should
1238 * not be called by generic code. Also note that by using this function the
1239 * nameidata argument is passed to the filesystem methods and a filesystem
1240 * using this helper needs to be prepared for that.
1242 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1247 err = __lookup_one_len(name, &this, base, len);
1249 return ERR_PTR(err);
1251 err = inode_permission(base->d_inode, MAY_EXEC);
1253 return ERR_PTR(err);
1254 return __lookup_hash(&this, base, NULL);
1258 * lookup_one_noperm - bad hack for sysfs
1259 * @name: pathname component to lookup
1260 * @base: base directory to lookup from
1262 * This is a variant of lookup_one_len that doesn't perform any permission
1263 * checks. It's a horrible hack to work around the braindead sysfs
1264 * architecture and should not be used anywhere else.
1266 * DON'T USE THIS FUNCTION EVER, thanks.
1268 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1273 err = __lookup_one_len(name, &this, base, strlen(name));
1275 return ERR_PTR(err);
1276 return __lookup_hash(&this, base, NULL);
1279 int user_path_at(int dfd, const char __user *name, unsigned flags,
1282 struct nameidata nd;
1283 char *tmp = getname(name);
1284 int err = PTR_ERR(tmp);
1287 BUG_ON(flags & LOOKUP_PARENT);
1289 err = do_path_lookup(dfd, tmp, flags, &nd);
1297 static int user_path_parent(int dfd, const char __user *path,
1298 struct nameidata *nd, char **name)
1300 char *s = getname(path);
1306 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1316 * It's inline, so penalty for filesystems that don't use sticky bit is
1319 static inline int check_sticky(struct inode *dir, struct inode *inode)
1321 uid_t fsuid = current_fsuid();
1323 if (!(dir->i_mode & S_ISVTX))
1325 if (inode->i_uid == fsuid)
1327 if (dir->i_uid == fsuid)
1329 return !capable(CAP_FOWNER);
1333 * Check whether we can remove a link victim from directory dir, check
1334 * whether the type of victim is right.
1335 * 1. We can't do it if dir is read-only (done in permission())
1336 * 2. We should have write and exec permissions on dir
1337 * 3. We can't remove anything from append-only dir
1338 * 4. We can't do anything with immutable dir (done in permission())
1339 * 5. If the sticky bit on dir is set we should either
1340 * a. be owner of dir, or
1341 * b. be owner of victim, or
1342 * c. have CAP_FOWNER capability
1343 * 6. If the victim is append-only or immutable we can't do antyhing with
1344 * links pointing to it.
1345 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1346 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1347 * 9. We can't remove a root or mountpoint.
1348 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1349 * nfs_async_unlink().
1351 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1355 if (!victim->d_inode)
1358 BUG_ON(victim->d_parent->d_inode != dir);
1359 audit_inode_child(victim->d_name.name, victim, dir);
1361 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1366 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1367 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1370 if (!S_ISDIR(victim->d_inode->i_mode))
1372 if (IS_ROOT(victim))
1374 } else if (S_ISDIR(victim->d_inode->i_mode))
1376 if (IS_DEADDIR(dir))
1378 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1383 /* Check whether we can create an object with dentry child in directory
1385 * 1. We can't do it if child already exists (open has special treatment for
1386 * this case, but since we are inlined it's OK)
1387 * 2. We can't do it if dir is read-only (done in permission())
1388 * 3. We should have write and exec permissions on dir
1389 * 4. We can't do it if dir is immutable (done in permission())
1391 static inline int may_create(struct inode *dir, struct dentry *child)
1395 if (IS_DEADDIR(dir))
1397 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1401 * O_DIRECTORY translates into forcing a directory lookup.
1403 static inline int lookup_flags(unsigned int f)
1405 unsigned long retval = LOOKUP_FOLLOW;
1408 retval &= ~LOOKUP_FOLLOW;
1410 if (f & O_DIRECTORY)
1411 retval |= LOOKUP_DIRECTORY;
1417 * p1 and p2 should be directories on the same fs.
