4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 static int __link_path_walk(const char *name, struct nameidata *nd);
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static int do_getname(const char __user *filename, char *page)
123 unsigned long len = PATH_MAX;
125 if (!segment_eq(get_fs(), KERNEL_DS)) {
126 if ((unsigned long) filename >= TASK_SIZE)
128 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
129 len = TASK_SIZE - (unsigned long) filename;
132 retval = strncpy_from_user(page, filename, len);
136 return -ENAMETOOLONG;
142 char * getname(const char __user * filename)
146 result = ERR_PTR(-ENOMEM);
149 int retval = do_getname(filename, tmp);
154 result = ERR_PTR(retval);
157 audit_getname(result);
161 #ifdef CONFIG_AUDITSYSCALL
162 void putname(const char *name)
164 if (unlikely(!audit_dummy_context()))
169 EXPORT_SYMBOL(putname);
174 * generic_permission - check for access rights on a Posix-like filesystem
175 * @inode: inode to check access rights for
176 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
177 * @check_acl: optional callback to check for Posix ACLs
179 * Used to check for read/write/execute permissions on a file.
180 * We use "fsuid" for this, letting us set arbitrary permissions
181 * for filesystem access without changing the "normal" uids which
182 * are used for other things..
184 int generic_permission(struct inode *inode, int mask,
185 int (*check_acl)(struct inode *inode, int mask))
187 umode_t mode = inode->i_mode;
189 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
191 if (current_fsuid() == inode->i_uid)
194 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
195 int error = check_acl(inode, mask);
196 if (error == -EACCES)
197 goto check_capabilities;
198 else if (error != -EAGAIN)
202 if (in_group_p(inode->i_gid))
207 * If the DACs are ok we don't need any capability check.
209 if ((mask & ~mode) == 0)
214 * Read/write DACs are always overridable.
215 * Executable DACs are overridable if at least one exec bit is set.
217 if (!(mask & MAY_EXEC) || execute_ok(inode))
218 if (capable(CAP_DAC_OVERRIDE))
222 * Searching includes executable on directories, else just read.
224 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
225 if (capable(CAP_DAC_READ_SEARCH))
232 * inode_permission - check for access rights to a given inode
233 * @inode: inode to check permission on
234 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
236 * Used to check for read/write/execute permissions on an inode.
237 * We use "fsuid" for this, letting us set arbitrary permissions
238 * for filesystem access without changing the "normal" uids which
239 * are used for other things.
241 int inode_permission(struct inode *inode, int mask)
245 if (mask & MAY_WRITE) {
246 umode_t mode = inode->i_mode;
249 * Nobody gets write access to a read-only fs.
251 if (IS_RDONLY(inode) &&
252 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
256 * Nobody gets write access to an immutable file.
258 if (IS_IMMUTABLE(inode))
262 if (inode->i_op->permission)
263 retval = inode->i_op->permission(inode, mask);
265 retval = generic_permission(inode, mask, NULL);
270 retval = devcgroup_inode_permission(inode, mask);
274 return security_inode_permission(inode,
275 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
279 * file_permission - check for additional access rights to a given file
280 * @file: file to check access rights for
281 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
283 * Used to check for read/write/execute permissions on an already opened
287 * Do not use this function in new code. All access checks should
288 * be done using inode_permission().
290 int file_permission(struct file *file, int mask)
292 return inode_permission(file->f_path.dentry->d_inode, mask);
296 * get_write_access() gets write permission for a file.
297 * put_write_access() releases this write permission.
298 * This is used for regular files.
299 * We cannot support write (and maybe mmap read-write shared) accesses and
300 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
301 * can have the following values:
302 * 0: no writers, no VM_DENYWRITE mappings
303 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
304 * > 0: (i_writecount) users are writing to the file.
306 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
307 * except for the cases where we don't hold i_writecount yet. Then we need to
308 * use {get,deny}_write_access() - these functions check the sign and refuse
309 * to do the change if sign is wrong. Exclusion between them is provided by
310 * the inode->i_lock spinlock.
313 int get_write_access(struct inode * inode)
315 spin_lock(&inode->i_lock);
316 if (atomic_read(&inode->i_writecount) < 0) {
317 spin_unlock(&inode->i_lock);
320 atomic_inc(&inode->i_writecount);
321 spin_unlock(&inode->i_lock);
326 int deny_write_access(struct file * file)
328 struct inode *inode = file->f_path.dentry->d_inode;
330 spin_lock(&inode->i_lock);
331 if (atomic_read(&inode->i_writecount) > 0) {
332 spin_unlock(&inode->i_lock);
335 atomic_dec(&inode->i_writecount);
336 spin_unlock(&inode->i_lock);
342 * path_get - get a reference to a path
343 * @path: path to get the reference to
345 * Given a path increment the reference count to the dentry and the vfsmount.
347 void path_get(struct path *path)
352 EXPORT_SYMBOL(path_get);
355 * path_put - put a reference to a path
356 * @path: path to put the reference to
358 * Given a path decrement the reference count to the dentry and the vfsmount.
360 void path_put(struct path *path)
365 EXPORT_SYMBOL(path_put);
368 * release_open_intent - free up open intent resources
369 * @nd: pointer to nameidata
371 void release_open_intent(struct nameidata *nd)
373 if (nd->intent.open.file->f_path.dentry == NULL)
374 put_filp(nd->intent.open.file);
376 fput(nd->intent.open.file);
379 static inline struct dentry *
380 do_revalidate(struct dentry *dentry, struct nameidata *nd)
382 int status = dentry->d_op->d_revalidate(dentry, nd);
383 if (unlikely(status <= 0)) {
385 * The dentry failed validation.
386 * If d_revalidate returned 0 attempt to invalidate
387 * the dentry otherwise d_revalidate is asking us
388 * to return a fail status.
391 if (!d_invalidate(dentry)) {
397 dentry = ERR_PTR(status);
404 * Internal lookup() using the new generic dcache.
407 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
409 struct dentry * dentry = __d_lookup(parent, name);
411 /* lockess __d_lookup may fail due to concurrent d_move()
412 * in some unrelated directory, so try with d_lookup
415 dentry = d_lookup(parent, name);
417 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
418 dentry = do_revalidate(dentry, nd);
424 * Short-cut version of permission(), for calling by
425 * path_walk(), when dcache lock is held. Combines parts
426 * of permission() and generic_permission(), and tests ONLY for
427 * MAY_EXEC permission.
429 * If appropriate, check DAC only. If not appropriate, or
430 * short-cut DAC fails, then call permission() to do more
431 * complete permission check.
