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;
599 br_read_lock(vfsmount_lock);
600 parent = path->mnt->mnt_parent;
601 if (parent == path->mnt) {
602 br_read_unlock(vfsmount_lock);
606 mountpoint = dget(path->mnt->mnt_mountpoint);
607 br_read_unlock(vfsmount_lock);
609 path->dentry = mountpoint;
616 * serialization is taken care of in namespace.c
618 static int __follow_mount(struct path *path)
621 while (d_mountpoint(path->dentry)) {
622 struct vfsmount *mounted = lookup_mnt(path);
629 path->dentry = dget(mounted->mnt_root);
635 static void follow_mount(struct path *path)
637 while (d_mountpoint(path->dentry)) {
638 struct vfsmount *mounted = lookup_mnt(path);
644 path->dentry = dget(mounted->mnt_root);
648 int follow_down(struct path *path)
650 struct vfsmount *mounted;
652 mounted = lookup_mnt(path);
657 path->dentry = dget(mounted->mnt_root);
663 static __always_inline void follow_dotdot(struct nameidata *nd)
668 struct dentry *old = nd->path.dentry;
670 if (nd->path.dentry == nd->root.dentry &&
671 nd->path.mnt == nd->root.mnt) {
674 if (nd->path.dentry != nd->path.mnt->mnt_root) {
675 /* rare case of legitimate dget_parent()... */
676 nd->path.dentry = dget_parent(nd->path.dentry);
680 if (!follow_up(&nd->path))
683 follow_mount(&nd->path);
687 * Allocate a dentry with name and parent, and perform a parent
688 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
689 * on error. parent->d_inode->i_mutex must be held. d_lookup must
690 * have verified that no child exists while under i_mutex.
692 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
693 struct qstr *name, struct nameidata *nd)
695 struct inode *inode = parent->d_inode;
696 struct dentry *dentry;
699 /* Don't create child dentry for a dead directory. */
700 if (unlikely(IS_DEADDIR(inode)))
701 return ERR_PTR(-ENOENT);
703 dentry = d_alloc(parent, name);
704 if (unlikely(!dentry))
705 return ERR_PTR(-ENOMEM);
707 old = inode->i_op->lookup(inode, dentry, nd);
716 * It's more convoluted than I'd like it to be, but... it's still fairly
717 * small and for now I'd prefer to have fast path as straight as possible.
718 * It _is_ time-critical.
720 static int do_lookup(struct nameidata *nd, struct qstr *name,
723 struct vfsmount *mnt = nd->path.mnt;
724 struct dentry *dentry, *parent;
727 * See if the low-level filesystem might want
728 * to use its own hash..
730 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
731 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
732 nd->path.dentry->d_inode, name);
738 * Rename seqlock is not required here because in the off chance
739 * of a false negative due to a concurrent rename, we're going to
740 * do the non-racy lookup, below.
742 dentry = __d_lookup(nd->path.dentry, name);
746 if (dentry->d_op && dentry->d_op->d_revalidate)
747 goto need_revalidate;
750 path->dentry = dentry;
751 __follow_mount(path);
755 parent = nd->path.dentry;
756 dir = parent->d_inode;
758 mutex_lock(&dir->i_mutex);
760 * First re-do the cached lookup just in case it was created
761 * while we waited for the directory semaphore, or the first
762 * lookup failed due to an unrelated rename.
764 * This could use version numbering or similar to avoid unnecessary
765 * cache lookups, but then we'd have to do the first lookup in the
766 * non-racy way. However in the common case here, everything should
767 * be hot in cache, so would it be a big win?
769 dentry = d_lookup(parent, name);
770 if (likely(!dentry)) {
771 dentry = d_alloc_and_lookup(parent, name, nd);
772 mutex_unlock(&dir->i_mutex);
778 * Uhhuh! Nasty case: the cache was re-populated while
779 * we waited on the semaphore. Need to revalidate.
781 mutex_unlock(&dir->i_mutex);
785 dentry = do_revalidate(dentry, nd);
793 return PTR_ERR(dentry);
797 * This is a temporary kludge to deal with "automount" symlinks; proper
798 * solution is to trigger them on follow_mount(), so that do_lookup()
799 * would DTRT. To be killed before 2.6.34-final.
801 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
803 return inode && unlikely(inode->i_op->follow_link) &&
804 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
809 * This is the basic name resolution function, turning a pathname into
810 * the final dentry. We expect 'base' to be positive and a directory.
812 * Returns 0 and nd will have valid dentry and mnt on success.
813 * Returns error and drops reference to input namei data on failure.
815 static int link_path_walk(const char *name, struct nameidata *nd)
820 unsigned int lookup_flags = nd->flags;
827 inode = nd->path.dentry->d_inode;
829 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
831 /* At this point we know we have a real path component. */
837 nd->flags |= LOOKUP_CONTINUE;
838 err = exec_permission(inode);
843 c = *(const unsigned char *)name;
845 hash = init_name_hash();
848 hash = partial_name_hash(c, hash);
849 c = *(const unsigned char *)name;
850 } while (c && (c != '/'));
851 this.len = name - (const char *) this.name;
852 this.hash = end_name_hash(hash);
854 /* remove trailing slashes? */
857 while (*++name == '/');
859 goto last_with_slashes;
862 * "." and ".." are special - ".." especially so because it has
863 * to be able to know about the current root directory and
864 * parent relationships.
866 if (this.name[0] == '.') switch (this.len) {
870 if (this.name[1] != '.')
