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/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user *filename, char *page)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 char * getname(const char __user * filename)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask))
175 umode_t mode = inode->i_mode;
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179 if (current_fsuid() == inode->i_uid)
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask);
184 if (error != -EAGAIN)
188 if (in_group_p(inode->i_gid))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask & ~mode) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
206 * Used to check for read/write/execute permissions on a file.
207 * We use "fsuid" for this, letting us set arbitrary permissions
208 * for filesystem access without changing the "normal" uids which
209 * are used for other things..
211 int generic_permission(struct inode *inode, int mask,
212 int (*check_acl)(struct inode *inode, int mask))
217 * Do the basic POSIX ACL permission checks.
219 ret = acl_permission_check(inode, mask, check_acl);
224 * Read/write DACs are always overridable.
225 * Executable DACs are overridable if at least one exec bit is set.
227 if (!(mask & MAY_EXEC) || execute_ok(inode))
228 if (capable(CAP_DAC_OVERRIDE))
232 * Searching includes executable on directories, else just read.
234 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
281 retval = devcgroup_inode_permission(inode, mask);
285 return security_inode_permission(inode, mask);
289 * file_permission - check for additional access rights to a given file
290 * @file: file to check access rights for
291 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
293 * Used to check for read/write/execute permissions on an already opened
297 * Do not use this function in new code. All access checks should
298 * be done using inode_permission().
300 int file_permission(struct file *file, int mask)
302 return inode_permission(file->f_path.dentry->d_inode, mask);
306 * get_write_access() gets write permission for a file.
307 * put_write_access() releases this write permission.
308 * This is used for regular files.
309 * We cannot support write (and maybe mmap read-write shared) accesses and
310 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
311 * can have the following values:
312 * 0: no writers, no VM_DENYWRITE mappings
313 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
314 * > 0: (i_writecount) users are writing to the file.
316 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
317 * except for the cases where we don't hold i_writecount yet. Then we need to
318 * use {get,deny}_write_access() - these functions check the sign and refuse
319 * to do the change if sign is wrong. Exclusion between them is provided by
320 * the inode->i_lock spinlock.
323 int get_write_access(struct inode * inode)
325 spin_lock(&inode->i_lock);
326 if (atomic_read(&inode->i_writecount) < 0) {
327 spin_unlock(&inode->i_lock);
330 atomic_inc(&inode->i_writecount);
331 spin_unlock(&inode->i_lock);
336 int deny_write_access(struct file * file)
338 struct inode *inode = file->f_path.dentry->d_inode;
340 spin_lock(&inode->i_lock);
341 if (atomic_read(&inode->i_writecount) > 0) {
342 spin_unlock(&inode->i_lock);
345 atomic_dec(&inode->i_writecount);
346 spin_unlock(&inode->i_lock);
352 * path_get - get a reference to a path
353 * @path: path to get the reference to
355 * Given a path increment the reference count to the dentry and the vfsmount.
357 void path_get(struct path *path)
362 EXPORT_SYMBOL(path_get);
365 * path_put - put a reference to a path
366 * @path: path to put the reference to
368 * Given a path decrement the reference count to the dentry and the vfsmount.
370 void path_put(struct path *path)
375 EXPORT_SYMBOL(path_put);
378 * release_open_intent - free up open intent resources
379 * @nd: pointer to nameidata
381 void release_open_intent(struct nameidata *nd)
383 if (nd->intent.open.file->f_path.dentry == NULL)
384 put_filp(nd->intent.open.file);
386 fput(nd->intent.open.file);
389 static inline struct dentry *
390 do_revalidate(struct dentry *dentry, struct nameidata *nd)
392 int status = dentry->d_op->d_revalidate(dentry, nd);
393 if (unlikely(status <= 0)) {
395 * The dentry failed validation.
396 * If d_revalidate returned 0 attempt to invalidate
397 * the dentry otherwise d_revalidate is asking us
398 * to return a fail status.
401 if (!d_invalidate(dentry)) {
407 dentry = ERR_PTR(status);
414 * force_reval_path - force revalidation of a dentry
416 * In some situations the path walking code will trust dentries without
417 * revalidating them. This causes problems for filesystems that depend on
418 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
419 * (which indicates that it's possible for the dentry to go stale), force
420 * a d_revalidate call before proceeding.
422 * Returns 0 if the revalidation was successful. If the revalidation fails,
423 * either return the error returned by d_revalidate or -ESTALE if the
424 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
425 * invalidate the dentry. It's up to the caller to handle putting references
426 * to the path if necessary.
429 force_reval_path(struct path *path, struct nameidata *nd)
432 struct dentry *dentry = path->dentry;
435 * only check on filesystems where it's possible for the dentry to
436 * become stale. It's assumed that if this flag is set then the
437 * d_revalidate op will also be defined.
439 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
442 status = dentry->d_op->d_revalidate(dentry, nd);
447 d_invalidate(dentry);
454 * Short-cut version of permission(), for calling on directories
455 * during pathname resolution. Combines parts of permission()
456 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
458 * If appropriate, check DAC only. If not appropriate, or
459 * short-cut DAC fails, then call ->permission() to do more
460 * complete permission check.
462 static int exec_permission(struct inode *inode)
466 if (inode->i_op->permission) {
467 ret = inode->i_op->permission(inode, MAY_EXEC);
472 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
476 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
481 return security_inode_permission(inode, MAY_EXEC);
484 static __always_inline void set_root(struct nameidata *nd)
487 get_fs_root(current->fs, &nd->root);
490 static int link_path_walk(const char *, struct nameidata *);
492 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
504 return link_path_walk(link, nd);
507 return PTR_ERR(link);
510 static void path_put_conditional(struct path *path, struct nameidata *nd)
513 if (path->mnt != nd->path.mnt)
517 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
519 dput(nd->path.dentry);
520 if (nd->path.mnt != path->mnt) {
521 mntput(nd->path.mnt);
522 nd->path.mnt = path->mnt;
524 nd->path.dentry = path->dentry;
527 static __always_inline int
528 __do_follow_link(struct path *path, struct nameidata *nd, void **p)
531 struct dentry *dentry = path->dentry;
533 touch_atime(path->mnt, dentry);
534 nd_set_link(nd, NULL);
536 if (path->mnt != nd->path.mnt) {
537 path_to_nameidata(path, nd);
541 nd->last_type = LAST_BIND;
542 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
545 char *s = nd_get_link(nd);
548 error = __vfs_follow_link(nd, s);
549 else if (nd->last_type == LAST_BIND) {
550 error = force_reval_path(&nd->path, nd);
559 * This limits recursive symlink follows to 8, while
560 * limiting consecutive symlinks to 40.