1419 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1424 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1428 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1430 p = d_ancestor(p2, p1);
1432 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1433 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1437 p = d_ancestor(p1, p2);
1439 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1440 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1444 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1445 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1449 void unlock_rename(struct dentry *p1, struct dentry *p2)
1451 mutex_unlock(&p1->d_inode->i_mutex);
1453 mutex_unlock(&p2->d_inode->i_mutex);
1454 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1458 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1459 struct nameidata *nd)
1461 int error = may_create(dir, dentry);
1466 if (!dir->i_op->create)
1467 return -EACCES; /* shouldn't it be ENOSYS? */
1470 error = security_inode_create(dir, dentry, mode);
1474 error = dir->i_op->create(dir, dentry, mode, nd);
1476 fsnotify_create(dir, dentry);
1480 int may_open(struct path *path, int acc_mode, int flag)
1482 struct dentry *dentry = path->dentry;
1483 struct inode *inode = dentry->d_inode;
1489 if (S_ISLNK(inode->i_mode))
1492 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1496 * FIFO's, sockets and device files are special: they don't
1497 * actually live on the filesystem itself, and as such you
1498 * can write to them even if the filesystem is read-only.
1500 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1502 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1503 if (path->mnt->mnt_flags & MNT_NODEV)
1509 error = inode_permission(inode, acc_mode);
1513 * An append-only file must be opened in append mode for writing.
1515 if (IS_APPEND(inode)) {
1516 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1522 /* O_NOATIME can only be set by the owner or superuser */
1523 if (flag & O_NOATIME)
1524 if (!is_owner_or_cap(inode))
1528 * Ensure there are no outstanding leases on the file.
1530 error = break_lease(inode, flag);
1534 if (flag & O_TRUNC) {
1535 error = get_write_access(inode);
1540 * Refuse to truncate files with mandatory locks held on them.
1542 error = locks_verify_locked(inode);
1544 error = security_path_truncate(path, 0,
1545 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1549 error = do_truncate(dentry, 0,
1550 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1553 put_write_access(inode);
1557 if (flag & FMODE_WRITE)
1564 * Be careful about ever adding any more callers of this
1565 * function. Its flags must be in the namei format, not
1566 * what get passed to sys_open().
1568 static int __open_namei_create(struct nameidata *nd, struct path *path,
1572 struct dentry *dir = nd->path.dentry;
1574 if (!IS_POSIXACL(dir->d_inode))
1575 mode &= ~current->fs->umask;
1576 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1579 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1581 mutex_unlock(&dir->d_inode->i_mutex);
1582 dput(nd->path.dentry);
1583 nd->path.dentry = path->dentry;
1586 /* Don't check for write permission, don't truncate */
1587 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1591 * Note that while the flag value (low two bits) for sys_open means:
1596 * it is changed into
1597 * 00 - no permissions needed
1598 * 01 - read-permission
1599 * 10 - write-permission
1601 * for the internal routines (ie open_namei()/follow_link() etc)
1602 * This is more logical, and also allows the 00 "no perm needed"
1603 * to be used for symlinks (where the permissions are checked
1607 static inline int open_to_namei_flags(int flag)
1609 if ((flag+1) & O_ACCMODE)
1614 static int open_will_write_to_fs(int flag, struct inode *inode)
1617 * We'll never write to the fs underlying
1620 if (special_file(inode->i_mode))
1622 return (flag & O_TRUNC);
1626 * Note that the low bits of the passed in "open_flag"
1627 * are not the same as in the local variable "flag". See
1628 * open_to_namei_flags() for more details.
1630 struct file *do_filp_open(int dfd, const char *pathname,
1631 int open_flag, int mode)
1634 struct nameidata nd;
1635 int acc_mode, error;
1640 int flag = open_to_namei_flags(open_flag);
1642 acc_mode = MAY_OPEN | ACC_MODE(flag);
1644 /* O_TRUNC implies we need access checks for write permissions */
1646 acc_mode |= MAY_WRITE;
1648 /* Allow the LSM permission hook to distinguish append
1649 access from general write access. */
1650 if (flag & O_APPEND)
1651 acc_mode |= MAY_APPEND;
1654 * The simplest case - just a plain lookup.
1656 if (!(flag & O_CREAT)) {
1657 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1660 return ERR_PTR(error);
1665 * Create - we need to know the parent.
