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_put - put a reference to a path
367 * @path: path to put the reference to
369 * Given a path decrement the reference count to the dentry and the vfsmount.
371 void path_put(struct path *path)
376 EXPORT_SYMBOL(path_put);
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
382 void release_open_intent(struct nameidata *nd)
384 if (nd->intent.open.file->f_path.dentry == NULL)
385 put_filp(nd->intent.open.file);
387 fput(nd->intent.open.file);
390 static inline struct dentry *
391 do_revalidate(struct dentry *dentry, struct nameidata *nd)
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
402 if (!d_invalidate(dentry)) {
408 dentry = ERR_PTR(status);
415 * Internal lookup() using the new generic dcache.
418 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
420 struct dentry * dentry = __d_lookup(parent, name);
422 /* lockess __d_lookup may fail due to concurrent d_move()
423 * in some unrelated directory, so try with d_lookup
426 dentry = d_lookup(parent, name);
428 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
429 dentry = do_revalidate(dentry, nd);
435 * Short-cut version of permission(), for calling by
436 * path_walk(), when dcache lock is held. Combines parts
437 * of permission() and generic_permission(), and tests ONLY for
438 * MAY_EXEC permission.
440 * If appropriate, check DAC only. If not appropriate, or
441 * short-cut DAC fails, then call permission() to do more
442 * complete permission check.
444 static int exec_permission_lite(struct inode *inode,
445 struct nameidata *nd)
447 umode_t mode = inode->i_mode;
449 if (inode->i_op && inode->i_op->permission)
452 if (current->fsuid == inode->i_uid)
454 else if (in_group_p(inode->i_gid))
460 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
463 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
466 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
471 return security_inode_permission(inode, MAY_EXEC, nd);
475 * This is called when everything else fails, and we actually have
476 * to go to the low-level filesystem to find out what we should do..
478 * We get the directory semaphore, and after getting that we also
479 * make sure that nobody added the entry to the dcache in the meantime..
482 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
484 struct dentry * result;
485 struct inode *dir = parent->d_inode;
487 mutex_lock(&dir->i_mutex);
489 * First re-do the cached lookup just in case it was created
490 * while we waited for the directory semaphore..
492 * FIXME! This could use version numbering or similar to
493 * avoid unnecessary cache lookups.
495 * The "dcache_lock" is purely to protect the RCU list walker
496 * from concurrent renames at this point (we mustn't get false
497 * negatives from the RCU list walk here, unlike the optimistic
500 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
502 result = d_lookup(parent, name);
504 struct dentry * dentry = d_alloc(parent, name);
505 result = ERR_PTR(-ENOMEM);
507 result = dir->i_op->lookup(dir, dentry, nd);
513 mutex_unlock(&dir->i_mutex);
518 * Uhhuh! Nasty case: the cache was re-populated while
519 * we waited on the semaphore. Need to revalidate.
521 mutex_unlock(&dir->i_mutex);
522 if (result->d_op && result->d_op->d_revalidate) {
523 result = do_revalidate(result, nd);
525 result = ERR_PTR(-ENOENT);
530 static int __emul_lookup_dentry(const char *, struct nameidata *);
533 static __always_inline int
534 walk_init_root(const char *name, struct nameidata *nd)
536 struct fs_struct *fs = current->fs;
538 read_lock(&fs->lock);
539 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
540 nd->path.mnt = mntget(fs->altrootmnt);
541 nd->path.dentry = dget(fs->altroot);
542 read_unlock(&fs->lock);
543 if (__emul_lookup_dentry(name,nd))
545 read_lock(&fs->lock);
547 nd->path.mnt = mntget(fs->rootmnt);
548 nd->path.dentry = dget(fs->root);
549 read_unlock(&fs->lock);
553 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
562 if (!walk_init_root(link, nd))
563 /* weird __emul_prefix() stuff did it */
566 res = link_path_walk(link, nd);
568 if (nd->depth || res || nd->last_type!=LAST_NORM)
571 * If it is an iterative symlinks resolution in open_namei() we
572 * have to copy the last component. And all that crap because of
573 * bloody create() on broken symlinks. Furrfu...
576 if (unlikely(!name)) {
580 strcpy(name, nd->last.name);
581 nd->last.name = name;
585 return PTR_ERR(link);
588 static void path_put_conditional(struct path *path, struct nameidata *nd)
591 if (path->mnt != nd->path.mnt)
595 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
597 dput(nd->path.dentry);
598 if (nd->path.mnt != path->mnt)
599 mntput(nd->path.mnt);
600 nd->path.mnt = path->mnt;
601 nd->path.dentry = path->dentry;
604 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
608 struct dentry *dentry = path->dentry;
610 touch_atime(path->mnt, dentry);
611 nd_set_link(nd, NULL);
613 if (path->mnt != nd->path.mnt) {
614 path_to_nameidata(path, nd);
618 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
619 error = PTR_ERR(cookie);
620 if (!IS_ERR(cookie)) {
621 char *s = nd_get_link(nd);
624 error = __vfs_follow_link(nd, s);
625 if (dentry->d_inode->i_op->put_link)
626 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
634 * This limits recursive symlink follows to 8, while
635 * limiting consecutive symlinks to 40.
637 * Without that kind of total limit, nasty chains of consecutive
638 * symlinks can cause almost arbitrarily long lookups.