562 * Without that kind of total limit, nasty chains of consecutive
563 * symlinks can cause almost arbitrarily long lookups.
565 static inline int do_follow_link(struct path *path, struct nameidata *nd)
569 if (current->link_count >= MAX_NESTED_LINKS)
571 if (current->total_link_count >= 40)
573 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
575 err = security_inode_follow_link(path->dentry, nd);
578 current->link_count++;
579 current->total_link_count++;
581 err = __do_follow_link(path, nd, &cookie);
582 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
583 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
585 current->link_count--;
589 path_put_conditional(path, nd);
594 int follow_up(struct path *path)
596 struct vfsmount *parent;
597 struct dentry *mountpoint;
598 spin_lock(&vfsmount_lock);
599 parent = path->mnt->mnt_parent;
600 if (parent == path->mnt) {
601 spin_unlock(&vfsmount_lock);
605 mountpoint = dget(path->mnt->mnt_mountpoint);
606 spin_unlock(&vfsmount_lock);
608 path->dentry = mountpoint;
614 /* no need for dcache_lock, as serialization is taken care in
617 static int __follow_mount(struct path *path)
620 while (d_mountpoint(path->dentry)) {
621 struct vfsmount *mounted = lookup_mnt(path);
628 path->dentry = dget(mounted->mnt_root);
634 static void follow_mount(struct path *path)
636 while (d_mountpoint(path->dentry)) {
637 struct vfsmount *mounted = lookup_mnt(path);
643 path->dentry = dget(mounted->mnt_root);
647 /* no need for dcache_lock, as serialization is taken care in
650 int follow_down(struct path *path)
652 struct vfsmount *mounted;
654 mounted = lookup_mnt(path);
659 path->dentry = dget(mounted->mnt_root);
665 static __always_inline void follow_dotdot(struct nameidata *nd)
670 struct dentry *old = nd->path.dentry;
672 if (nd->path.dentry == nd->root.dentry &&
673 nd->path.mnt == nd->root.mnt) {
676 if (nd->path.dentry != nd->path.mnt->mnt_root) {
677 /* rare case of legitimate dget_parent()... */
678 nd->path.dentry = dget_parent(nd->path.dentry);
682 if (!follow_up(&nd->path))
685 follow_mount(&nd->path);
689 * Allocate a dentry with name and parent, and perform a parent
690 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
691 * on error. parent->d_inode->i_mutex must be held. d_lookup must
692 * have verified that no child exists while under i_mutex.
694 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
695 struct qstr *name, struct nameidata *nd)
697 struct inode *inode = parent->d_inode;
698 struct dentry *dentry;
701 /* Don't create child dentry for a dead directory. */
702 if (unlikely(IS_DEADDIR(inode)))
703 return ERR_PTR(-ENOENT);
705 dentry = d_alloc(parent, name);
706 if (unlikely(!dentry))
707 return ERR_PTR(-ENOMEM);
709 old = inode->i_op->lookup(inode, dentry, nd);
718 * It's more convoluted than I'd like it to be, but... it's still fairly
719 * small and for now I'd prefer to have fast path as straight as possible.
720 * It _is_ time-critical.
722 static int do_lookup(struct nameidata *nd, struct qstr *name,
725 struct vfsmount *mnt = nd->path.mnt;
726 struct dentry *dentry, *parent;
729 * See if the low-level filesystem might want
730 * to use its own hash..
732 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
733 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
739 * Rename seqlock is not required here because in the off chance
740 * of a false negative due to a concurrent rename, we're going to
741 * do the non-racy lookup, below.
743 dentry = __d_lookup(nd->path.dentry, name);
747 if (dentry->d_op && dentry->d_op->d_revalidate)
748 goto need_revalidate;
751 path->dentry = dentry;
752 __follow_mount(path);
756 parent = nd->path.dentry;
757 dir = parent->d_inode;
759 mutex_lock(&dir->i_mutex);
761 * First re-do the cached lookup just in case it was created
762 * while we waited for the directory semaphore, or the first
763 * lookup failed due to an unrelated rename.
765 * This could use version numbering or similar to avoid unnecessary
766 * cache lookups, but then we'd have to do the first lookup in the
767 * non-racy way. However in the common case here, everything should
768 * be hot in cache, so would it be a big win?
770 dentry = d_lookup(parent, name);
771 if (likely(!dentry)) {
772 dentry = d_alloc_and_lookup(parent, name, nd);
773 mutex_unlock(&dir->i_mutex);
779 * Uhhuh! Nasty case: the cache was re-populated while
780 * we waited on the semaphore. Need to revalidate.
782 mutex_unlock(&dir->i_mutex);
786 dentry = do_revalidate(dentry, nd);
794 return PTR_ERR(dentry);
798 * This is a temporary kludge to deal with "automount" symlinks; proper
799 * solution is to trigger them on follow_mount(), so that do_lookup()
800 * would DTRT. To be killed before 2.6.34-final.
802 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
804 return inode && unlikely(inode->i_op->follow_link) &&
805 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
810 * This is the basic name resolution function, turning a pathname into
811 * the final dentry. We expect 'base' to be positive and a directory.
813 * Returns 0 and nd will have valid dentry and mnt on success.
814 * Returns error and drops reference to input namei data on failure.
816 static int link_path_walk(const char *name, struct nameidata *nd)
821 unsigned int lookup_flags = nd->flags;
828 inode = nd->path.dentry->d_inode;
830 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
832 /* At this point we know we have a real path component. */
838 nd->flags |= LOOKUP_CONTINUE;
839 err = exec_permission(inode);
844 c = *(const unsigned char *)name;
846 hash = init_name_hash();
849 hash = partial_name_hash(c, hash);
850 c = *(const unsigned char *)name;
851 } while (c && (c != '/'));
852 this.len = name - (const char *) this.name;
853 this.hash = end_name_hash(hash);
855 /* remove trailing slashes? */
858 while (*++name == '/');
860 goto last_with_slashes;
863 * "." and ".." are special - ".." especially so because it has
864 * to be able to know about the current root directory and
865 * parent relationships.
867 if (this.name[0] == '.') switch (this.len) {
871 if (this.name[1] != '.')
874 inode = nd->path.dentry->d_inode;
879 /* This does the actual lookups.. */
880 err = do_lookup(nd, &this, &next);
885 inode = next.dentry->d_inode;
889 if (inode->i_op->follow_link) {
890 err = do_follow_link(&next, nd);
894 inode = nd->path.dentry->d_inode;
898 path_to_nameidata(&next, nd);
900 if (!inode->i_op->lookup)
903 /* here ends the main loop */
906 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
908 /* Clear LOOKUP_CONTINUE iff it was previously unset */
909 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
910 if (lookup_flags & LOOKUP_PARENT)
912 if (this.name[0] == '.') switch (this.len) {
916 if (this.name[1] != '.')
