4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <asm/namei.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name, struct nameidata *nd);
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
185 umode_t mode = inode->i_mode;
187 if (current->fsuid == inode->i_uid)
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
198 if (in_group_p(inode->i_gid))
203 * If the DACs are ok we don't need any capability check.
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
219 * Searching includes executable on directories, else just read.
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
228 int permission(struct inode *inode, int mask, struct nameidata *nd)
231 struct vfsmount *mnt = NULL;
236 if (mask & MAY_WRITE) {
237 umode_t mode = inode->i_mode;
240 * Nobody gets write access to a read-only fs.
242 if (IS_RDONLY(inode) &&
243 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
247 * Nobody gets write access to an immutable file.
249 if (IS_IMMUTABLE(inode))
253 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
255 * MAY_EXEC on regular files is denied if the fs is mounted
256 * with the "noexec" flag.
258 if (mnt && (mnt->mnt_flags & MNT_NOEXEC))
262 /* Ordinary permission routines do not understand MAY_APPEND. */
263 submask = mask & ~MAY_APPEND;
264 if (inode->i_op && inode->i_op->permission) {
265 retval = inode->i_op->permission(inode, submask, nd);
268 * Exec permission on a regular file is denied if none
269 * of the execute bits are set.
271 * This check should be done by the ->permission()
274 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
275 !(inode->i_mode & S_IXUGO))
279 retval = generic_permission(inode, submask, NULL);
284 return security_inode_permission(inode, mask, nd);
288 * vfs_permission - check for access rights to a given path
289 * @nd: lookup result that describes the path
290 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
292 * Used to check for read/write/execute permissions on a path.
293 * We use "fsuid" for this, letting us set arbitrary permissions
294 * for filesystem access without changing the "normal" uids which
295 * are used for other things.
297 int vfs_permission(struct nameidata *nd, int mask)
299 return permission(nd->path.dentry->d_inode, mask, nd);
303 * file_permission - check for additional access rights to a given file
304 * @file: file to check access rights for
305 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
307 * Used to check for read/write/execute permissions on an already opened
311 * Do not use this function in new code. All access checks should
312 * be done using vfs_permission().
314 int file_permission(struct file *file, int mask)
316 return permission(file->f_path.dentry->d_inode, mask, NULL);
320 * get_write_access() gets write permission for a file.
321 * put_write_access() releases this write permission.
322 * This is used for regular files.
323 * We cannot support write (and maybe mmap read-write shared) accesses and
324 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
325 * can have the following values:
326 * 0: no writers, no VM_DENYWRITE mappings
327 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
328 * > 0: (i_writecount) users are writing to the file.
330 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
331 * except for the cases where we don't hold i_writecount yet. Then we need to
332 * use {get,deny}_write_access() - these functions check the sign and refuse
333 * to do the change if sign is wrong. Exclusion between them is provided by
334 * the inode->i_lock spinlock.
337 int get_write_access(struct inode * inode)
339 spin_lock(&inode->i_lock);
340 if (atomic_read(&inode->i_writecount) < 0) {
341 spin_unlock(&inode->i_lock);
344 atomic_inc(&inode->i_writecount);
345 spin_unlock(&inode->i_lock);
350 int deny_write_access(struct file * file)
352 struct inode *inode = file->f_path.dentry->d_inode;
354 spin_lock(&inode->i_lock);
355 if (atomic_read(&inode->i_writecount) > 0) {
356 spin_unlock(&inode->i_lock);
359 atomic_dec(&inode->i_writecount);
360 spin_unlock(&inode->i_lock);
366 * path_get - get a reference to a path
367 * @path: path to get the reference to
369 * Given a path increment the reference count to the dentry and the vfsmount.
371 void path_get(struct path *path)
376 EXPORT_SYMBOL(path_get);
379 * path_put - put a reference to a path
380 * @path: path to put the reference to
382 * Given a path decrement the reference count to the dentry and the vfsmount.
384 void path_put(struct path *path)
389 EXPORT_SYMBOL(path_put);
392 * release_open_intent - free up open intent resources
393 * @nd: pointer to nameidata
395 void release_open_intent(struct nameidata *nd)
397 if (nd->intent.open.file->f_path.dentry == NULL)
398 put_filp(nd->intent.open.file);
400 fput(nd->intent.open.file);
403 static inline struct dentry *
404 do_revalidate(struct dentry *dentry, struct nameidata *nd)
406 int status = dentry->d_op->d_revalidate(dentry, nd);
407 if (unlikely(status <= 0)) {
409 * The dentry failed validation.
410 * If d_revalidate returned 0 attempt to invalidate
411 * the dentry otherwise d_revalidate is asking us
412 * to return a fail status.
415 if (!d_invalidate(dentry)) {
421 dentry = ERR_PTR(status);
428 * Internal lookup() using the new generic dcache.
431 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
433 struct dentry * dentry = __d_lookup(parent, name);
435 /* lockess __d_lookup may fail due to concurrent d_move()
436 * in some unrelated directory, so try with d_lookup
439 dentry = d_lookup(parent, name);
441 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
442 dentry = do_revalidate(dentry, nd);
448 * Short-cut version of permission(), for calling by
449 * path_walk(), when dcache lock is held. Combines parts
450 * of permission() and generic_permission(), and tests ONLY for
451 * MAY_EXEC permission.
453 * If appropriate, check DAC only. If not appropriate, or
454 * short-cut DAC fails, then call permission() to do more
455 * complete permission check.
457 static int exec_permission_lite(struct inode *inode,
458 struct nameidata *nd)
460 umode_t mode = inode->i_mode;
462 if (inode->i_op && inode->i_op->permission)
465 if (current->fsuid == inode->i_uid)
467 else if (in_group_p(inode->i_gid))
473 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
476 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
479 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
484 return security_inode_permission(inode, MAY_EXEC, nd);
488 * This is called when everything else fails, and we actually have
489 * to go to the low-level filesystem to find out what we should do..
491 * We get the directory semaphore, and after getting that we also
492 * make sure that nobody added the entry to the dcache in the meantime..
495 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
497 struct dentry * result;
498 struct inode *dir = parent->d_inode;
500 mutex_lock(&dir->i_mutex);
502 * First re-do the cached lookup just in case it was created
503 * while we waited for the directory semaphore..
505 * FIXME! This could use version numbering or similar to
506 * avoid unnecessary cache lookups.