640 static inline int do_follow_link(struct path *path, struct nameidata *nd)
643 if (current->link_count >= MAX_NESTED_LINKS)
645 if (current->total_link_count >= 40)
647 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
649 err = security_inode_follow_link(path->dentry, nd);
652 current->link_count++;
653 current->total_link_count++;
655 err = __do_follow_link(path, nd);
656 current->link_count--;
660 path_put_conditional(path, nd);
665 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
667 struct vfsmount *parent;
668 struct dentry *mountpoint;
669 spin_lock(&vfsmount_lock);
670 parent=(*mnt)->mnt_parent;
671 if (parent == *mnt) {
672 spin_unlock(&vfsmount_lock);
676 mountpoint=dget((*mnt)->mnt_mountpoint);
677 spin_unlock(&vfsmount_lock);
679 *dentry = mountpoint;
685 /* no need for dcache_lock, as serialization is taken care in
688 static int __follow_mount(struct path *path)
691 while (d_mountpoint(path->dentry)) {
692 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
699 path->dentry = dget(mounted->mnt_root);
705 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
707 while (d_mountpoint(*dentry)) {
708 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
714 *dentry = dget(mounted->mnt_root);
718 /* no need for dcache_lock, as serialization is taken care in
721 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
723 struct vfsmount *mounted;
725 mounted = lookup_mnt(*mnt, *dentry);
730 *dentry = dget(mounted->mnt_root);
736 static __always_inline void follow_dotdot(struct nameidata *nd)
738 struct fs_struct *fs = current->fs;
741 struct vfsmount *parent;
742 struct dentry *old = nd->path.dentry;
744 read_lock(&fs->lock);
745 if (nd->path.dentry == fs->root &&
746 nd->path.mnt == fs->rootmnt) {
747 read_unlock(&fs->lock);
750 read_unlock(&fs->lock);
751 spin_lock(&dcache_lock);
752 if (nd->path.dentry != nd->path.mnt->mnt_root) {
753 nd->path.dentry = dget(nd->path.dentry->d_parent);
754 spin_unlock(&dcache_lock);
758 spin_unlock(&dcache_lock);
759 spin_lock(&vfsmount_lock);
760 parent = nd->path.mnt->mnt_parent;
761 if (parent == nd->path.mnt) {
762 spin_unlock(&vfsmount_lock);
766 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
767 spin_unlock(&vfsmount_lock);
769 mntput(nd->path.mnt);
770 nd->path.mnt = parent;
772 follow_mount(&nd->path.mnt, &nd->path.dentry);
776 * It's more convoluted than I'd like it to be, but... it's still fairly
777 * small and for now I'd prefer to have fast path as straight as possible.
778 * It _is_ time-critical.
780 static int do_lookup(struct nameidata *nd, struct qstr *name,
783 struct vfsmount *mnt = nd->path.mnt;
784 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
788 if (dentry->d_op && dentry->d_op->d_revalidate)
789 goto need_revalidate;
792 path->dentry = dentry;
793 __follow_mount(path);
797 dentry = real_lookup(nd->path.dentry, name, nd);
803 dentry = do_revalidate(dentry, nd);
811 return PTR_ERR(dentry);
816 * This is the basic name resolution function, turning a pathname into
817 * the final dentry. We expect 'base' to be positive and a directory.
819 * Returns 0 and nd will have valid dentry and mnt on success.
820 * Returns error and drops reference to input namei data on failure.
822 static int __link_path_walk(const char *name, struct nameidata *nd)
827 unsigned int lookup_flags = nd->flags;
834 inode = nd->path.dentry->d_inode;
836 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
838 /* At this point we know we have a real path component. */
844 nd->flags |= LOOKUP_CONTINUE;
845 err = exec_permission_lite(inode, nd);
847 err = vfs_permission(nd, MAY_EXEC);
852 c = *(const unsigned char *)name;
854 hash = init_name_hash();
857 hash = partial_name_hash(c, hash);
858 c = *(const unsigned char *)name;
859 } while (c && (c != '/'));
860 this.len = name - (const char *) this.name;
861 this.hash = end_name_hash(hash);
863 /* remove trailing slashes? */
866 while (*++name == '/');
868 goto last_with_slashes;
871 * "." and ".." are special - ".." especially so because it has
872 * to be able to know about the current root directory and
873 * parent relationships.
875 if (this.name[0] == '.') switch (this.len) {
879 if (this.name[1] != '.')
882 inode = nd->path.dentry->d_inode;
888 * See if the low-level filesystem might want
889 * to use its own hash..
891 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
892 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
897 /* This does the actual lookups.. */
898 err = do_lookup(nd, &this, &next);
903 inode = next.dentry->d_inode;
910 if (inode->i_op->follow_link) {
911 err = do_follow_link(&next, nd);
915 inode = nd->path.dentry->d_inode;
922 path_to_nameidata(&next, nd);
924 if (!inode->i_op->lookup)
927 /* here ends the main loop */
930 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
932 /* Clear LOOKUP_CONTINUE iff it was previously unset */
933 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
934 if (lookup_flags & LOOKUP_PARENT)
936 if (this.name[0] == '.') switch (this.len) {
940 if (this.name[1] != '.')
943 inode = nd->path.dentry->d_inode;
948 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
949 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
954 err = do_lookup(nd, &this, &next);
957 inode = next.dentry->d_inode;
958 if ((lookup_flags & LOOKUP_FOLLOW)
959 && inode && inode->i_op && inode->i_op->follow_link) {
960 err = do_follow_link(&next, nd);
963 inode = nd->path.dentry->d_inode;
965 path_to_nameidata(&next, nd);
969 if (lookup_flags & LOOKUP_DIRECTORY) {
971 if (!inode->i_op || !inode->i_op->lookup)
977 nd->last_type = LAST_NORM;
978 if (this.name[0] != '.')
981 nd->last_type = LAST_DOT;
982 else if (this.len == 2 && this.name[1] == '.')
983 nd->last_type = LAST_DOTDOT;
988 * We bypassed the ordinary revalidation routines.
989 * We may need to check the cached dentry for staleness.
991 if (nd->path.dentry && nd->path.dentry->d_sb &&
992 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
994 /* Note: we do not d_invalidate() */
995 if (!nd->path.dentry->d_op->d_revalidate(
996 nd->path.dentry, nd))
1002 path_put_conditional(&next, nd);
1005 path_put(&nd->path);
1011 * Wrapper to retry pathname resolution whenever the underlying
1012 * file system returns an ESTALE.
1014 * Retry the whole path once, forcing real lookup requests
1015 * instead of relying on the dcache.