1667 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
1669 return ERR_PTR(error);
1672 * We have the parent and last component. First of all, check
1673 * that we are not asked to creat(2) an obvious directory - that
1677 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1681 filp = get_empty_filp();
1684 nd.intent.open.file = filp;
1685 nd.intent.open.flags = flag;
1686 nd.intent.open.create_mode = mode;
1687 dir = nd.path.dentry;
1688 nd.flags &= ~LOOKUP_PARENT;
1689 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1691 nd.flags |= LOOKUP_EXCL;
1692 mutex_lock(&dir->d_inode->i_mutex);
1693 path.dentry = lookup_hash(&nd);
1694 path.mnt = nd.path.mnt;
1697 error = PTR_ERR(path.dentry);
1698 if (IS_ERR(path.dentry)) {
1699 mutex_unlock(&dir->d_inode->i_mutex);
1703 if (IS_ERR(nd.intent.open.file)) {
1704 error = PTR_ERR(nd.intent.open.file);
1705 goto exit_mutex_unlock;
1708 /* Negative dentry, just create the file */
1709 if (!path.dentry->d_inode) {
1711 * This write is needed to ensure that a
1712 * ro->rw transition does not occur between
1713 * the time when the file is created and when
1714 * a permanent write count is taken through
1715 * the 'struct file' in nameidata_to_filp().
1717 error = mnt_want_write(nd.path.mnt);
1719 goto exit_mutex_unlock;
1720 error = __open_namei_create(&nd, &path, flag, mode);
1722 mnt_drop_write(nd.path.mnt);
1725 filp = nameidata_to_filp(&nd, open_flag);
1726 mnt_drop_write(nd.path.mnt);
1731 * It already exists.
1733 mutex_unlock(&dir->d_inode->i_mutex);
1734 audit_inode(pathname, path.dentry);
1740 if (__follow_mount(&path)) {
1742 if (flag & O_NOFOLLOW)
1747 if (!path.dentry->d_inode)
1749 if (path.dentry->d_inode->i_op->follow_link)
1752 path_to_nameidata(&path, &nd);
1754 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1759 * 1. may_open() truncates a file
1760 * 2. a rw->ro mount transition occurs
1761 * 3. nameidata_to_filp() fails due to
1763 * That would be inconsistent, and should
1764 * be avoided. Taking this mnt write here
1765 * ensures that (2) can not occur.
1767 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1769 error = mnt_want_write(nd.path.mnt);
1773 error = may_open(&nd.path, acc_mode, flag);
1776 mnt_drop_write(nd.path.mnt);
1779 filp = nameidata_to_filp(&nd, open_flag);
1781 * It is now safe to drop the mnt write
1782 * because the filp has had a write taken
1786 mnt_drop_write(nd.path.mnt);
1790 mutex_unlock(&dir->d_inode->i_mutex);
1792 path_put_conditional(&path, &nd);
1794 if (!IS_ERR(nd.intent.open.file))
1795 release_open_intent(&nd);
1798 return ERR_PTR(error);
1802 if (flag & O_NOFOLLOW)
1805 * This is subtle. Instead of calling do_follow_link() we do the
1806 * thing by hands. The reason is that this way we have zero link_count
1807 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1808 * After that we have the parent and last component, i.e.
1809 * we are in the same situation as after the first path_walk().
1810 * Well, almost - if the last component is normal we get its copy
1811 * stored in nd->last.name and we will have to putname() it when we
1812 * are done. Procfs-like symlinks just set LAST_BIND.
1814 nd.flags |= LOOKUP_PARENT;
1815 error = security_inode_follow_link(path.dentry, &nd);
1818 error = __do_follow_link(&path, &nd);
1820 /* Does someone understand code flow here? Or it is only
1821 * me so stupid? Anathema to whoever designed this non-sense
1822 * with "intent.open".