508 * The "dcache_lock" is purely to protect the RCU list walker
509 * from concurrent renames at this point (we mustn't get false
510 * negatives from the RCU list walk here, unlike the optimistic
513 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
515 result = d_lookup(parent, name);
517 struct dentry *dentry;
519 /* Don't create child dentry for a dead directory. */
520 result = ERR_PTR(-ENOENT);
524 dentry = d_alloc(parent, name);
525 result = ERR_PTR(-ENOMEM);
527 result = dir->i_op->lookup(dir, dentry, nd);
534 mutex_unlock(&dir->i_mutex);
539 * Uhhuh! Nasty case: the cache was re-populated while
540 * we waited on the semaphore. Need to revalidate.
542 mutex_unlock(&dir->i_mutex);
543 if (result->d_op && result->d_op->d_revalidate) {
544 result = do_revalidate(result, nd);
546 result = ERR_PTR(-ENOENT);
551 static int __emul_lookup_dentry(const char *, struct nameidata *);
554 static __always_inline int
555 walk_init_root(const char *name, struct nameidata *nd)
557 struct fs_struct *fs = current->fs;
559 read_lock(&fs->lock);
560 if (fs->altroot.dentry && !(nd->flags & LOOKUP_NOALT)) {
561 nd->path = fs->altroot;
562 path_get(&fs->altroot);
563 read_unlock(&fs->lock);
564 if (__emul_lookup_dentry(name,nd))
566 read_lock(&fs->lock);
570 read_unlock(&fs->lock);
575 * Wrapper to retry pathname resolution whenever the underlying
576 * file system returns an ESTALE.
578 * Retry the whole path once, forcing real lookup requests
579 * instead of relying on the dcache.
581 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
583 struct path save = nd->path;
586 /* make sure the stuff we saved doesn't go away */
590 result = __link_path_walk(name, nd);
591 if (result == -ESTALE) {
592 /* nd->path had been dropped */
594 dget(nd->path.dentry);
595 mntget(nd->path.mnt);
596 nd->flags |= LOOKUP_REVAL;
597 result = __link_path_walk(name, nd);
605 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
614 if (!walk_init_root(link, nd))
615 /* weird __emul_prefix() stuff did it */
618 res = link_path_walk(link, nd);
620 if (nd->depth || res || nd->last_type!=LAST_NORM)
623 * If it is an iterative symlinks resolution in open_namei() we
624 * have to copy the last component. And all that crap because of
625 * bloody create() on broken symlinks. Furrfu...
628 if (unlikely(!name)) {
632 strcpy(name, nd->last.name);
633 nd->last.name = name;
637 return PTR_ERR(link);
640 static void path_put_conditional(struct path *path, struct nameidata *nd)
643 if (path->mnt != nd->path.mnt)
647 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
649 dput(nd->path.dentry);
650 if (nd->path.mnt != path->mnt)
651 mntput(nd->path.mnt);
652 nd->path.mnt = path->mnt;
653 nd->path.dentry = path->dentry;
656 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
660 struct dentry *dentry = path->dentry;
662 touch_atime(path->mnt, dentry);
663 nd_set_link(nd, NULL);
665 if (path->mnt != nd->path.mnt) {
666 path_to_nameidata(path, nd);
670 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
671 error = PTR_ERR(cookie);
672 if (!IS_ERR(cookie)) {
673 char *s = nd_get_link(nd);
676 error = __vfs_follow_link(nd, s);
677 if (dentry->d_inode->i_op->put_link)
678 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
686 * This limits recursive symlink follows to 8, while
687 * limiting consecutive symlinks to 40.
689 * Without that kind of total limit, nasty chains of consecutive
690 * symlinks can cause almost arbitrarily long lookups.
692 static inline int do_follow_link(struct path *path, struct nameidata *nd)
695 if (current->link_count >= MAX_NESTED_LINKS)
697 if (current->total_link_count >= 40)
699 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
701 err = security_inode_follow_link(path->dentry, nd);
704 current->link_count++;
705 current->total_link_count++;
707 err = __do_follow_link(path, nd);
708 current->link_count--;
712 path_put_conditional(path, nd);
717 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
719 struct vfsmount *parent;
720 struct dentry *mountpoint;
721 spin_lock(&vfsmount_lock);
722 parent=(*mnt)->mnt_parent;
723 if (parent == *mnt) {
724 spin_unlock(&vfsmount_lock);
728 mountpoint=dget((*mnt)->mnt_mountpoint);
729 spin_unlock(&vfsmount_lock);
731 *dentry = mountpoint;
737 /* no need for dcache_lock, as serialization is taken care in
740 static int __follow_mount(struct path *path)
743 while (d_mountpoint(path->dentry)) {
744 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
751 path->dentry = dget(mounted->mnt_root);
757 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
759 while (d_mountpoint(*dentry)) {
760 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
766 *dentry = dget(mounted->mnt_root);
770 /* no need for dcache_lock, as serialization is taken care in
773 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
775 struct vfsmount *mounted;
777 mounted = lookup_mnt(*mnt, *dentry);
782 *dentry = dget(mounted->mnt_root);
788 static __always_inline void follow_dotdot(struct nameidata *nd)
790 struct fs_struct *fs = current->fs;
793 struct vfsmount *parent;
794 struct dentry *old = nd->path.dentry;
796 read_lock(&fs->lock);
797 if (nd->path.dentry == fs->root.dentry &&
798 nd->path.mnt == fs->root.mnt) {
799 read_unlock(&fs->lock);
802 read_unlock(&fs->lock);
803 spin_lock(&dcache_lock);
804 if (nd->path.dentry != nd->path.mnt->mnt_root) {
805 nd->path.dentry = dget(nd->path.dentry->d_parent);
806 spin_unlock(&dcache_lock);
810 spin_unlock(&dcache_lock);
811 spin_lock(&vfsmount_lock);
812 parent = nd->path.mnt->mnt_parent;
813 if (parent == nd->path.mnt) {
814 spin_unlock(&vfsmount_lock);
818 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
819 spin_unlock(&vfsmount_lock);
821 mntput(nd->path.mnt);
822 nd->path.mnt = parent;
824 follow_mount(&nd->path.mnt, &nd->path.dentry);
828 * It's more convoluted than I'd like it to be, but... it's still fairly
829 * small and for now I'd prefer to have fast path as straight as possible.
830 * It _is_ time-critical.
832 static int do_lookup(struct nameidata *nd, struct qstr *name,
835 struct vfsmount *mnt = nd->path.mnt;
836 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
840 if (dentry->d_op && dentry->d_op->d_revalidate)
841 goto need_revalidate;
844 path->dentry = dentry;
845 __follow_mount(path);
849 dentry = real_lookup(nd->path.dentry, name, nd);
855 dentry = do_revalidate(dentry, nd);
863 return PTR_ERR(dentry);
868 * This is the basic name resolution function, turning a pathname into
869 * the final dentry. We expect 'base' to be positive and a directory.
