4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name, struct nameidata *nd);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
185 umode_t mode = inode->i_mode;
187 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
189 if (current->fsuid == inode->i_uid)
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
200 if (in_group_p(inode->i_gid))
205 * If the DACs are ok we don't need any capability check.
207 if ((mask & ~mode) == 0)
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask & MAY_EXEC) ||
216 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
217 if (capable(CAP_DAC_OVERRIDE))
221 * Searching includes executable on directories, else just read.
223 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
224 if (capable(CAP_DAC_READ_SEARCH))
230 int inode_permission(struct inode *inode, int mask)
234 if (mask & MAY_WRITE) {
235 umode_t mode = inode->i_mode;
238 * Nobody gets write access to a read-only fs.
240 if (IS_RDONLY(inode) &&
241 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
245 * Nobody gets write access to an immutable file.
247 if (IS_IMMUTABLE(inode))
251 /* Ordinary permission routines do not understand MAY_APPEND. */
252 if (inode->i_op && inode->i_op->permission) {
253 retval = inode->i_op->permission(inode, mask);
256 * Exec permission on a regular file is denied if none
257 * of the execute bits are set.
259 * This check should be done by the ->permission()
262 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
263 !(inode->i_mode & S_IXUGO))
267 retval = generic_permission(inode, mask, NULL);
272 retval = devcgroup_inode_permission(inode, mask);
276 return security_inode_permission(inode,
277 mask & (MAY_READ|MAY_WRITE|MAY_EXEC));
281 * vfs_permission - check for access rights to a given path
282 * @nd: lookup result that describes the path
283 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
285 * Used to check for read/write/execute permissions on a path.
286 * We use "fsuid" for this, letting us set arbitrary permissions
287 * for filesystem access without changing the "normal" uids which
288 * are used for other things.
290 int vfs_permission(struct nameidata *nd, int mask)
292 return inode_permission(nd->path.dentry->d_inode, mask);
296 * file_permission - check for additional access rights to a given file
297 * @file: file to check access rights for
298 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
300 * Used to check for read/write/execute permissions on an already opened
304 * Do not use this function in new code. All access checks should
305 * be done using vfs_permission().
307 int file_permission(struct file *file, int mask)
309 return inode_permission(file->f_path.dentry->d_inode, mask);
313 * get_write_access() gets write permission for a file.
314 * put_write_access() releases this write permission.
315 * This is used for regular files.
316 * We cannot support write (and maybe mmap read-write shared) accesses and
317 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
318 * can have the following values:
319 * 0: no writers, no VM_DENYWRITE mappings
320 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
321 * > 0: (i_writecount) users are writing to the file.
323 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
324 * except for the cases where we don't hold i_writecount yet. Then we need to
325 * use {get,deny}_write_access() - these functions check the sign and refuse
326 * to do the change if sign is wrong. Exclusion between them is provided by
327 * the inode->i_lock spinlock.
330 int get_write_access(struct inode * inode)
332 spin_lock(&inode->i_lock);
333 if (atomic_read(&inode->i_writecount) < 0) {
334 spin_unlock(&inode->i_lock);
337 atomic_inc(&inode->i_writecount);
338 spin_unlock(&inode->i_lock);
343 int deny_write_access(struct file * file)
345 struct inode *inode = file->f_path.dentry->d_inode;
347 spin_lock(&inode->i_lock);
348 if (atomic_read(&inode->i_writecount) > 0) {
349 spin_unlock(&inode->i_lock);
352 atomic_dec(&inode->i_writecount);
353 spin_unlock(&inode->i_lock);
359 * path_get - get a reference to a path
360 * @path: path to get the reference to
362 * Given a path increment the reference count to the dentry and the vfsmount.
364 void path_get(struct path *path)
369 EXPORT_SYMBOL(path_get);
372 * path_put - put a reference to a path
373 * @path: path to put the reference to
375 * Given a path decrement the reference count to the dentry and the vfsmount.
377 void path_put(struct path *path)
382 EXPORT_SYMBOL(path_put);
385 * release_open_intent - free up open intent resources
386 * @nd: pointer to nameidata
388 void release_open_intent(struct nameidata *nd)
390 if (nd->intent.open.file->f_path.dentry == NULL)
391 put_filp(nd->intent.open.file);
393 fput(nd->intent.open.file);
396 static inline struct dentry *
397 do_revalidate(struct dentry *dentry, struct nameidata *nd)
399 int status = dentry->d_op->d_revalidate(dentry, nd);
400 if (unlikely(status <= 0)) {
402 * The dentry failed validation.
403 * If d_revalidate returned 0 attempt to invalidate
404 * the dentry otherwise d_revalidate is asking us
405 * to return a fail status.
408 if (!d_invalidate(dentry)) {
414 dentry = ERR_PTR(status);
421 * Internal lookup() using the new generic dcache.
424 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
426 struct dentry * dentry = __d_lookup(parent, name);
428 /* lockess __d_lookup may fail due to concurrent d_move()
429 * in some unrelated directory, so try with d_lookup
432 dentry = d_lookup(parent, name);
434 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
435 dentry = do_revalidate(dentry, nd);
441 * Short-cut version of permission(), for calling by
442 * path_walk(), when dcache lock is held. Combines parts
443 * of permission() and generic_permission(), and tests ONLY for
444 * MAY_EXEC permission.
446 * If appropriate, check DAC only. If not appropriate, or
447 * short-cut DAC fails, then call permission() to do more
448 * complete permission check.
450 static int exec_permission_lite(struct inode *inode)
452 umode_t mode = inode->i_mode;
454 if (inode->i_op && inode->i_op->permission)
457 if (current->fsuid == inode->i_uid)
459 else if (in_group_p(inode->i_gid))
465 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
468 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
471 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
476 return security_inode_permission(inode, MAY_EXEC);
480 * This is called when everything else fails, and we actually have
481 * to go to the low-level filesystem to find out what we should do..
483 * We get the directory semaphore, and after getting that we also
484 * make sure that nobody added the entry to the dcache in the meantime..
487 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
489 struct dentry * result;
490 struct inode *dir = parent->d_inode;
492 mutex_lock(&dir->i_mutex);
494 * First re-do the cached lookup just in case it was created
495 * while we waited for the directory semaphore..
497 * FIXME! This could use version numbering or similar to
498 * avoid unnecessary cache lookups.