919 inode = nd->path.dentry->d_inode;
924 err = do_lookup(nd, &this, &next);
927 inode = next.dentry->d_inode;
928 if (follow_on_final(inode, lookup_flags)) {
929 err = do_follow_link(&next, nd);
932 inode = nd->path.dentry->d_inode;
934 path_to_nameidata(&next, nd);
938 if (lookup_flags & LOOKUP_DIRECTORY) {
940 if (!inode->i_op->lookup)
946 nd->last_type = LAST_NORM;
947 if (this.name[0] != '.')
950 nd->last_type = LAST_DOT;
951 else if (this.len == 2 && this.name[1] == '.')
952 nd->last_type = LAST_DOTDOT;
957 * We bypassed the ordinary revalidation routines.
958 * We may need to check the cached dentry for staleness.
960 if (nd->path.dentry && nd->path.dentry->d_sb &&
961 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
963 /* Note: we do not d_invalidate() */
964 if (!nd->path.dentry->d_op->d_revalidate(
965 nd->path.dentry, nd))
971 path_put_conditional(&next, nd);
979 static int path_walk(const char *name, struct nameidata *nd)
981 struct path save = nd->path;
984 current->total_link_count = 0;
986 /* make sure the stuff we saved doesn't go away */
989 result = link_path_walk(name, nd);
990 if (result == -ESTALE) {
991 /* nd->path had been dropped */
992 current->total_link_count = 0;
995 nd->flags |= LOOKUP_REVAL;
996 result = link_path_walk(name, nd);
1004 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1010 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1013 nd->root.mnt = NULL;
1017 nd->path = nd->root;
1018 path_get(&nd->root);
1019 } else if (dfd == AT_FDCWD) {
1020 get_fs_pwd(current->fs, &nd->path);
1022 struct dentry *dentry;
1024 file = fget_light(dfd, &fput_needed);
1029 dentry = file->f_path.dentry;
1032 if (!S_ISDIR(dentry->d_inode->i_mode))
1035 retval = file_permission(file, MAY_EXEC);
1039 nd->path = file->f_path;
1040 path_get(&file->f_path);
1042 fput_light(file, fput_needed);
1047 fput_light(file, fput_needed);
1052 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1053 static int do_path_lookup(int dfd, const char *name,
1054 unsigned int flags, struct nameidata *nd)
1056 int retval = path_init(dfd, name, flags, nd);
1058 retval = path_walk(name, nd);
1059 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1060 nd->path.dentry->d_inode))
1061 audit_inode(name, nd->path.dentry);
1063 path_put(&nd->root);
1064 nd->root.mnt = NULL;
1069 int path_lookup(const char *name, unsigned int flags,
1070 struct nameidata *nd)
1072 return do_path_lookup(AT_FDCWD, name, flags, nd);
1075 int kern_path(const char *name, unsigned int flags, struct path *path)
1077 struct nameidata nd;
1078 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1085 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1086 * @dentry: pointer to dentry of the base directory
1087 * @mnt: pointer to vfs mount of the base directory
1088 * @name: pointer to file name
1089 * @flags: lookup flags
1090 * @nd: pointer to nameidata
1092 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1093 const char *name, unsigned int flags,
1094 struct nameidata *nd)
1098 /* same as do_path_lookup */
1099 nd->last_type = LAST_ROOT;
1103 nd->path.dentry = dentry;
1105 path_get(&nd->path);
1106 nd->root = nd->path;
1107 path_get(&nd->root);
1109 retval = path_walk(name, nd);
1110 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1111 nd->path.dentry->d_inode))
1112 audit_inode(name, nd->path.dentry);
1114 path_put(&nd->root);
1115 nd->root.mnt = NULL;
1120 static struct dentry *__lookup_hash(struct qstr *name,
1121 struct dentry *base, struct nameidata *nd)
1123 struct dentry *dentry;
1124 struct inode *inode;
1127 inode = base->d_inode;
1130 * See if the low-level filesystem might want
1131 * to use its own hash..
1133 if (base->d_op && base->d_op->d_hash) {
1134 err = base->d_op->d_hash(base, name);
1135 dentry = ERR_PTR(err);
1141 * Don't bother with __d_lookup: callers are for creat as
1142 * well as unlink, so a lot of the time it would cost
1145 dentry = d_lookup(base, name);
1147 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1148 dentry = do_revalidate(dentry, nd);
1151 dentry = d_alloc_and_lookup(base, name, nd);
1157 * Restricted form of lookup. Doesn't follow links, single-component only,
1158 * needs parent already locked. Doesn't follow mounts.
1161 static struct dentry *lookup_hash(struct nameidata *nd)
1165 err = exec_permission(nd->path.dentry->d_inode);
1167 return ERR_PTR(err);
1168 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1171 static int __lookup_one_len(const char *name, struct qstr *this,
1172 struct dentry *base, int len)
1182 hash = init_name_hash();
1184 c = *(const unsigned char *)name++;
1185 if (c == '/' || c == '\0')
1187 hash = partial_name_hash(c, hash);
1189 this->hash = end_name_hash(hash);
1194 * lookup_one_len - filesystem helper to lookup single pathname component
1195 * @name: pathname component to lookup
1196 * @base: base directory to lookup from
1197 * @len: maximum length @len should be interpreted to
1199 * Note that this routine is purely a helper for filesystem usage and should
1200 * not be called by generic code. Also note that by using this function the
1201 * nameidata argument is passed to the filesystem methods and a filesystem
1202 * using this helper needs to be prepared for that.
1204 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1209 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1211 err = __lookup_one_len(name, &this, base, len);
1213 return ERR_PTR(err);
1215 err = exec_permission(base->d_inode);
1217 return ERR_PTR(err);
1218 return __lookup_hash(&this, base, NULL);
1221 int user_path_at(int dfd, const char __user *name, unsigned flags,
1224 struct nameidata nd;
1225 char *tmp = getname(name);
1226 int err = PTR_ERR(tmp);
1229 BUG_ON(flags & LOOKUP_PARENT);
1231 err = do_path_lookup(dfd, tmp, flags, &nd);
1239 static int user_path_parent(int dfd, const char __user *path,
1240 struct nameidata *nd, char **name)
1242 char *s = getname(path);
1248 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1258 * It's inline, so penalty for filesystems that don't use sticky bit is
1261 static inline int check_sticky(struct inode *dir, struct inode *inode)
1263 uid_t fsuid = current_fsuid();
1265 if (!(dir->i_mode & S_ISVTX))
1267 if (inode->i_uid == fsuid)
1269 if (dir->i_uid == fsuid)
1271 return !capable(CAP_FOWNER);
1275 * Check whether we can remove a link victim from directory dir, check
1276 * whether the type of victim is right.