1824 release_open_intent(&nd);
1825 return ERR_PTR(error);
1827 nd.flags &= ~LOOKUP_PARENT;
1828 if (nd.last_type == LAST_BIND)
1831 if (nd.last_type != LAST_NORM)
1833 if (nd.last.name[nd.last.len]) {
1834 __putname(nd.last.name);
1839 __putname(nd.last.name);
1842 dir = nd.path.dentry;
1843 mutex_lock(&dir->d_inode->i_mutex);
1844 path.dentry = lookup_hash(&nd);
1845 path.mnt = nd.path.mnt;
1846 __putname(nd.last.name);
1851 * filp_open - open file and return file pointer
1853 * @filename: path to open
1854 * @flags: open flags as per the open(2) second argument
1855 * @mode: mode for the new file if O_CREAT is set, else ignored
1857 * This is the helper to open a file from kernelspace if you really
1858 * have to. But in generally you should not do this, so please move
1859 * along, nothing to see here..
1861 struct file *filp_open(const char *filename, int flags, int mode)
1863 return do_filp_open(AT_FDCWD, filename, flags, mode);
1865 EXPORT_SYMBOL(filp_open);
1868 * lookup_create - lookup a dentry, creating it if it doesn't exist
1869 * @nd: nameidata info
1870 * @is_dir: directory flag
1872 * Simple function to lookup and return a dentry and create it
1873 * if it doesn't exist. Is SMP-safe.
1875 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1877 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1879 struct dentry *dentry = ERR_PTR(-EEXIST);
1881 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1883 * Yucky last component or no last component at all?
1884 * (foo/., foo/.., /////)
1886 if (nd->last_type != LAST_NORM)
1888 nd->flags &= ~LOOKUP_PARENT;
1889 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1890 nd->intent.open.flags = O_EXCL;
1893 * Do the final lookup.
1895 dentry = lookup_hash(nd);
1899 if (dentry->d_inode)
1902 * Special case - lookup gave negative, but... we had foo/bar/
1903 * From the vfs_mknod() POV we just have a negative dentry -
1904 * all is fine. Let's be bastards - you had / on the end, you've
1905 * been asking for (non-existent) directory. -ENOENT for you.
1907 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1909 dentry = ERR_PTR(-ENOENT);
1914 dentry = ERR_PTR(-EEXIST);
1918 EXPORT_SYMBOL_GPL(lookup_create);
1920 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1922 int error = may_create(dir, dentry);
1927 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1930 if (!dir->i_op->mknod)
1933 error = devcgroup_inode_mknod(mode, dev);
1937 error = security_inode_mknod(dir, dentry, mode, dev);
1942 error = dir->i_op->mknod(dir, dentry, mode, dev);
1944 fsnotify_create(dir, dentry);
1948 static int may_mknod(mode_t mode)
1950 switch (mode & S_IFMT) {
1956 case 0: /* zero mode translates to S_IFREG */
1965 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1970 struct dentry *dentry;
1971 struct nameidata nd;
1976 error = user_path_parent(dfd, filename, &nd, &tmp);
1980 dentry = lookup_create(&nd, 0);
1981 if (IS_ERR(dentry)) {
1982 error = PTR_ERR(dentry);
1985 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1986 mode &= ~current->fs->umask;
1987 error = may_mknod(mode);
1990 error = mnt_want_write(nd.path.mnt);
1993 error = security_path_mknod(&nd.path, dentry, mode, dev);
1995 goto out_drop_write;
1996 switch (mode & S_IFMT) {
1997 case 0: case S_IFREG:
1998 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2000 case S_IFCHR: case S_IFBLK:
2001 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2002 new_decode_dev(dev));
2004 case S_IFIFO: case S_IFSOCK:
2005 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2009 mnt_drop_write(nd.path.mnt);
2013 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2020 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2022 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2025 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2027 int error = may_create(dir, dentry);
2032 if (!dir->i_op->mkdir)
2035 mode &= (S_IRWXUGO|S_ISVTX);
2036 error = security_inode_mkdir(dir, dentry, mode);
2041 error = dir->i_op->mkdir(dir, dentry, mode);
2043 fsnotify_mkdir(dir, dentry);
2047 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2051 struct dentry *dentry;
2052 struct nameidata nd;
2054 error = user_path_parent(dfd, pathname, &nd, &tmp);
2058 dentry = lookup_create(&nd, 1);
2059 error = PTR_ERR(dentry);
2063 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2064 mode &= ~current->fs->umask;
2065 error = mnt_want_write(nd.path.mnt);
2068 error = security_path_mkdir(&nd.path, dentry, mode);
2070 goto out_drop_write;
2071 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2073 mnt_drop_write(nd.path.mnt);
2077 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2084 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2086 return sys_mkdirat(AT_FDCWD, pathname, mode);
2090 * We try to drop the dentry early: we should have
2091 * a usage count of 2 if we're the only user of this
2092 * dentry, and if that is true (possibly after pruning
2093 * the dcache), then we drop the dentry now.