871 * Returns 0 and nd will have valid dentry and mnt on success.
872 * Returns error and drops reference to input namei data on failure.
874 static int __link_path_walk(const char *name, struct nameidata *nd)
879 unsigned int lookup_flags = nd->flags;
886 inode = nd->path.dentry->d_inode;
888 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
890 /* At this point we know we have a real path component. */
896 nd->flags |= LOOKUP_CONTINUE;
897 err = exec_permission_lite(inode, nd);
899 err = vfs_permission(nd, MAY_EXEC);
904 c = *(const unsigned char *)name;
906 hash = init_name_hash();
909 hash = partial_name_hash(c, hash);
910 c = *(const unsigned char *)name;
911 } while (c && (c != '/'));
912 this.len = name - (const char *) this.name;
913 this.hash = end_name_hash(hash);
915 /* remove trailing slashes? */
918 while (*++name == '/');
920 goto last_with_slashes;
923 * "." and ".." are special - ".." especially so because it has
924 * to be able to know about the current root directory and
925 * parent relationships.
927 if (this.name[0] == '.') switch (this.len) {
931 if (this.name[1] != '.')
934 inode = nd->path.dentry->d_inode;
940 * See if the low-level filesystem might want
941 * to use its own hash..
943 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
944 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
949 /* This does the actual lookups.. */
950 err = do_lookup(nd, &this, &next);
955 inode = next.dentry->d_inode;
962 if (inode->i_op->follow_link) {
963 err = do_follow_link(&next, nd);
967 inode = nd->path.dentry->d_inode;
974 path_to_nameidata(&next, nd);
976 if (!inode->i_op->lookup)
979 /* here ends the main loop */
982 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
984 /* Clear LOOKUP_CONTINUE iff it was previously unset */
985 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
986 if (lookup_flags & LOOKUP_PARENT)
988 if (this.name[0] == '.') switch (this.len) {
992 if (this.name[1] != '.')
995 inode = nd->path.dentry->d_inode;
1000 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
1001 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
1006 err = do_lookup(nd, &this, &next);
1009 inode = next.dentry->d_inode;
1010 if ((lookup_flags & LOOKUP_FOLLOW)
1011 && inode && inode->i_op && inode->i_op->follow_link) {
1012 err = do_follow_link(&next, nd);
1015 inode = nd->path.dentry->d_inode;
1017 path_to_nameidata(&next, nd);
1021 if (lookup_flags & LOOKUP_DIRECTORY) {
1023 if (!inode->i_op || !inode->i_op->lookup)
1029 nd->last_type = LAST_NORM;
1030 if (this.name[0] != '.')
1033 nd->last_type = LAST_DOT;
1034 else if (this.len == 2 && this.name[1] == '.')
1035 nd->last_type = LAST_DOTDOT;
1040 * We bypassed the ordinary revalidation routines.
1041 * We may need to check the cached dentry for staleness.
1043 if (nd->path.dentry && nd->path.dentry->d_sb &&
1044 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1046 /* Note: we do not d_invalidate() */
1047 if (!nd->path.dentry->d_op->d_revalidate(
1048 nd->path.dentry, nd))
1054 path_put_conditional(&next, nd);
1057 path_put(&nd->path);
1062 static int path_walk(const char *name, struct nameidata *nd)
1064 current->total_link_count = 0;
1065 return link_path_walk(name, nd);
1069 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1070 * everything is done. Returns 0 and drops input nd, if lookup failed;
1072 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1074 if (path_walk(name, nd))
1075 return 0; /* something went wrong... */
1077 if (!nd->path.dentry->d_inode ||
1078 S_ISDIR(nd->path.dentry->d_inode->i_mode)) {
1079 struct path old_path = nd->path;
1080 struct qstr last = nd->last;
1081 int last_type = nd->last_type;
1082 struct fs_struct *fs = current->fs;
1085 * NAME was not found in alternate root or it's a directory.
1086 * Try to find it in the normal root:
1088 nd->last_type = LAST_ROOT;
1089 read_lock(&fs->lock);
1090 nd->path = fs->root;
1091 path_get(&fs->root);
1092 read_unlock(&fs->lock);
1093 if (path_walk(name, nd) == 0) {
1094 if (nd->path.dentry->d_inode) {
1095 path_put(&old_path);
1098 path_put(&nd->path);
1100 nd->path = old_path;
1102 nd->last_type = last_type;
1107 void set_fs_altroot(void)
1109 char *emul = __emul_prefix();
1110 struct nameidata nd;
1111 struct path path = {}, old_path;
1113 struct fs_struct *fs = current->fs;
1117 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1121 write_lock(&fs->lock);
1122 old_path = fs->altroot;
1124 write_unlock(&fs->lock);
1125 if (old_path.dentry)
1126 path_put(&old_path);
1129 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1130 static int do_path_lookup(int dfd, const char *name,
1131 unsigned int flags, struct nameidata *nd)
1136 struct fs_struct *fs = current->fs;
1138 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1143 read_lock(&fs->lock);
1144 if (fs->altroot.dentry && !(nd->flags & LOOKUP_NOALT)) {
1145 nd->path = fs->altroot;
1146 path_get(&fs->altroot);
1147 read_unlock(&fs->lock);
1148 if (__emul_lookup_dentry(name,nd))
1149 goto out; /* found in altroot */
1150 read_lock(&fs->lock);
1152 nd->path = fs->root;
1153 path_get(&fs->root);
1154 read_unlock(&fs->lock);
1155 } else if (dfd == AT_FDCWD) {
1156 read_lock(&fs->lock);
1159 read_unlock(&fs->lock);
1161 struct dentry *dentry;
1163 file = fget_light(dfd, &fput_needed);
1168 dentry = file->f_path.dentry;
1171 if (!S_ISDIR(dentry->d_inode->i_mode))
1174 retval = file_permission(file, MAY_EXEC);
1178 nd->path = file->f_path;
1179 path_get(&file->f_path);
1181 fput_light(file, fput_needed);
1184 retval = path_walk(name, nd);
1186 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1187 nd->path.dentry->d_inode))
1188 audit_inode(name, nd->path.dentry);
1193 fput_light(file, fput_needed);
1197 int path_lookup(const char *name, unsigned int flags,
1198 struct nameidata *nd)
1200 return do_path_lookup(AT_FDCWD, name, flags, nd);
1204 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1205 * @dentry: pointer to dentry of the base directory
1206 * @mnt: pointer to vfs mount of the base directory
1207 * @name: pointer to file name
1208 * @flags: lookup flags
1209 * @nd: pointer to nameidata
1211 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1212 const char *name, unsigned int flags,