1017 static int link_path_walk(const char *name, struct nameidata *nd)
1019 struct nameidata save = *nd;
1022 /* make sure the stuff we saved doesn't go away */
1023 dget(save.path.dentry);
1024 mntget(save.path.mnt);
1026 result = __link_path_walk(name, nd);
1027 if (result == -ESTALE) {
1029 dget(nd->path.dentry);
1030 mntget(nd->path.mnt);
1031 nd->flags |= LOOKUP_REVAL;
1032 result = __link_path_walk(name, nd);
1035 path_put(&save.path);
1040 static int path_walk(const char *name, struct nameidata *nd)
1042 current->total_link_count = 0;
1043 return link_path_walk(name, nd);
1047 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1048 * everything is done. Returns 0 and drops input nd, if lookup failed;
1050 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1052 if (path_walk(name, nd))
1053 return 0; /* something went wrong... */
1055 if (!nd->path.dentry->d_inode ||
1056 S_ISDIR(nd->path.dentry->d_inode->i_mode)) {
1057 struct path old_path = nd->path;
1058 struct qstr last = nd->last;
1059 int last_type = nd->last_type;
1060 struct fs_struct *fs = current->fs;
1063 * NAME was not found in alternate root or it's a directory.
1064 * Try to find it in the normal root:
1066 nd->last_type = LAST_ROOT;
1067 read_lock(&fs->lock);
1068 nd->path.mnt = mntget(fs->rootmnt);
1069 nd->path.dentry = dget(fs->root);
1070 read_unlock(&fs->lock);
1071 if (path_walk(name, nd) == 0) {
1072 if (nd->path.dentry->d_inode) {
1073 path_put(&old_path);
1076 path_put(&nd->path);
1078 nd->path = old_path;
1080 nd->last_type = last_type;
1085 void set_fs_altroot(void)
1087 char *emul = __emul_prefix();
1088 struct nameidata nd;
1089 struct vfsmount *mnt = NULL, *oldmnt;
1090 struct dentry *dentry = NULL, *olddentry;
1092 struct fs_struct *fs = current->fs;
1096 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1099 dentry = nd.path.dentry;
1102 write_lock(&fs->lock);
1103 oldmnt = fs->altrootmnt;
1104 olddentry = fs->altroot;
1105 fs->altrootmnt = mnt;
1106 fs->altroot = dentry;
1107 write_unlock(&fs->lock);
1114 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1115 static int do_path_lookup(int dfd, const char *name,
1116 unsigned int flags, struct nameidata *nd)
1121 struct fs_struct *fs = current->fs;
1123 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1128 read_lock(&fs->lock);
1129 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1130 nd->path.mnt = mntget(fs->altrootmnt);
1131 nd->path.dentry = dget(fs->altroot);
1132 read_unlock(&fs->lock);
1133 if (__emul_lookup_dentry(name,nd))
1134 goto out; /* found in altroot */
1135 read_lock(&fs->lock);
1137 nd->path.mnt = mntget(fs->rootmnt);
1138 nd->path.dentry = dget(fs->root);
1139 read_unlock(&fs->lock);
1140 } else if (dfd == AT_FDCWD) {
1141 read_lock(&fs->lock);
1142 nd->path.mnt = mntget(fs->pwdmnt);
1143 nd->path.dentry = dget(fs->pwd);
1144 read_unlock(&fs->lock);
1146 struct dentry *dentry;
1148 file = fget_light(dfd, &fput_needed);
1153 dentry = file->f_path.dentry;
1156 if (!S_ISDIR(dentry->d_inode->i_mode))
1159 retval = file_permission(file, MAY_EXEC);
1163 nd->path.mnt = mntget(file->f_path.mnt);
1164 nd->path.dentry = dget(dentry);
1166 fput_light(file, fput_needed);
1169 retval = path_walk(name, nd);
1171 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1172 nd->path.dentry->d_inode))
1173 audit_inode(name, nd->path.dentry);
1178 fput_light(file, fput_needed);
1182 int path_lookup(const char *name, unsigned int flags,
1183 struct nameidata *nd)
1185 return do_path_lookup(AT_FDCWD, name, flags, nd);
1189 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1190 * @dentry: pointer to dentry of the base directory
1191 * @mnt: pointer to vfs mount of the base directory
1192 * @name: pointer to file name
1193 * @flags: lookup flags
1194 * @nd: pointer to nameidata
1196 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1197 const char *name, unsigned int flags,
1198 struct nameidata *nd)
1202 /* same as do_path_lookup */
1203 nd->last_type = LAST_ROOT;
1207 nd->path.mnt = mntget(mnt);
1208 nd->path.dentry = dget(dentry);
1210 retval = path_walk(name, nd);
1211 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1212 nd->path.dentry->d_inode))
1213 audit_inode(name, nd->path.dentry);
1219 static int __path_lookup_intent_open(int dfd, const char *name,
1220 unsigned int lookup_flags, struct nameidata *nd,
1221 int open_flags, int create_mode)
1223 struct file *filp = get_empty_filp();
1228 nd->intent.open.file = filp;
1229 nd->intent.open.flags = open_flags;
1230 nd->intent.open.create_mode = create_mode;
1231 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1232 if (IS_ERR(nd->intent.open.file)) {
1234 err = PTR_ERR(nd->intent.open.file);
1235 path_put(&nd->path);
1237 } else if (err != 0)
1238 release_open_intent(nd);
1243 * path_lookup_open - lookup a file path with open intent
1244 * @dfd: the directory to use as base, or AT_FDCWD
1245 * @name: pointer to file name
1246 * @lookup_flags: lookup intent flags
1247 * @nd: pointer to nameidata
1248 * @open_flags: open intent flags
1250 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1251 struct nameidata *nd, int open_flags)
1253 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1258 * path_lookup_create - lookup a file path with open + create 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
1264 * @create_mode: create intent flags
1266 static int path_lookup_create(int dfd, const char *name,
1267 unsigned int lookup_flags, struct nameidata *nd,
1268 int open_flags, int create_mode)
1270 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1271 nd, open_flags, create_mode);
1274 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1275 struct nameidata *nd, int open_flags)
1277 char *tmp = getname(name);
1278 int err = PTR_ERR(tmp);
1281 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1287 static struct dentry *__lookup_hash(struct qstr *name,
1288 struct dentry *base, struct nameidata *nd)
1290 struct dentry *dentry;
1291 struct inode *inode;
1294 inode = base->d_inode;
1297 * See if the low-level filesystem might want
1298 * to use its own hash..