433 static int exec_permission_lite(struct inode *inode)
435 umode_t mode = inode->i_mode;
437 if (inode->i_op->permission) {
438 int ret = inode->i_op->permission(inode, MAY_EXEC);
444 if (current_fsuid() == inode->i_uid)
446 else if (in_group_p(inode->i_gid))
452 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
457 return security_inode_permission(inode, MAY_EXEC);
461 * This is called when everything else fails, and we actually have
462 * to go to the low-level filesystem to find out what we should do..
464 * We get the directory semaphore, and after getting that we also
465 * make sure that nobody added the entry to the dcache in the meantime..
468 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
470 struct dentry * result;
471 struct inode *dir = parent->d_inode;
473 mutex_lock(&dir->i_mutex);
475 * First re-do the cached lookup just in case it was created
476 * while we waited for the directory semaphore..
478 * FIXME! This could use version numbering or similar to
479 * avoid unnecessary cache lookups.
481 * The "dcache_lock" is purely to protect the RCU list walker
482 * from concurrent renames at this point (we mustn't get false
483 * negatives from the RCU list walk here, unlike the optimistic
486 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
488 result = d_lookup(parent, name);
490 struct dentry *dentry;
492 /* Don't create child dentry for a dead directory. */
493 result = ERR_PTR(-ENOENT);
497 dentry = d_alloc(parent, name);
498 result = ERR_PTR(-ENOMEM);
500 result = dir->i_op->lookup(dir, dentry, nd);
507 mutex_unlock(&dir->i_mutex);
512 * Uhhuh! Nasty case: the cache was re-populated while
513 * we waited on the semaphore. Need to revalidate.
515 mutex_unlock(&dir->i_mutex);
516 if (result->d_op && result->d_op->d_revalidate) {
517 result = do_revalidate(result, nd);
519 result = ERR_PTR(-ENOENT);
525 * Wrapper to retry pathname resolution whenever the underlying
526 * file system returns an ESTALE.
528 * Retry the whole path once, forcing real lookup requests
529 * instead of relying on the dcache.
531 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
533 struct path save = nd->path;
536 /* make sure the stuff we saved doesn't go away */
539 result = __link_path_walk(name, nd);
540 if (result == -ESTALE) {
541 /* nd->path had been dropped */
544 nd->flags |= LOOKUP_REVAL;
545 result = __link_path_walk(name, nd);
553 static __always_inline void set_root(struct nameidata *nd)
556 struct fs_struct *fs = current->fs;
557 read_lock(&fs->lock);
560 read_unlock(&fs->lock);
564 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
578 res = link_path_walk(link, nd);
579 if (nd->depth || res || nd->last_type!=LAST_NORM)
582 * If it is an iterative symlinks resolution in open_namei() we
583 * have to copy the last component. And all that crap because of
584 * bloody create() on broken symlinks. Furrfu...
587 if (unlikely(!name)) {
591 strcpy(name, nd->last.name);
592 nd->last.name = name;
596 return PTR_ERR(link);
599 static void path_put_conditional(struct path *path, struct nameidata *nd)
602 if (path->mnt != nd->path.mnt)
606 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
608 dput(nd->path.dentry);
609 if (nd->path.mnt != path->mnt)
610 mntput(nd->path.mnt);
611 nd->path.mnt = path->mnt;
612 nd->path.dentry = path->dentry;
615 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
619 struct dentry *dentry = path->dentry;
621 touch_atime(path->mnt, dentry);
622 nd_set_link(nd, NULL);
624 if (path->mnt != nd->path.mnt) {
625 path_to_nameidata(path, nd);
629 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
630 error = PTR_ERR(cookie);
631 if (!IS_ERR(cookie)) {
632 char *s = nd_get_link(nd);
635 error = __vfs_follow_link(nd, s);
636 if (dentry->d_inode->i_op->put_link)
637 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
645 * This limits recursive symlink follows to 8, while
646 * limiting consecutive symlinks to 40.
648 * Without that kind of total limit, nasty chains of consecutive
649 * symlinks can cause almost arbitrarily long lookups.
651 static inline int do_follow_link(struct path *path, struct nameidata *nd)
654 if (current->link_count >= MAX_NESTED_LINKS)
656 if (current->total_link_count >= 40)
658 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
660 err = security_inode_follow_link(path->dentry, nd);
663 current->link_count++;
664 current->total_link_count++;
666 err = __do_follow_link(path, nd);
667 current->link_count--;
671 path_put_conditional(path, nd);
676 int follow_up(struct path *path)
678 struct vfsmount *parent;
679 struct dentry *mountpoint;
680 spin_lock(&vfsmount_lock);
681 parent = path->mnt->mnt_parent;
682 if (parent == path->mnt) {
683 spin_unlock(&vfsmount_lock);
687 mountpoint = dget(path->mnt->mnt_mountpoint);
688 spin_unlock(&vfsmount_lock);
690 path->dentry = mountpoint;
696 /* no need for dcache_lock, as serialization is taken care in
699 static int __follow_mount(struct path *path)
702 while (d_mountpoint(path->dentry)) {
703 struct vfsmount *mounted = lookup_mnt(path);
710 path->dentry = dget(mounted->mnt_root);
716 static void follow_mount(struct path *path)
718 while (d_mountpoint(path->dentry)) {
719 struct vfsmount *mounted = lookup_mnt(path);
725 path->dentry = dget(mounted->mnt_root);
729 /* no need for dcache_lock, as serialization is taken care in
732 int follow_down(struct path *path)
734 struct vfsmount *mounted;
736 mounted = lookup_mnt(path);
741 path->dentry = dget(mounted->mnt_root);
747 static __always_inline void follow_dotdot(struct nameidata *nd)
752 struct vfsmount *parent;
753 struct dentry *old = nd->path.dentry;
755 if (nd->path.dentry == nd->root.dentry &&
756 nd->path.mnt == nd->root.mnt) {
759 spin_lock(&dcache_lock);
760 if (nd->path.dentry != nd->path.mnt->mnt_root) {
761 nd->path.dentry = dget(nd->path.dentry->d_parent);
762 spin_unlock(&dcache_lock);
766 spin_unlock(&dcache_lock);
767 spin_lock(&vfsmount_lock);
768 parent = nd->path.mnt->mnt_parent;
769 if (parent == nd->path.mnt) {
770 spin_unlock(&vfsmount_lock);
774 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
775 spin_unlock(&vfsmount_lock);
777 mntput(nd->path.mnt);
778 nd->path.mnt = parent;
780 follow_mount(&nd->path);
784 * It's more convoluted than I'd like it to be, but... it's still fairly
785 * small and for now I'd prefer to have fast path as straight as possible.
786 * It _is_ time-critical.
788 static int do_lookup(struct nameidata *nd, struct qstr *name,
791 struct vfsmount *mnt = nd->path.mnt;
792 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
796 if (dentry->d_op && dentry->d_op->d_revalidate)
797 goto need_revalidate;
800 path->dentry = dentry;
801 __follow_mount(path);
805 dentry = real_lookup(nd->path.dentry, name, nd);
811 dentry = do_revalidate(dentry, nd);
819 return PTR_ERR(dentry);
824 * This is the basic name resolution function, turning a pathname into
825 * the final dentry. We expect 'base' to be positive and a directory.