2095 * A low-level filesystem can, if it choses, legally
2098 * if (!d_unhashed(dentry))
2101 * if it cannot handle the case of removing a directory
2102 * that is still in use by something else..
2104 void dentry_unhash(struct dentry *dentry)
2107 shrink_dcache_parent(dentry);
2108 spin_lock(&dcache_lock);
2109 spin_lock(&dentry->d_lock);
2110 if (atomic_read(&dentry->d_count) == 2)
2112 spin_unlock(&dentry->d_lock);
2113 spin_unlock(&dcache_lock);
2116 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2118 int error = may_delete(dir, dentry, 1);
2123 if (!dir->i_op->rmdir)
2128 mutex_lock(&dentry->d_inode->i_mutex);
2129 dentry_unhash(dentry);
2130 if (d_mountpoint(dentry))
2133 error = security_inode_rmdir(dir, dentry);
2135 error = dir->i_op->rmdir(dir, dentry);
2137 dentry->d_inode->i_flags |= S_DEAD;
2140 mutex_unlock(&dentry->d_inode->i_mutex);
2149 static long do_rmdir(int dfd, const char __user *pathname)
2153 struct dentry *dentry;
2154 struct nameidata nd;
2156 error = user_path_parent(dfd, pathname, &nd, &name);
2160 switch(nd.last_type) {
2172 nd.flags &= ~LOOKUP_PARENT;
2174 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2175 dentry = lookup_hash(&nd);
2176 error = PTR_ERR(dentry);
2179 error = mnt_want_write(nd.path.mnt);
2182 error = security_path_rmdir(&nd.path, dentry);
2185 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2187 mnt_drop_write(nd.path.mnt);
2191 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2198 asmlinkage long sys_rmdir(const char __user *pathname)
2200 return do_rmdir(AT_FDCWD, pathname);
2203 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2205 int error = may_delete(dir, dentry, 0);
2210 if (!dir->i_op->unlink)
2215 mutex_lock(&dentry->d_inode->i_mutex);
2216 if (d_mountpoint(dentry))
2219 error = security_inode_unlink(dir, dentry);
2221 error = dir->i_op->unlink(dir, dentry);
2223 mutex_unlock(&dentry->d_inode->i_mutex);
2225 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2226 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2227 fsnotify_link_count(dentry->d_inode);
2235 * Make sure that the actual truncation of the file will occur outside its
2236 * directory's i_mutex. Truncate can take a long time if there is a lot of
2237 * writeout happening, and we don't want to prevent access to the directory
2238 * while waiting on the I/O.