873 inode = nd->path.dentry->d_inode;
878 /* This does the actual lookups.. */
879 err = do_lookup(nd, &this, &next);
884 inode = next.dentry->d_inode;
888 if (inode->i_op->follow_link) {
889 err = do_follow_link(&next, nd);
893 inode = nd->path.dentry->d_inode;
897 path_to_nameidata(&next, nd);
899 if (!inode->i_op->lookup)
902 /* here ends the main loop */
905 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
907 /* Clear LOOKUP_CONTINUE iff it was previously unset */
908 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
909 if (lookup_flags & LOOKUP_PARENT)
911 if (this.name[0] == '.') switch (this.len) {
915 if (this.name[1] != '.')
918 inode = nd->path.dentry->d_inode;
923 err = do_lookup(nd, &this, &next);
926 inode = next.dentry->d_inode;
927 if (follow_on_final(inode, lookup_flags)) {
928 err = do_follow_link(&next, nd);
931 inode = nd->path.dentry->d_inode;
933 path_to_nameidata(&next, nd);
937 if (lookup_flags & LOOKUP_DIRECTORY) {
939 if (!inode->i_op->lookup)
945 nd->last_type = LAST_NORM;
946 if (this.name[0] != '.')
949 nd->last_type = LAST_DOT;
950 else if (this.len == 2 && this.name[1] == '.')
951 nd->last_type = LAST_DOTDOT;
956 * We bypassed the ordinary revalidation routines.
957 * We may need to check the cached dentry for staleness.
959 if (nd->path.dentry && nd->path.dentry->d_sb &&
960 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
962 /* Note: we do not d_invalidate() */
963 if (!nd->path.dentry->d_op->d_revalidate(
964 nd->path.dentry, nd))
970 path_put_conditional(&next, nd);
978 static int path_walk(const char *name, struct nameidata *nd)
980 struct path save = nd->path;
983 current->total_link_count = 0;
985 /* make sure the stuff we saved doesn't go away */
988 result = link_path_walk(name, nd);
989 if (result == -ESTALE) {
990 /* nd->path had been dropped */
991 current->total_link_count = 0;
994 nd->flags |= LOOKUP_REVAL;
995 result = link_path_walk(name, nd);
1003 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1009 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1012 nd->root.mnt = NULL;
1016 nd->path = nd->root;
1017 path_get(&nd->root);
1018 } else if (dfd == AT_FDCWD) {
1019 get_fs_pwd(current->fs, &nd->path);
1021 struct dentry *dentry;
1023 file = fget_light(dfd, &fput_needed);
1028 dentry = file->f_path.dentry;
1031 if (!S_ISDIR(dentry->d_inode->i_mode))
1034 retval = file_permission(file, MAY_EXEC);
1038 nd->path = file->f_path;
1039 path_get(&file->f_path);
1041 fput_light(file, fput_needed);
1046 fput_light(file, fput_needed);
1051 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1052 static int do_path_lookup(int dfd, const char *name,
1053 unsigned int flags, struct nameidata *nd)
1055 int retval = path_init(dfd, name, flags, nd);
1057 retval = path_walk(name, nd);
1058 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1059 nd->path.dentry->d_inode))
1060 audit_inode(name, nd->path.dentry);
1062 path_put(&nd->root);
1063 nd->root.mnt = NULL;
1068 int path_lookup(const char *name, unsigned int flags,
1069 struct nameidata *nd)
1071 return do_path_lookup(AT_FDCWD, name, flags, nd);
1074 int kern_path(const char *name, unsigned int flags, struct path *path)
1076 struct nameidata nd;
1077 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1084 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1085 * @dentry: pointer to dentry of the base directory
1086 * @mnt: pointer to vfs mount of the base directory
1087 * @name: pointer to file name
1088 * @flags: lookup flags
1089 * @nd: pointer to nameidata
1091 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1092 const char *name, unsigned int flags,
1093 struct nameidata *nd)
1097 /* same as do_path_lookup */
1098 nd->last_type = LAST_ROOT;
1102 nd->path.dentry = dentry;
1104 path_get(&nd->path);
1105 nd->root = nd->path;
1106 path_get(&nd->root);
1108 retval = path_walk(name, nd);
1109 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1110 nd->path.dentry->d_inode))
1111 audit_inode(name, nd->path.dentry);
1113 path_put(&nd->root);
1114 nd->root.mnt = NULL;
1119 static struct dentry *__lookup_hash(struct qstr *name,
1120 struct dentry *base, struct nameidata *nd)
1122 struct inode *inode = base->d_inode;
1123 struct dentry *dentry;
1126 err = exec_permission(inode);
1128 return ERR_PTR(err);
1131 * See if the low-level filesystem might want
1132 * to use its own hash..
1134 if (base->d_op && base->d_op->d_hash) {
1135 err = base->d_op->d_hash(base, inode, name);
1136 dentry = ERR_PTR(err);
1142 * Don't bother with __d_lookup: callers are for creat as
1143 * well as unlink, so a lot of the time it would cost
1146 dentry = d_lookup(base, name);
1148 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1149 dentry = do_revalidate(dentry, nd);
1152 dentry = d_alloc_and_lookup(base, name, nd);
1158 * Restricted form of lookup. Doesn't follow links, single-component only,
1159 * needs parent already locked. Doesn't follow mounts.
1162 static struct dentry *lookup_hash(struct nameidata *nd)
1164 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1167 static int __lookup_one_len(const char *name, struct qstr *this,
1168 struct dentry *base, int len)
1178 hash = init_name_hash();
1180 c = *(const unsigned char *)name++;
1181 if (c == '/' || c == '\0')
1183 hash = partial_name_hash(c, hash);
1185 this->hash = end_name_hash(hash);
1190 * lookup_one_len - filesystem helper to lookup single pathname component
1191 * @name: pathname component to lookup
1192 * @base: base directory to lookup from
1193 * @len: maximum length @len should be interpreted to
1195 * Note that this routine is purely a helper for filesystem usage and should
1196 * not be called by generic code. Also note that by using this function the
1197 * nameidata argument is passed to the filesystem methods and a filesystem
1198 * using this helper needs to be prepared for that.