500 * The "dcache_lock" is purely to protect the RCU list walker
501 * from concurrent renames at this point (we mustn't get false
502 * negatives from the RCU list walk here, unlike the optimistic
505 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
507 result = d_lookup(parent, name);
509 struct dentry *dentry;
511 /* Don't create child dentry for a dead directory. */
512 result = ERR_PTR(-ENOENT);
516 dentry = d_alloc(parent, name);
517 result = ERR_PTR(-ENOMEM);
519 result = dir->i_op->lookup(dir, dentry, nd);
526 mutex_unlock(&dir->i_mutex);
531 * Uhhuh! Nasty case: the cache was re-populated while
532 * we waited on the semaphore. Need to revalidate.
534 mutex_unlock(&dir->i_mutex);
535 if (result->d_op && result->d_op->d_revalidate) {
536 result = do_revalidate(result, nd);
538 result = ERR_PTR(-ENOENT);
544 static __always_inline void
545 walk_init_root(const char *name, struct nameidata *nd)
547 struct fs_struct *fs = current->fs;
549 read_lock(&fs->lock);
552 read_unlock(&fs->lock);
556 * Wrapper to retry pathname resolution whenever the underlying
557 * file system returns an ESTALE.
559 * Retry the whole path once, forcing real lookup requests
560 * instead of relying on the dcache.
562 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
564 struct path save = nd->path;
567 /* make sure the stuff we saved doesn't go away */
570 result = __link_path_walk(name, nd);
571 if (result == -ESTALE) {
572 /* nd->path had been dropped */
575 nd->flags |= LOOKUP_REVAL;
576 result = __link_path_walk(name, nd);
584 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
593 walk_init_root(link, nd);
595 res = link_path_walk(link, nd);
596 if (nd->depth || res || nd->last_type!=LAST_NORM)
599 * If it is an iterative symlinks resolution in open_namei() we
600 * have to copy the last component. And all that crap because of
601 * bloody create() on broken symlinks. Furrfu...
604 if (unlikely(!name)) {
608 strcpy(name, nd->last.name);
609 nd->last.name = name;
613 return PTR_ERR(link);
616 static void path_put_conditional(struct path *path, struct nameidata *nd)
619 if (path->mnt != nd->path.mnt)
623 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
625 dput(nd->path.dentry);
626 if (nd->path.mnt != path->mnt)
627 mntput(nd->path.mnt);
628 nd->path.mnt = path->mnt;
629 nd->path.dentry = path->dentry;
632 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
636 struct dentry *dentry = path->dentry;
638 touch_atime(path->mnt, dentry);
639 nd_set_link(nd, NULL);
641 if (path->mnt != nd->path.mnt) {
642 path_to_nameidata(path, nd);
646 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
647 error = PTR_ERR(cookie);
648 if (!IS_ERR(cookie)) {
649 char *s = nd_get_link(nd);
652 error = __vfs_follow_link(nd, s);
653 if (dentry->d_inode->i_op->put_link)
654 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
662 * This limits recursive symlink follows to 8, while
663 * limiting consecutive symlinks to 40.
665 * Without that kind of total limit, nasty chains of consecutive
666 * symlinks can cause almost arbitrarily long lookups.
668 static inline int do_follow_link(struct path *path, struct nameidata *nd)
671 if (current->link_count >= MAX_NESTED_LINKS)
673 if (current->total_link_count >= 40)
675 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
677 err = security_inode_follow_link(path->dentry, nd);
680 current->link_count++;
681 current->total_link_count++;
683 err = __do_follow_link(path, nd);
684 current->link_count--;
688 path_put_conditional(path, nd);
693 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
695 struct vfsmount *parent;
696 struct dentry *mountpoint;
697 spin_lock(&vfsmount_lock);
698 parent=(*mnt)->mnt_parent;
699 if (parent == *mnt) {
700 spin_unlock(&vfsmount_lock);
704 mountpoint=dget((*mnt)->mnt_mountpoint);
705 spin_unlock(&vfsmount_lock);
707 *dentry = mountpoint;
713 /* no need for dcache_lock, as serialization is taken care in
716 static int __follow_mount(struct path *path)
719 while (d_mountpoint(path->dentry)) {
720 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
727 path->dentry = dget(mounted->mnt_root);
733 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
735 while (d_mountpoint(*dentry)) {
736 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
742 *dentry = dget(mounted->mnt_root);
746 /* no need for dcache_lock, as serialization is taken care in
749 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
751 struct vfsmount *mounted;
753 mounted = lookup_mnt(*mnt, *dentry);
758 *dentry = dget(mounted->mnt_root);
764 static __always_inline void follow_dotdot(struct nameidata *nd)
766 struct fs_struct *fs = current->fs;
769 struct vfsmount *parent;
770 struct dentry *old = nd->path.dentry;
772 read_lock(&fs->lock);
773 if (nd->path.dentry == fs->root.dentry &&
774 nd->path.mnt == fs->root.mnt) {
775 read_unlock(&fs->lock);
778 read_unlock(&fs->lock);
779 spin_lock(&dcache_lock);
780 if (nd->path.dentry != nd->path.mnt->mnt_root) {
781 nd->path.dentry = dget(nd->path.dentry->d_parent);
782 spin_unlock(&dcache_lock);
786 spin_unlock(&dcache_lock);
787 spin_lock(&vfsmount_lock);
788 parent = nd->path.mnt->mnt_parent;
789 if (parent == nd->path.mnt) {
790 spin_unlock(&vfsmount_lock);
794 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
795 spin_unlock(&vfsmount_lock);
797 mntput(nd->path.mnt);
798 nd->path.mnt = parent;
800 follow_mount(&nd->path.mnt, &nd->path.dentry);
804 * It's more convoluted than I'd like it to be, but... it's still fairly
805 * small and for now I'd prefer to have fast path as straight as possible.
806 * It _is_ time-critical.
808 static int do_lookup(struct nameidata *nd, struct qstr *name,
811 struct vfsmount *mnt = nd->path.mnt;
812 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
816 if (dentry->d_op && dentry->d_op->d_revalidate)
817 goto need_revalidate;
820 path->dentry = dentry;
821 __follow_mount(path);
825 dentry = real_lookup(nd->path.dentry, name, nd);
831 dentry = do_revalidate(dentry, nd);
839 return PTR_ERR(dentry);
844 * This is the basic name resolution function, turning a pathname into
845 * the final dentry. We expect 'base' to be positive and a directory.
847 * Returns 0 and nd will have valid dentry and mnt on success.
848 * Returns error and drops reference to input namei data on failure.