1213 struct nameidata *nd)
1217 /* same as do_path_lookup */
1218 nd->last_type = LAST_ROOT;
1222 nd->path.mnt = mntget(mnt);
1223 nd->path.dentry = dget(dentry);
1225 retval = path_walk(name, nd);
1226 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1227 nd->path.dentry->d_inode))
1228 audit_inode(name, nd->path.dentry);
1234 static int __path_lookup_intent_open(int dfd, const char *name,
1235 unsigned int lookup_flags, struct nameidata *nd,
1236 int open_flags, int create_mode)
1238 struct file *filp = get_empty_filp();
1243 nd->intent.open.file = filp;
1244 nd->intent.open.flags = open_flags;
1245 nd->intent.open.create_mode = create_mode;
1246 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1247 if (IS_ERR(nd->intent.open.file)) {
1249 err = PTR_ERR(nd->intent.open.file);
1250 path_put(&nd->path);
1252 } else if (err != 0)
1253 release_open_intent(nd);
1258 * path_lookup_open - lookup a file path with open intent
1259 * @dfd: the directory to use as base, or AT_FDCWD
1260 * @name: pointer to file name
1261 * @lookup_flags: lookup intent flags
1262 * @nd: pointer to nameidata
1263 * @open_flags: open intent flags
1265 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1266 struct nameidata *nd, int open_flags)
1268 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1273 * path_lookup_create - lookup a file path with open + create intent
1274 * @dfd: the directory to use as base, or AT_FDCWD
1275 * @name: pointer to file name
1276 * @lookup_flags: lookup intent flags
1277 * @nd: pointer to nameidata
1278 * @open_flags: open intent flags
1279 * @create_mode: create intent flags
1281 static int path_lookup_create(int dfd, const char *name,
1282 unsigned int lookup_flags, struct nameidata *nd,
1283 int open_flags, int create_mode)
1285 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1286 nd, open_flags, create_mode);
1289 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1290 struct nameidata *nd, int open_flags)
1292 char *tmp = getname(name);
1293 int err = PTR_ERR(tmp);
1296 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1302 static struct dentry *__lookup_hash(struct qstr *name,
1303 struct dentry *base, struct nameidata *nd)
1305 struct dentry *dentry;
1306 struct inode *inode;
1309 inode = base->d_inode;
1312 * See if the low-level filesystem might want
1313 * to use its own hash..
1315 if (base->d_op && base->d_op->d_hash) {
1316 err = base->d_op->d_hash(base, name);
1317 dentry = ERR_PTR(err);
1322 dentry = cached_lookup(base, name, nd);
1326 /* Don't create child dentry for a dead directory. */
1327 dentry = ERR_PTR(-ENOENT);
1328 if (IS_DEADDIR(inode))
1331 new = d_alloc(base, name);
1332 dentry = ERR_PTR(-ENOMEM);
1335 dentry = inode->i_op->lookup(inode, new, nd);
1346 * Restricted form of lookup. Doesn't follow links, single-component only,
1347 * needs parent already locked. Doesn't follow mounts.
1350 static struct dentry *lookup_hash(struct nameidata *nd)
1354 err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
1356 return ERR_PTR(err);
1357 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1360 static int __lookup_one_len(const char *name, struct qstr *this,
1361 struct dentry *base, int len)
1371 hash = init_name_hash();
1373 c = *(const unsigned char *)name++;
1374 if (c == '/' || c == '\0')
1376 hash = partial_name_hash(c, hash);
1378 this->hash = end_name_hash(hash);
1383 * lookup_one_len - filesystem helper to lookup single pathname component
1384 * @name: pathname component to lookup
1385 * @base: base directory to lookup from
1386 * @len: maximum length @len should be interpreted to
1388 * Note that this routine is purely a helper for filesystem usage and should
1389 * not be called by generic code. Also note that by using this function the
1390 * nameidata argument is passed to the filesystem methods and a filesystem
1391 * using this helper needs to be prepared for that.
1393 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1398 err = __lookup_one_len(name, &this, base, len);
1400 return ERR_PTR(err);
1402 err = permission(base->d_inode, MAY_EXEC, NULL);
1404 return ERR_PTR(err);
1405 return __lookup_hash(&this, base, NULL);
1409 * lookup_one_noperm - bad hack for sysfs
1410 * @name: pathname component to lookup
1411 * @base: base directory to lookup from
1413 * This is a variant of lookup_one_len that doesn't perform any permission
1414 * checks. It's a horrible hack to work around the braindead sysfs
1415 * architecture and should not be used anywhere else.
1417 * DON'T USE THIS FUNCTION EVER, thanks.
1419 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1424 err = __lookup_one_len(name, &this, base, strlen(name));
1426 return ERR_PTR(err);
1427 return __lookup_hash(&this, base, NULL);
1430 int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1431 struct nameidata *nd)
1433 char *tmp = getname(name);
1434 int err = PTR_ERR(tmp);
1437 err = do_path_lookup(dfd, tmp, flags, nd);
1443 int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1445 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1449 * It's inline, so penalty for filesystems that don't use sticky bit is
1452 static inline int check_sticky(struct inode *dir, struct inode *inode)
1454 if (!(dir->i_mode & S_ISVTX))
1456 if (inode->i_uid == current->fsuid)
1458 if (dir->i_uid == current->fsuid)
1460 return !capable(CAP_FOWNER);
1464 * Check whether we can remove a link victim from directory dir, check
1465 * whether the type of victim is right.
1466 * 1. We can't do it if dir is read-only (done in permission())
1467 * 2. We should have write and exec permissions on dir
1468 * 3. We can't remove anything from append-only dir
1469 * 4. We can't do anything with immutable dir (done in permission())
1470 * 5. If the sticky bit on dir is set we should either
1471 * a. be owner of dir, or
1472 * b. be owner of victim, or
1473 * c. have CAP_FOWNER capability
1474 * 6. If the victim is append-only or immutable we can't do antyhing with
1475 * links pointing to it.
1476 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1477 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1478 * 9. We can't remove a root or mountpoint.
1479 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1480 * nfs_async_unlink().