2240 static long do_unlinkat(int dfd, const char __user *pathname)
2244 struct dentry *dentry;
2245 struct nameidata nd;
2246 struct inode *inode = NULL;
2248 error = user_path_parent(dfd, pathname, &nd, &name);
2253 if (nd.last_type != LAST_NORM)
2256 nd.flags &= ~LOOKUP_PARENT;
2258 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2259 dentry = lookup_hash(&nd);
2260 error = PTR_ERR(dentry);
2261 if (!IS_ERR(dentry)) {
2262 /* Why not before? Because we want correct error value */
2263 if (nd.last.name[nd.last.len])
2265 inode = dentry->d_inode;
2267 atomic_inc(&inode->i_count);
2268 error = mnt_want_write(nd.path.mnt);
2271 error = security_path_unlink(&nd.path, dentry);
2274 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2276 mnt_drop_write(nd.path.mnt);
2280 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2282 iput(inode); /* truncate the inode here */
2289 error = !dentry->d_inode ? -ENOENT :
2290 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2294 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2296 if ((flag & ~AT_REMOVEDIR) != 0)
2299 if (flag & AT_REMOVEDIR)
2300 return do_rmdir(dfd, pathname);
2302 return do_unlinkat(dfd, pathname);
2305 asmlinkage long sys_unlink(const char __user *pathname)
2307 return do_unlinkat(AT_FDCWD, pathname);
2310 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2312 int error = may_create(dir, dentry);
2317 if (!dir->i_op->symlink)
2320 error = security_inode_symlink(dir, dentry, oldname);
2325 error = dir->i_op->symlink(dir, dentry, oldname);
2327 fsnotify_create(dir, dentry);
2331 asmlinkage long sys_symlinkat(const char __user *oldname,
2332 int newdfd, const char __user *newname)
2337 struct dentry *dentry;
2338 struct nameidata nd;
2340 from = getname(oldname);
2342 return PTR_ERR(from);
2344 error = user_path_parent(newdfd, newname, &nd, &to);
2348 dentry = lookup_create(&nd, 0);
2349 error = PTR_ERR(dentry);
2353 error = mnt_want_write(nd.path.mnt);
2356 error = security_path_symlink(&nd.path, dentry, from);
2358 goto out_drop_write;
2359 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2361 mnt_drop_write(nd.path.mnt);
2365 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2373 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2375 return sys_symlinkat(oldname, AT_FDCWD, newname);
2378 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2380 struct inode *inode = old_dentry->d_inode;
2386 error = may_create(dir, new_dentry);
2390 if (dir->i_sb != inode->i_sb)
2394 * A link to an append-only or immutable file cannot be created.
2396 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2398 if (!dir->i_op->link)
2400 if (S_ISDIR(inode->i_mode))
2403 error = security_inode_link(old_dentry, dir, new_dentry);
2407 mutex_lock(&inode->i_mutex);
2409 error = dir->i_op->link(old_dentry, dir, new_dentry);
2410 mutex_unlock(&inode->i_mutex);
2412 fsnotify_link(dir, inode, new_dentry);
2417 * Hardlinks are often used in delicate situations. We avoid
2418 * security-related surprises by not following symlinks on the
2421 * We don't follow them on the oldname either to be compatible
2422 * with linux 2.0, and to avoid hard-linking to directories
2423 * and other special files. --ADM
2425 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2426 int newdfd, const char __user *newname,
2429 struct dentry *new_dentry;
2430 struct nameidata nd;
2431 struct path old_path;
2435 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2438 error = user_path_at(olddfd, oldname,
2439 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2444 error = user_path_parent(newdfd, newname, &nd, &to);
2448 if (old_path.mnt != nd.path.mnt)
2450 new_dentry = lookup_create(&nd, 0);
2451 error = PTR_ERR(new_dentry);
2452 if (IS_ERR(new_dentry))
2454 error = mnt_want_write(nd.path.mnt);
2457 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2459 goto out_drop_write;
2460 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2462 mnt_drop_write(nd.path.mnt);
2466 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2471 path_put(&old_path);
2476 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2478 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2482 * The worst of all namespace operations - renaming directory. "Perverted"
2483 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2485 * a) we can get into loop creation. Check is done in is_subdir().
2486 * b) race potential - two innocent renames can create a loop together.
2487 * That's where 4.4 screws up. Current fix: serialization on
2488 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2490 * c) we have to lock _three_ objects - parents and victim (if it exists).
2491 * And that - after we got ->i_mutex on parents (until then we don't know
2492 * whether the target exists). Solution: try to be smart with locking
2493 * order for inodes. We rely on the fact that tree topology may change
2494 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2495 * move will be locked. Thus we can rank directories by the tree
2496 * (ancestors first) and rank all non-directories after them.
2497 * That works since everybody except rename does "lock parent, lookup,
2498 * lock child" and rename is under ->s_vfs_rename_mutex.
2499 * HOWEVER, it relies on the assumption that any object with ->lookup()
2500 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2501 * we'd better make sure that there's no link(2) for them.