850 static int __link_path_walk(const char *name, struct nameidata *nd)
855 unsigned int lookup_flags = nd->flags;
862 inode = nd->path.dentry->d_inode;
864 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
866 /* At this point we know we have a real path component. */
872 nd->flags |= LOOKUP_CONTINUE;
873 err = exec_permission_lite(inode);
875 err = vfs_permission(nd, MAY_EXEC);
880 c = *(const unsigned char *)name;
882 hash = init_name_hash();
885 hash = partial_name_hash(c, hash);
886 c = *(const unsigned char *)name;
887 } while (c && (c != '/'));
888 this.len = name - (const char *) this.name;
889 this.hash = end_name_hash(hash);
891 /* remove trailing slashes? */
894 while (*++name == '/');
896 goto last_with_slashes;
899 * "." and ".." are special - ".." especially so because it has
900 * to be able to know about the current root directory and
901 * parent relationships.
903 if (this.name[0] == '.') switch (this.len) {
907 if (this.name[1] != '.')
910 inode = nd->path.dentry->d_inode;
916 * See if the low-level filesystem might want
917 * to use its own hash..
919 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
920 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
925 /* This does the actual lookups.. */
926 err = do_lookup(nd, &this, &next);
931 inode = next.dentry->d_inode;
938 if (inode->i_op->follow_link) {
939 err = do_follow_link(&next, nd);
943 inode = nd->path.dentry->d_inode;
950 path_to_nameidata(&next, nd);
952 if (!inode->i_op->lookup)
955 /* here ends the main loop */
958 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
960 /* Clear LOOKUP_CONTINUE iff it was previously unset */
961 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
962 if (lookup_flags & LOOKUP_PARENT)
964 if (this.name[0] == '.') switch (this.len) {
968 if (this.name[1] != '.')
971 inode = nd->path.dentry->d_inode;
976 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
977 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
982 err = do_lookup(nd, &this, &next);
985 inode = next.dentry->d_inode;
986 if ((lookup_flags & LOOKUP_FOLLOW)
987 && inode && inode->i_op && inode->i_op->follow_link) {
988 err = do_follow_link(&next, nd);
991 inode = nd->path.dentry->d_inode;
993 path_to_nameidata(&next, nd);
997 if (lookup_flags & LOOKUP_DIRECTORY) {
999 if (!inode->i_op || !inode->i_op->lookup)
1005 nd->last_type = LAST_NORM;
1006 if (this.name[0] != '.')
1009 nd->last_type = LAST_DOT;
1010 else if (this.len == 2 && this.name[1] == '.')
1011 nd->last_type = LAST_DOTDOT;
1016 * We bypassed the ordinary revalidation routines.
1017 * We may need to check the cached dentry for staleness.
1019 if (nd->path.dentry && nd->path.dentry->d_sb &&
1020 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1022 /* Note: we do not d_invalidate() */
1023 if (!nd->path.dentry->d_op->d_revalidate(
1024 nd->path.dentry, nd))
1030 path_put_conditional(&next, nd);
1033 path_put(&nd->path);
1038 static int path_walk(const char *name, struct nameidata *nd)
1040 current->total_link_count = 0;
1041 return link_path_walk(name, nd);
1044 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1045 static int do_path_lookup(int dfd, const char *name,
1046 unsigned int flags, struct nameidata *nd)
1051 struct fs_struct *fs = current->fs;
1053 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1058 read_lock(&fs->lock);
1059 nd->path = fs->root;
1060 path_get(&fs->root);
1061 read_unlock(&fs->lock);
1062 } else if (dfd == AT_FDCWD) {
1063 read_lock(&fs->lock);
1066 read_unlock(&fs->lock);
1068 struct dentry *dentry;
1070 file = fget_light(dfd, &fput_needed);
1075 dentry = file->f_path.dentry;
1078 if (!S_ISDIR(dentry->d_inode->i_mode))
1081 retval = file_permission(file, MAY_EXEC);
1085 nd->path = file->f_path;
1086 path_get(&file->f_path);
1088 fput_light(file, fput_needed);
1091 retval = path_walk(name, nd);
1092 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1093 nd->path.dentry->d_inode))
1094 audit_inode(name, nd->path.dentry);
1099 fput_light(file, fput_needed);
1103 int path_lookup(const char *name, unsigned int flags,
1104 struct nameidata *nd)
1106 return do_path_lookup(AT_FDCWD, name, flags, nd);
1110 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1111 * @dentry: pointer to dentry of the base directory
1112 * @mnt: pointer to vfs mount of the base directory
1113 * @name: pointer to file name
1114 * @flags: lookup flags
1115 * @nd: pointer to nameidata
1117 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1118 const char *name, unsigned int flags,
1119 struct nameidata *nd)
1123 /* same as do_path_lookup */
1124 nd->last_type = LAST_ROOT;
1128 nd->path.dentry = dentry;
1130 path_get(&nd->path);
1132 retval = path_walk(name, nd);
1133 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1134 nd->path.dentry->d_inode))
1135 audit_inode(name, nd->path.dentry);
1141 static int __path_lookup_intent_open(int dfd, const char *name,
1142 unsigned int lookup_flags, struct nameidata *nd,
1143 int open_flags, int create_mode)
1145 struct file *filp = get_empty_filp();
1150 nd->intent.open.file = filp;
1151 nd->intent.open.flags = open_flags;
1152 nd->intent.open.create_mode = create_mode;
1153 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1154 if (IS_ERR(nd->intent.open.file)) {
1156 err = PTR_ERR(nd->intent.open.file);
1157 path_put(&nd->path);
1159 } else if (err != 0)
1160 release_open_intent(nd);
1165 * path_lookup_open - lookup a file path with open intent
1166 * @dfd: the directory to use as base, or AT_FDCWD
1167 * @name: pointer to file name
1168 * @lookup_flags: lookup intent flags
1169 * @nd: pointer to nameidata
1170 * @open_flags: open intent flags
1172 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1173 struct nameidata *nd, int open_flags)
1175 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1180 * path_lookup_create - lookup a file path with open + create intent
1181 * @dfd: the directory to use as base, or AT_FDCWD
1182 * @name: pointer to file name
1183 * @lookup_flags: lookup intent flags
1184 * @nd: pointer to nameidata
1185 * @open_flags: open intent flags
1186 * @create_mode: create intent flags
1188 static int path_lookup_create(int dfd, const char *name,
1189 unsigned int lookup_flags, struct nameidata *nd,
1190 int open_flags, int create_mode)
1192 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1193 nd, open_flags, create_mode);
1196 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1197 struct nameidata *nd, int open_flags)
1199 char *tmp = getname(name);
1200 int err = PTR_ERR(tmp);
1203 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1209 static struct dentry *__lookup_hash(struct qstr *name,
1210 struct dentry *base, struct nameidata *nd)
1212 struct dentry *dentry;
1213 struct inode *inode;
1216 inode = base->d_inode;
1219 * See if the low-level filesystem might want
1220 * to use its own hash..