1482 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1486 if (!victim->d_inode)
1489 BUG_ON(victim->d_parent->d_inode != dir);
1490 audit_inode_child(victim->d_name.name, victim, dir);
1492 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1497 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1498 IS_IMMUTABLE(victim->d_inode))
1501 if (!S_ISDIR(victim->d_inode->i_mode))
1503 if (IS_ROOT(victim))
1505 } else if (S_ISDIR(victim->d_inode->i_mode))
1507 if (IS_DEADDIR(dir))
1509 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1514 /* Check whether we can create an object with dentry child in directory
1516 * 1. We can't do it if child already exists (open has special treatment for
1517 * this case, but since we are inlined it's OK)
1518 * 2. We can't do it if dir is read-only (done in permission())
1519 * 3. We should have write and exec permissions on dir
1520 * 4. We can't do it if dir is immutable (done in permission())
1522 static inline int may_create(struct inode *dir, struct dentry *child,
1523 struct nameidata *nd)
1527 if (IS_DEADDIR(dir))
1529 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1533 * O_DIRECTORY translates into forcing a directory lookup.
1535 static inline int lookup_flags(unsigned int f)
1537 unsigned long retval = LOOKUP_FOLLOW;
1540 retval &= ~LOOKUP_FOLLOW;
1542 if (f & O_DIRECTORY)
1543 retval |= LOOKUP_DIRECTORY;
1549 * p1 and p2 should be directories on the same fs.
1551 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1556 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1560 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1562 for (p = p1; p->d_parent != p; p = p->d_parent) {
1563 if (p->d_parent == p2) {
1564 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1565 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1570 for (p = p2; p->d_parent != p; p = p->d_parent) {
1571 if (p->d_parent == p1) {
1572 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1573 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1578 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1579 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1583 void unlock_rename(struct dentry *p1, struct dentry *p2)
1585 mutex_unlock(&p1->d_inode->i_mutex);
1587 mutex_unlock(&p2->d_inode->i_mutex);
1588 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1592 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1593 struct nameidata *nd)
1595 int error = may_create(dir, dentry, nd);
1600 if (!dir->i_op || !dir->i_op->create)
1601 return -EACCES; /* shouldn't it be ENOSYS? */
1604 error = security_inode_create(dir, dentry, mode);
1608 error = dir->i_op->create(dir, dentry, mode, nd);
1610 fsnotify_create(dir, dentry);
1614 int may_open(struct nameidata *nd, int acc_mode, int flag)
1616 struct dentry *dentry = nd->path.dentry;
1617 struct inode *inode = dentry->d_inode;
1623 if (S_ISLNK(inode->i_mode))
1626 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1630 * FIFO's, sockets and device files are special: they don't
1631 * actually live on the filesystem itself, and as such you
1632 * can write to them even if the filesystem is read-only.
1634 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1636 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1637 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1641 } else if (IS_RDONLY(inode) && (acc_mode & MAY_WRITE))
1644 error = vfs_permission(nd, acc_mode);
1648 * An append-only file must be opened in append mode for writing.
1650 if (IS_APPEND(inode)) {
1651 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1657 /* O_NOATIME can only be set by the owner or superuser */
1658 if (flag & O_NOATIME)
1659 if (!is_owner_or_cap(inode))
1663 * Ensure there are no outstanding leases on the file.
1665 error = break_lease(inode, flag);
1669 if (flag & O_TRUNC) {
1670 error = get_write_access(inode);
1675 * Refuse to truncate files with mandatory locks held on them.
1677 error = locks_verify_locked(inode);
1681 error = do_truncate(dentry, 0,
1682 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1685 put_write_access(inode);
1689 if (flag & FMODE_WRITE)
1695 static int open_namei_create(struct nameidata *nd, struct path *path,
1699 struct dentry *dir = nd->path.dentry;
1701 if (!IS_POSIXACL(dir->d_inode))
1702 mode &= ~current->fs->umask;
1703 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1704 mutex_unlock(&dir->d_inode->i_mutex);
1705 dput(nd->path.dentry);
1706 nd->path.dentry = path->dentry;
1709 /* Don't check for write permission, don't truncate */
1710 return may_open(nd, 0, flag & ~O_TRUNC);
1716 * namei for open - this is in fact almost the whole open-routine.
1718 * Note that the low bits of "flag" aren't the same as in the open
1719 * system call - they are 00 - no permissions needed
1720 * 01 - read permission needed
1721 * 10 - write permission needed
1722 * 11 - read/write permissions needed
1723 * which is a lot more logical, and also allows the "no perm" needed
1724 * for symlinks (where the permissions are checked later).
1727 int open_namei(int dfd, const char *pathname, int flag,
1728 int mode, struct nameidata *nd)
1730 int acc_mode, error;
1735 acc_mode = ACC_MODE(flag);
1737 /* O_TRUNC implies we need access checks for write permissions */
1739 acc_mode |= MAY_WRITE;
1741 /* Allow the LSM permission hook to distinguish append
1742 access from general write access. */
1743 if (flag & O_APPEND)
1744 acc_mode |= MAY_APPEND;
1747 * The simplest case - just a plain lookup.
1749 if (!(flag & O_CREAT)) {
1750 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1758 * Create - we need to know the parent.
1760 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1765 * We have the parent and last component. First of all, check
1766 * that we are not asked to creat(2) an obvious directory - that
1770 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1773 dir = nd->path.dentry;
1774 nd->flags &= ~LOOKUP_PARENT;
1775 mutex_lock(&dir->d_inode->i_mutex);
1776 path.dentry = lookup_hash(nd);
1777 path.mnt = nd->path.mnt;
1780 error = PTR_ERR(path.dentry);
1781 if (IS_ERR(path.dentry)) {
1782 mutex_unlock(&dir->d_inode->i_mutex);
1786 if (IS_ERR(nd->intent.open.file)) {
1787 mutex_unlock(&dir->d_inode->i_mutex);
1788 error = PTR_ERR(nd->intent.open.file);
1792 /* Negative dentry, just create the file */
1793 if (!path.dentry->d_inode) {
1794 error = open_namei_create(nd, &path, flag, mode);
1801 * It already exists.
1803 mutex_unlock(&dir->d_inode->i_mutex);
1804 audit_inode(pathname, path.dentry);
1810 if (__follow_mount(&path)) {
1812 if (flag & O_NOFOLLOW)
1817 if (!path.dentry->d_inode)
1819 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1822 path_to_nameidata(&path, nd);
1824 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1827 error = may_open(nd, acc_mode, flag);
1833 path_put_conditional(&path, nd);
1835 if (!IS_ERR(nd->intent.open.file))
1836 release_open_intent(nd);
1837 path_put(&nd->path);
1842 if (flag & O_NOFOLLOW)
1845 * This is subtle. Instead of calling do_follow_link() we do the
1846 * thing by hands. The reason is that this way we have zero link_count
1847 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1848 * After that we have the parent and last component, i.e.