1300 if (base->d_op && base->d_op->d_hash) {
1301 err = base->d_op->d_hash(base, name);
1302 dentry = ERR_PTR(err);
1307 dentry = cached_lookup(base, name, nd);
1309 struct dentry *new = d_alloc(base, name);
1310 dentry = ERR_PTR(-ENOMEM);
1313 dentry = inode->i_op->lookup(inode, new, nd);
1324 * Restricted form of lookup. Doesn't follow links, single-component only,
1325 * needs parent already locked. Doesn't follow mounts.
1328 static struct dentry *lookup_hash(struct nameidata *nd)
1332 err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
1334 return ERR_PTR(err);
1335 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1338 static int __lookup_one_len(const char *name, struct qstr *this,
1339 struct dentry *base, int len)
1349 hash = init_name_hash();
1351 c = *(const unsigned char *)name++;
1352 if (c == '/' || c == '\0')
1354 hash = partial_name_hash(c, hash);
1356 this->hash = end_name_hash(hash);
1361 * lookup_one_len: filesystem helper to lookup single pathname component
1362 * @name: pathname component to lookup
1363 * @base: base directory to lookup from
1364 * @len: maximum length @len should be interpreted to
1366 * Note that this routine is purely a helper for filesystem useage and should
1367 * not be called by generic code. Also note that by using this function to
1368 * nameidata argument is passed to the filesystem methods and a filesystem
1369 * using this helper needs to be prepared for that.
1371 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1376 err = __lookup_one_len(name, &this, base, len);
1378 return ERR_PTR(err);
1380 err = permission(base->d_inode, MAY_EXEC, NULL);
1382 return ERR_PTR(err);
1383 return __lookup_hash(&this, base, NULL);
1387 * lookup_one_noperm - bad hack for sysfs
1388 * @name: pathname component to lookup
1389 * @base: base directory to lookup from
1391 * This is a variant of lookup_one_len that doesn't perform any permission
1392 * checks. It's a horrible hack to work around the braindead sysfs
1393 * architecture and should not be used anywhere else.
1395 * DON'T USE THIS FUNCTION EVER, thanks.
1397 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1402 err = __lookup_one_len(name, &this, base, strlen(name));
1404 return ERR_PTR(err);
1405 return __lookup_hash(&this, base, NULL);
1408 int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1409 struct nameidata *nd)
1411 char *tmp = getname(name);
1412 int err = PTR_ERR(tmp);
1415 err = do_path_lookup(dfd, tmp, flags, nd);
1421 int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1423 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1427 * It's inline, so penalty for filesystems that don't use sticky bit is
1430 static inline int check_sticky(struct inode *dir, struct inode *inode)
1432 if (!(dir->i_mode & S_ISVTX))
1434 if (inode->i_uid == current->fsuid)
1436 if (dir->i_uid == current->fsuid)
1438 return !capable(CAP_FOWNER);
1442 * Check whether we can remove a link victim from directory dir, check
1443 * whether the type of victim is right.
1444 * 1. We can't do it if dir is read-only (done in permission())
1445 * 2. We should have write and exec permissions on dir
1446 * 3. We can't remove anything from append-only dir
1447 * 4. We can't do anything with immutable dir (done in permission())
1448 * 5. If the sticky bit on dir is set we should either
1449 * a. be owner of dir, or
1450 * b. be owner of victim, or
1451 * c. have CAP_FOWNER capability
1452 * 6. If the victim is append-only or immutable we can't do antyhing with
1453 * links pointing to it.
1454 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1455 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1456 * 9. We can't remove a root or mountpoint.
1457 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1458 * nfs_async_unlink().
1460 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1464 if (!victim->d_inode)
1467 BUG_ON(victim->d_parent->d_inode != dir);
1468 audit_inode_child(victim->d_name.name, victim, dir);
1470 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1475 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1476 IS_IMMUTABLE(victim->d_inode))
1479 if (!S_ISDIR(victim->d_inode->i_mode))
1481 if (IS_ROOT(victim))
1483 } else if (S_ISDIR(victim->d_inode->i_mode))
1485 if (IS_DEADDIR(dir))
1487 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1492 /* Check whether we can create an object with dentry child in directory
1494 * 1. We can't do it if child already exists (open has special treatment for
1495 * this case, but since we are inlined it's OK)
1496 * 2. We can't do it if dir is read-only (done in permission())
1497 * 3. We should have write and exec permissions on dir
1498 * 4. We can't do it if dir is immutable (done in permission())
1500 static inline int may_create(struct inode *dir, struct dentry *child,
1501 struct nameidata *nd)
1505 if (IS_DEADDIR(dir))
1507 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1511 * O_DIRECTORY translates into forcing a directory lookup.
1513 static inline int lookup_flags(unsigned int f)
1515 unsigned long retval = LOOKUP_FOLLOW;
1518 retval &= ~LOOKUP_FOLLOW;
1520 if (f & O_DIRECTORY)
1521 retval |= LOOKUP_DIRECTORY;
1527 * p1 and p2 should be directories on the same fs.
1529 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1534 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1538 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1540 for (p = p1; p->d_parent != p; p = p->d_parent) {
1541 if (p->d_parent == p2) {
1542 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1543 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1548 for (p = p2; p->d_parent != p; p = p->d_parent) {
1549 if (p->d_parent == p1) {
1550 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1551 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1556 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1557 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1561 void unlock_rename(struct dentry *p1, struct dentry *p2)
1563 mutex_unlock(&p1->d_inode->i_mutex);
1565 mutex_unlock(&p2->d_inode->i_mutex);
1566 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1570 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1571 struct nameidata *nd)
1573 int error = may_create(dir, dentry, nd);
1578 if (!dir->i_op || !dir->i_op->create)
1579 return -EACCES; /* shouldn't it be ENOSYS? */
1582 error = security_inode_create(dir, dentry, mode);
1586 error = dir->i_op->create(dir, dentry, mode, nd);
1588 fsnotify_create(dir, dentry);
1592 int may_open(struct nameidata *nd, int acc_mode, int flag)
1594 struct dentry *dentry = nd->path.dentry;
1595 struct inode *inode = dentry->d_inode;
1601 if (S_ISLNK(inode->i_mode))
1604 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1608 * FIFO's, sockets and device files are special: they don't
1609 * actually live on the filesystem itself, and as such you
1610 * can write to them even if the filesystem is read-only.