1200 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1205 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1207 err = __lookup_one_len(name, &this, base, len);
1209 return ERR_PTR(err);
1211 return __lookup_hash(&this, base, NULL);
1214 int user_path_at(int dfd, const char __user *name, unsigned flags,
1217 struct nameidata nd;
1218 char *tmp = getname(name);
1219 int err = PTR_ERR(tmp);
1222 BUG_ON(flags & LOOKUP_PARENT);
1224 err = do_path_lookup(dfd, tmp, flags, &nd);
1232 static int user_path_parent(int dfd, const char __user *path,
1233 struct nameidata *nd, char **name)
1235 char *s = getname(path);
1241 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1251 * It's inline, so penalty for filesystems that don't use sticky bit is
1254 static inline int check_sticky(struct inode *dir, struct inode *inode)
1256 uid_t fsuid = current_fsuid();
1258 if (!(dir->i_mode & S_ISVTX))
1260 if (inode->i_uid == fsuid)
1262 if (dir->i_uid == fsuid)
1264 return !capable(CAP_FOWNER);
1268 * Check whether we can remove a link victim from directory dir, check
1269 * whether the type of victim is right.
1270 * 1. We can't do it if dir is read-only (done in permission())
1271 * 2. We should have write and exec permissions on dir
1272 * 3. We can't remove anything from append-only dir
1273 * 4. We can't do anything with immutable dir (done in permission())
1274 * 5. If the sticky bit on dir is set we should either
1275 * a. be owner of dir, or
1276 * b. be owner of victim, or
1277 * c. have CAP_FOWNER capability
1278 * 6. If the victim is append-only or immutable we can't do antyhing with
1279 * links pointing to it.
1280 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1281 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1282 * 9. We can't remove a root or mountpoint.
1283 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1284 * nfs_async_unlink().
1286 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1290 if (!victim->d_inode)
1293 BUG_ON(victim->d_parent->d_inode != dir);
1294 audit_inode_child(victim, dir);
1296 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1301 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1302 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1305 if (!S_ISDIR(victim->d_inode->i_mode))
1307 if (IS_ROOT(victim))
1309 } else if (S_ISDIR(victim->d_inode->i_mode))
1311 if (IS_DEADDIR(dir))
1313 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1318 /* Check whether we can create an object with dentry child in directory
1320 * 1. We can't do it if child already exists (open has special treatment for
1321 * this case, but since we are inlined it's OK)
1322 * 2. We can't do it if dir is read-only (done in permission())
1323 * 3. We should have write and exec permissions on dir
1324 * 4. We can't do it if dir is immutable (done in permission())
1326 static inline int may_create(struct inode *dir, struct dentry *child)
1330 if (IS_DEADDIR(dir))
1332 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1336 * p1 and p2 should be directories on the same fs.
1338 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1343 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1347 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1349 p = d_ancestor(p2, p1);
1351 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1352 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1356 p = d_ancestor(p1, p2);
1358 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1359 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1363 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1364 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1368 void unlock_rename(struct dentry *p1, struct dentry *p2)
1370 mutex_unlock(&p1->d_inode->i_mutex);
1372 mutex_unlock(&p2->d_inode->i_mutex);
1373 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1377 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1378 struct nameidata *nd)
1380 int error = may_create(dir, dentry);
1385 if (!dir->i_op->create)
1386 return -EACCES; /* shouldn't it be ENOSYS? */
1389 error = security_inode_create(dir, dentry, mode);
1392 error = dir->i_op->create(dir, dentry, mode, nd);
1394 fsnotify_create(dir, dentry);
1398 int may_open(struct path *path, int acc_mode, int flag)
1400 struct dentry *dentry = path->dentry;
1401 struct inode *inode = dentry->d_inode;
1407 switch (inode->i_mode & S_IFMT) {
1411 if (acc_mode & MAY_WRITE)
1416 if (path->mnt->mnt_flags & MNT_NODEV)
1425 error = inode_permission(inode, acc_mode);
1430 * An append-only file must be opened in append mode for writing.
1432 if (IS_APPEND(inode)) {
1433 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1439 /* O_NOATIME can only be set by the owner or superuser */
1440 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1444 * Ensure there are no outstanding leases on the file.
1446 return break_lease(inode, flag);
1449 static int handle_truncate(struct path *path)
1451 struct inode *inode = path->dentry->d_inode;
1452 int error = get_write_access(inode);
1456 * Refuse to truncate files with mandatory locks held on them.
1458 error = locks_verify_locked(inode);
1460 error = security_path_truncate(path);
1462 error = do_truncate(path->dentry, 0,
1463 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1466 put_write_access(inode);
1471 * Be careful about ever adding any more callers of this
1472 * function. Its flags must be in the namei format, not
1473 * what get passed to sys_open().