1277 * 1. We can't do it if dir is read-only (done in permission())
1278 * 2. We should have write and exec permissions on dir
1279 * 3. We can't remove anything from append-only dir
1280 * 4. We can't do anything with immutable dir (done in permission())
1281 * 5. If the sticky bit on dir is set we should either
1282 * a. be owner of dir, or
1283 * b. be owner of victim, or
1284 * c. have CAP_FOWNER capability
1285 * 6. If the victim is append-only or immutable we can't do antyhing with
1286 * links pointing to it.
1287 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1288 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1289 * 9. We can't remove a root or mountpoint.
1290 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1291 * nfs_async_unlink().
1293 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1297 if (!victim->d_inode)
1300 BUG_ON(victim->d_parent->d_inode != dir);
1301 audit_inode_child(victim, dir);
1303 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1308 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1309 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1312 if (!S_ISDIR(victim->d_inode->i_mode))
1314 if (IS_ROOT(victim))
1316 } else if (S_ISDIR(victim->d_inode->i_mode))
1318 if (IS_DEADDIR(dir))
1320 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1325 /* Check whether we can create an object with dentry child in directory
1327 * 1. We can't do it if child already exists (open has special treatment for
1328 * this case, but since we are inlined it's OK)
1329 * 2. We can't do it if dir is read-only (done in permission())
1330 * 3. We should have write and exec permissions on dir
1331 * 4. We can't do it if dir is immutable (done in permission())
1333 static inline int may_create(struct inode *dir, struct dentry *child)
1337 if (IS_DEADDIR(dir))
1339 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1343 * p1 and p2 should be directories on the same fs.
1345 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1350 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1354 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1356 p = d_ancestor(p2, p1);
1358 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1359 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1363 p = d_ancestor(p1, p2);
1365 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1366 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1370 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1371 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1375 void unlock_rename(struct dentry *p1, struct dentry *p2)
1377 mutex_unlock(&p1->d_inode->i_mutex);
1379 mutex_unlock(&p2->d_inode->i_mutex);
1380 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1384 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1385 struct nameidata *nd)
1387 int error = may_create(dir, dentry);
1392 if (!dir->i_op->create)
1393 return -EACCES; /* shouldn't it be ENOSYS? */
1396 error = security_inode_create(dir, dentry, mode);
1399 error = dir->i_op->create(dir, dentry, mode, nd);
1401 fsnotify_create(dir, dentry);
1405 int may_open(struct path *path, int acc_mode, int flag)
1407 struct dentry *dentry = path->dentry;
1408 struct inode *inode = dentry->d_inode;
1414 switch (inode->i_mode & S_IFMT) {
1418 if (acc_mode & MAY_WRITE)
1423 if (path->mnt->mnt_flags & MNT_NODEV)
1432 error = inode_permission(inode, acc_mode);
1437 * An append-only file must be opened in append mode for writing.
1439 if (IS_APPEND(inode)) {
1440 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1446 /* O_NOATIME can only be set by the owner or superuser */
1447 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1451 * Ensure there are no outstanding leases on the file.
1453 return break_lease(inode, flag);
1456 static int handle_truncate(struct path *path)
1458 struct inode *inode = path->dentry->d_inode;
1459 int error = get_write_access(inode);
1463 * Refuse to truncate files with mandatory locks held on them.
1465 error = locks_verify_locked(inode);
1467 error = security_path_truncate(path);
1469 error = do_truncate(path->dentry, 0,
1470 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1473 put_write_access(inode);
1478 * Be careful about ever adding any more callers of this
1479 * function. Its flags must be in the namei format, not
1480 * what get passed to sys_open().
1482 static int __open_namei_create(struct nameidata *nd, struct path *path,
1483 int open_flag, int mode)
1486 struct dentry *dir = nd->path.dentry;
1488 if (!IS_POSIXACL(dir->d_inode))
1489 mode &= ~current_umask();
1490 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1493 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1495 mutex_unlock(&dir->d_inode->i_mutex);
1496 dput(nd->path.dentry);
1497 nd->path.dentry = path->dentry;
1500 /* Don't check for write permission, don't truncate */
1501 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
1505 * Note that while the flag value (low two bits) for sys_open means:
1510 * it is changed into
1511 * 00 - no permissions needed
1512 * 01 - read-permission
1513 * 10 - write-permission
1515 * for the internal routines (ie open_namei()/follow_link() etc)
1516 * This is more logical, and also allows the 00 "no perm needed"
1517 * to be used for symlinks (where the permissions are checked
1521 static inline int open_to_namei_flags(int flag)
1523 if ((flag+1) & O_ACCMODE)
1528 static int open_will_truncate(int flag, struct inode *inode)
1531 * We'll never write to the fs underlying
1534 if (special_file(inode->i_mode))
1536 return (flag & O_TRUNC);
1539 static struct file *finish_open(struct nameidata *nd,
1540 int open_flag, int acc_mode)
1546 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
1547 if (will_truncate) {
1548 error = mnt_want_write(nd->path.mnt);
1552 error = may_open(&nd->path, acc_mode, open_flag);
1555 mnt_drop_write(nd->path.mnt);
1558 filp = nameidata_to_filp(nd);
1559 if (!IS_ERR(filp)) {
1560 error = ima_file_check(filp, acc_mode);
1563 filp = ERR_PTR(error);
1566 if (!IS_ERR(filp)) {
1567 if (will_truncate) {
1568 error = handle_truncate(&nd->path);
1571 filp = ERR_PTR(error);
1576 * It is now safe to drop the mnt write
1577 * because the filp has had a write taken
1581 mnt_drop_write(nd->path.mnt);
1585 if (!IS_ERR(nd->intent.open.file))
1586 release_open_intent(nd);
1587 path_put(&nd->path);
1588 return ERR_PTR(error);
1591 static struct file *do_last(struct nameidata *nd, struct path *path,
1592 int open_flag, int acc_mode,
1593 int mode, const char *pathname)
1595 struct dentry *dir = nd->path.dentry;
1597 int error = -EISDIR;
1599 switch (nd->last_type) {
1602 dir = nd->path.dentry;
1604 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1605 if (!dir->d_op->d_revalidate(dir, nd)) {
1612 if (open_flag & O_CREAT)
1616 audit_inode(pathname, dir);
1620 /* trailing slashes? */
1621 if (nd->last.name[nd->last.len]) {
1622 if (open_flag & O_CREAT)
1624 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
1627 /* just plain open? */
1628 if (!(open_flag & O_CREAT)) {
1629 error = do_lookup(nd, &nd->last, path);
1633 if (!path->dentry->d_inode)
1635 if (path->dentry->d_inode->i_op->follow_link)
1638 if (nd->flags & LOOKUP_DIRECTORY) {
1639 if (!path->dentry->d_inode->i_op->lookup)
1642 path_to_nameidata(path, nd);
1643 audit_inode(pathname, nd->path.dentry);
1647 /* OK, it's O_CREAT */
1648 mutex_lock(&dir->d_inode->i_mutex);
1650 path->dentry = lookup_hash(nd);
1651 path->mnt = nd->path.mnt;
1653 error = PTR_ERR(path->dentry);
1654 if (IS_ERR(path->dentry)) {
1655 mutex_unlock(&dir->d_inode->i_mutex);
1659 if (IS_ERR(nd->intent.open.file)) {
1660 error = PTR_ERR(nd->intent.open.file);
1661 goto exit_mutex_unlock;
1664 /* Negative dentry, just create the file */
1665 if (!path->dentry->d_inode) {
1667 * This write is needed to ensure that a
1668 * ro->rw transition does not occur between
1669 * the time when the file is created and when
1670 * a permanent write count is taken through
1671 * the 'struct file' in nameidata_to_filp().