827 * Returns 0 and nd will have valid dentry and mnt on success.
828 * Returns error and drops reference to input namei data on failure.
830 static int __link_path_walk(const char *name, struct nameidata *nd)
835 unsigned int lookup_flags = nd->flags;
842 inode = nd->path.dentry->d_inode;
844 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
846 /* At this point we know we have a real path component. */
852 nd->flags |= LOOKUP_CONTINUE;
853 err = exec_permission_lite(inode);
858 c = *(const unsigned char *)name;
860 hash = init_name_hash();
863 hash = partial_name_hash(c, hash);
864 c = *(const unsigned char *)name;
865 } while (c && (c != '/'));
866 this.len = name - (const char *) this.name;
867 this.hash = end_name_hash(hash);
869 /* remove trailing slashes? */
872 while (*++name == '/');
874 goto last_with_slashes;
877 * "." and ".." are special - ".." especially so because it has
878 * to be able to know about the current root directory and
879 * parent relationships.
881 if (this.name[0] == '.') switch (this.len) {
885 if (this.name[1] != '.')
888 inode = nd->path.dentry->d_inode;
894 * See if the low-level filesystem might want
895 * to use its own hash..
897 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
898 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
903 /* This does the actual lookups.. */
904 err = do_lookup(nd, &this, &next);
909 inode = next.dentry->d_inode;
913 if (inode->i_op->follow_link) {
914 err = do_follow_link(&next, nd);
918 inode = nd->path.dentry->d_inode;
922 path_to_nameidata(&next, nd);
924 if (!inode->i_op->lookup)
927 /* here ends the main loop */
930 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
932 /* Clear LOOKUP_CONTINUE iff it was previously unset */
933 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
934 if (lookup_flags & LOOKUP_PARENT)
936 if (this.name[0] == '.') switch (this.len) {
940 if (this.name[1] != '.')
943 inode = nd->path.dentry->d_inode;
948 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
949 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
954 err = do_lookup(nd, &this, &next);
957 inode = next.dentry->d_inode;
958 if ((lookup_flags & LOOKUP_FOLLOW)
959 && inode && inode->i_op->follow_link) {
960 err = do_follow_link(&next, nd);
963 inode = nd->path.dentry->d_inode;
965 path_to_nameidata(&next, nd);
969 if (lookup_flags & LOOKUP_DIRECTORY) {
971 if (!inode->i_op->lookup)
977 nd->last_type = LAST_NORM;
978 if (this.name[0] != '.')
981 nd->last_type = LAST_DOT;
982 else if (this.len == 2 && this.name[1] == '.')
983 nd->last_type = LAST_DOTDOT;
988 * We bypassed the ordinary revalidation routines.
989 * We may need to check the cached dentry for staleness.
991 if (nd->path.dentry && nd->path.dentry->d_sb &&
992 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
994 /* Note: we do not d_invalidate() */
995 if (!nd->path.dentry->d_op->d_revalidate(
996 nd->path.dentry, nd))
1002 path_put_conditional(&next, nd);
1005 path_put(&nd->path);
1010 static int path_walk(const char *name, struct nameidata *nd)
1012 current->total_link_count = 0;
1013 return link_path_walk(name, nd);
1016 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1022 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1025 nd->root.mnt = NULL;
1029 nd->path = nd->root;
1030 path_get(&nd->root);
1031 } else if (dfd == AT_FDCWD) {
1032 struct fs_struct *fs = current->fs;
1033 read_lock(&fs->lock);
1036 read_unlock(&fs->lock);
1038 struct dentry *dentry;
1040 file = fget_light(dfd, &fput_needed);
1045 dentry = file->f_path.dentry;
1048 if (!S_ISDIR(dentry->d_inode->i_mode))
1051 retval = file_permission(file, MAY_EXEC);
1055 nd->path = file->f_path;
1056 path_get(&file->f_path);
1058 fput_light(file, fput_needed);
1063 fput_light(file, fput_needed);
1068 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1069 static int do_path_lookup(int dfd, const char *name,
1070 unsigned int flags, struct nameidata *nd)
1072 int retval = path_init(dfd, name, flags, nd);
1074 retval = path_walk(name, nd);
1075 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1076 nd->path.dentry->d_inode))
1077 audit_inode(name, nd->path.dentry);
1079 path_put(&nd->root);
1080 nd->root.mnt = NULL;
1085 int path_lookup(const char *name, unsigned int flags,
1086 struct nameidata *nd)
1088 return do_path_lookup(AT_FDCWD, name, flags, nd);
1091 int kern_path(const char *name, unsigned int flags, struct path *path)
1093 struct nameidata nd;
1094 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1101 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1102 * @dentry: pointer to dentry of the base directory
1103 * @mnt: pointer to vfs mount of the base directory
1104 * @name: pointer to file name
1105 * @flags: lookup flags
1106 * @nd: pointer to nameidata
1108 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1109 const char *name, unsigned int flags,
1110 struct nameidata *nd)
1114 /* same as do_path_lookup */
1115 nd->last_type = LAST_ROOT;
1119 nd->path.dentry = dentry;
1121 path_get(&nd->path);
1122 nd->root = nd->path;
1123 path_get(&nd->root);
1125 retval = path_walk(name, nd);
1126 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1127 nd->path.dentry->d_inode))
1128 audit_inode(name, nd->path.dentry);
1130 path_put(&nd->root);
1131 nd->root.mnt = NULL;
1137 * path_lookup_open - lookup a file path with open intent
1138 * @dfd: the directory to use as base, or AT_FDCWD
1139 * @name: pointer to file name
1140 * @lookup_flags: lookup intent flags
1141 * @nd: pointer to nameidata
1142 * @open_flags: open intent flags
1144 static int path_lookup_open(int dfd, const char *name,
1145 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1147 struct file *filp = get_empty_filp();
1152 nd->intent.open.file = filp;
1153 nd->intent.open.flags = open_flags;
1154 nd->intent.open.create_mode = 0;
1155 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1156 if (IS_ERR(nd->intent.open.file)) {
1158 err = PTR_ERR(nd->intent.open.file);
1159 path_put(&nd->path);
1161 } else if (err != 0)
1162 release_open_intent(nd);
1166 static struct dentry *__lookup_hash(struct qstr *name,
1167 struct dentry *base, struct nameidata *nd)
1169 struct dentry *dentry;
1170 struct inode *inode;
1173 inode = base->d_inode;
1176 * See if the low-level filesystem might want
1177 * to use its own hash..