1475 static int __open_namei_create(struct nameidata *nd, struct path *path,
1476 int open_flag, int mode)
1479 struct dentry *dir = nd->path.dentry;
1481 if (!IS_POSIXACL(dir->d_inode))
1482 mode &= ~current_umask();
1483 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1486 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1488 mutex_unlock(&dir->d_inode->i_mutex);
1489 dput(nd->path.dentry);
1490 nd->path.dentry = path->dentry;
1493 /* Don't check for write permission, don't truncate */
1494 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
1498 * Note that while the flag value (low two bits) for sys_open means:
1503 * it is changed into
1504 * 00 - no permissions needed
1505 * 01 - read-permission
1506 * 10 - write-permission
1508 * for the internal routines (ie open_namei()/follow_link() etc)
1509 * This is more logical, and also allows the 00 "no perm needed"
1510 * to be used for symlinks (where the permissions are checked
1514 static inline int open_to_namei_flags(int flag)
1516 if ((flag+1) & O_ACCMODE)
1521 static int open_will_truncate(int flag, struct inode *inode)
1524 * We'll never write to the fs underlying
1527 if (special_file(inode->i_mode))
1529 return (flag & O_TRUNC);
1532 static struct file *finish_open(struct nameidata *nd,
1533 int open_flag, int acc_mode)
1539 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
1540 if (will_truncate) {
1541 error = mnt_want_write(nd->path.mnt);
1545 error = may_open(&nd->path, acc_mode, open_flag);
1548 mnt_drop_write(nd->path.mnt);
1551 filp = nameidata_to_filp(nd);
1552 if (!IS_ERR(filp)) {
1553 error = ima_file_check(filp, acc_mode);
1556 filp = ERR_PTR(error);
1559 if (!IS_ERR(filp)) {
1560 if (will_truncate) {
1561 error = handle_truncate(&nd->path);
1564 filp = ERR_PTR(error);
1569 * It is now safe to drop the mnt write
1570 * because the filp has had a write taken
1574 mnt_drop_write(nd->path.mnt);
1575 path_put(&nd->path);
1579 if (!IS_ERR(nd->intent.open.file))
1580 release_open_intent(nd);
1581 path_put(&nd->path);
1582 return ERR_PTR(error);
1585 static struct file *do_last(struct nameidata *nd, struct path *path,
1586 int open_flag, int acc_mode,
1587 int mode, const char *pathname)
1589 struct dentry *dir = nd->path.dentry;
1591 int error = -EISDIR;
1593 switch (nd->last_type) {
1596 dir = nd->path.dentry;
1598 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1599 if (!dir->d_op->d_revalidate(dir, nd)) {
1606 if (open_flag & O_CREAT)
1610 audit_inode(pathname, dir);
1614 /* trailing slashes? */
1615 if (nd->last.name[nd->last.len]) {
1616 if (open_flag & O_CREAT)
1618 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
1621 /* just plain open? */
1622 if (!(open_flag & O_CREAT)) {
1623 error = do_lookup(nd, &nd->last, path);
1627 if (!path->dentry->d_inode)
1629 if (path->dentry->d_inode->i_op->follow_link)
1632 if (nd->flags & LOOKUP_DIRECTORY) {
1633 if (!path->dentry->d_inode->i_op->lookup)
1636 path_to_nameidata(path, nd);
1637 audit_inode(pathname, nd->path.dentry);
1641 /* OK, it's O_CREAT */
1642 mutex_lock(&dir->d_inode->i_mutex);
1644 path->dentry = lookup_hash(nd);
1645 path->mnt = nd->path.mnt;
1647 error = PTR_ERR(path->dentry);
1648 if (IS_ERR(path->dentry)) {
1649 mutex_unlock(&dir->d_inode->i_mutex);
1653 if (IS_ERR(nd->intent.open.file)) {
1654 error = PTR_ERR(nd->intent.open.file);
1655 goto exit_mutex_unlock;
1658 /* Negative dentry, just create the file */
1659 if (!path->dentry->d_inode) {
1661 * This write is needed to ensure that a
1662 * ro->rw transition does not occur between
1663 * the time when the file is created and when
1664 * a permanent write count is taken through
1665 * the 'struct file' in nameidata_to_filp().
1667 error = mnt_want_write(nd->path.mnt);
1669 goto exit_mutex_unlock;
1670 error = __open_namei_create(nd, path, open_flag, mode);
1672 mnt_drop_write(nd->path.mnt);
1675 filp = nameidata_to_filp(nd);
1676 mnt_drop_write(nd->path.mnt);
1677 path_put(&nd->path);
1678 if (!IS_ERR(filp)) {
1679 error = ima_file_check(filp, acc_mode);
1682 filp = ERR_PTR(error);
1689 * It already exists.
1691 mutex_unlock(&dir->d_inode->i_mutex);
1692 audit_inode(pathname, path->dentry);
1695 if (open_flag & O_EXCL)
1698 if (__follow_mount(path)) {
1700 if (open_flag & O_NOFOLLOW)
1705 if (!path->dentry->d_inode)
1708 if (path->dentry->d_inode->i_op->follow_link)
1711 path_to_nameidata(path, nd);
1713 if (S_ISDIR(path->dentry->d_inode->i_mode))
1716 filp = finish_open(nd, open_flag, acc_mode);
1720 mutex_unlock(&dir->d_inode->i_mutex);
1722 path_put_conditional(path, nd);
1724 if (!IS_ERR(nd->intent.open.file))
1725 release_open_intent(nd);
1726 path_put(&nd->path);
1727 return ERR_PTR(error);
1731 * Note that the low bits of the passed in "open_flag"
1732 * are not the same as in the local variable "flag". See
1733 * open_to_namei_flags() for more details.
1735 struct file *do_filp_open(int dfd, const char *pathname,
1736 int open_flag, int mode, int acc_mode)
1739 struct nameidata nd;
1743 int flag = open_to_namei_flags(open_flag);
1744 int force_reval = 0;
1746 if (!(open_flag & O_CREAT))
1749 /* Must never be set by userspace */
1750 open_flag &= ~FMODE_NONOTIFY;
1753 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1754 * check for O_DSYNC if the need any syncing at all we enforce it's
1755 * always set instead of having to deal with possibly weird behaviour
1756 * for malicious applications setting only __O_SYNC.