2502 * d) some filesystems don't support opened-but-unlinked directories,
2503 * either because of layout or because they are not ready to deal with
2504 * all cases correctly. The latter will be fixed (taking this sort of
2505 * stuff into VFS), but the former is not going away. Solution: the same
2506 * trick as in rmdir().
2507 * e) conversion from fhandle to dentry may come in the wrong moment - when
2508 * we are removing the target. Solution: we will have to grab ->i_mutex
2509 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2510 * ->i_mutex on parents, which works but leads to some truely excessive
2513 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2514 struct inode *new_dir, struct dentry *new_dentry)
2517 struct inode *target;
2520 * If we are going to change the parent - check write permissions,
2521 * we'll need to flip '..'.
2523 if (new_dir != old_dir) {
2524 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2529 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2533 target = new_dentry->d_inode;
2535 mutex_lock(&target->i_mutex);
2536 dentry_unhash(new_dentry);
2538 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2541 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2544 target->i_flags |= S_DEAD;
2545 mutex_unlock(&target->i_mutex);
2546 if (d_unhashed(new_dentry))
2547 d_rehash(new_dentry);
2551 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2552 d_move(old_dentry,new_dentry);
2556 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2557 struct inode *new_dir, struct dentry *new_dentry)
2559 struct inode *target;
2562 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2567 target = new_dentry->d_inode;
2569 mutex_lock(&target->i_mutex);
2570 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2573 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2575 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2576 d_move(old_dentry, new_dentry);
2579 mutex_unlock(&target->i_mutex);
2584 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2585 struct inode *new_dir, struct dentry *new_dentry)
2588 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2589 const char *old_name;
2591 if (old_dentry->d_inode == new_dentry->d_inode)
2594 error = may_delete(old_dir, old_dentry, is_dir);
2598 if (!new_dentry->d_inode)
2599 error = may_create(new_dir, new_dentry);
2601 error = may_delete(new_dir, new_dentry, is_dir);
2605 if (!old_dir->i_op->rename)
2608 DQUOT_INIT(old_dir);
2609 DQUOT_INIT(new_dir);
2611 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2614 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2616 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2618 const char *new_name = old_dentry->d_name.name;
2619 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2620 new_dentry->d_inode, old_dentry);
2622 fsnotify_oldname_free(old_name);
2627 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2628 int newdfd, const char __user *newname)
2630 struct dentry *old_dir, *new_dir;
2631 struct dentry *old_dentry, *new_dentry;
2632 struct dentry *trap;
2633 struct nameidata oldnd, newnd;
2638 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2642 error = user_path_parent(newdfd, newname, &newnd, &to);
2647 if (oldnd.path.mnt != newnd.path.mnt)
2650 old_dir = oldnd.path.dentry;
2652 if (oldnd.last_type != LAST_NORM)
2655 new_dir = newnd.path.dentry;
2656 if (newnd.last_type != LAST_NORM)
2659 oldnd.flags &= ~LOOKUP_PARENT;
2660 newnd.flags &= ~LOOKUP_PARENT;
2661 newnd.flags |= LOOKUP_RENAME_TARGET;
2663 trap = lock_rename(new_dir, old_dir);
2665 old_dentry = lookup_hash(&oldnd);
2666 error = PTR_ERR(old_dentry);
2667 if (IS_ERR(old_dentry))
2669 /* source must exist */
2671 if (!old_dentry->d_inode)
2673 /* unless the source is a directory trailing slashes give -ENOTDIR */
2674 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2676 if (oldnd.last.name[oldnd.last.len])
2678 if (newnd.last.name[newnd.last.len])
2681 /* source should not be ancestor of target */
2683 if (old_dentry == trap)
2685 new_dentry = lookup_hash(&newnd);
2686 error = PTR_ERR(new_dentry);
2687 if (IS_ERR(new_dentry))
2689 /* target should not be an ancestor of source */
2691 if (new_dentry == trap)
2694 error = mnt_want_write(oldnd.path.mnt);
2697 error = security_path_rename(&oldnd.path, old_dentry,
2698 &newnd.path, new_dentry);
2701 error = vfs_rename(old_dir->d_inode, old_dentry,
2702 new_dir->d_inode, new_dentry);
2704 mnt_drop_write(oldnd.path.mnt);
2710 unlock_rename(new_dir, old_dir);
2712 path_put(&newnd.path);
2715 path_put(&oldnd.path);
2721 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2723 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2726 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2730 len = PTR_ERR(link);
2735 if (len > (unsigned) buflen)
2737 if (copy_to_user(buffer, link, len))
2744 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2745 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2746 * using) it for any given inode is up to filesystem.