1179 if (base->d_op && base->d_op->d_hash) {
1180 err = base->d_op->d_hash(base, name);
1181 dentry = ERR_PTR(err);
1186 dentry = cached_lookup(base, name, nd);
1190 /* Don't create child dentry for a dead directory. */
1191 dentry = ERR_PTR(-ENOENT);
1192 if (IS_DEADDIR(inode))
1195 new = d_alloc(base, name);
1196 dentry = ERR_PTR(-ENOMEM);
1199 dentry = inode->i_op->lookup(inode, new, nd);
1210 * Restricted form of lookup. Doesn't follow links, single-component only,
1211 * needs parent already locked. Doesn't follow mounts.
1214 static struct dentry *lookup_hash(struct nameidata *nd)
1218 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1220 return ERR_PTR(err);
1221 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1224 static int __lookup_one_len(const char *name, struct qstr *this,
1225 struct dentry *base, int len)
1235 hash = init_name_hash();
1237 c = *(const unsigned char *)name++;
1238 if (c == '/' || c == '\0')
1240 hash = partial_name_hash(c, hash);
1242 this->hash = end_name_hash(hash);
1247 * lookup_one_len - filesystem helper to lookup single pathname component
1248 * @name: pathname component to lookup
1249 * @base: base directory to lookup from
1250 * @len: maximum length @len should be interpreted to
1252 * Note that this routine is purely a helper for filesystem usage and should
1253 * not be called by generic code. Also note that by using this function the
1254 * nameidata argument is passed to the filesystem methods and a filesystem
1255 * using this helper needs to be prepared for that.
1257 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1262 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1264 err = __lookup_one_len(name, &this, base, len);
1266 return ERR_PTR(err);
1268 err = inode_permission(base->d_inode, MAY_EXEC);
1270 return ERR_PTR(err);
1271 return __lookup_hash(&this, base, NULL);
1275 * lookup_one_noperm - bad hack for sysfs
1276 * @name: pathname component to lookup
1277 * @base: base directory to lookup from
1279 * This is a variant of lookup_one_len that doesn't perform any permission
1280 * checks. It's a horrible hack to work around the braindead sysfs
1281 * architecture and should not be used anywhere else.
1283 * DON'T USE THIS FUNCTION EVER, thanks.
1285 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1290 err = __lookup_one_len(name, &this, base, strlen(name));
1292 return ERR_PTR(err);
1293 return __lookup_hash(&this, base, NULL);
1296 int user_path_at(int dfd, const char __user *name, unsigned flags,
1299 struct nameidata nd;
1300 char *tmp = getname(name);
1301 int err = PTR_ERR(tmp);
1304 BUG_ON(flags & LOOKUP_PARENT);
1306 err = do_path_lookup(dfd, tmp, flags, &nd);
1314 static int user_path_parent(int dfd, const char __user *path,
1315 struct nameidata *nd, char **name)
1317 char *s = getname(path);
1323 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1333 * It's inline, so penalty for filesystems that don't use sticky bit is
1336 static inline int check_sticky(struct inode *dir, struct inode *inode)
1338 uid_t fsuid = current_fsuid();
1340 if (!(dir->i_mode & S_ISVTX))
1342 if (inode->i_uid == fsuid)
1344 if (dir->i_uid == fsuid)
1346 return !capable(CAP_FOWNER);
1350 * Check whether we can remove a link victim from directory dir, check
1351 * whether the type of victim is right.
1352 * 1. We can't do it if dir is read-only (done in permission())
1353 * 2. We should have write and exec permissions on dir
1354 * 3. We can't remove anything from append-only dir
1355 * 4. We can't do anything with immutable dir (done in permission())
1356 * 5. If the sticky bit on dir is set we should either
1357 * a. be owner of dir, or
1358 * b. be owner of victim, or
1359 * c. have CAP_FOWNER capability
1360 * 6. If the victim is append-only or immutable we can't do antyhing with
1361 * links pointing to it.
1362 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1363 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1364 * 9. We can't remove a root or mountpoint.
1365 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1366 * nfs_async_unlink().
1368 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1372 if (!victim->d_inode)
1375 BUG_ON(victim->d_parent->d_inode != dir);
1376 audit_inode_child(victim->d_name.name, victim, dir);
1378 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1383 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1384 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1387 if (!S_ISDIR(victim->d_inode->i_mode))
1389 if (IS_ROOT(victim))
1391 } else if (S_ISDIR(victim->d_inode->i_mode))
1393 if (IS_DEADDIR(dir))
1395 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1400 /* Check whether we can create an object with dentry child in directory
1402 * 1. We can't do it if child already exists (open has special treatment for
1403 * this case, but since we are inlined it's OK)
1404 * 2. We can't do it if dir is read-only (done in permission())
1405 * 3. We should have write and exec permissions on dir
1406 * 4. We can't do it if dir is immutable (done in permission())
1408 static inline int may_create(struct inode *dir, struct dentry *child)
1412 if (IS_DEADDIR(dir))
1414 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1418 * O_DIRECTORY translates into forcing a directory lookup.
1420 static inline int lookup_flags(unsigned int f)
1422 unsigned long retval = LOOKUP_FOLLOW;
1425 retval &= ~LOOKUP_FOLLOW;
1427 if (f & O_DIRECTORY)
1428 retval |= LOOKUP_DIRECTORY;
1434 * p1 and p2 should be directories on the same fs.
1436 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1441 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1445 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1447 p = d_ancestor(p2, p1);
1449 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1450 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1454 p = d_ancestor(p1, p2);
1456 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1457 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1461 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1462 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1466 void unlock_rename(struct dentry *p1, struct dentry *p2)
1468 mutex_unlock(&p1->d_inode->i_mutex);
1470 mutex_unlock(&p2->d_inode->i_mutex);
1471 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1475 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1476 struct nameidata *nd)
1478 int error = may_create(dir, dentry);
1483 if (!dir->i_op->create)
1484 return -EACCES; /* shouldn't it be ENOSYS? */
1487 error = security_inode_create(dir, dentry, mode);
1491 error = dir->i_op->create(dir, dentry, mode, nd);
1493 fsnotify_create(dir, dentry);
1497 int may_open(struct path *path, int acc_mode, int flag)
1499 struct dentry *dentry = path->dentry;
1500 struct inode *inode = dentry->d_inode;
1506 switch (inode->i_mode & S_IFMT) {
1510 if (acc_mode & MAY_WRITE)
1515 if (path->mnt->mnt_flags & MNT_NODEV)
1524 error = inode_permission(inode, acc_mode);
1528 error = ima_path_check(path,
1529 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC),
1534 * An append-only file must be opened in append mode for writing.
1536 if (IS_APPEND(inode)) {
1538 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1544 /* O_NOATIME can only be set by the owner or superuser */
1545 if (flag & O_NOATIME)
1546 if (!is_owner_or_cap(inode)) {
1552 * Ensure there are no outstanding leases on the file.