1758 if (open_flag & __O_SYNC)
1759 open_flag |= O_DSYNC;
1762 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1764 /* O_TRUNC implies we need access checks for write permissions */
1765 if (open_flag & O_TRUNC)
1766 acc_mode |= MAY_WRITE;
1768 /* Allow the LSM permission hook to distinguish append
1769 access from general write access. */
1770 if (open_flag & O_APPEND)
1771 acc_mode |= MAY_APPEND;
1773 /* find the parent */
1775 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1777 return ERR_PTR(error);
1779 nd.flags |= LOOKUP_REVAL;
1781 current->total_link_count = 0;
1782 error = link_path_walk(pathname, &nd);
1784 filp = ERR_PTR(error);
1787 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
1788 audit_inode(pathname, nd.path.dentry);
1791 * We have the parent and last component.
1795 filp = get_empty_filp();
1798 nd.intent.open.file = filp;
1799 filp->f_flags = open_flag;
1800 nd.intent.open.flags = flag;
1801 nd.intent.open.create_mode = mode;
1802 nd.flags &= ~LOOKUP_PARENT;
1803 nd.flags |= LOOKUP_OPEN;
1804 if (open_flag & O_CREAT) {
1805 nd.flags |= LOOKUP_CREATE;
1806 if (open_flag & O_EXCL)
1807 nd.flags |= LOOKUP_EXCL;
1809 if (open_flag & O_DIRECTORY)
1810 nd.flags |= LOOKUP_DIRECTORY;
1811 if (!(open_flag & O_NOFOLLOW))
1812 nd.flags |= LOOKUP_FOLLOW;
1813 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1814 while (unlikely(!filp)) { /* trailing symlink */
1816 struct inode *inode = path.dentry->d_inode;
1819 /* S_ISDIR part is a temporary automount kludge */
1820 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
1825 * This is subtle. Instead of calling do_follow_link() we do
1826 * the thing by hands. The reason is that this way we have zero
1827 * link_count and path_walk() (called from ->follow_link)
1828 * honoring LOOKUP_PARENT. After that we have the parent and
1829 * last component, i.e. we are in the same situation as after
1830 * the first path_walk(). Well, almost - if the last component
1831 * is normal we get its copy stored in nd->last.name and we will
1832 * have to putname() it when we are done. Procfs-like symlinks
1833 * just set LAST_BIND.
1835 nd.flags |= LOOKUP_PARENT;
1836 error = security_inode_follow_link(path.dentry, &nd);
1839 error = __do_follow_link(&path, &nd, &cookie);
1840 if (unlikely(error)) {
1841 /* nd.path had been dropped */
1842 if (!IS_ERR(cookie) && inode->i_op->put_link)
1843 inode->i_op->put_link(path.dentry, &nd, cookie);
1845 release_open_intent(&nd);
1846 filp = ERR_PTR(error);
1850 nd.flags &= ~LOOKUP_PARENT;
1851 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1852 if (inode->i_op->put_link)
1853 inode->i_op->put_link(holder.dentry, &nd, cookie);
1859 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1866 path_put_conditional(&path, &nd);
1867 if (!IS_ERR(nd.intent.open.file))
1868 release_open_intent(&nd);
1871 filp = ERR_PTR(error);
1876 * filp_open - open file and return file pointer
1878 * @filename: path to open
1879 * @flags: open flags as per the open(2) second argument
1880 * @mode: mode for the new file if O_CREAT is set, else ignored
1882 * This is the helper to open a file from kernelspace if you really
1883 * have to. But in generally you should not do this, so please move
1884 * along, nothing to see here..
1886 struct file *filp_open(const char *filename, int flags, int mode)
1888 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1890 EXPORT_SYMBOL(filp_open);
1893 * lookup_create - lookup a dentry, creating it if it doesn't exist
1894 * @nd: nameidata info
1895 * @is_dir: directory flag
1897 * Simple function to lookup and return a dentry and create it
1898 * if it doesn't exist. Is SMP-safe.
1900 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1902 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1904 struct dentry *dentry = ERR_PTR(-EEXIST);
1906 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1908 * Yucky last component or no last component at all?
1909 * (foo/., foo/.., /////)
1911 if (nd->last_type != LAST_NORM)
1913 nd->flags &= ~LOOKUP_PARENT;
1914 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1915 nd->intent.open.flags = O_EXCL;
1918 * Do the final lookup.
1920 dentry = lookup_hash(nd);
1924 if (dentry->d_inode)
1927 * Special case - lookup gave negative, but... we had foo/bar/
1928 * From the vfs_mknod() POV we just have a negative dentry -
1929 * all is fine. Let's be bastards - you had / on the end, you've
1930 * been asking for (non-existent) directory. -ENOENT for you.