1222 if (base->d_op && base->d_op->d_hash) {
1223 err = base->d_op->d_hash(base, name);
1224 dentry = ERR_PTR(err);
1229 dentry = cached_lookup(base, name, nd);
1233 /* Don't create child dentry for a dead directory. */
1234 dentry = ERR_PTR(-ENOENT);
1235 if (IS_DEADDIR(inode))
1238 new = d_alloc(base, name);
1239 dentry = ERR_PTR(-ENOMEM);
1242 dentry = inode->i_op->lookup(inode, new, nd);
1253 * Restricted form of lookup. Doesn't follow links, single-component only,
1254 * needs parent already locked. Doesn't follow mounts.
1257 static struct dentry *lookup_hash(struct nameidata *nd)
1261 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1263 return ERR_PTR(err);
1264 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1267 static int __lookup_one_len(const char *name, struct qstr *this,
1268 struct dentry *base, int len)
1278 hash = init_name_hash();
1280 c = *(const unsigned char *)name++;
1281 if (c == '/' || c == '\0')
1283 hash = partial_name_hash(c, hash);
1285 this->hash = end_name_hash(hash);
1290 * lookup_one_len - filesystem helper to lookup single pathname component
1291 * @name: pathname component to lookup
1292 * @base: base directory to lookup from
1293 * @len: maximum length @len should be interpreted to
1295 * Note that this routine is purely a helper for filesystem usage and should
1296 * not be called by generic code. Also note that by using this function the
1297 * nameidata argument is passed to the filesystem methods and a filesystem
1298 * using this helper needs to be prepared for that.
1300 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1305 err = __lookup_one_len(name, &this, base, len);
1307 return ERR_PTR(err);
1309 err = inode_permission(base->d_inode, MAY_EXEC);
1311 return ERR_PTR(err);
1312 return __lookup_hash(&this, base, NULL);
1316 * lookup_one_noperm - bad hack for sysfs
1317 * @name: pathname component to lookup
1318 * @base: base directory to lookup from
1320 * This is a variant of lookup_one_len that doesn't perform any permission
1321 * checks. It's a horrible hack to work around the braindead sysfs
1322 * architecture and should not be used anywhere else.
1324 * DON'T USE THIS FUNCTION EVER, thanks.
1326 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1331 err = __lookup_one_len(name, &this, base, strlen(name));
1333 return ERR_PTR(err);
1334 return __lookup_hash(&this, base, NULL);
1337 int user_path_at(int dfd, const char __user *name, unsigned flags,
1340 struct nameidata nd;
1341 char *tmp = getname(name);
1342 int err = PTR_ERR(tmp);
1345 BUG_ON(flags & LOOKUP_PARENT);
1347 err = do_path_lookup(dfd, tmp, flags, &nd);
1355 static int user_path_parent(int dfd, const char __user *path,
1356 struct nameidata *nd, char **name)
1358 char *s = getname(path);
1364 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1374 * It's inline, so penalty for filesystems that don't use sticky bit is
1377 static inline int check_sticky(struct inode *dir, struct inode *inode)
1379 if (!(dir->i_mode & S_ISVTX))
1381 if (inode->i_uid == current->fsuid)
1383 if (dir->i_uid == current->fsuid)
1385 return !capable(CAP_FOWNER);
1389 * Check whether we can remove a link victim from directory dir, check
1390 * whether the type of victim is right.
1391 * 1. We can't do it if dir is read-only (done in permission())
1392 * 2. We should have write and exec permissions on dir
1393 * 3. We can't remove anything from append-only dir
1394 * 4. We can't do anything with immutable dir (done in permission())
1395 * 5. If the sticky bit on dir is set we should either
1396 * a. be owner of dir, or
1397 * b. be owner of victim, or
1398 * c. have CAP_FOWNER capability
1399 * 6. If the victim is append-only or immutable we can't do antyhing with
1400 * links pointing to it.
1401 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1402 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1403 * 9. We can't remove a root or mountpoint.
1404 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1405 * nfs_async_unlink().
1407 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1411 if (!victim->d_inode)
1414 BUG_ON(victim->d_parent->d_inode != dir);
1415 audit_inode_child(victim->d_name.name, victim, dir);
1417 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1422 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1423 IS_IMMUTABLE(victim->d_inode))
1426 if (!S_ISDIR(victim->d_inode->i_mode))
1428 if (IS_ROOT(victim))
1430 } else if (S_ISDIR(victim->d_inode->i_mode))
1432 if (IS_DEADDIR(dir))
1434 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1439 /* Check whether we can create an object with dentry child in directory
1441 * 1. We can't do it if child already exists (open has special treatment for
1442 * this case, but since we are inlined it's OK)
1443 * 2. We can't do it if dir is read-only (done in permission())
1444 * 3. We should have write and exec permissions on dir
1445 * 4. We can't do it if dir is immutable (done in permission())
1447 static inline int may_create(struct inode *dir, struct dentry *child,
1448 struct nameidata *nd)
1452 if (IS_DEADDIR(dir))
1454 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1458 * O_DIRECTORY translates into forcing a directory lookup.
1460 static inline int lookup_flags(unsigned int f)
1462 unsigned long retval = LOOKUP_FOLLOW;
1465 retval &= ~LOOKUP_FOLLOW;
1467 if (f & O_DIRECTORY)
1468 retval |= LOOKUP_DIRECTORY;
1474 * p1 and p2 should be directories on the same fs.
1476 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1481 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1485 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1487 for (p = p1; p->d_parent != p; p = p->d_parent) {
1488 if (p->d_parent == p2) {
1489 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1490 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1495 for (p = p2; p->d_parent != p; p = p->d_parent) {
1496 if (p->d_parent == p1) {
1497 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1498 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1503 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1504 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1508 void unlock_rename(struct dentry *p1, struct dentry *p2)
1510 mutex_unlock(&p1->d_inode->i_mutex);
1512 mutex_unlock(&p2->d_inode->i_mutex);
1513 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1517 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1518 struct nameidata *nd)
1520 int error = may_create(dir, dentry, nd);
1525 if (!dir->i_op || !dir->i_op->create)
1526 return -EACCES; /* shouldn't it be ENOSYS? */
1529 error = security_inode_create(dir, dentry, mode);
1533 error = dir->i_op->create(dir, dentry, mode, nd);
1535 fsnotify_create(dir, dentry);
1539 int may_open(struct nameidata *nd, int acc_mode, int flag)
1541 struct dentry *dentry = nd->path.dentry;
1542 struct inode *inode = dentry->d_inode;
1548 if (S_ISLNK(inode->i_mode))
1551 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1555 * FIFO's, sockets and device files are special: they don't
1556 * actually live on the filesystem itself, and as such you
1557 * can write to them even if the filesystem is read-only.