1673 error = mnt_want_write(nd->path.mnt);
1675 goto exit_mutex_unlock;
1676 error = __open_namei_create(nd, path, open_flag, mode);
1678 mnt_drop_write(nd->path.mnt);
1681 filp = nameidata_to_filp(nd);
1682 mnt_drop_write(nd->path.mnt);
1683 if (!IS_ERR(filp)) {
1684 error = ima_file_check(filp, acc_mode);
1687 filp = ERR_PTR(error);
1694 * It already exists.
1696 mutex_unlock(&dir->d_inode->i_mutex);
1697 audit_inode(pathname, path->dentry);
1700 if (open_flag & O_EXCL)
1703 if (__follow_mount(path)) {
1705 if (open_flag & O_NOFOLLOW)
1710 if (!path->dentry->d_inode)
1713 if (path->dentry->d_inode->i_op->follow_link)
1716 path_to_nameidata(path, nd);
1718 if (S_ISDIR(path->dentry->d_inode->i_mode))
1721 filp = finish_open(nd, open_flag, acc_mode);
1725 mutex_unlock(&dir->d_inode->i_mutex);
1727 path_put_conditional(path, nd);
1729 if (!IS_ERR(nd->intent.open.file))
1730 release_open_intent(nd);
1731 path_put(&nd->path);
1732 return ERR_PTR(error);
1736 * Note that the low bits of the passed in "open_flag"
1737 * are not the same as in the local variable "flag". See
1738 * open_to_namei_flags() for more details.
1740 struct file *do_filp_open(int dfd, const char *pathname,
1741 int open_flag, int mode, int acc_mode)
1744 struct nameidata nd;
1748 int flag = open_to_namei_flags(open_flag);
1749 int force_reval = 0;
1751 if (!(open_flag & O_CREAT))
1755 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1756 * check for O_DSYNC if the need any syncing at all we enforce it's
1757 * always set instead of having to deal with possibly weird behaviour
1758 * for malicious applications setting only __O_SYNC.
1760 if (open_flag & __O_SYNC)
1761 open_flag |= O_DSYNC;
1764 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1766 /* O_TRUNC implies we need access checks for write permissions */
1767 if (open_flag & O_TRUNC)
1768 acc_mode |= MAY_WRITE;
1770 /* Allow the LSM permission hook to distinguish append
1771 access from general write access. */
1772 if (open_flag & O_APPEND)
1773 acc_mode |= MAY_APPEND;
1775 /* find the parent */
1777 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1779 return ERR_PTR(error);
1781 nd.flags |= LOOKUP_REVAL;
1783 current->total_link_count = 0;
1784 error = link_path_walk(pathname, &nd);
1786 filp = ERR_PTR(error);
1789 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
1790 audit_inode(pathname, nd.path.dentry);
1793 * We have the parent and last component.
1797 filp = get_empty_filp();
1800 nd.intent.open.file = filp;
1801 filp->f_flags = open_flag;
1802 nd.intent.open.flags = flag;
1803 nd.intent.open.create_mode = mode;
1804 nd.flags &= ~LOOKUP_PARENT;
1805 nd.flags |= LOOKUP_OPEN;
1806 if (open_flag & O_CREAT) {
1807 nd.flags |= LOOKUP_CREATE;
1808 if (open_flag & O_EXCL)
1809 nd.flags |= LOOKUP_EXCL;
1811 if (open_flag & O_DIRECTORY)
1812 nd.flags |= LOOKUP_DIRECTORY;
1813 if (!(open_flag & O_NOFOLLOW))
1814 nd.flags |= LOOKUP_FOLLOW;
1815 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1816 while (unlikely(!filp)) { /* trailing symlink */
1818 struct inode *inode = path.dentry->d_inode;
1821 /* S_ISDIR part is a temporary automount kludge */
1822 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
1827 * This is subtle. Instead of calling do_follow_link() we do
1828 * the thing by hands. The reason is that this way we have zero
1829 * link_count and path_walk() (called from ->follow_link)
1830 * honoring LOOKUP_PARENT. After that we have the parent and
1831 * last component, i.e. we are in the same situation as after
1832 * the first path_walk(). Well, almost - if the last component
1833 * is normal we get its copy stored in nd->last.name and we will
1834 * have to putname() it when we are done. Procfs-like symlinks
1835 * just set LAST_BIND.
1837 nd.flags |= LOOKUP_PARENT;
1838 error = security_inode_follow_link(path.dentry, &nd);
1841 error = __do_follow_link(&path, &nd, &cookie);
1842 if (unlikely(error)) {
1843 /* nd.path had been dropped */
1844 if (!IS_ERR(cookie) && inode->i_op->put_link)
1845 inode->i_op->put_link(path.dentry, &nd, cookie);
1847 release_open_intent(&nd);
1848 filp = ERR_PTR(error);
1852 nd.flags &= ~LOOKUP_PARENT;
1853 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1854 if (inode->i_op->put_link)
1855 inode->i_op->put_link(holder.dentry, &nd, cookie);
1861 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1868 path_put_conditional(&path, &nd);
1869 if (!IS_ERR(nd.intent.open.file))
1870 release_open_intent(&nd);
1873 filp = ERR_PTR(error);
1878 * filp_open - open file and return file pointer
1880 * @filename: path to open
1881 * @flags: open flags as per the open(2) second argument
1882 * @mode: mode for the new file if O_CREAT is set, else ignored
1884 * This is the helper to open a file from kernelspace if you really
1885 * have to. But in generally you should not do this, so please move
1886 * along, nothing to see here..