1554 error = break_lease(inode, flag);
1558 if (flag & O_TRUNC) {
1559 error = get_write_access(inode);
1564 * Refuse to truncate files with mandatory locks held on them.
1566 error = locks_verify_locked(inode);
1568 error = security_path_truncate(path, 0,
1569 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1573 error = do_truncate(dentry, 0,
1574 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1577 put_write_access(inode);
1581 if (flag & FMODE_WRITE)
1586 ima_counts_put(path, acc_mode ?
1587 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1588 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1593 * Be careful about ever adding any more callers of this
1594 * function. Its flags must be in the namei format, not
1595 * what get passed to sys_open().
1597 static int __open_namei_create(struct nameidata *nd, struct path *path,
1601 struct dentry *dir = nd->path.dentry;
1603 if (!IS_POSIXACL(dir->d_inode))
1604 mode &= ~current_umask();
1605 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1608 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1610 mutex_unlock(&dir->d_inode->i_mutex);
1611 dput(nd->path.dentry);
1612 nd->path.dentry = path->dentry;
1615 /* Don't check for write permission, don't truncate */
1616 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1620 * Note that while the flag value (low two bits) for sys_open means:
1625 * it is changed into
1626 * 00 - no permissions needed
1627 * 01 - read-permission
1628 * 10 - write-permission
1630 * for the internal routines (ie open_namei()/follow_link() etc)
1631 * This is more logical, and also allows the 00 "no perm needed"
1632 * to be used for symlinks (where the permissions are checked
1636 static inline int open_to_namei_flags(int flag)
1638 if ((flag+1) & O_ACCMODE)
1643 static int open_will_write_to_fs(int flag, struct inode *inode)
1646 * We'll never write to the fs underlying
1649 if (special_file(inode->i_mode))
1651 return (flag & O_TRUNC);
1655 * Note that the low bits of the passed in "open_flag"
1656 * are not the same as in the local variable "flag". See
1657 * open_to_namei_flags() for more details.
1659 struct file *do_filp_open(int dfd, const char *pathname,
1660 int open_flag, int mode, int acc_mode)
1663 struct nameidata nd;
1669 int flag = open_to_namei_flags(open_flag);
1672 acc_mode = MAY_OPEN | ACC_MODE(flag);
1674 /* O_TRUNC implies we need access checks for write permissions */
1676 acc_mode |= MAY_WRITE;
1678 /* Allow the LSM permission hook to distinguish append
1679 access from general write access. */
1680 if (flag & O_APPEND)
1681 acc_mode |= MAY_APPEND;
1684 * The simplest case - just a plain lookup.
1686 if (!(flag & O_CREAT)) {
1687 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1690 return ERR_PTR(error);
1695 * Create - we need to know the parent.
1697 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1699 return ERR_PTR(error);
1700 error = path_walk(pathname, &nd);
1704 return ERR_PTR(error);
1706 if (unlikely(!audit_dummy_context()))
1707 audit_inode(pathname, nd.path.dentry);
1710 * We have the parent and last component. First of all, check
1711 * that we are not asked to creat(2) an obvious directory - that
1715 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1719 filp = get_empty_filp();
1722 nd.intent.open.file = filp;
1723 nd.intent.open.flags = flag;
1724 nd.intent.open.create_mode = mode;
1725 dir = nd.path.dentry;
1726 nd.flags &= ~LOOKUP_PARENT;
1727 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1729 nd.flags |= LOOKUP_EXCL;
1730 mutex_lock(&dir->d_inode->i_mutex);
1731 path.dentry = lookup_hash(&nd);
1732 path.mnt = nd.path.mnt;
1735 error = PTR_ERR(path.dentry);
1736 if (IS_ERR(path.dentry)) {
1737 mutex_unlock(&dir->d_inode->i_mutex);
1741 if (IS_ERR(nd.intent.open.file)) {
1742 error = PTR_ERR(nd.intent.open.file);
1743 goto exit_mutex_unlock;
1746 /* Negative dentry, just create the file */
1747 if (!path.dentry->d_inode) {
1749 * This write is needed to ensure that a
1750 * ro->rw transition does not occur between
1751 * the time when the file is created and when
1752 * a permanent write count is taken through
1753 * the 'struct file' in nameidata_to_filp().
1755 error = mnt_want_write(nd.path.mnt);
1757 goto exit_mutex_unlock;
1758 error = __open_namei_create(&nd, &path, flag, mode);
1760 mnt_drop_write(nd.path.mnt);
1763 filp = nameidata_to_filp(&nd, open_flag);
1765 ima_counts_put(&nd.path,
1766 acc_mode & (MAY_READ | MAY_WRITE |
1768 mnt_drop_write(nd.path.mnt);
1775 * It already exists.
1777 mutex_unlock(&dir->d_inode->i_mutex);
1778 audit_inode(pathname, path.dentry);
1784 if (__follow_mount(&path)) {
1786 if (flag & O_NOFOLLOW)
1791 if (!path.dentry->d_inode)
1793 if (path.dentry->d_inode->i_op->follow_link)
1796 path_to_nameidata(&path, &nd);
1798 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1803 * 1. may_open() truncates a file
1804 * 2. a rw->ro mount transition occurs
1805 * 3. nameidata_to_filp() fails due to
1807 * That would be inconsistent, and should
1808 * be avoided. Taking this mnt write here
1809 * ensures that (2) can not occur.
1811 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1813 error = mnt_want_write(nd.path.mnt);
1817 error = may_open(&nd.path, acc_mode, flag);
1820 mnt_drop_write(nd.path.mnt);
1823 filp = nameidata_to_filp(&nd, open_flag);
1825 ima_counts_put(&nd.path,
1826 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1828 * It is now safe to drop the mnt write
1829 * because the filp has had a write taken
1833 mnt_drop_write(nd.path.mnt);
1839 mutex_unlock(&dir->d_inode->i_mutex);
1841 path_put_conditional(&path, &nd);
1843 if (!IS_ERR(nd.intent.open.file))
1844 release_open_intent(&nd);
1849 return ERR_PTR(error);
1853 if (flag & O_NOFOLLOW)
1856 * This is subtle. Instead of calling do_follow_link() we do the
1857 * thing by hands. The reason is that this way we have zero link_count
1858 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1859 * After that we have the parent and last component, i.e.
1860 * we are in the same situation as after the first path_walk().
1861 * Well, almost - if the last component is normal we get its copy
1862 * stored in nd->last.name and we will have to putname() it when we
1863 * are done. Procfs-like symlinks just set LAST_BIND.
1865 nd.flags |= LOOKUP_PARENT;
1866 error = security_inode_follow_link(path.dentry, &nd);
1869 error = __do_follow_link(&path, &nd);
1871 /* Does someone understand code flow here? Or it is only
1872 * me so stupid? Anathema to whoever designed this non-sense
1873 * with "intent.open".