1559 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1561 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1562 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1568 error = vfs_permission(nd, acc_mode);
1572 * An append-only file must be opened in append mode for writing.
1574 if (IS_APPEND(inode)) {
1575 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1581 /* O_NOATIME can only be set by the owner or superuser */
1582 if (flag & O_NOATIME)
1583 if (!is_owner_or_cap(inode))
1587 * Ensure there are no outstanding leases on the file.
1589 error = break_lease(inode, flag);
1593 if (flag & O_TRUNC) {
1594 error = get_write_access(inode);
1599 * Refuse to truncate files with mandatory locks held on them.
1601 error = locks_verify_locked(inode);
1605 error = do_truncate(dentry, 0,
1606 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1609 put_write_access(inode);
1613 if (flag & FMODE_WRITE)
1620 * Be careful about ever adding any more callers of this
1621 * function. Its flags must be in the namei format, not
1622 * what get passed to sys_open().
1624 static int __open_namei_create(struct nameidata *nd, struct path *path,
1628 struct dentry *dir = nd->path.dentry;
1630 if (!IS_POSIXACL(dir->d_inode))
1631 mode &= ~current->fs->umask;
1632 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1633 mutex_unlock(&dir->d_inode->i_mutex);
1634 dput(nd->path.dentry);
1635 nd->path.dentry = path->dentry;
1638 /* Don't check for write permission, don't truncate */
1639 return may_open(nd, 0, flag & ~O_TRUNC);
1643 * Note that while the flag value (low two bits) for sys_open means:
1648 * it is changed into
1649 * 00 - no permissions needed
1650 * 01 - read-permission
1651 * 10 - write-permission
1653 * for the internal routines (ie open_namei()/follow_link() etc)
1654 * This is more logical, and also allows the 00 "no perm needed"
1655 * to be used for symlinks (where the permissions are checked
1659 static inline int open_to_namei_flags(int flag)
1661 if ((flag+1) & O_ACCMODE)
1666 static int open_will_write_to_fs(int flag, struct inode *inode)
1669 * We'll never write to the fs underlying
1672 if (special_file(inode->i_mode))
1674 return (flag & O_TRUNC);
1678 * Note that the low bits of the passed in "open_flag"
1679 * are not the same as in the local variable "flag". See
1680 * open_to_namei_flags() for more details.
1682 struct file *do_filp_open(int dfd, const char *pathname,
1683 int open_flag, int mode)
1686 struct nameidata nd;
1687 int acc_mode, error;
1692 int flag = open_to_namei_flags(open_flag);
1694 acc_mode = MAY_OPEN | ACC_MODE(flag);
1696 /* O_TRUNC implies we need access checks for write permissions */
1698 acc_mode |= MAY_WRITE;
1700 /* Allow the LSM permission hook to distinguish append
1701 access from general write access. */
1702 if (flag & O_APPEND)
1703 acc_mode |= MAY_APPEND;
1706 * The simplest case - just a plain lookup.
1708 if (!(flag & O_CREAT)) {
1709 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1712 return ERR_PTR(error);
1717 * Create - we need to know the parent.
1719 error = path_lookup_create(dfd, pathname, LOOKUP_PARENT,
1722 return ERR_PTR(error);
1725 * We have the parent and last component. First of all, check
1726 * that we are not asked to creat(2) an obvious directory - that
1730 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1733 dir = nd.path.dentry;
1734 nd.flags &= ~LOOKUP_PARENT;
1735 mutex_lock(&dir->d_inode->i_mutex);
1736 path.dentry = lookup_hash(&nd);
1737 path.mnt = nd.path.mnt;
1740 error = PTR_ERR(path.dentry);
1741 if (IS_ERR(path.dentry)) {
1742 mutex_unlock(&dir->d_inode->i_mutex);
1746 if (IS_ERR(nd.intent.open.file)) {
1747 error = PTR_ERR(nd.intent.open.file);
1748 goto exit_mutex_unlock;
1751 /* Negative dentry, just create the file */
1752 if (!path.dentry->d_inode) {
1754 * This write is needed to ensure that a
1755 * ro->rw transition does not occur between
1756 * the time when the file is created and when
1757 * a permanent write count is taken through
1758 * the 'struct file' in nameidata_to_filp().
1760 error = mnt_want_write(nd.path.mnt);
1762 goto exit_mutex_unlock;
1763 error = __open_namei_create(&nd, &path, flag, mode);
1765 mnt_drop_write(nd.path.mnt);
1768 filp = nameidata_to_filp(&nd, open_flag);
1769 mnt_drop_write(nd.path.mnt);
1774 * It already exists.
1776 mutex_unlock(&dir->d_inode->i_mutex);
1777 audit_inode(pathname, path.dentry);
1783 if (__follow_mount(&path)) {
1785 if (flag & O_NOFOLLOW)
1790 if (!path.dentry->d_inode)
1792 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1795 path_to_nameidata(&path, &nd);
1797 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1802 * 1. may_open() truncates a file
1803 * 2. a rw->ro mount transition occurs
1804 * 3. nameidata_to_filp() fails due to
1806 * That would be inconsistent, and should
1807 * be avoided. Taking this mnt write here
1808 * ensures that (2) can not occur.
1810 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1812 error = mnt_want_write(nd.path.mnt);
1816 error = may_open(&nd, acc_mode, flag);
1819 mnt_drop_write(nd.path.mnt);
1822 filp = nameidata_to_filp(&nd, open_flag);
1824 * It is now safe to drop the mnt write
1825 * because the filp has had a write taken
1829 mnt_drop_write(nd.path.mnt);
1833 mutex_unlock(&dir->d_inode->i_mutex);
1835 path_put_conditional(&path, &nd);
1837 if (!IS_ERR(nd.intent.open.file))
1838 release_open_intent(&nd);
1840 return ERR_PTR(error);
1844 if (flag & O_NOFOLLOW)
1847 * This is subtle. Instead of calling do_follow_link() we do the
1848 * thing by hands. The reason is that this way we have zero link_count
1849 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1850 * After that we have the parent and last component, i.e.
1851 * we are in the same situation as after the first path_walk().
1852 * Well, almost - if the last component is normal we get its copy
1853 * stored in nd->last.name and we will have to putname() it when we
1854 * are done. Procfs-like symlinks just set LAST_BIND.