1888 struct file *filp_open(const char *filename, int flags, int mode)
1890 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1892 EXPORT_SYMBOL(filp_open);
1895 * lookup_create - lookup a dentry, creating it if it doesn't exist
1896 * @nd: nameidata info
1897 * @is_dir: directory flag
1899 * Simple function to lookup and return a dentry and create it
1900 * if it doesn't exist. Is SMP-safe.
1902 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1904 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1906 struct dentry *dentry = ERR_PTR(-EEXIST);
1908 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1910 * Yucky last component or no last component at all?
1911 * (foo/., foo/.., /////)
1913 if (nd->last_type != LAST_NORM)
1915 nd->flags &= ~LOOKUP_PARENT;
1916 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1917 nd->intent.open.flags = O_EXCL;
1920 * Do the final lookup.
1922 dentry = lookup_hash(nd);
1926 if (dentry->d_inode)
1929 * Special case - lookup gave negative, but... we had foo/bar/
1930 * From the vfs_mknod() POV we just have a negative dentry -
1931 * all is fine. Let's be bastards - you had / on the end, you've
1932 * been asking for (non-existent) directory. -ENOENT for you.
1934 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1936 dentry = ERR_PTR(-ENOENT);
1941 dentry = ERR_PTR(-EEXIST);
1945 EXPORT_SYMBOL_GPL(lookup_create);
1947 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1949 int error = may_create(dir, dentry);
1954 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1957 if (!dir->i_op->mknod)
1960 error = devcgroup_inode_mknod(mode, dev);
1964 error = security_inode_mknod(dir, dentry, mode, dev);
1968 error = dir->i_op->mknod(dir, dentry, mode, dev);
1970 fsnotify_create(dir, dentry);
1974 static int may_mknod(mode_t mode)
1976 switch (mode & S_IFMT) {
1982 case 0: /* zero mode translates to S_IFREG */
1991 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1996 struct dentry *dentry;
1997 struct nameidata nd;
2002 error = user_path_parent(dfd, filename, &nd, &tmp);
2006 dentry = lookup_create(&nd, 0);
2007 if (IS_ERR(dentry)) {
2008 error = PTR_ERR(dentry);
2011 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2012 mode &= ~current_umask();
2013 error = may_mknod(mode);
2016 error = mnt_want_write(nd.path.mnt);
2019 error = security_path_mknod(&nd.path, dentry, mode, dev);
2021 goto out_drop_write;
2022 switch (mode & S_IFMT) {
2023 case 0: case S_IFREG:
2024 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2026 case S_IFCHR: case S_IFBLK:
2027 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2028 new_decode_dev(dev));
2030 case S_IFIFO: case S_IFSOCK:
2031 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2035 mnt_drop_write(nd.path.mnt);
2039 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2046 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2048 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2051 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2053 int error = may_create(dir, dentry);
2058 if (!dir->i_op->mkdir)
2061 mode &= (S_IRWXUGO|S_ISVTX);
2062 error = security_inode_mkdir(dir, dentry, mode);
2066 error = dir->i_op->mkdir(dir, dentry, mode);
2068 fsnotify_mkdir(dir, dentry);
2072 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2076 struct dentry *dentry;
2077 struct nameidata nd;
2079 error = user_path_parent(dfd, pathname, &nd, &tmp);
2083 dentry = lookup_create(&nd, 1);
2084 error = PTR_ERR(dentry);
2088 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2089 mode &= ~current_umask();
2090 error = mnt_want_write(nd.path.mnt);
2093 error = security_path_mkdir(&nd.path, dentry, mode);
2095 goto out_drop_write;
2096 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2098 mnt_drop_write(nd.path.mnt);
2102 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2109 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2111 return sys_mkdirat(AT_FDCWD, pathname, mode);
2115 * We try to drop the dentry early: we should have
2116 * a usage count of 2 if we're the only user of this
2117 * dentry, and if that is true (possibly after pruning
2118 * the dcache), then we drop the dentry now.
2120 * A low-level filesystem can, if it choses, legally
2123 * if (!d_unhashed(dentry))
2126 * if it cannot handle the case of removing a directory
2127 * that is still in use by something else..
2129 void dentry_unhash(struct dentry *dentry)
2132 shrink_dcache_parent(dentry);
2133 spin_lock(&dcache_lock);
2134 spin_lock(&dentry->d_lock);
2135 if (atomic_read(&dentry->d_count) == 2)
2137 spin_unlock(&dentry->d_lock);
2138 spin_unlock(&dcache_lock);
2141 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2143 int error = may_delete(dir, dentry, 1);
2148 if (!dir->i_op->rmdir)
2151 mutex_lock(&dentry->d_inode->i_mutex);
2152 dentry_unhash(dentry);
2153 if (d_mountpoint(dentry))
2156 error = security_inode_rmdir(dir, dentry);
2158 error = dir->i_op->rmdir(dir, dentry);
2160 dentry->d_inode->i_flags |= S_DEAD;
2165 mutex_unlock(&dentry->d_inode->i_mutex);
2174 static long do_rmdir(int dfd, const char __user *pathname)
2178 struct dentry *dentry;
2179 struct nameidata nd;
2181 error = user_path_parent(dfd, pathname, &nd, &name);
2185 switch(nd.last_type) {
2197 nd.flags &= ~LOOKUP_PARENT;
2199 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2200 dentry = lookup_hash(&nd);
2201 error = PTR_ERR(dentry);
2204 error = mnt_want_write(nd.path.mnt);
2207 error = security_path_rmdir(&nd.path, dentry);
2210 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2212 mnt_drop_write(nd.path.mnt);
2216 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2223 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2225 return do_rmdir(AT_FDCWD, pathname);
2228 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2230 int error = may_delete(dir, dentry, 0);
2235 if (!dir->i_op->unlink)
2238 mutex_lock(&dentry->d_inode->i_mutex);
2239 if (d_mountpoint(dentry))
2242 error = security_inode_unlink(dir, dentry);
2244 error = dir->i_op->unlink(dir, dentry);
2249 mutex_unlock(&dentry->d_inode->i_mutex);
2251 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2252 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2253 fsnotify_link_count(dentry->d_inode);
2261 * Make sure that the actual truncation of the file will occur outside its
2262 * directory's i_mutex. Truncate can take a long time if there is a lot of
2263 * writeout happening, and we don't want to prevent access to the directory
2264 * while waiting on the I/O.