1875 release_open_intent(&nd);
1878 return ERR_PTR(error);
1880 nd.flags &= ~LOOKUP_PARENT;
1881 if (nd.last_type == LAST_BIND)
1884 if (nd.last_type != LAST_NORM)
1886 if (nd.last.name[nd.last.len]) {
1887 __putname(nd.last.name);
1892 __putname(nd.last.name);
1895 dir = nd.path.dentry;
1896 mutex_lock(&dir->d_inode->i_mutex);
1897 path.dentry = lookup_hash(&nd);
1898 path.mnt = nd.path.mnt;
1899 __putname(nd.last.name);
1904 * filp_open - open file and return file pointer
1906 * @filename: path to open
1907 * @flags: open flags as per the open(2) second argument
1908 * @mode: mode for the new file if O_CREAT is set, else ignored
1910 * This is the helper to open a file from kernelspace if you really
1911 * have to. But in generally you should not do this, so please move
1912 * along, nothing to see here..
1914 struct file *filp_open(const char *filename, int flags, int mode)
1916 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1918 EXPORT_SYMBOL(filp_open);
1921 * lookup_create - lookup a dentry, creating it if it doesn't exist
1922 * @nd: nameidata info
1923 * @is_dir: directory flag
1925 * Simple function to lookup and return a dentry and create it
1926 * if it doesn't exist. Is SMP-safe.
1928 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1930 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1932 struct dentry *dentry = ERR_PTR(-EEXIST);
1934 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1936 * Yucky last component or no last component at all?
1937 * (foo/., foo/.., /////)
1939 if (nd->last_type != LAST_NORM)
1941 nd->flags &= ~LOOKUP_PARENT;
1942 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1943 nd->intent.open.flags = O_EXCL;
1946 * Do the final lookup.
1948 dentry = lookup_hash(nd);
1952 if (dentry->d_inode)
1955 * Special case - lookup gave negative, but... we had foo/bar/
1956 * From the vfs_mknod() POV we just have a negative dentry -
1957 * all is fine. Let's be bastards - you had / on the end, you've
1958 * been asking for (non-existent) directory. -ENOENT for you.
1960 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1962 dentry = ERR_PTR(-ENOENT);
1967 dentry = ERR_PTR(-EEXIST);
1971 EXPORT_SYMBOL_GPL(lookup_create);
1973 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1975 int error = may_create(dir, dentry);
1980 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1983 if (!dir->i_op->mknod)
1986 error = devcgroup_inode_mknod(mode, dev);
1990 error = security_inode_mknod(dir, dentry, mode, dev);
1995 error = dir->i_op->mknod(dir, dentry, mode, dev);
1997 fsnotify_create(dir, dentry);
2001 static int may_mknod(mode_t mode)
2003 switch (mode & S_IFMT) {
2009 case 0: /* zero mode translates to S_IFREG */
2018 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2023 struct dentry *dentry;
2024 struct nameidata nd;
2029 error = user_path_parent(dfd, filename, &nd, &tmp);
2033 dentry = lookup_create(&nd, 0);
2034 if (IS_ERR(dentry)) {
2035 error = PTR_ERR(dentry);
2038 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2039 mode &= ~current_umask();
2040 error = may_mknod(mode);
2043 error = mnt_want_write(nd.path.mnt);
2046 error = security_path_mknod(&nd.path, dentry, mode, dev);
2048 goto out_drop_write;
2049 switch (mode & S_IFMT) {
2050 case 0: case S_IFREG:
2051 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2053 case S_IFCHR: case S_IFBLK:
2054 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2055 new_decode_dev(dev));
2057 case S_IFIFO: case S_IFSOCK:
2058 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2062 mnt_drop_write(nd.path.mnt);
2066 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2073 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2075 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2078 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2080 int error = may_create(dir, dentry);
2085 if (!dir->i_op->mkdir)
2088 mode &= (S_IRWXUGO|S_ISVTX);
2089 error = security_inode_mkdir(dir, dentry, mode);
2094 error = dir->i_op->mkdir(dir, dentry, mode);
2096 fsnotify_mkdir(dir, dentry);
2100 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2104 struct dentry *dentry;
2105 struct nameidata nd;
2107 error = user_path_parent(dfd, pathname, &nd, &tmp);
2111 dentry = lookup_create(&nd, 1);
2112 error = PTR_ERR(dentry);
2116 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2117 mode &= ~current_umask();
2118 error = mnt_want_write(nd.path.mnt);
2121 error = security_path_mkdir(&nd.path, dentry, mode);
2123 goto out_drop_write;
2124 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2126 mnt_drop_write(nd.path.mnt);
2130 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2137 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2139 return sys_mkdirat(AT_FDCWD, pathname, mode);
2143 * We try to drop the dentry early: we should have
2144 * a usage count of 2 if we're the only user of this
2145 * dentry, and if that is true (possibly after pruning
2146 * the dcache), then we drop the dentry now.
2148 * A low-level filesystem can, if it choses, legally
2151 * if (!d_unhashed(dentry))
2154 * if it cannot handle the case of removing a directory
2155 * that is still in use by something else..
2157 void dentry_unhash(struct dentry *dentry)
2160 shrink_dcache_parent(dentry);
2161 spin_lock(&dcache_lock);
2162 spin_lock(&dentry->d_lock);
2163 if (atomic_read(&dentry->d_count) == 2)
2165 spin_unlock(&dentry->d_lock);
2166 spin_unlock(&dcache_lock);
2169 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2171 int error = may_delete(dir, dentry, 1);
2176 if (!dir->i_op->rmdir)
2181 mutex_lock(&dentry->d_inode->i_mutex);
2182 dentry_unhash(dentry);
2183 if (d_mountpoint(dentry))
2186 error = security_inode_rmdir(dir, dentry);
2188 error = dir->i_op->rmdir(dir, dentry);
2190 dentry->d_inode->i_flags |= S_DEAD;
2193 mutex_unlock(&dentry->d_inode->i_mutex);
2202 static long do_rmdir(int dfd, const char __user *pathname)
2206 struct dentry *dentry;
2207 struct nameidata nd;
2209 error = user_path_parent(dfd, pathname, &nd, &name);
2213 switch(nd.last_type) {
2225 nd.flags &= ~LOOKUP_PARENT;
2227 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2228 dentry = lookup_hash(&nd);
2229 error = PTR_ERR(dentry);
2232 error = mnt_want_write(nd.path.mnt);
2235 error = security_path_rmdir(&nd.path, dentry);
2238 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2240 mnt_drop_write(nd.path.mnt);
2244 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2251 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2253 return do_rmdir(AT_FDCWD, pathname);
2256 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2258 int error = may_delete(dir, dentry, 0);
2263 if (!dir->i_op->unlink)
2268 mutex_lock(&dentry->d_inode->i_mutex);
2269 if (d_mountpoint(dentry))
2272 error = security_inode_unlink(dir, dentry);
2274 error = dir->i_op->unlink(dir, dentry);
2276 mutex_unlock(&dentry->d_inode->i_mutex);
2278 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2279 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2280 fsnotify_link_count(dentry->d_inode);
2288 * Make sure that the actual truncation of the file will occur outside its
2289 * directory's i_mutex. Truncate can take a long time if there is a lot of
2290 * writeout happening, and we don't want to prevent access to the directory
2291 * while waiting on the I/O.