1932 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1934 dentry = ERR_PTR(-ENOENT);
1939 dentry = ERR_PTR(-EEXIST);
1943 EXPORT_SYMBOL_GPL(lookup_create);
1945 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1947 int error = may_create(dir, dentry);
1952 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1955 if (!dir->i_op->mknod)
1958 error = devcgroup_inode_mknod(mode, dev);
1962 error = security_inode_mknod(dir, dentry, mode, dev);
1966 error = dir->i_op->mknod(dir, dentry, mode, dev);
1968 fsnotify_create(dir, dentry);
1972 static int may_mknod(mode_t mode)
1974 switch (mode & S_IFMT) {
1980 case 0: /* zero mode translates to S_IFREG */
1989 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1994 struct dentry *dentry;
1995 struct nameidata nd;
2000 error = user_path_parent(dfd, filename, &nd, &tmp);
2004 dentry = lookup_create(&nd, 0);
2005 if (IS_ERR(dentry)) {
2006 error = PTR_ERR(dentry);
2009 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2010 mode &= ~current_umask();
2011 error = may_mknod(mode);
2014 error = mnt_want_write(nd.path.mnt);
2017 error = security_path_mknod(&nd.path, dentry, mode, dev);
2019 goto out_drop_write;
2020 switch (mode & S_IFMT) {
2021 case 0: case S_IFREG:
2022 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2024 case S_IFCHR: case S_IFBLK:
2025 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2026 new_decode_dev(dev));
2028 case S_IFIFO: case S_IFSOCK:
2029 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2033 mnt_drop_write(nd.path.mnt);
2037 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2044 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2046 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2049 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2051 int error = may_create(dir, dentry);
2056 if (!dir->i_op->mkdir)
2059 mode &= (S_IRWXUGO|S_ISVTX);
2060 error = security_inode_mkdir(dir, dentry, mode);
2064 error = dir->i_op->mkdir(dir, dentry, mode);
2066 fsnotify_mkdir(dir, dentry);
2070 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2074 struct dentry *dentry;
2075 struct nameidata nd;
2077 error = user_path_parent(dfd, pathname, &nd, &tmp);
2081 dentry = lookup_create(&nd, 1);
2082 error = PTR_ERR(dentry);
2086 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2087 mode &= ~current_umask();
2088 error = mnt_want_write(nd.path.mnt);
2091 error = security_path_mkdir(&nd.path, dentry, mode);
2093 goto out_drop_write;
2094 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2096 mnt_drop_write(nd.path.mnt);
2100 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2107 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2109 return sys_mkdirat(AT_FDCWD, pathname, mode);
2113 * We try to drop the dentry early: we should have
2114 * a usage count of 2 if we're the only user of this
2115 * dentry, and if that is true (possibly after pruning
2116 * the dcache), then we drop the dentry now.
2118 * A low-level filesystem can, if it choses, legally
2121 * if (!d_unhashed(dentry))
2124 * if it cannot handle the case of removing a directory
2125 * that is still in use by something else..
2127 void dentry_unhash(struct dentry *dentry)
2130 shrink_dcache_parent(dentry);
2131 spin_lock(&dentry->d_lock);
2132 if (dentry->d_count == 2)
2134 spin_unlock(&dentry->d_lock);
2137 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2139 int error = may_delete(dir, dentry, 1);
2144 if (!dir->i_op->rmdir)
2147 mutex_lock(&dentry->d_inode->i_mutex);
2148 dentry_unhash(dentry);
2149 if (d_mountpoint(dentry))
2152 error = security_inode_rmdir(dir, dentry);
2154 error = dir->i_op->rmdir(dir, dentry);
2156 dentry->d_inode->i_flags |= S_DEAD;
2161 mutex_unlock(&dentry->d_inode->i_mutex);
2170 static long do_rmdir(int dfd, const char __user *pathname)
2174 struct dentry *dentry;
2175 struct nameidata nd;
2177 error = user_path_parent(dfd, pathname, &nd, &name);
2181 switch(nd.last_type) {
2193 nd.flags &= ~LOOKUP_PARENT;
2195 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2196 dentry = lookup_hash(&nd);
2197 error = PTR_ERR(dentry);
2200 error = mnt_want_write(nd.path.mnt);
2203 error = security_path_rmdir(&nd.path, dentry);
2206 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2208 mnt_drop_write(nd.path.mnt);
2212 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2219 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2221 return do_rmdir(AT_FDCWD, pathname);
2224 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2226 int error = may_delete(dir, dentry, 0);
2231 if (!dir->i_op->unlink)
2234 mutex_lock(&dentry->d_inode->i_mutex);
2235 if (d_mountpoint(dentry))
2238 error = security_inode_unlink(dir, dentry);
2240 error = dir->i_op->unlink(dir, dentry);
2245 mutex_unlock(&dentry->d_inode->i_mutex);
2247 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2248 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2249 fsnotify_link_count(dentry->d_inode);
2257 * Make sure that the actual truncation of the file will occur outside its
2258 * directory's i_mutex. Truncate can take a long time if there is a lot of
2259 * writeout happening, and we don't want to prevent access to the directory
2260 * while waiting on the I/O.
2262 static long do_unlinkat(int dfd, const char __user *pathname)
2266 struct dentry *dentry;
2267 struct nameidata nd;
2268 struct inode *inode = NULL;
2270 error = user_path_parent(dfd, pathname, &nd, &name);
2275 if (nd.last_type != LAST_NORM)
2278 nd.flags &= ~LOOKUP_PARENT;
2280 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2281 dentry = lookup_hash(&nd);
2282 error = PTR_ERR(dentry);
2283 if (!IS_ERR(dentry)) {
2284 /* Why not before? Because we want correct error value */
2285 if (nd.last.name[nd.last.len])
2287 inode = dentry->d_inode;
2290 error = mnt_want_write(nd.path.mnt);
2293 error = security_path_unlink(&nd.path, dentry);
2296 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2298 mnt_drop_write(nd.path.mnt);
2302 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2304 iput(inode); /* truncate the inode here */
2311 error = !dentry->d_inode ? -ENOENT :
2312 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2316 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2318 if ((flag & ~AT_REMOVEDIR) != 0)
2321 if (flag & AT_REMOVEDIR)
2322 return do_rmdir(dfd, pathname);
2324 return do_unlinkat(dfd, pathname);
2327 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2329 return do_unlinkat(AT_FDCWD, pathname);
2332 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2334 int error = may_create(dir, dentry);
2339 if (!dir->i_op->symlink)
2342 error = security_inode_symlink(dir, dentry, oldname);
2346 error = dir->i_op->symlink(dir, dentry, oldname);
2348 fsnotify_create(dir, dentry);
2352 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2353 int, newdfd, const char __user *, newname)
2358 struct dentry *dentry;
2359 struct nameidata nd;
2361 from = getname(oldname);
2363 return PTR_ERR(from);
2365 error = user_path_parent(newdfd, newname, &nd, &to);
2369 dentry = lookup_create(&nd, 0);
2370 error = PTR_ERR(dentry);
2374 error = mnt_want_write(nd.path.mnt);
2377 error = security_path_symlink(&nd.path, dentry, from);
2379 goto out_drop_write;
2380 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2382 mnt_drop_write(nd.path.mnt);
2386 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2394 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2396 return sys_symlinkat(oldname, AT_FDCWD, newname);
2399 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2401 struct inode *inode = old_dentry->d_inode;
2407 error = may_create(dir, new_dentry);
2411 if (dir->i_sb != inode->i_sb)
2415 * A link to an append-only or immutable file cannot be created.