2266 static long do_unlinkat(int dfd, const char __user *pathname)
2270 struct dentry *dentry;
2271 struct nameidata nd;
2272 struct inode *inode = NULL;
2274 error = user_path_parent(dfd, pathname, &nd, &name);
2279 if (nd.last_type != LAST_NORM)
2282 nd.flags &= ~LOOKUP_PARENT;
2284 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2285 dentry = lookup_hash(&nd);
2286 error = PTR_ERR(dentry);
2287 if (!IS_ERR(dentry)) {
2288 /* Why not before? Because we want correct error value */
2289 if (nd.last.name[nd.last.len])
2291 inode = dentry->d_inode;
2293 atomic_inc(&inode->i_count);
2294 error = mnt_want_write(nd.path.mnt);
2297 error = security_path_unlink(&nd.path, dentry);
2300 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2302 mnt_drop_write(nd.path.mnt);
2306 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2308 iput(inode); /* truncate the inode here */
2315 error = !dentry->d_inode ? -ENOENT :
2316 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2320 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2322 if ((flag & ~AT_REMOVEDIR) != 0)
2325 if (flag & AT_REMOVEDIR)
2326 return do_rmdir(dfd, pathname);
2328 return do_unlinkat(dfd, pathname);
2331 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2333 return do_unlinkat(AT_FDCWD, pathname);
2336 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2338 int error = may_create(dir, dentry);
2343 if (!dir->i_op->symlink)
2346 error = security_inode_symlink(dir, dentry, oldname);
2350 error = dir->i_op->symlink(dir, dentry, oldname);
2352 fsnotify_create(dir, dentry);
2356 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2357 int, newdfd, const char __user *, newname)
2362 struct dentry *dentry;
2363 struct nameidata nd;
2365 from = getname(oldname);
2367 return PTR_ERR(from);
2369 error = user_path_parent(newdfd, newname, &nd, &to);
2373 dentry = lookup_create(&nd, 0);
2374 error = PTR_ERR(dentry);
2378 error = mnt_want_write(nd.path.mnt);
2381 error = security_path_symlink(&nd.path, dentry, from);
2383 goto out_drop_write;
2384 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2386 mnt_drop_write(nd.path.mnt);
2390 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2398 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2400 return sys_symlinkat(oldname, AT_FDCWD, newname);
2403 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2405 struct inode *inode = old_dentry->d_inode;
2411 error = may_create(dir, new_dentry);
2415 if (dir->i_sb != inode->i_sb)
2419 * A link to an append-only or immutable file cannot be created.
2421 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2423 if (!dir->i_op->link)
2425 if (S_ISDIR(inode->i_mode))
2428 error = security_inode_link(old_dentry, dir, new_dentry);
2432 mutex_lock(&inode->i_mutex);
2433 error = dir->i_op->link(old_dentry, dir, new_dentry);
2434 mutex_unlock(&inode->i_mutex);
2436 fsnotify_link(dir, inode, new_dentry);
2441 * Hardlinks are often used in delicate situations. We avoid
2442 * security-related surprises by not following symlinks on the
2445 * We don't follow them on the oldname either to be compatible
2446 * with linux 2.0, and to avoid hard-linking to directories
2447 * and other special files. --ADM
2449 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2450 int, newdfd, const char __user *, newname, int, flags)
2452 struct dentry *new_dentry;
2453 struct nameidata nd;
2454 struct path old_path;
2458 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2461 error = user_path_at(olddfd, oldname,
2462 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2467 error = user_path_parent(newdfd, newname, &nd, &to);
2471 if (old_path.mnt != nd.path.mnt)
2473 new_dentry = lookup_create(&nd, 0);
2474 error = PTR_ERR(new_dentry);
2475 if (IS_ERR(new_dentry))
2477 error = mnt_want_write(nd.path.mnt);
2480 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2482 goto out_drop_write;
2483 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2485 mnt_drop_write(nd.path.mnt);
2489 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2494 path_put(&old_path);
2499 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2501 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2505 * The worst of all namespace operations - renaming directory. "Perverted"
2506 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2508 * a) we can get into loop creation. Check is done in is_subdir().
2509 * b) race potential - two innocent renames can create a loop together.
2510 * That's where 4.4 screws up. Current fix: serialization on
2511 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2513 * c) we have to lock _three_ objects - parents and victim (if it exists).
2514 * And that - after we got ->i_mutex on parents (until then we don't know
2515 * whether the target exists). Solution: try to be smart with locking
2516 * order for inodes. We rely on the fact that tree topology may change
2517 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2518 * move will be locked. Thus we can rank directories by the tree
2519 * (ancestors first) and rank all non-directories after them.
2520 * That works since everybody except rename does "lock parent, lookup,
2521 * lock child" and rename is under ->s_vfs_rename_mutex.
2522 * HOWEVER, it relies on the assumption that any object with ->lookup()
2523 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2524 * we'd better make sure that there's no link(2) for them.
2525 * d) some filesystems don't support opened-but-unlinked directories,
2526 * either because of layout or because they are not ready to deal with
2527 * all cases correctly. The latter will be fixed (taking this sort of
2528 * stuff into VFS), but the former is not going away. Solution: the same
2529 * trick as in rmdir().
2530 * e) conversion from fhandle to dentry may come in the wrong moment - when
2531 * we are removing the target. Solution: we will have to grab ->i_mutex
2532 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2533 * ->i_mutex on parents, which works but leads to some truly excessive
2536 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2537 struct inode *new_dir, struct dentry *new_dentry)
2540 struct inode *target;
2543 * If we are going to change the parent - check write permissions,
2544 * we'll need to flip '..'.