2293 static long do_unlinkat(int dfd, const char __user *pathname)
2297 struct dentry *dentry;
2298 struct nameidata nd;
2299 struct inode *inode = NULL;
2301 error = user_path_parent(dfd, pathname, &nd, &name);
2306 if (nd.last_type != LAST_NORM)
2309 nd.flags &= ~LOOKUP_PARENT;
2311 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2312 dentry = lookup_hash(&nd);
2313 error = PTR_ERR(dentry);
2314 if (!IS_ERR(dentry)) {
2315 /* Why not before? Because we want correct error value */
2316 if (nd.last.name[nd.last.len])
2318 inode = dentry->d_inode;
2320 atomic_inc(&inode->i_count);
2321 error = mnt_want_write(nd.path.mnt);
2324 error = security_path_unlink(&nd.path, dentry);
2327 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2329 mnt_drop_write(nd.path.mnt);
2333 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2335 iput(inode); /* truncate the inode here */
2342 error = !dentry->d_inode ? -ENOENT :
2343 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2347 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2349 if ((flag & ~AT_REMOVEDIR) != 0)
2352 if (flag & AT_REMOVEDIR)
2353 return do_rmdir(dfd, pathname);
2355 return do_unlinkat(dfd, pathname);
2358 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2360 return do_unlinkat(AT_FDCWD, pathname);
2363 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2365 int error = may_create(dir, dentry);
2370 if (!dir->i_op->symlink)
2373 error = security_inode_symlink(dir, dentry, oldname);
2378 error = dir->i_op->symlink(dir, dentry, oldname);
2380 fsnotify_create(dir, dentry);
2384 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2385 int, newdfd, const char __user *, newname)
2390 struct dentry *dentry;
2391 struct nameidata nd;
2393 from = getname(oldname);
2395 return PTR_ERR(from);
2397 error = user_path_parent(newdfd, newname, &nd, &to);
2401 dentry = lookup_create(&nd, 0);
2402 error = PTR_ERR(dentry);
2406 error = mnt_want_write(nd.path.mnt);
2409 error = security_path_symlink(&nd.path, dentry, from);
2411 goto out_drop_write;
2412 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2414 mnt_drop_write(nd.path.mnt);
2418 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2426 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2428 return sys_symlinkat(oldname, AT_FDCWD, newname);
2431 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2433 struct inode *inode = old_dentry->d_inode;
2439 error = may_create(dir, new_dentry);
2443 if (dir->i_sb != inode->i_sb)
2447 * A link to an append-only or immutable file cannot be created.
2449 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2451 if (!dir->i_op->link)
2453 if (S_ISDIR(inode->i_mode))
2456 error = security_inode_link(old_dentry, dir, new_dentry);
2460 mutex_lock(&inode->i_mutex);
2462 error = dir->i_op->link(old_dentry, dir, new_dentry);
2463 mutex_unlock(&inode->i_mutex);
2465 fsnotify_link(dir, inode, new_dentry);
2470 * Hardlinks are often used in delicate situations. We avoid
2471 * security-related surprises by not following symlinks on the
2474 * We don't follow them on the oldname either to be compatible
2475 * with linux 2.0, and to avoid hard-linking to directories
2476 * and other special files. --ADM
2478 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2479 int, newdfd, const char __user *, newname, int, flags)
2481 struct dentry *new_dentry;
2482 struct nameidata nd;
2483 struct path old_path;
2487 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2490 error = user_path_at(olddfd, oldname,
2491 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2496 error = user_path_parent(newdfd, newname, &nd, &to);
2500 if (old_path.mnt != nd.path.mnt)
2502 new_dentry = lookup_create(&nd, 0);
2503 error = PTR_ERR(new_dentry);
2504 if (IS_ERR(new_dentry))
2506 error = mnt_want_write(nd.path.mnt);
2509 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2511 goto out_drop_write;
2512 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2514 mnt_drop_write(nd.path.mnt);
2518 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2523 path_put(&old_path);
2528 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2530 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2534 * The worst of all namespace operations - renaming directory. "Perverted"
2535 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2537 * a) we can get into loop creation. Check is done in is_subdir().
2538 * b) race potential - two innocent renames can create a loop together.
2539 * That's where 4.4 screws up. Current fix: serialization on
2540 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2542 * c) we have to lock _three_ objects - parents and victim (if it exists).
2543 * And that - after we got ->i_mutex on parents (until then we don't know
2544 * whether the target exists). Solution: try to be smart with locking
2545 * order for inodes. We rely on the fact that tree topology may change
2546 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2547 * move will be locked. Thus we can rank directories by the tree
2548 * (ancestors first) and rank all non-directories after them.
2549 * That works since everybody except rename does "lock parent, lookup,
2550 * lock child" and rename is under ->s_vfs_rename_mutex.
2551 * HOWEVER, it relies on the assumption that any object with ->lookup()
2552 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2553 * we'd better make sure that there's no link(2) for them.
2554 * d) some filesystems don't support opened-but-unlinked directories,
2555 * either because of layout or because they are not ready to deal with
2556 * all cases correctly. The latter will be fixed (taking this sort of
2557 * stuff into VFS), but the former is not going away. Solution: the same
2558 * trick as in rmdir().
2559 * e) conversion from fhandle to dentry may come in the wrong moment - when
2560 * we are removing the target. Solution: we will have to grab ->i_mutex
2561 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2562 * ->i_mutex on parents, which works but leads to some truely excessive
2565 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2566 struct inode *new_dir, struct dentry *new_dentry)
2569 struct inode *target;
2572 * If we are going to change the parent - check write permissions,
2573 * we'll need to flip '..'.