1849 * we are in the same situation as after the first path_walk().
1850 * Well, almost - if the last component is normal we get its copy
1851 * stored in nd->last.name and we will have to putname() it when we
1852 * are done. Procfs-like symlinks just set LAST_BIND.
1854 nd->flags |= LOOKUP_PARENT;
1855 error = security_inode_follow_link(path.dentry, nd);
1858 error = __do_follow_link(&path, nd);
1860 /* Does someone understand code flow here? Or it is only
1861 * me so stupid? Anathema to whoever designed this non-sense
1862 * with "intent.open".
1864 release_open_intent(nd);
1867 nd->flags &= ~LOOKUP_PARENT;
1868 if (nd->last_type == LAST_BIND)
1871 if (nd->last_type != LAST_NORM)
1873 if (nd->last.name[nd->last.len]) {
1874 __putname(nd->last.name);
1879 __putname(nd->last.name);
1882 dir = nd->path.dentry;
1883 mutex_lock(&dir->d_inode->i_mutex);
1884 path.dentry = lookup_hash(nd);
1885 path.mnt = nd->path.mnt;
1886 __putname(nd->last.name);
1891 * lookup_create - lookup a dentry, creating it if it doesn't exist
1892 * @nd: nameidata info
1893 * @is_dir: directory flag
1895 * Simple function to lookup and return a dentry and create it
1896 * if it doesn't exist. Is SMP-safe.
1898 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1900 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1902 struct dentry *dentry = ERR_PTR(-EEXIST);
1904 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1906 * Yucky last component or no last component at all?
1907 * (foo/., foo/.., /////)
1909 if (nd->last_type != LAST_NORM)
1911 nd->flags &= ~LOOKUP_PARENT;
1912 nd->flags |= LOOKUP_CREATE;
1913 nd->intent.open.flags = O_EXCL;
1916 * Do the final lookup.
1918 dentry = lookup_hash(nd);
1922 if (dentry->d_inode)
1925 * Special case - lookup gave negative, but... we had foo/bar/
1926 * From the vfs_mknod() POV we just have a negative dentry -
1927 * all is fine. Let's be bastards - you had / on the end, you've
1928 * been asking for (non-existent) directory. -ENOENT for you.
1930 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1932 dentry = ERR_PTR(-ENOENT);
1937 dentry = ERR_PTR(-EEXIST);
1941 EXPORT_SYMBOL_GPL(lookup_create);
1943 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1945 int error = may_create(dir, dentry, NULL);
1950 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1953 if (!dir->i_op || !dir->i_op->mknod)
1956 error = security_inode_mknod(dir, dentry, mode, dev);
1961 error = dir->i_op->mknod(dir, dentry, mode, dev);
1963 fsnotify_create(dir, dentry);
1967 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1972 struct dentry * dentry;
1973 struct nameidata nd;
1977 tmp = getname(filename);
1979 return PTR_ERR(tmp);
1981 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1984 dentry = lookup_create(&nd, 0);
1985 error = PTR_ERR(dentry);
1987 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1988 mode &= ~current->fs->umask;
1989 if (!IS_ERR(dentry)) {
1990 switch (mode & S_IFMT) {
1991 case 0: case S_IFREG:
1992 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1994 case S_IFCHR: case S_IFBLK:
1995 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1996 new_decode_dev(dev));
1998 case S_IFIFO: case S_IFSOCK:
1999 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2009 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2017 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2019 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2022 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2024 int error = may_create(dir, dentry, NULL);
2029 if (!dir->i_op || !dir->i_op->mkdir)
2032 mode &= (S_IRWXUGO|S_ISVTX);
2033 error = security_inode_mkdir(dir, dentry, mode);
2038 error = dir->i_op->mkdir(dir, dentry, mode);
2040 fsnotify_mkdir(dir, dentry);
2044 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2048 struct dentry *dentry;
2049 struct nameidata nd;
2051 tmp = getname(pathname);
2052 error = PTR_ERR(tmp);
2056 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2059 dentry = lookup_create(&nd, 1);
2060 error = PTR_ERR(dentry);
2064 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2065 mode &= ~current->fs->umask;
2066 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2069 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2077 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2079 return sys_mkdirat(AT_FDCWD, pathname, mode);
2083 * We try to drop the dentry early: we should have
2084 * a usage count of 2 if we're the only user of this
2085 * dentry, and if that is true (possibly after pruning
2086 * the dcache), then we drop the dentry now.
2088 * A low-level filesystem can, if it choses, legally
2091 * if (!d_unhashed(dentry))
2094 * if it cannot handle the case of removing a directory
2095 * that is still in use by something else..
2097 void dentry_unhash(struct dentry *dentry)
2100 shrink_dcache_parent(dentry);
2101 spin_lock(&dcache_lock);
2102 spin_lock(&dentry->d_lock);
2103 if (atomic_read(&dentry->d_count) == 2)
2105 spin_unlock(&dentry->d_lock);
2106 spin_unlock(&dcache_lock);
2109 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2111 int error = may_delete(dir, dentry, 1);
2116 if (!dir->i_op || !dir->i_op->rmdir)
2121 mutex_lock(&dentry->d_inode->i_mutex);
2122 dentry_unhash(dentry);
2123 if (d_mountpoint(dentry))
2126 error = security_inode_rmdir(dir, dentry);
2128 error = dir->i_op->rmdir(dir, dentry);
2130 dentry->d_inode->i_flags |= S_DEAD;
2133 mutex_unlock(&dentry->d_inode->i_mutex);
2142 static long do_rmdir(int dfd, const char __user *pathname)
2146 struct dentry *dentry;
2147 struct nameidata nd;
2149 name = getname(pathname);
2151 return PTR_ERR(name);
2153 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2157 switch(nd.last_type) {
2168 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2169 dentry = lookup_hash(&nd);
2170 error = PTR_ERR(dentry);
2173 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2176 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2184 asmlinkage long sys_rmdir(const char __user *pathname)
2186 return do_rmdir(AT_FDCWD, pathname);
2189 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2191 int error = may_delete(dir, dentry, 0);
2196 if (!dir->i_op || !dir->i_op->unlink)
2201 mutex_lock(&dentry->d_inode->i_mutex);
2202 if (d_mountpoint(dentry))
2205 error = security_inode_unlink(dir, dentry);
2207 error = dir->i_op->unlink(dir, dentry);
2209 mutex_unlock(&dentry->d_inode->i_mutex);
2211 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2212 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2213 fsnotify_link_count(dentry->d_inode);
2221 * Make sure that the actual truncation of the file will occur outside its
2222 * directory's i_mutex. Truncate can take a long time if there is a lot of
2223 * writeout happening, and we don't want to prevent access to the directory
2224 * while waiting on the I/O.