2748 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2750 struct nameidata nd;
2755 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2757 return PTR_ERR(cookie);
2759 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2760 if (dentry->d_inode->i_op->put_link)
2761 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2765 int vfs_follow_link(struct nameidata *nd, const char *link)
2767 return __vfs_follow_link(nd, link);
2770 /* get the link contents into pagecache */
2771 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2775 struct address_space *mapping = dentry->d_inode->i_mapping;
2776 page = read_mapping_page(mapping, 0, NULL);
2781 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2785 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2787 struct page *page = NULL;
2788 char *s = page_getlink(dentry, &page);
2789 int res = vfs_readlink(dentry,buffer,buflen,s);
2792 page_cache_release(page);
2797 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2799 struct page *page = NULL;
2800 nd_set_link(nd, page_getlink(dentry, &page));
2804 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2806 struct page *page = cookie;
2810 page_cache_release(page);
2815 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2817 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2819 struct address_space *mapping = inode->i_mapping;
2824 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2826 flags |= AOP_FLAG_NOFS;
2829 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2830 flags, &page, &fsdata);
2834 kaddr = kmap_atomic(page, KM_USER0);
2835 memcpy(kaddr, symname, len-1);
2836 kunmap_atomic(kaddr, KM_USER0);
2838 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2845 mark_inode_dirty(inode);
2851 int page_symlink(struct inode *inode, const char *symname, int len)
2853 return __page_symlink(inode, symname, len,
2854 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2857 const struct inode_operations page_symlink_inode_operations = {
2858 .readlink = generic_readlink,
2859 .follow_link = page_follow_link_light,
2860 .put_link = page_put_link,
2863 EXPORT_SYMBOL(user_path_at);
2864 EXPORT_SYMBOL(follow_down);
2865 EXPORT_SYMBOL(follow_up);
2866 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2867 EXPORT_SYMBOL(getname);
2868 EXPORT_SYMBOL(lock_rename);
2869 EXPORT_SYMBOL(lookup_one_len);
2870 EXPORT_SYMBOL(page_follow_link_light);
2871 EXPORT_SYMBOL(page_put_link);
2872 EXPORT_SYMBOL(page_readlink);
2873 EXPORT_SYMBOL(__page_symlink);
2874 EXPORT_SYMBOL(page_symlink);
2875 EXPORT_SYMBOL(page_symlink_inode_operations);
2876 EXPORT_SYMBOL(path_lookup);
2877 EXPORT_SYMBOL(kern_path);
2878 EXPORT_SYMBOL(vfs_path_lookup);
2879 EXPORT_SYMBOL(inode_permission);
2880 EXPORT_SYMBOL(file_permission);
2881 EXPORT_SYMBOL(unlock_rename);
2882 EXPORT_SYMBOL(vfs_create);
2883 EXPORT_SYMBOL(vfs_follow_link);
2884 EXPORT_SYMBOL(vfs_link);
2885 EXPORT_SYMBOL(vfs_mkdir);
2886 EXPORT_SYMBOL(vfs_mknod);
2887 EXPORT_SYMBOL(generic_permission);
2888 EXPORT_SYMBOL(vfs_readlink);
2889 EXPORT_SYMBOL(vfs_rename);
2890 EXPORT_SYMBOL(vfs_rmdir);
2891 EXPORT_SYMBOL(vfs_symlink);
2892 EXPORT_SYMBOL(vfs_unlink);
2893 EXPORT_SYMBOL(dentry_unhash);
2894 EXPORT_SYMBOL(generic_readlink);
2896 /* to be mentioned only in INIT_TASK */
2897 struct fs_struct init_fs = {
2898 .count = ATOMIC_INIT(1),
2899 .lock = __RW_LOCK_UNLOCKED(init_fs.lock),