2417 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2419 if (!dir->i_op->link)
2421 if (S_ISDIR(inode->i_mode))
2424 error = security_inode_link(old_dentry, dir, new_dentry);
2428 mutex_lock(&inode->i_mutex);
2429 error = dir->i_op->link(old_dentry, dir, new_dentry);
2430 mutex_unlock(&inode->i_mutex);
2432 fsnotify_link(dir, inode, new_dentry);
2437 * Hardlinks are often used in delicate situations. We avoid
2438 * security-related surprises by not following symlinks on the
2441 * We don't follow them on the oldname either to be compatible
2442 * with linux 2.0, and to avoid hard-linking to directories
2443 * and other special files. --ADM
2445 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2446 int, newdfd, const char __user *, newname, int, flags)
2448 struct dentry *new_dentry;
2449 struct nameidata nd;
2450 struct path old_path;
2454 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2457 error = user_path_at(olddfd, oldname,
2458 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2463 error = user_path_parent(newdfd, newname, &nd, &to);
2467 if (old_path.mnt != nd.path.mnt)
2469 new_dentry = lookup_create(&nd, 0);
2470 error = PTR_ERR(new_dentry);
2471 if (IS_ERR(new_dentry))
2473 error = mnt_want_write(nd.path.mnt);
2476 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2478 goto out_drop_write;
2479 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2481 mnt_drop_write(nd.path.mnt);
2485 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2490 path_put(&old_path);
2495 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2497 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2501 * The worst of all namespace operations - renaming directory. "Perverted"
2502 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2504 * a) we can get into loop creation. Check is done in is_subdir().
2505 * b) race potential - two innocent renames can create a loop together.
2506 * That's where 4.4 screws up. Current fix: serialization on
2507 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2509 * c) we have to lock _three_ objects - parents and victim (if it exists).
2510 * And that - after we got ->i_mutex on parents (until then we don't know
2511 * whether the target exists). Solution: try to be smart with locking
2512 * order for inodes. We rely on the fact that tree topology may change
2513 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2514 * move will be locked. Thus we can rank directories by the tree
2515 * (ancestors first) and rank all non-directories after them.
2516 * That works since everybody except rename does "lock parent, lookup,
2517 * lock child" and rename is under ->s_vfs_rename_mutex.
2518 * HOWEVER, it relies on the assumption that any object with ->lookup()
2519 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2520 * we'd better make sure that there's no link(2) for them.
2521 * d) some filesystems don't support opened-but-unlinked directories,
2522 * either because of layout or because they are not ready to deal with
2523 * all cases correctly. The latter will be fixed (taking this sort of
2524 * stuff into VFS), but the former is not going away. Solution: the same
2525 * trick as in rmdir().
2526 * e) conversion from fhandle to dentry may come in the wrong moment - when
2527 * we are removing the target. Solution: we will have to grab ->i_mutex
2528 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2529 * ->i_mutex on parents, which works but leads to some truly excessive
2532 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2533 struct inode *new_dir, struct dentry *new_dentry)
2536 struct inode *target;
2539 * If we are going to change the parent - check write permissions,
2540 * we'll need to flip '..'.
2542 if (new_dir != old_dir) {
2543 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2548 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2552 target = new_dentry->d_inode;
2554 mutex_lock(&target->i_mutex);
2555 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2559 dentry_unhash(new_dentry);
2560 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2564 target->i_flags |= S_DEAD;
2565 dont_mount(new_dentry);
2567 mutex_unlock(&target->i_mutex);
2568 if (d_unhashed(new_dentry))
2569 d_rehash(new_dentry);
2573 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2574 d_move(old_dentry,new_dentry);
2578 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2579 struct inode *new_dir, struct dentry *new_dentry)
2581 struct inode *target;
2584 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2589 target = new_dentry->d_inode;
2591 mutex_lock(&target->i_mutex);
2592 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2595 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2598 dont_mount(new_dentry);
2599 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2600 d_move(old_dentry, new_dentry);
2603 mutex_unlock(&target->i_mutex);
2608 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2609 struct inode *new_dir, struct dentry *new_dentry)
2612 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2613 const unsigned char *old_name;
2615 if (old_dentry->d_inode == new_dentry->d_inode)
2618 error = may_delete(old_dir, old_dentry, is_dir);
2622 if (!new_dentry->d_inode)
2623 error = may_create(new_dir, new_dentry);
2625 error = may_delete(new_dir, new_dentry, is_dir);
2629 if (!old_dir->i_op->rename)
2632 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2635 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2637 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2639 fsnotify_move(old_dir, new_dir, old_name, is_dir,