2546 if (new_dir != old_dir) {
2547 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2552 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2556 target = new_dentry->d_inode;
2558 mutex_lock(&target->i_mutex);
2559 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2563 dentry_unhash(new_dentry);
2564 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2568 target->i_flags |= S_DEAD;
2569 dont_mount(new_dentry);
2571 mutex_unlock(&target->i_mutex);
2572 if (d_unhashed(new_dentry))
2573 d_rehash(new_dentry);
2577 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2578 d_move(old_dentry,new_dentry);
2582 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2583 struct inode *new_dir, struct dentry *new_dentry)
2585 struct inode *target;
2588 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2593 target = new_dentry->d_inode;
2595 mutex_lock(&target->i_mutex);
2596 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2599 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2602 dont_mount(new_dentry);
2603 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2604 d_move(old_dentry, new_dentry);
2607 mutex_unlock(&target->i_mutex);
2612 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2613 struct inode *new_dir, struct dentry *new_dentry)
2616 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2617 const unsigned char *old_name;
2619 if (old_dentry->d_inode == new_dentry->d_inode)
2622 error = may_delete(old_dir, old_dentry, is_dir);
2626 if (!new_dentry->d_inode)
2627 error = may_create(new_dir, new_dentry);
2629 error = may_delete(new_dir, new_dentry, is_dir);
2633 if (!old_dir->i_op->rename)
2636 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2639 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2641 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2643 fsnotify_move(old_dir, new_dir, old_name, is_dir,
2644 new_dentry->d_inode, old_dentry);
2645 fsnotify_oldname_free(old_name);
2650 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2651 int, newdfd, const char __user *, newname)
2653 struct dentry *old_dir, *new_dir;
2654 struct dentry *old_dentry, *new_dentry;
2655 struct dentry *trap;
2656 struct nameidata oldnd, newnd;
2661 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2665 error = user_path_parent(newdfd, newname, &newnd, &to);
2670 if (oldnd.path.mnt != newnd.path.mnt)
2673 old_dir = oldnd.path.dentry;
2675 if (oldnd.last_type != LAST_NORM)
2678 new_dir = newnd.path.dentry;
2679 if (newnd.last_type != LAST_NORM)
2682 oldnd.flags &= ~LOOKUP_PARENT;
2683 newnd.flags &= ~LOOKUP_PARENT;
2684 newnd.flags |= LOOKUP_RENAME_TARGET;
2686 trap = lock_rename(new_dir, old_dir);
2688 old_dentry = lookup_hash(&oldnd);
2689 error = PTR_ERR(old_dentry);
2690 if (IS_ERR(old_dentry))
2692 /* source must exist */
2694 if (!old_dentry->d_inode)
2696 /* unless the source is a directory trailing slashes give -ENOTDIR */
2697 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2699 if (oldnd.last.name[oldnd.last.len])
2701 if (newnd.last.name[newnd.last.len])
2704 /* source should not be ancestor of target */
2706 if (old_dentry == trap)
2708 new_dentry = lookup_hash(&newnd);
2709 error = PTR_ERR(new_dentry);
2710 if (IS_ERR(new_dentry))
2712 /* target should not be an ancestor of source */
2714 if (new_dentry == trap)
2717 error = mnt_want_write(oldnd.path.mnt);
2720 error = security_path_rename(&oldnd.path, old_dentry,
2721 &newnd.path, new_dentry);
2724 error = vfs_rename(old_dir->d_inode, old_dentry,
2725 new_dir->d_inode, new_dentry);
2727 mnt_drop_write(oldnd.path.mnt);
2733 unlock_rename(new_dir, old_dir);
2735 path_put(&newnd.path);
2738 path_put(&oldnd.path);
2744 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2746 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2749 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2753 len = PTR_ERR(link);
2758 if (len > (unsigned) buflen)
2760 if (copy_to_user(buffer, link, len))
2767 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2768 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2769 * using) it for any given inode is up to filesystem.
2771 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2773 struct nameidata nd;
2778 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2780 return PTR_ERR(cookie);
2782 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2783 if (dentry->d_inode->i_op->put_link)
2784 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2788 int vfs_follow_link(struct nameidata *nd, const char *link)
2790 return __vfs_follow_link(nd, link);
2793 /* get the link contents into pagecache */
2794 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2798 struct address_space *mapping = dentry->d_inode->i_mapping;
2799 page = read_mapping_page(mapping, 0, NULL);
2804 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2808 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2810 struct page *page = NULL;
2811 char *s = page_getlink(dentry, &page);
2812 int res = vfs_readlink(dentry,buffer,buflen,s);
2815 page_cache_release(page);
2820 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2822 struct page *page = NULL;
2823 nd_set_link(nd, page_getlink(dentry, &page));
2827 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2829 struct page *page = cookie;
2833 page_cache_release(page);
2838 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2840 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2842 struct address_space *mapping = inode->i_mapping;
2847 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2849 flags |= AOP_FLAG_NOFS;
2852 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2853 flags, &page, &fsdata);
2857 kaddr = kmap_atomic(page, KM_USER0);
2858 memcpy(kaddr, symname, len-1);
2859 kunmap_atomic(kaddr, KM_USER0);
2861 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2868 mark_inode_dirty(inode);
2874 int page_symlink(struct inode *inode, const char *symname, int len)
2876 return __page_symlink(inode, symname, len,
2877 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2880 const struct inode_operations page_symlink_inode_operations = {
2881 .readlink = generic_readlink,
2882 .follow_link = page_follow_link_light,
2883 .put_link = page_put_link,
2886 EXPORT_SYMBOL(user_path_at);
2887 EXPORT_SYMBOL(follow_down);
2888 EXPORT_SYMBOL(follow_up);
2889 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2890 EXPORT_SYMBOL(getname);
2891 EXPORT_SYMBOL(lock_rename);
2892 EXPORT_SYMBOL(lookup_one_len);
2893 EXPORT_SYMBOL(page_follow_link_light);
2894 EXPORT_SYMBOL(page_put_link);
2895 EXPORT_SYMBOL(page_readlink);
2896 EXPORT_SYMBOL(__page_symlink);
2897 EXPORT_SYMBOL(page_symlink);
2898 EXPORT_SYMBOL(page_symlink_inode_operations);
2899 EXPORT_SYMBOL(path_lookup);
2900 EXPORT_SYMBOL(kern_path);
2901 EXPORT_SYMBOL(vfs_path_lookup);
2902 EXPORT_SYMBOL(inode_permission);
2903 EXPORT_SYMBOL(file_permission);
2904 EXPORT_SYMBOL(unlock_rename);
2905 EXPORT_SYMBOL(vfs_create);
2906 EXPORT_SYMBOL(vfs_follow_link);
2907 EXPORT_SYMBOL(vfs_link);
2908 EXPORT_SYMBOL(vfs_mkdir);
2909 EXPORT_SYMBOL(vfs_mknod);
2910 EXPORT_SYMBOL(generic_permission);
2911 EXPORT_SYMBOL(vfs_readlink);
2912 EXPORT_SYMBOL(vfs_rename);
2913 EXPORT_SYMBOL(vfs_rmdir);
2914 EXPORT_SYMBOL(vfs_symlink);
2915 EXPORT_SYMBOL(vfs_unlink);
2916 EXPORT_SYMBOL(dentry_unhash);
2917 EXPORT_SYMBOL(generic_readlink);