2575 if (new_dir != old_dir) {
2576 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2581 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2585 target = new_dentry->d_inode;
2587 mutex_lock(&target->i_mutex);
2588 dentry_unhash(new_dentry);
2590 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2593 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2596 target->i_flags |= S_DEAD;
2597 mutex_unlock(&target->i_mutex);
2598 if (d_unhashed(new_dentry))
2599 d_rehash(new_dentry);
2603 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2604 d_move(old_dentry,new_dentry);
2608 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2609 struct inode *new_dir, struct dentry *new_dentry)
2611 struct inode *target;
2614 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2619 target = new_dentry->d_inode;
2621 mutex_lock(&target->i_mutex);
2622 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2625 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2627 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2628 d_move(old_dentry, new_dentry);
2631 mutex_unlock(&target->i_mutex);
2636 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2637 struct inode *new_dir, struct dentry *new_dentry)
2640 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2641 const char *old_name;
2643 if (old_dentry->d_inode == new_dentry->d_inode)
2646 error = may_delete(old_dir, old_dentry, is_dir);
2650 if (!new_dentry->d_inode)
2651 error = may_create(new_dir, new_dentry);
2653 error = may_delete(new_dir, new_dentry, is_dir);
2657 if (!old_dir->i_op->rename)
2660 vfs_dq_init(old_dir);
2661 vfs_dq_init(new_dir);
2663 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2666 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2668 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2670 const char *new_name = old_dentry->d_name.name;
2671 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2672 new_dentry->d_inode, old_dentry);
2674 fsnotify_oldname_free(old_name);
2679 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2680 int, newdfd, const char __user *, newname)
2682 struct dentry *old_dir, *new_dir;
2683 struct dentry *old_dentry, *new_dentry;
2684 struct dentry *trap;
2685 struct nameidata oldnd, newnd;
2690 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2694 error = user_path_parent(newdfd, newname, &newnd, &to);
2699 if (oldnd.path.mnt != newnd.path.mnt)
2702 old_dir = oldnd.path.dentry;
2704 if (oldnd.last_type != LAST_NORM)
2707 new_dir = newnd.path.dentry;
2708 if (newnd.last_type != LAST_NORM)
2711 oldnd.flags &= ~LOOKUP_PARENT;
2712 newnd.flags &= ~LOOKUP_PARENT;
2713 newnd.flags |= LOOKUP_RENAME_TARGET;
2715 trap = lock_rename(new_dir, old_dir);
2717 old_dentry = lookup_hash(&oldnd);
2718 error = PTR_ERR(old_dentry);
2719 if (IS_ERR(old_dentry))
2721 /* source must exist */
2723 if (!old_dentry->d_inode)
2725 /* unless the source is a directory trailing slashes give -ENOTDIR */
2726 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2728 if (oldnd.last.name[oldnd.last.len])
2730 if (newnd.last.name[newnd.last.len])
2733 /* source should not be ancestor of target */
2735 if (old_dentry == trap)
2737 new_dentry = lookup_hash(&newnd);
2738 error = PTR_ERR(new_dentry);
2739 if (IS_ERR(new_dentry))
2741 /* target should not be an ancestor of source */
2743 if (new_dentry == trap)
2746 error = mnt_want_write(oldnd.path.mnt);
2749 error = security_path_rename(&oldnd.path, old_dentry,
2750 &newnd.path, new_dentry);
2753 error = vfs_rename(old_dir->d_inode, old_dentry,
2754 new_dir->d_inode, new_dentry);
2756 mnt_drop_write(oldnd.path.mnt);
2762 unlock_rename(new_dir, old_dir);
2764 path_put(&newnd.path);
2767 path_put(&oldnd.path);
2773 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2775 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2778 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2782 len = PTR_ERR(link);
2787 if (len > (unsigned) buflen)
2789 if (copy_to_user(buffer, link, len))
2796 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2797 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2798 * using) it for any given inode is up to filesystem.
2800 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2802 struct nameidata nd;
2807 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2809 return PTR_ERR(cookie);
2811 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2812 if (dentry->d_inode->i_op->put_link)
2813 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2817 int vfs_follow_link(struct nameidata *nd, const char *link)
2819 return __vfs_follow_link(nd, link);
2822 /* get the link contents into pagecache */
2823 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2827 struct address_space *mapping = dentry->d_inode->i_mapping;
2828 page = read_mapping_page(mapping, 0, NULL);
2833 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2837 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2839 struct page *page = NULL;
2840 char *s = page_getlink(dentry, &page);
2841 int res = vfs_readlink(dentry,buffer,buflen,s);
2844 page_cache_release(page);
2849 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2851 struct page *page = NULL;
2852 nd_set_link(nd, page_getlink(dentry, &page));
2856 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2858 struct page *page = cookie;
2862 page_cache_release(page);
2867 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2869 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2871 struct address_space *mapping = inode->i_mapping;
2876 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2878 flags |= AOP_FLAG_NOFS;
2881 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2882 flags, &page, &fsdata);
2886 kaddr = kmap_atomic(page, KM_USER0);
2887 memcpy(kaddr, symname, len-1);
2888 kunmap_atomic(kaddr, KM_USER0);
2890 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2897 mark_inode_dirty(inode);
2903 int page_symlink(struct inode *inode, const char *symname, int len)
2905 return __page_symlink(inode, symname, len,
2906 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2909 const struct inode_operations page_symlink_inode_operations = {
2910 .readlink = generic_readlink,
2911 .follow_link = page_follow_link_light,
2912 .put_link = page_put_link,
2915 EXPORT_SYMBOL(user_path_at);
2916 EXPORT_SYMBOL(follow_down);
2917 EXPORT_SYMBOL(follow_up);
2918 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2919 EXPORT_SYMBOL(getname);
2920 EXPORT_SYMBOL(lock_rename);
2921 EXPORT_SYMBOL(lookup_one_len);
2922 EXPORT_SYMBOL(page_follow_link_light);
2923 EXPORT_SYMBOL(page_put_link);
2924 EXPORT_SYMBOL(page_readlink);
2925 EXPORT_SYMBOL(__page_symlink);
2926 EXPORT_SYMBOL(page_symlink);
2927 EXPORT_SYMBOL(page_symlink_inode_operations);
2928 EXPORT_SYMBOL(path_lookup);
2929 EXPORT_SYMBOL(kern_path);
2930 EXPORT_SYMBOL(vfs_path_lookup);
2931 EXPORT_SYMBOL(inode_permission);
2932 EXPORT_SYMBOL(file_permission);
2933 EXPORT_SYMBOL(unlock_rename);
2934 EXPORT_SYMBOL(vfs_create);
2935 EXPORT_SYMBOL(vfs_follow_link);
2936 EXPORT_SYMBOL(vfs_link);
2937 EXPORT_SYMBOL(vfs_mkdir);
2938 EXPORT_SYMBOL(vfs_mknod);
2939 EXPORT_SYMBOL(generic_permission);
2940 EXPORT_SYMBOL(vfs_readlink);
2941 EXPORT_SYMBOL(vfs_rename);
2942 EXPORT_SYMBOL(vfs_rmdir);
2943 EXPORT_SYMBOL(vfs_symlink);
2944 EXPORT_SYMBOL(vfs_unlink);
2945 EXPORT_SYMBOL(dentry_unhash);
2946 EXPORT_SYMBOL(generic_readlink);