2226 static long do_unlinkat(int dfd, const char __user *pathname)
2230 struct dentry *dentry;
2231 struct nameidata nd;
2232 struct inode *inode = NULL;
2234 name = getname(pathname);
2236 return PTR_ERR(name);
2238 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2242 if (nd.last_type != LAST_NORM)
2244 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2245 dentry = lookup_hash(&nd);
2246 error = PTR_ERR(dentry);
2247 if (!IS_ERR(dentry)) {
2248 /* Why not before? Because we want correct error value */
2249 if (nd.last.name[nd.last.len])
2251 inode = dentry->d_inode;
2253 atomic_inc(&inode->i_count);
2254 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2258 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2260 iput(inode); /* truncate the inode here */
2268 error = !dentry->d_inode ? -ENOENT :
2269 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2273 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2275 if ((flag & ~AT_REMOVEDIR) != 0)
2278 if (flag & AT_REMOVEDIR)
2279 return do_rmdir(dfd, pathname);
2281 return do_unlinkat(dfd, pathname);
2284 asmlinkage long sys_unlink(const char __user *pathname)
2286 return do_unlinkat(AT_FDCWD, pathname);
2289 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2291 int error = may_create(dir, dentry, NULL);
2296 if (!dir->i_op || !dir->i_op->symlink)
2299 error = security_inode_symlink(dir, dentry, oldname);
2304 error = dir->i_op->symlink(dir, dentry, oldname);
2306 fsnotify_create(dir, dentry);
2310 asmlinkage long sys_symlinkat(const char __user *oldname,
2311 int newdfd, const char __user *newname)
2316 struct dentry *dentry;
2317 struct nameidata nd;
2319 from = getname(oldname);
2321 return PTR_ERR(from);
2322 to = getname(newname);
2323 error = PTR_ERR(to);
2327 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2330 dentry = lookup_create(&nd, 0);
2331 error = PTR_ERR(dentry);
2335 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from, S_IALLUGO);
2338 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2347 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2349 return sys_symlinkat(oldname, AT_FDCWD, newname);
2352 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2354 struct inode *inode = old_dentry->d_inode;
2360 error = may_create(dir, new_dentry, NULL);
2364 if (dir->i_sb != inode->i_sb)
2368 * A link to an append-only or immutable file cannot be created.
2370 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2372 if (!dir->i_op || !dir->i_op->link)
2374 if (S_ISDIR(old_dentry->d_inode->i_mode))
2377 error = security_inode_link(old_dentry, dir, new_dentry);
2381 mutex_lock(&old_dentry->d_inode->i_mutex);
2383 error = dir->i_op->link(old_dentry, dir, new_dentry);
2384 mutex_unlock(&old_dentry->d_inode->i_mutex);
2386 fsnotify_link(dir, old_dentry->d_inode, new_dentry);
2391 * Hardlinks are often used in delicate situations. We avoid
2392 * security-related surprises by not following symlinks on the
2395 * We don't follow them on the oldname either to be compatible
2396 * with linux 2.0, and to avoid hard-linking to directories
2397 * and other special files. --ADM
2399 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2400 int newdfd, const char __user *newname,
2403 struct dentry *new_dentry;
2404 struct nameidata nd, old_nd;
2408 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2411 to = getname(newname);
2415 error = __user_walk_fd(olddfd, oldname,
2416 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2420 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2424 if (old_nd.path.mnt != nd.path.mnt)
2426 new_dentry = lookup_create(&nd, 0);
2427 error = PTR_ERR(new_dentry);
2428 if (IS_ERR(new_dentry))
2430 error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
2433 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2437 path_put(&old_nd.path);
2444 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2446 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2450 * The worst of all namespace operations - renaming directory. "Perverted"
2451 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2453 * a) we can get into loop creation. Check is done in is_subdir().
2454 * b) race potential - two innocent renames can create a loop together.
2455 * That's where 4.4 screws up. Current fix: serialization on
2456 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2458 * c) we have to lock _three_ objects - parents and victim (if it exists).
2459 * And that - after we got ->i_mutex on parents (until then we don't know
2460 * whether the target exists). Solution: try to be smart with locking
2461 * order for inodes. We rely on the fact that tree topology may change
2462 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2463 * move will be locked. Thus we can rank directories by the tree
2464 * (ancestors first) and rank all non-directories after them.
2465 * That works since everybody except rename does "lock parent, lookup,
2466 * lock child" and rename is under ->s_vfs_rename_mutex.
2467 * HOWEVER, it relies on the assumption that any object with ->lookup()
2468 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2469 * we'd better make sure that there's no link(2) for them.
2470 * d) some filesystems don't support opened-but-unlinked directories,
2471 * either because of layout or because they are not ready to deal with
2472 * all cases correctly. The latter will be fixed (taking this sort of
2473 * stuff into VFS), but the former is not going away. Solution: the same
2474 * trick as in rmdir().
2475 * e) conversion from fhandle to dentry may come in the wrong moment - when
2476 * we are removing the target. Solution: we will have to grab ->i_mutex
2477 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2478 * ->i_mutex on parents, which works but leads to some truely excessive
2481 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2482 struct inode *new_dir, struct dentry *new_dentry)
2485 struct inode *target;
2488 * If we are going to change the parent - check write permissions,
2489 * we'll need to flip '..'.