1856 nd.flags |= LOOKUP_PARENT;
1857 error = security_inode_follow_link(path.dentry, &nd);
1860 error = __do_follow_link(&path, &nd);
1862 /* Does someone understand code flow here? Or it is only
1863 * me so stupid? Anathema to whoever designed this non-sense
1864 * with "intent.open".
1866 release_open_intent(&nd);
1867 return ERR_PTR(error);
1869 nd.flags &= ~LOOKUP_PARENT;
1870 if (nd.last_type == LAST_BIND)
1873 if (nd.last_type != LAST_NORM)
1875 if (nd.last.name[nd.last.len]) {
1876 __putname(nd.last.name);
1881 __putname(nd.last.name);
1884 dir = nd.path.dentry;
1885 mutex_lock(&dir->d_inode->i_mutex);
1886 path.dentry = lookup_hash(&nd);
1887 path.mnt = nd.path.mnt;
1888 __putname(nd.last.name);
1893 * filp_open - open file and return file pointer
1895 * @filename: path to open
1896 * @flags: open flags as per the open(2) second argument
1897 * @mode: mode for the new file if O_CREAT is set, else ignored
1899 * This is the helper to open a file from kernelspace if you really
1900 * have to. But in generally you should not do this, so please move
1901 * along, nothing to see here..
1903 struct file *filp_open(const char *filename, int flags, int mode)
1905 return do_filp_open(AT_FDCWD, filename, flags, mode);
1907 EXPORT_SYMBOL(filp_open);
1910 * lookup_create - lookup a dentry, creating it if it doesn't exist
1911 * @nd: nameidata info
1912 * @is_dir: directory flag
1914 * Simple function to lookup and return a dentry and create it
1915 * if it doesn't exist. Is SMP-safe.
1917 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1919 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1921 struct dentry *dentry = ERR_PTR(-EEXIST);
1923 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1925 * Yucky last component or no last component at all?
1926 * (foo/., foo/.., /////)
1928 if (nd->last_type != LAST_NORM)
1930 nd->flags &= ~LOOKUP_PARENT;
1931 nd->flags |= LOOKUP_CREATE;
1932 nd->intent.open.flags = O_EXCL;
1935 * Do the final lookup.
1937 dentry = lookup_hash(nd);
1941 if (dentry->d_inode)
1944 * Special case - lookup gave negative, but... we had foo/bar/
1945 * From the vfs_mknod() POV we just have a negative dentry -
1946 * all is fine. Let's be bastards - you had / on the end, you've
1947 * been asking for (non-existent) directory. -ENOENT for you.
1949 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1951 dentry = ERR_PTR(-ENOENT);
1956 dentry = ERR_PTR(-EEXIST);
1960 EXPORT_SYMBOL_GPL(lookup_create);
1962 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1964 int error = may_create(dir, dentry, NULL);
1969 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1972 if (!dir->i_op || !dir->i_op->mknod)
1975 error = devcgroup_inode_mknod(mode, dev);
1979 error = security_inode_mknod(dir, dentry, mode, dev);
1984 error = dir->i_op->mknod(dir, dentry, mode, dev);
1986 fsnotify_create(dir, dentry);
1990 static int may_mknod(mode_t mode)
1992 switch (mode & S_IFMT) {
1998 case 0: /* zero mode translates to S_IFREG */
2007 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
2012 struct dentry *dentry;
2013 struct nameidata nd;
2018 error = user_path_parent(dfd, filename, &nd, &tmp);
2022 dentry = lookup_create(&nd, 0);
2023 if (IS_ERR(dentry)) {
2024 error = PTR_ERR(dentry);
2027 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2028 mode &= ~current->fs->umask;
2029 error = may_mknod(mode);
2032 error = mnt_want_write(nd.path.mnt);
2035 switch (mode & S_IFMT) {
2036 case 0: case S_IFREG:
2037 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2039 case S_IFCHR: case S_IFBLK:
2040 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2041 new_decode_dev(dev));
2043 case S_IFIFO: case S_IFSOCK:
2044 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2047 mnt_drop_write(nd.path.mnt);
2051 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2058 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2060 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2063 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2065 int error = may_create(dir, dentry, NULL);
2070 if (!dir->i_op || !dir->i_op->mkdir)
2073 mode &= (S_IRWXUGO|S_ISVTX);
2074 error = security_inode_mkdir(dir, dentry, mode);
2079 error = dir->i_op->mkdir(dir, dentry, mode);
2081 fsnotify_mkdir(dir, dentry);
2085 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2089 struct dentry *dentry;
2090 struct nameidata nd;
2092 error = user_path_parent(dfd, pathname, &nd, &tmp);
2096 dentry = lookup_create(&nd, 1);
2097 error = PTR_ERR(dentry);
2101 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2102 mode &= ~current->fs->umask;
2103 error = mnt_want_write(nd.path.mnt);
2106 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2107 mnt_drop_write(nd.path.mnt);
2111 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2118 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2120 return sys_mkdirat(AT_FDCWD, pathname, mode);
2124 * We try to drop the dentry early: we should have
2125 * a usage count of 2 if we're the only user of this
2126 * dentry, and if that is true (possibly after pruning
2127 * the dcache), then we drop the dentry now.
2129 * A low-level filesystem can, if it choses, legally
2132 * if (!d_unhashed(dentry))
2135 * if it cannot handle the case of removing a directory
2136 * that is still in use by something else..
2138 void dentry_unhash(struct dentry *dentry)
2141 shrink_dcache_parent(dentry);
2142 spin_lock(&dcache_lock);
2143 spin_lock(&dentry->d_lock);
2144 if (atomic_read(&dentry->d_count) == 2)
2146 spin_unlock(&dentry->d_lock);
2147 spin_unlock(&dcache_lock);
2150 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2152 int error = may_delete(dir, dentry, 1);
2157 if (!dir->i_op || !dir->i_op->rmdir)
2162 mutex_lock(&dentry->d_inode->i_mutex);
2163 dentry_unhash(dentry);
2164 if (d_mountpoint(dentry))
2167 error = security_inode_rmdir(dir, dentry);
2169 error = dir->i_op->rmdir(dir, dentry);
2171 dentry->d_inode->i_flags |= S_DEAD;
2174 mutex_unlock(&dentry->d_inode->i_mutex);
2183 static long do_rmdir(int dfd, const char __user *pathname)
2187 struct dentry *dentry;
2188 struct nameidata nd;
2190 error = user_path_parent(dfd, pathname, &nd, &name);
2194 switch(nd.last_type) {
2205 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2206 dentry = lookup_hash(&nd);
2207 error = PTR_ERR(dentry);
2210 error = mnt_want_write(nd.path.mnt);
2213 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2214 mnt_drop_write(nd.path.mnt);
2218 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2225 asmlinkage long sys_rmdir(const char __user *pathname)
2227 return do_rmdir(AT_FDCWD, pathname);
2230 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2232 int error = may_delete(dir, dentry, 0);
2237 if (!dir->i_op || !dir->i_op->unlink)
2242 mutex_lock(&dentry->d_inode->i_mutex);
2243 if (d_mountpoint(dentry))
2246 error = security_inode_unlink(dir, dentry);
2248 error = dir->i_op->unlink(dir, dentry);
2250 mutex_unlock(&dentry->d_inode->i_mutex);
2252 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2253 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2254 fsnotify_link_count(dentry->d_inode);
2262 * Make sure that the actual truncation of the file will occur outside its
2263 * directory's i_mutex. Truncate can take a long time if there is a lot of
2264 * writeout happening, and we don't want to prevent access to the directory
2265 * while waiting on the I/O.