2640 new_dentry->d_inode, old_dentry);
2641 fsnotify_oldname_free(old_name);
2646 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2647 int, newdfd, const char __user *, newname)
2649 struct dentry *old_dir, *new_dir;
2650 struct dentry *old_dentry, *new_dentry;
2651 struct dentry *trap;
2652 struct nameidata oldnd, newnd;
2657 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2661 error = user_path_parent(newdfd, newname, &newnd, &to);
2666 if (oldnd.path.mnt != newnd.path.mnt)
2669 old_dir = oldnd.path.dentry;
2671 if (oldnd.last_type != LAST_NORM)
2674 new_dir = newnd.path.dentry;
2675 if (newnd.last_type != LAST_NORM)
2678 oldnd.flags &= ~LOOKUP_PARENT;
2679 newnd.flags &= ~LOOKUP_PARENT;
2680 newnd.flags |= LOOKUP_RENAME_TARGET;
2682 trap = lock_rename(new_dir, old_dir);
2684 old_dentry = lookup_hash(&oldnd);
2685 error = PTR_ERR(old_dentry);
2686 if (IS_ERR(old_dentry))
2688 /* source must exist */
2690 if (!old_dentry->d_inode)
2692 /* unless the source is a directory trailing slashes give -ENOTDIR */
2693 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2695 if (oldnd.last.name[oldnd.last.len])
2697 if (newnd.last.name[newnd.last.len])
2700 /* source should not be ancestor of target */
2702 if (old_dentry == trap)
2704 new_dentry = lookup_hash(&newnd);
2705 error = PTR_ERR(new_dentry);
2706 if (IS_ERR(new_dentry))
2708 /* target should not be an ancestor of source */
2710 if (new_dentry == trap)
2713 error = mnt_want_write(oldnd.path.mnt);
2716 error = security_path_rename(&oldnd.path, old_dentry,
2717 &newnd.path, new_dentry);
2720 error = vfs_rename(old_dir->d_inode, old_dentry,
2721 new_dir->d_inode, new_dentry);
2723 mnt_drop_write(oldnd.path.mnt);
2729 unlock_rename(new_dir, old_dir);
2731 path_put(&newnd.path);
2734 path_put(&oldnd.path);
2740 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2742 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2745 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2749 len = PTR_ERR(link);
2754 if (len > (unsigned) buflen)
2756 if (copy_to_user(buffer, link, len))
2763 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2764 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2765 * using) it for any given inode is up to filesystem.
2767 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2769 struct nameidata nd;
2774 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2776 return PTR_ERR(cookie);
2778 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2779 if (dentry->d_inode->i_op->put_link)
2780 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2784 int vfs_follow_link(struct nameidata *nd, const char *link)
2786 return __vfs_follow_link(nd, link);
2789 /* get the link contents into pagecache */
2790 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2794 struct address_space *mapping = dentry->d_inode->i_mapping;
2795 page = read_mapping_page(mapping, 0, NULL);
2800 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2804 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2806 struct page *page = NULL;
2807 char *s = page_getlink(dentry, &page);
2808 int res = vfs_readlink(dentry,buffer,buflen,s);
2811 page_cache_release(page);
2816 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2818 struct page *page = NULL;
2819 nd_set_link(nd, page_getlink(dentry, &page));
2823 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2825 struct page *page = cookie;
2829 page_cache_release(page);
2834 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2836 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2838 struct address_space *mapping = inode->i_mapping;
2843 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2845 flags |= AOP_FLAG_NOFS;
2848 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2849 flags, &page, &fsdata);
2853 kaddr = kmap_atomic(page, KM_USER0);
2854 memcpy(kaddr, symname, len-1);
2855 kunmap_atomic(kaddr, KM_USER0);
2857 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2864 mark_inode_dirty(inode);
2870 int page_symlink(struct inode *inode, const char *symname, int len)
2872 return __page_symlink(inode, symname, len,
2873 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2876 const struct inode_operations page_symlink_inode_operations = {
2877 .readlink = generic_readlink,
2878 .follow_link = page_follow_link_light,
2879 .put_link = page_put_link,
2882 EXPORT_SYMBOL(user_path_at);
2883 EXPORT_SYMBOL(follow_down);
2884 EXPORT_SYMBOL(follow_up);
2885 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2886 EXPORT_SYMBOL(getname);
2887 EXPORT_SYMBOL(lock_rename);
2888 EXPORT_SYMBOL(lookup_one_len);
2889 EXPORT_SYMBOL(page_follow_link_light);
2890 EXPORT_SYMBOL(page_put_link);
2891 EXPORT_SYMBOL(page_readlink);
2892 EXPORT_SYMBOL(__page_symlink);
2893 EXPORT_SYMBOL(page_symlink);
2894 EXPORT_SYMBOL(page_symlink_inode_operations);
2895 EXPORT_SYMBOL(path_lookup);
2896 EXPORT_SYMBOL(kern_path);
2897 EXPORT_SYMBOL(vfs_path_lookup);
2898 EXPORT_SYMBOL(inode_permission);
2899 EXPORT_SYMBOL(file_permission);
2900 EXPORT_SYMBOL(unlock_rename);
2901 EXPORT_SYMBOL(vfs_create);
2902 EXPORT_SYMBOL(vfs_follow_link);
2903 EXPORT_SYMBOL(vfs_link);
2904 EXPORT_SYMBOL(vfs_mkdir);
2905 EXPORT_SYMBOL(vfs_mknod);
2906 EXPORT_SYMBOL(generic_permission);
2907 EXPORT_SYMBOL(vfs_readlink);
2908 EXPORT_SYMBOL(vfs_rename);
2909 EXPORT_SYMBOL(vfs_rmdir);
2910 EXPORT_SYMBOL(vfs_symlink);
2911 EXPORT_SYMBOL(vfs_unlink);
2912 EXPORT_SYMBOL(dentry_unhash);
2913 EXPORT_SYMBOL(generic_readlink);