2491 if (new_dir != old_dir) {
2492 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2497 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2501 target = new_dentry->d_inode;
2503 mutex_lock(&target->i_mutex);
2504 dentry_unhash(new_dentry);
2506 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2509 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2512 target->i_flags |= S_DEAD;
2513 mutex_unlock(&target->i_mutex);
2514 if (d_unhashed(new_dentry))
2515 d_rehash(new_dentry);
2519 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2520 d_move(old_dentry,new_dentry);
2524 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2525 struct inode *new_dir, struct dentry *new_dentry)
2527 struct inode *target;
2530 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2535 target = new_dentry->d_inode;
2537 mutex_lock(&target->i_mutex);
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);
2543 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2544 d_move(old_dentry, new_dentry);
2547 mutex_unlock(&target->i_mutex);
2552 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2553 struct inode *new_dir, struct dentry *new_dentry)
2556 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2557 const char *old_name;
2559 if (old_dentry->d_inode == new_dentry->d_inode)
2562 error = may_delete(old_dir, old_dentry, is_dir);
2566 if (!new_dentry->d_inode)
2567 error = may_create(new_dir, new_dentry, NULL);
2569 error = may_delete(new_dir, new_dentry, is_dir);
2573 if (!old_dir->i_op || !old_dir->i_op->rename)
2576 DQUOT_INIT(old_dir);
2577 DQUOT_INIT(new_dir);
2579 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2582 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2584 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2586 const char *new_name = old_dentry->d_name.name;
2587 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2588 new_dentry->d_inode, old_dentry);
2590 fsnotify_oldname_free(old_name);
2595 static int do_rename(int olddfd, const char *oldname,
2596 int newdfd, const char *newname)
2599 struct dentry * old_dir, * new_dir;
2600 struct dentry * old_dentry, *new_dentry;
2601 struct dentry * trap;
2602 struct nameidata oldnd, newnd;
2604 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2608 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2613 if (oldnd.path.mnt != newnd.path.mnt)
2616 old_dir = oldnd.path.dentry;
2618 if (oldnd.last_type != LAST_NORM)
2621 new_dir = newnd.path.dentry;
2622 if (newnd.last_type != LAST_NORM)
2625 trap = lock_rename(new_dir, old_dir);
2627 old_dentry = lookup_hash(&oldnd);
2628 error = PTR_ERR(old_dentry);
2629 if (IS_ERR(old_dentry))
2631 /* source must exist */
2633 if (!old_dentry->d_inode)
2635 /* unless the source is a directory trailing slashes give -ENOTDIR */
2636 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2638 if (oldnd.last.name[oldnd.last.len])
2640 if (newnd.last.name[newnd.last.len])
2643 /* source should not be ancestor of target */
2645 if (old_dentry == trap)
2647 new_dentry = lookup_hash(&newnd);
2648 error = PTR_ERR(new_dentry);
2649 if (IS_ERR(new_dentry))
2651 /* target should not be an ancestor of source */
2653 if (new_dentry == trap)
2656 error = vfs_rename(old_dir->d_inode, old_dentry,
2657 new_dir->d_inode, new_dentry);
2663 unlock_rename(new_dir, old_dir);
2665 path_put(&newnd.path);
2667 path_put(&oldnd.path);
2672 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2673 int newdfd, const char __user *newname)
2679 from = getname(oldname);
2681 return PTR_ERR(from);
2682 to = getname(newname);
2683 error = PTR_ERR(to);
2685 error = do_rename(olddfd, from, newdfd, to);
2692 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2694 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2697 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2701 len = PTR_ERR(link);
2706 if (len > (unsigned) buflen)
2708 if (copy_to_user(buffer, link, len))
2715 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2716 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2717 * using) it for any given inode is up to filesystem.
2719 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2721 struct nameidata nd;
2725 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2726 if (!IS_ERR(cookie)) {
2727 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2728 if (dentry->d_inode->i_op->put_link)
2729 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2730 cookie = ERR_PTR(res);
2732 return PTR_ERR(cookie);
2735 int vfs_follow_link(struct nameidata *nd, const char *link)
2737 return __vfs_follow_link(nd, link);
2740 /* get the link contents into pagecache */
2741 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2744 struct address_space *mapping = dentry->d_inode->i_mapping;
2745 page = read_mapping_page(mapping, 0, NULL);
2752 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2754 struct page *page = NULL;
2755 char *s = page_getlink(dentry, &page);
2756 int res = vfs_readlink(dentry,buffer,buflen,s);
2759 page_cache_release(page);
2764 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2766 struct page *page = NULL;
2767 nd_set_link(nd, page_getlink(dentry, &page));
2771 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2773 struct page *page = cookie;
2777 page_cache_release(page);
2781 int __page_symlink(struct inode *inode, const char *symname, int len,
2784 struct address_space *mapping = inode->i_mapping;
2791 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2792 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2796 kaddr = kmap_atomic(page, KM_USER0);
2797 memcpy(kaddr, symname, len-1);
2798 kunmap_atomic(kaddr, KM_USER0);
2800 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2807 mark_inode_dirty(inode);
2813 int page_symlink(struct inode *inode, const char *symname, int len)
2815 return __page_symlink(inode, symname, len,
2816 mapping_gfp_mask(inode->i_mapping));
2819 const struct inode_operations page_symlink_inode_operations = {
2820 .readlink = generic_readlink,
2821 .follow_link = page_follow_link_light,
2822 .put_link = page_put_link,
2825 EXPORT_SYMBOL(__user_walk);
2826 EXPORT_SYMBOL(__user_walk_fd);
2827 EXPORT_SYMBOL(follow_down);
2828 EXPORT_SYMBOL(follow_up);
2829 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2830 EXPORT_SYMBOL(getname);
2831 EXPORT_SYMBOL(lock_rename);
2832 EXPORT_SYMBOL(lookup_one_len);
2833 EXPORT_SYMBOL(page_follow_link_light);
2834 EXPORT_SYMBOL(page_put_link);
2835 EXPORT_SYMBOL(page_readlink);
2836 EXPORT_SYMBOL(__page_symlink);
2837 EXPORT_SYMBOL(page_symlink);
2838 EXPORT_SYMBOL(page_symlink_inode_operations);
2839 EXPORT_SYMBOL(path_lookup);
2840 EXPORT_SYMBOL(vfs_path_lookup);
2841 EXPORT_SYMBOL(permission);
2842 EXPORT_SYMBOL(vfs_permission);
2843 EXPORT_SYMBOL(file_permission);
2844 EXPORT_SYMBOL(unlock_rename);
2845 EXPORT_SYMBOL(vfs_create);
2846 EXPORT_SYMBOL(vfs_follow_link);
2847 EXPORT_SYMBOL(vfs_link);
2848 EXPORT_SYMBOL(vfs_mkdir);
2849 EXPORT_SYMBOL(vfs_mknod);
2850 EXPORT_SYMBOL(generic_permission);
2851 EXPORT_SYMBOL(vfs_readlink);
2852 EXPORT_SYMBOL(vfs_rename);
2853 EXPORT_SYMBOL(vfs_rmdir);
2854 EXPORT_SYMBOL(vfs_symlink);
2855 EXPORT_SYMBOL(vfs_unlink);
2856 EXPORT_SYMBOL(dentry_unhash);
2857 EXPORT_SYMBOL(generic_readlink);