2267 static long do_unlinkat(int dfd, const char __user *pathname)
2271 struct dentry *dentry;
2272 struct nameidata nd;
2273 struct inode *inode = NULL;
2275 error = user_path_parent(dfd, pathname, &nd, &name);
2280 if (nd.last_type != LAST_NORM)
2282 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2283 dentry = lookup_hash(&nd);
2284 error = PTR_ERR(dentry);
2285 if (!IS_ERR(dentry)) {
2286 /* Why not before? Because we want correct error value */
2287 if (nd.last.name[nd.last.len])
2289 inode = dentry->d_inode;
2291 atomic_inc(&inode->i_count);
2292 error = mnt_want_write(nd.path.mnt);
2295 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2296 mnt_drop_write(nd.path.mnt);
2300 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2302 iput(inode); /* truncate the inode here */
2309 error = !dentry->d_inode ? -ENOENT :
2310 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2314 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2316 if ((flag & ~AT_REMOVEDIR) != 0)
2319 if (flag & AT_REMOVEDIR)
2320 return do_rmdir(dfd, pathname);
2322 return do_unlinkat(dfd, pathname);
2325 asmlinkage long sys_unlink(const char __user *pathname)
2327 return do_unlinkat(AT_FDCWD, pathname);
2330 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2332 int error = may_create(dir, dentry, NULL);
2337 if (!dir->i_op || !dir->i_op->symlink)
2340 error = security_inode_symlink(dir, dentry, oldname);
2345 error = dir->i_op->symlink(dir, dentry, oldname);
2347 fsnotify_create(dir, dentry);
2351 asmlinkage long sys_symlinkat(const char __user *oldname,
2352 int newdfd, const char __user *newname)
2357 struct dentry *dentry;
2358 struct nameidata nd;
2360 from = getname(oldname);
2362 return PTR_ERR(from);
2364 error = user_path_parent(newdfd, newname, &nd, &to);
2368 dentry = lookup_create(&nd, 0);
2369 error = PTR_ERR(dentry);
2373 error = mnt_want_write(nd.path.mnt);
2376 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2377 mnt_drop_write(nd.path.mnt);
2381 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2389 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2391 return sys_symlinkat(oldname, AT_FDCWD, newname);
2394 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2396 struct inode *inode = old_dentry->d_inode;
2402 error = may_create(dir, new_dentry, NULL);
2406 if (dir->i_sb != inode->i_sb)
2410 * A link to an append-only or immutable file cannot be created.
2412 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2414 if (!dir->i_op || !dir->i_op->link)
2416 if (S_ISDIR(inode->i_mode))
2419 error = security_inode_link(old_dentry, dir, new_dentry);
2423 mutex_lock(&inode->i_mutex);
2425 error = dir->i_op->link(old_dentry, dir, new_dentry);
2426 mutex_unlock(&inode->i_mutex);
2428 fsnotify_link(dir, inode, new_dentry);
2433 * Hardlinks are often used in delicate situations. We avoid
2434 * security-related surprises by not following symlinks on the
2437 * We don't follow them on the oldname either to be compatible
2438 * with linux 2.0, and to avoid hard-linking to directories
2439 * and other special files. --ADM
2441 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2442 int newdfd, const char __user *newname,
2445 struct dentry *new_dentry;
2446 struct nameidata nd;
2447 struct path old_path;
2451 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2454 error = user_path_at(olddfd, oldname,
2455 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2460 error = user_path_parent(newdfd, newname, &nd, &to);
2464 if (old_path.mnt != nd.path.mnt)
2466 new_dentry = lookup_create(&nd, 0);
2467 error = PTR_ERR(new_dentry);
2468 if (IS_ERR(new_dentry))
2470 error = mnt_want_write(nd.path.mnt);
2473 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2474 mnt_drop_write(nd.path.mnt);
2478 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2483 path_put(&old_path);
2488 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2490 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2494 * The worst of all namespace operations - renaming directory. "Perverted"
2495 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2497 * a) we can get into loop creation. Check is done in is_subdir().
2498 * b) race potential - two innocent renames can create a loop together.
2499 * That's where 4.4 screws up. Current fix: serialization on
2500 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2502 * c) we have to lock _three_ objects - parents and victim (if it exists).
2503 * And that - after we got ->i_mutex on parents (until then we don't know
2504 * whether the target exists). Solution: try to be smart with locking
2505 * order for inodes. We rely on the fact that tree topology may change
2506 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2507 * move will be locked. Thus we can rank directories by the tree
2508 * (ancestors first) and rank all non-directories after them.
2509 * That works since everybody except rename does "lock parent, lookup,
2510 * lock child" and rename is under ->s_vfs_rename_mutex.
2511 * HOWEVER, it relies on the assumption that any object with ->lookup()
2512 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2513 * we'd better make sure that there's no link(2) for them.
2514 * d) some filesystems don't support opened-but-unlinked directories,
2515 * either because of layout or because they are not ready to deal with
2516 * all cases correctly. The latter will be fixed (taking this sort of
2517 * stuff into VFS), but the former is not going away. Solution: the same
2518 * trick as in rmdir().
2519 * e) conversion from fhandle to dentry may come in the wrong moment - when
2520 * we are removing the target. Solution: we will have to grab ->i_mutex
2521 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2522 * ->i_mutex on parents, which works but leads to some truely excessive
2525 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2526 struct inode *new_dir, struct dentry *new_dentry)
2529 struct inode *target;
2532 * If we are going to change the parent - check write permissions,
2533 * we'll need to flip '..'.