1612 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1614 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1615 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1619 } else if (IS_RDONLY(inode) && (acc_mode & MAY_WRITE))
1622 error = vfs_permission(nd, acc_mode);
1626 * An append-only file must be opened in append mode for writing.
1628 if (IS_APPEND(inode)) {
1629 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1635 /* O_NOATIME can only be set by the owner or superuser */
1636 if (flag & O_NOATIME)
1637 if (!is_owner_or_cap(inode))
1641 * Ensure there are no outstanding leases on the file.
1643 error = break_lease(inode, flag);
1647 if (flag & O_TRUNC) {
1648 error = get_write_access(inode);
1653 * Refuse to truncate files with mandatory locks held on them.
1655 error = locks_verify_locked(inode);
1659 error = do_truncate(dentry, 0,
1660 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1663 put_write_access(inode);
1667 if (flag & FMODE_WRITE)
1673 static int open_namei_create(struct nameidata *nd, struct path *path,
1677 struct dentry *dir = nd->path.dentry;
1679 if (!IS_POSIXACL(dir->d_inode))
1680 mode &= ~current->fs->umask;
1681 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1682 mutex_unlock(&dir->d_inode->i_mutex);
1683 dput(nd->path.dentry);
1684 nd->path.dentry = path->dentry;
1687 /* Don't check for write permission, don't truncate */
1688 return may_open(nd, 0, flag & ~O_TRUNC);
1694 * namei for open - this is in fact almost the whole open-routine.
1696 * Note that the low bits of "flag" aren't the same as in the open
1697 * system call - they are 00 - no permissions needed
1698 * 01 - read permission needed
1699 * 10 - write permission needed
1700 * 11 - read/write permissions needed
1701 * which is a lot more logical, and also allows the "no perm" needed
1702 * for symlinks (where the permissions are checked later).
1705 int open_namei(int dfd, const char *pathname, int flag,
1706 int mode, struct nameidata *nd)
1708 int acc_mode, error;
1713 acc_mode = ACC_MODE(flag);
1715 /* O_TRUNC implies we need access checks for write permissions */
1717 acc_mode |= MAY_WRITE;
1719 /* Allow the LSM permission hook to distinguish append
1720 access from general write access. */
1721 if (flag & O_APPEND)
1722 acc_mode |= MAY_APPEND;
1725 * The simplest case - just a plain lookup.
1727 if (!(flag & O_CREAT)) {
1728 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1736 * Create - we need to know the parent.
1738 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1743 * We have the parent and last component. First of all, check
1744 * that we are not asked to creat(2) an obvious directory - that
1748 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1751 dir = nd->path.dentry;
1752 nd->flags &= ~LOOKUP_PARENT;
1753 mutex_lock(&dir->d_inode->i_mutex);
1754 path.dentry = lookup_hash(nd);
1755 path.mnt = nd->path.mnt;
1758 error = PTR_ERR(path.dentry);
1759 if (IS_ERR(path.dentry)) {
1760 mutex_unlock(&dir->d_inode->i_mutex);
1764 if (IS_ERR(nd->intent.open.file)) {
1765 mutex_unlock(&dir->d_inode->i_mutex);
1766 error = PTR_ERR(nd->intent.open.file);
1770 /* Negative dentry, just create the file */
1771 if (!path.dentry->d_inode) {
1772 error = open_namei_create(nd, &path, flag, mode);
1779 * It already exists.
1781 mutex_unlock(&dir->d_inode->i_mutex);
1782 audit_inode(pathname, path.dentry);
1788 if (__follow_mount(&path)) {
1790 if (flag & O_NOFOLLOW)
1795 if (!path.dentry->d_inode)
1797 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1800 path_to_nameidata(&path, nd);
1802 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1805 error = may_open(nd, acc_mode, flag);
1811 path_put_conditional(&path, nd);
1813 if (!IS_ERR(nd->intent.open.file))
1814 release_open_intent(nd);
1815 path_put(&nd->path);
1820 if (flag & O_NOFOLLOW)
1823 * This is subtle. Instead of calling do_follow_link() we do the
1824 * thing by hands. The reason is that this way we have zero link_count
1825 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1826 * After that we have the parent and last component, i.e.
1827 * we are in the same situation as after the first path_walk().
1828 * Well, almost - if the last component is normal we get its copy
1829 * stored in nd->last.name and we will have to putname() it when we
1830 * are done. Procfs-like symlinks just set LAST_BIND.
1832 nd->flags |= LOOKUP_PARENT;
1833 error = security_inode_follow_link(path.dentry, nd);
1836 error = __do_follow_link(&path, nd);
1838 /* Does someone understand code flow here? Or it is only
1839 * me so stupid? Anathema to whoever designed this non-sense
1840 * with "intent.open".
1842 release_open_intent(nd);
1845 nd->flags &= ~LOOKUP_PARENT;
1846 if (nd->last_type == LAST_BIND)
1849 if (nd->last_type != LAST_NORM)
1851 if (nd->last.name[nd->last.len]) {
1852 __putname(nd->last.name);
1857 __putname(nd->last.name);
1860 dir = nd->path.dentry;
1861 mutex_lock(&dir->d_inode->i_mutex);
1862 path.dentry = lookup_hash(nd);
1863 path.mnt = nd->path.mnt;
1864 __putname(nd->last.name);
1869 * lookup_create - lookup a dentry, creating it if it doesn't exist
1870 * @nd: nameidata info
1871 * @is_dir: directory flag
1873 * Simple function to lookup and return a dentry and create it
1874 * if it doesn't exist. Is SMP-safe.
1876 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1878 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1880 struct dentry *dentry = ERR_PTR(-EEXIST);
1882 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1884 * Yucky last component or no last component at all?
1885 * (foo/., foo/.., /////)
1887 if (nd->last_type != LAST_NORM)
1889 nd->flags &= ~LOOKUP_PARENT;
1890 nd->flags |= LOOKUP_CREATE;
1891 nd->intent.open.flags = O_EXCL;
1894 * Do the final lookup.
1896 dentry = lookup_hash(nd);
1901 * Special case - lookup gave negative, but... we had foo/bar/
1902 * From the vfs_mknod() POV we just have a negative dentry -
1903 * all is fine. Let's be bastards - you had / on the end, you've
1904 * been asking for (non-existent) directory. -ENOENT for you.
1906 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1911 dentry = ERR_PTR(-ENOENT);
1915 EXPORT_SYMBOL_GPL(lookup_create);
1917 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1919 int error = may_create(dir, dentry, NULL);
1924 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1927 if (!dir->i_op || !dir->i_op->mknod)
1930 error = security_inode_mknod(dir, dentry, mode, dev);
1935 error = dir->i_op->mknod(dir, dentry, mode, dev);
1937 fsnotify_create(dir, dentry);
1941 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1946 struct dentry * dentry;
1947 struct nameidata nd;
1951 tmp = getname(filename);
1953 return PTR_ERR(tmp);
1955 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1958 dentry = lookup_create(&nd, 0);
1959 error = PTR_ERR(dentry);
1961 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1962 mode &= ~current->fs->umask;
1963 if (!IS_ERR(dentry)) {
1964 switch (mode & S_IFMT) {
1965 case 0: case S_IFREG:
1966 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1968 case S_IFCHR: case S_IFBLK:
1969 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1970 new_decode_dev(dev));
1972 case S_IFIFO: case S_IFSOCK:
1973 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1983 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1991 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1993 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1996 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1998 int error = may_create(dir, dentry, NULL);
2003 if (!dir->i_op || !dir->i_op->mkdir)
2006 mode &= (S_IRWXUGO|S_ISVTX);
2007 error = security_inode_mkdir(dir, dentry, mode);
2012 error = dir->i_op->mkdir(dir, dentry, mode);
2014 fsnotify_mkdir(dir, dentry);
2018 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2022 struct dentry *dentry;
2023 struct nameidata nd;
2025 tmp = getname(pathname);
2026 error = PTR_ERR(tmp);
2030 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2033 dentry = lookup_create(&nd, 1);
2034 error = PTR_ERR(dentry);
2038 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2039 mode &= ~current->fs->umask;
2040 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2043 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2051 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2053 return sys_mkdirat(AT_FDCWD, pathname, mode);
2057 * We try to drop the dentry early: we should have
2058 * a usage count of 2 if we're the only user of this
2059 * dentry, and if that is true (possibly after pruning
2060 * the dcache), then we drop the dentry now.
2062 * A low-level filesystem can, if it choses, legally
2065 * if (!d_unhashed(dentry))
2068 * if it cannot handle the case of removing a directory
2069 * that is still in use by something else..
2071 void dentry_unhash(struct dentry *dentry)
2074 shrink_dcache_parent(dentry);
2075 spin_lock(&dcache_lock);
2076 spin_lock(&dentry->d_lock);
2077 if (atomic_read(&dentry->d_count) == 2)
2079 spin_unlock(&dentry->d_lock);
2080 spin_unlock(&dcache_lock);
2083 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2085 int error = may_delete(dir, dentry, 1);
2090 if (!dir->i_op || !dir->i_op->rmdir)
2095 mutex_lock(&dentry->d_inode->i_mutex);
2096 dentry_unhash(dentry);
2097 if (d_mountpoint(dentry))
2100 error = security_inode_rmdir(dir, dentry);
2102 error = dir->i_op->rmdir(dir, dentry);
2104 dentry->d_inode->i_flags |= S_DEAD;
2107 mutex_unlock(&dentry->d_inode->i_mutex);
2116 static long do_rmdir(int dfd, const char __user *pathname)
2120 struct dentry *dentry;
2121 struct nameidata nd;
2123 name = getname(pathname);
2125 return PTR_ERR(name);
2127 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2131 switch(nd.last_type) {
2142 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2143 dentry = lookup_hash(&nd);
2144 error = PTR_ERR(dentry);
2147 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2150 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2158 asmlinkage long sys_rmdir(const char __user *pathname)
2160 return do_rmdir(AT_FDCWD, pathname);
2163 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2165 int error = may_delete(dir, dentry, 0);
2170 if (!dir->i_op || !dir->i_op->unlink)
2175 mutex_lock(&dentry->d_inode->i_mutex);
2176 if (d_mountpoint(dentry))
2179 error = security_inode_unlink(dir, dentry);
2181 error = dir->i_op->unlink(dir, dentry);
2183 mutex_unlock(&dentry->d_inode->i_mutex);
2185 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2186 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2187 fsnotify_link_count(dentry->d_inode);
2195 * Make sure that the actual truncation of the file will occur outside its
2196 * directory's i_mutex. Truncate can take a long time if there is a lot of
2197 * writeout happening, and we don't want to prevent access to the directory
2198 * while waiting on the I/O.