2535 if (new_dir != old_dir) {
2536 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2541 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2545 target = new_dentry->d_inode;
2547 mutex_lock(&target->i_mutex);
2548 dentry_unhash(new_dentry);
2550 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2553 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2556 target->i_flags |= S_DEAD;
2557 mutex_unlock(&target->i_mutex);
2558 if (d_unhashed(new_dentry))
2559 d_rehash(new_dentry);
2563 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2564 d_move(old_dentry,new_dentry);
2568 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2569 struct inode *new_dir, struct dentry *new_dentry)
2571 struct inode *target;
2574 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2579 target = new_dentry->d_inode;
2581 mutex_lock(&target->i_mutex);
2582 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2585 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2587 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2588 d_move(old_dentry, new_dentry);
2591 mutex_unlock(&target->i_mutex);
2596 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2597 struct inode *new_dir, struct dentry *new_dentry)
2600 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2601 const char *old_name;
2603 if (old_dentry->d_inode == new_dentry->d_inode)
2606 error = may_delete(old_dir, old_dentry, is_dir);
2610 if (!new_dentry->d_inode)
2611 error = may_create(new_dir, new_dentry, NULL);
2613 error = may_delete(new_dir, new_dentry, is_dir);
2617 if (!old_dir->i_op || !old_dir->i_op->rename)
2620 DQUOT_INIT(old_dir);
2621 DQUOT_INIT(new_dir);
2623 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2626 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2628 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2630 const char *new_name = old_dentry->d_name.name;
2631 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2632 new_dentry->d_inode, old_dentry);
2634 fsnotify_oldname_free(old_name);
2639 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2640 int newdfd, const char __user *newname)
2642 struct dentry *old_dir, *new_dir;
2643 struct dentry *old_dentry, *new_dentry;
2644 struct dentry *trap;
2645 struct nameidata oldnd, newnd;
2650 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2654 error = user_path_parent(newdfd, newname, &newnd, &to);
2659 if (oldnd.path.mnt != newnd.path.mnt)
2662 old_dir = oldnd.path.dentry;
2664 if (oldnd.last_type != LAST_NORM)
2667 new_dir = newnd.path.dentry;
2668 if (newnd.last_type != LAST_NORM)
2671 trap = lock_rename(new_dir, old_dir);
2673 old_dentry = lookup_hash(&oldnd);
2674 error = PTR_ERR(old_dentry);
2675 if (IS_ERR(old_dentry))
2677 /* source must exist */
2679 if (!old_dentry->d_inode)
2681 /* unless the source is a directory trailing slashes give -ENOTDIR */
2682 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2684 if (oldnd.last.name[oldnd.last.len])
2686 if (newnd.last.name[newnd.last.len])
2689 /* source should not be ancestor of target */
2691 if (old_dentry == trap)
2693 new_dentry = lookup_hash(&newnd);
2694 error = PTR_ERR(new_dentry);
2695 if (IS_ERR(new_dentry))
2697 /* target should not be an ancestor of source */
2699 if (new_dentry == trap)
2702 error = mnt_want_write(oldnd.path.mnt);
2705 error = vfs_rename(old_dir->d_inode, old_dentry,
2706 new_dir->d_inode, new_dentry);
2707 mnt_drop_write(oldnd.path.mnt);
2713 unlock_rename(new_dir, old_dir);
2715 path_put(&newnd.path);
2718 path_put(&oldnd.path);
2724 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2726 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2729 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2733 len = PTR_ERR(link);
2738 if (len > (unsigned) buflen)
2740 if (copy_to_user(buffer, link, len))
2747 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2748 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2749 * using) it for any given inode is up to filesystem.
2751 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2753 struct nameidata nd;
2758 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2760 return PTR_ERR(cookie);
2762 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2763 if (dentry->d_inode->i_op->put_link)
2764 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2768 int vfs_follow_link(struct nameidata *nd, const char *link)
2770 return __vfs_follow_link(nd, link);
2773 /* get the link contents into pagecache */
2774 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2777 struct address_space *mapping = dentry->d_inode->i_mapping;
2778 page = read_mapping_page(mapping, 0, NULL);
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);
2814 int __page_symlink(struct inode *inode, const char *symname, int len,
2817 struct address_space *mapping = inode->i_mapping;
2824 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2825 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2829 kaddr = kmap_atomic(page, KM_USER0);
2830 memcpy(kaddr, symname, len-1);
2831 kunmap_atomic(kaddr, KM_USER0);
2833 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2840 mark_inode_dirty(inode);
2846 int page_symlink(struct inode *inode, const char *symname, int len)
2848 return __page_symlink(inode, symname, len,
2849 mapping_gfp_mask(inode->i_mapping));
2852 const struct inode_operations page_symlink_inode_operations = {
2853 .readlink = generic_readlink,
2854 .follow_link = page_follow_link_light,
2855 .put_link = page_put_link,
2858 EXPORT_SYMBOL(user_path_at);
2859 EXPORT_SYMBOL(follow_down);
2860 EXPORT_SYMBOL(follow_up);
2861 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2862 EXPORT_SYMBOL(getname);
2863 EXPORT_SYMBOL(lock_rename);
2864 EXPORT_SYMBOL(lookup_one_len);
2865 EXPORT_SYMBOL(page_follow_link_light);
2866 EXPORT_SYMBOL(page_put_link);
2867 EXPORT_SYMBOL(page_readlink);
2868 EXPORT_SYMBOL(__page_symlink);
2869 EXPORT_SYMBOL(page_symlink);
2870 EXPORT_SYMBOL(page_symlink_inode_operations);
2871 EXPORT_SYMBOL(path_lookup);
2872 EXPORT_SYMBOL(vfs_path_lookup);
2873 EXPORT_SYMBOL(inode_permission);
2874 EXPORT_SYMBOL(vfs_permission);
2875 EXPORT_SYMBOL(file_permission);
2876 EXPORT_SYMBOL(unlock_rename);
2877 EXPORT_SYMBOL(vfs_create);
2878 EXPORT_SYMBOL(vfs_follow_link);
2879 EXPORT_SYMBOL(vfs_link);
2880 EXPORT_SYMBOL(vfs_mkdir);
2881 EXPORT_SYMBOL(vfs_mknod);
2882 EXPORT_SYMBOL(generic_permission);
2883 EXPORT_SYMBOL(vfs_readlink);
2884 EXPORT_SYMBOL(vfs_rename);
2885 EXPORT_SYMBOL(vfs_rmdir);
2886 EXPORT_SYMBOL(vfs_symlink);
2887 EXPORT_SYMBOL(vfs_unlink);
2888 EXPORT_SYMBOL(dentry_unhash);
2889 EXPORT_SYMBOL(generic_readlink);