2200 static long do_unlinkat(int dfd, const char __user *pathname)
2204 struct dentry *dentry;
2205 struct nameidata nd;
2206 struct inode *inode = NULL;
2208 name = getname(pathname);
2210 return PTR_ERR(name);
2212 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2216 if (nd.last_type != LAST_NORM)
2218 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2219 dentry = lookup_hash(&nd);
2220 error = PTR_ERR(dentry);
2221 if (!IS_ERR(dentry)) {
2222 /* Why not before? Because we want correct error value */
2223 if (nd.last.name[nd.last.len])
2225 inode = dentry->d_inode;
2227 atomic_inc(&inode->i_count);
2228 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2232 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2234 iput(inode); /* truncate the inode here */
2242 error = !dentry->d_inode ? -ENOENT :
2243 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2247 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2249 if ((flag & ~AT_REMOVEDIR) != 0)
2252 if (flag & AT_REMOVEDIR)
2253 return do_rmdir(dfd, pathname);
2255 return do_unlinkat(dfd, pathname);
2258 asmlinkage long sys_unlink(const char __user *pathname)
2260 return do_unlinkat(AT_FDCWD, pathname);
2263 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2265 int error = may_create(dir, dentry, NULL);
2270 if (!dir->i_op || !dir->i_op->symlink)
2273 error = security_inode_symlink(dir, dentry, oldname);
2278 error = dir->i_op->symlink(dir, dentry, oldname);
2280 fsnotify_create(dir, dentry);
2284 asmlinkage long sys_symlinkat(const char __user *oldname,
2285 int newdfd, const char __user *newname)
2290 struct dentry *dentry;
2291 struct nameidata nd;
2293 from = getname(oldname);
2295 return PTR_ERR(from);
2296 to = getname(newname);
2297 error = PTR_ERR(to);
2301 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2304 dentry = lookup_create(&nd, 0);
2305 error = PTR_ERR(dentry);
2309 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from, S_IALLUGO);
2312 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2321 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2323 return sys_symlinkat(oldname, AT_FDCWD, newname);
2326 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2328 struct inode *inode = old_dentry->d_inode;
2334 error = may_create(dir, new_dentry, NULL);
2338 if (dir->i_sb != inode->i_sb)
2342 * A link to an append-only or immutable file cannot be created.
2344 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2346 if (!dir->i_op || !dir->i_op->link)
2348 if (S_ISDIR(old_dentry->d_inode->i_mode))
2351 error = security_inode_link(old_dentry, dir, new_dentry);
2355 mutex_lock(&old_dentry->d_inode->i_mutex);
2357 error = dir->i_op->link(old_dentry, dir, new_dentry);
2358 mutex_unlock(&old_dentry->d_inode->i_mutex);
2360 fsnotify_link(dir, old_dentry->d_inode, new_dentry);
2365 * Hardlinks are often used in delicate situations. We avoid
2366 * security-related surprises by not following symlinks on the
2369 * We don't follow them on the oldname either to be compatible
2370 * with linux 2.0, and to avoid hard-linking to directories
2371 * and other special files. --ADM
2373 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2374 int newdfd, const char __user *newname,
2377 struct dentry *new_dentry;
2378 struct nameidata nd, old_nd;
2382 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2385 to = getname(newname);
2389 error = __user_walk_fd(olddfd, oldname,
2390 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2394 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2398 if (old_nd.path.mnt != nd.path.mnt)
2400 new_dentry = lookup_create(&nd, 0);
2401 error = PTR_ERR(new_dentry);
2402 if (IS_ERR(new_dentry))
2404 error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
2407 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2411 path_put(&old_nd.path);
2418 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2420 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2424 * The worst of all namespace operations - renaming directory. "Perverted"
2425 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2427 * a) we can get into loop creation. Check is done in is_subdir().
2428 * b) race potential - two innocent renames can create a loop together.
2429 * That's where 4.4 screws up. Current fix: serialization on
2430 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2432 * c) we have to lock _three_ objects - parents and victim (if it exists).
2433 * And that - after we got ->i_mutex on parents (until then we don't know
2434 * whether the target exists). Solution: try to be smart with locking
2435 * order for inodes. We rely on the fact that tree topology may change
2436 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2437 * move will be locked. Thus we can rank directories by the tree
2438 * (ancestors first) and rank all non-directories after them.
2439 * That works since everybody except rename does "lock parent, lookup,
2440 * lock child" and rename is under ->s_vfs_rename_mutex.
2441 * HOWEVER, it relies on the assumption that any object with ->lookup()
2442 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2443 * we'd better make sure that there's no link(2) for them.
2444 * d) some filesystems don't support opened-but-unlinked directories,
2445 * either because of layout or because they are not ready to deal with
2446 * all cases correctly. The latter will be fixed (taking this sort of
2447 * stuff into VFS), but the former is not going away. Solution: the same
2448 * trick as in rmdir().
2449 * e) conversion from fhandle to dentry may come in the wrong moment - when
2450 * we are removing the target. Solution: we will have to grab ->i_mutex
2451 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2452 * ->i_mutex on parents, which works but leads to some truely excessive
2455 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2456 struct inode *new_dir, struct dentry *new_dentry)
2459 struct inode *target;
2462 * If we are going to change the parent - check write permissions,
2463 * we'll need to flip '..'.
2465 if (new_dir != old_dir) {
2466 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2471 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2475 target = new_dentry->d_inode;
2477 mutex_lock(&target->i_mutex);
2478 dentry_unhash(new_dentry);
2480 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2483 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2486 target->i_flags |= S_DEAD;
2487 mutex_unlock(&target->i_mutex);
2488 if (d_unhashed(new_dentry))
2489 d_rehash(new_dentry);
2493 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2494 d_move(old_dentry,new_dentry);
2498 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2499 struct inode *new_dir, struct dentry *new_dentry)
2501 struct inode *target;
2504 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2509 target = new_dentry->d_inode;
2511 mutex_lock(&target->i_mutex);
2512 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2515 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2517 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2518 d_move(old_dentry, new_dentry);
2521 mutex_unlock(&target->i_mutex);
2526 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2527 struct inode *new_dir, struct dentry *new_dentry)
2530 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2531 const char *old_name;
2533 if (old_dentry->d_inode == new_dentry->d_inode)
2536 error = may_delete(old_dir, old_dentry, is_dir);
2540 if (!new_dentry->d_inode)
2541 error = may_create(new_dir, new_dentry, NULL);
2543 error = may_delete(new_dir, new_dentry, is_dir);
2547 if (!old_dir->i_op || !old_dir->i_op->rename)
2550 DQUOT_INIT(old_dir);
2551 DQUOT_INIT(new_dir);
2553 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2556 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2558 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2560 const char *new_name = old_dentry->d_name.name;
2561 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2562 new_dentry->d_inode, old_dentry);
2564 fsnotify_oldname_free(old_name);
2569 static int do_rename(int olddfd, const char *oldname,
2570 int newdfd, const char *newname)
2573 struct dentry * old_dir, * new_dir;
2574 struct dentry * old_dentry, *new_dentry;
2575 struct dentry * trap;
2576 struct nameidata oldnd, newnd;
2578 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2582 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2587 if (oldnd.path.mnt != newnd.path.mnt)
2590 old_dir = oldnd.path.dentry;
2592 if (oldnd.last_type != LAST_NORM)
2595 new_dir = newnd.path.dentry;
2596 if (newnd.last_type != LAST_NORM)
2599 trap = lock_rename(new_dir, old_dir);
2601 old_dentry = lookup_hash(&oldnd);
2602 error = PTR_ERR(old_dentry);
2603 if (IS_ERR(old_dentry))
2605 /* source must exist */
2607 if (!old_dentry->d_inode)
2609 /* unless the source is a directory trailing slashes give -ENOTDIR */
2610 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2612 if (oldnd.last.name[oldnd.last.len])
2614 if (newnd.last.name[newnd.last.len])
2617 /* source should not be ancestor of target */
2619 if (old_dentry == trap)
2621 new_dentry = lookup_hash(&newnd);
2622 error = PTR_ERR(new_dentry);
2623 if (IS_ERR(new_dentry))
2625 /* target should not be an ancestor of source */
2627 if (new_dentry == trap)
2630 error = vfs_rename(old_dir->d_inode, old_dentry,
2631 new_dir->d_inode, new_dentry);
2637 unlock_rename(new_dir, old_dir);
2639 path_put(&newnd.path);
2641 path_put(&oldnd.path);
2646 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2647 int newdfd, const char __user *newname)
2653 from = getname(oldname);
2655 return PTR_ERR(from);
2656 to = getname(newname);
2657 error = PTR_ERR(to);
2659 error = do_rename(olddfd, from, newdfd, to);
2666 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2668 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2671 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2675 len = PTR_ERR(link);
2680 if (len > (unsigned) buflen)
2682 if (copy_to_user(buffer, link, len))
2689 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2690 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2691 * using) it for any given inode is up to filesystem.
2693 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2695 struct nameidata nd;
2699 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2700 if (!IS_ERR(cookie)) {
2701 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2702 if (dentry->d_inode->i_op->put_link)
2703 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2704 cookie = ERR_PTR(res);
2706 return PTR_ERR(cookie);
2709 int vfs_follow_link(struct nameidata *nd, const char *link)
2711 return __vfs_follow_link(nd, link);
2714 /* get the link contents into pagecache */
2715 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2718 struct address_space *mapping = dentry->d_inode->i_mapping;
2719 page = read_mapping_page(mapping, 0, NULL);
2726 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2728 struct page *page = NULL;
2729 char *s = page_getlink(dentry, &page);
2730 int res = vfs_readlink(dentry,buffer,buflen,s);
2733 page_cache_release(page);
2738 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2740 struct page *page = NULL;
2741 nd_set_link(nd, page_getlink(dentry, &page));
2745 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2747 struct page *page = cookie;
2751 page_cache_release(page);
2755 int __page_symlink(struct inode *inode, const char *symname, int len,
2758 struct address_space *mapping = inode->i_mapping;
2765 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2766 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2770 kaddr = kmap_atomic(page, KM_USER0);
2771 memcpy(kaddr, symname, len-1);
2772 kunmap_atomic(kaddr, KM_USER0);
2774 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2781 mark_inode_dirty(inode);
2787 int page_symlink(struct inode *inode, const char *symname, int len)
2789 return __page_symlink(inode, symname, len,
2790 mapping_gfp_mask(inode->i_mapping));
2793 const struct inode_operations page_symlink_inode_operations = {
2794 .readlink = generic_readlink,
2795 .follow_link = page_follow_link_light,
2796 .put_link = page_put_link,
2799 EXPORT_SYMBOL(__user_walk);
2800 EXPORT_SYMBOL(__user_walk_fd);
2801 EXPORT_SYMBOL(follow_down);
2802 EXPORT_SYMBOL(follow_up);
2803 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2804 EXPORT_SYMBOL(getname);
2805 EXPORT_SYMBOL(lock_rename);
2806 EXPORT_SYMBOL(lookup_one_len);
2807 EXPORT_SYMBOL(page_follow_link_light);
2808 EXPORT_SYMBOL(page_put_link);
2809 EXPORT_SYMBOL(page_readlink);
2810 EXPORT_SYMBOL(__page_symlink);
2811 EXPORT_SYMBOL(page_symlink);
2812 EXPORT_SYMBOL(page_symlink_inode_operations);
2813 EXPORT_SYMBOL(path_lookup);
2814 EXPORT_SYMBOL(vfs_path_lookup);
2815 EXPORT_SYMBOL(permission);
2816 EXPORT_SYMBOL(vfs_permission);
2817 EXPORT_SYMBOL(file_permission);
2818 EXPORT_SYMBOL(unlock_rename);
2819 EXPORT_SYMBOL(vfs_create);
2820 EXPORT_SYMBOL(vfs_follow_link);
2821 EXPORT_SYMBOL(vfs_link);
2822 EXPORT_SYMBOL(vfs_mkdir);
2823 EXPORT_SYMBOL(vfs_mknod);
2824 EXPORT_SYMBOL(generic_permission);
2825 EXPORT_SYMBOL(vfs_readlink);
2826 EXPORT_SYMBOL(vfs_rename);
2827 EXPORT_SYMBOL(vfs_rmdir);
2828 EXPORT_SYMBOL(vfs_symlink);
2829 EXPORT_SYMBOL(vfs_unlink);
2830 EXPORT_SYMBOL(dentry_unhash);
2831 EXPORT_SYMBOL(generic_readlink);