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)
440 if (current_fsuid() == inode->i_uid)
442 else if (in_group_p(inode->i_gid))
448 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
451 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
454 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
459 return security_inode_permission(inode, MAY_EXEC);
463 * This is called when everything else fails, and we actually have
464 * to go to the low-level filesystem to find out what we should do..
466 * We get the directory semaphore, and after getting that we also
467 * make sure that nobody added the entry to the dcache in the meantime..
470 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
472 struct dentry * result;
473 struct inode *dir = parent->d_inode;
475 mutex_lock(&dir->i_mutex);
477 * First re-do the cached lookup just in case it was created
478 * while we waited for the directory semaphore..
480 * FIXME! This could use version numbering or similar to
481 * avoid unnecessary cache lookups.
483 * The "dcache_lock" is purely to protect the RCU list walker
484 * from concurrent renames at this point (we mustn't get false
485 * negatives from the RCU list walk here, unlike the optimistic
488 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
490 result = d_lookup(parent, name);
492 struct dentry *dentry;
494 /* Don't create child dentry for a dead directory. */
495 result = ERR_PTR(-ENOENT);
499 dentry = d_alloc(parent, name);
500 result = ERR_PTR(-ENOMEM);
502 result = dir->i_op->lookup(dir, dentry, nd);
509 mutex_unlock(&dir->i_mutex);
514 * Uhhuh! Nasty case: the cache was re-populated while
515 * we waited on the semaphore. Need to revalidate.
517 mutex_unlock(&dir->i_mutex);
518 if (result->d_op && result->d_op->d_revalidate) {
519 result = do_revalidate(result, nd);
521 result = ERR_PTR(-ENOENT);
527 * Wrapper to retry pathname resolution whenever the underlying
528 * file system returns an ESTALE.
530 * Retry the whole path once, forcing real lookup requests
531 * instead of relying on the dcache.
533 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
535 struct path save = nd->path;
538 /* make sure the stuff we saved doesn't go away */
541 result = __link_path_walk(name, nd);
542 if (result == -ESTALE) {
543 /* nd->path had been dropped */
546 nd->flags |= LOOKUP_REVAL;
547 result = __link_path_walk(name, nd);
555 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
563 struct fs_struct *fs = current->fs;
567 read_lock(&fs->lock);
570 read_unlock(&fs->lock);
573 res = link_path_walk(link, nd);
574 if (nd->depth || res || nd->last_type!=LAST_NORM)
577 * If it is an iterative symlinks resolution in open_namei() we
578 * have to copy the last component. And all that crap because of
579 * bloody create() on broken symlinks. Furrfu...
582 if (unlikely(!name)) {
586 strcpy(name, nd->last.name);
587 nd->last.name = name;
591 return PTR_ERR(link);
594 static void path_put_conditional(struct path *path, struct nameidata *nd)
597 if (path->mnt != nd->path.mnt)
601 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
603 dput(nd->path.dentry);
604 if (nd->path.mnt != path->mnt)
605 mntput(nd->path.mnt);
606 nd->path.mnt = path->mnt;
607 nd->path.dentry = path->dentry;
610 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
614 struct dentry *dentry = path->dentry;
616 touch_atime(path->mnt, dentry);
617 nd_set_link(nd, NULL);
619 if (path->mnt != nd->path.mnt) {
620 path_to_nameidata(path, nd);
624 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
625 error = PTR_ERR(cookie);
626 if (!IS_ERR(cookie)) {
627 char *s = nd_get_link(nd);
630 error = __vfs_follow_link(nd, s);
631 if (dentry->d_inode->i_op->put_link)
632 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
640 * This limits recursive symlink follows to 8, while
641 * limiting consecutive symlinks to 40.
643 * Without that kind of total limit, nasty chains of consecutive
644 * symlinks can cause almost arbitrarily long lookups.
646 static inline int do_follow_link(struct path *path, struct nameidata *nd)
649 if (current->link_count >= MAX_NESTED_LINKS)
651 if (current->total_link_count >= 40)
653 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
655 err = security_inode_follow_link(path->dentry, nd);
658 current->link_count++;
659 current->total_link_count++;
661 err = __do_follow_link(path, nd);
662 current->link_count--;
666 path_put_conditional(path, nd);
671 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
673 struct vfsmount *parent;
674 struct dentry *mountpoint;
675 spin_lock(&vfsmount_lock);
676 parent=(*mnt)->mnt_parent;
677 if (parent == *mnt) {
678 spin_unlock(&vfsmount_lock);
682 mountpoint=dget((*mnt)->mnt_mountpoint);
683 spin_unlock(&vfsmount_lock);
685 *dentry = mountpoint;
691 /* no need for dcache_lock, as serialization is taken care in
694 static int __follow_mount(struct path *path)
697 while (d_mountpoint(path->dentry)) {
698 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
705 path->dentry = dget(mounted->mnt_root);
711 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
713 while (d_mountpoint(*dentry)) {
714 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
720 *dentry = dget(mounted->mnt_root);
724 /* no need for dcache_lock, as serialization is taken care in
727 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
729 struct vfsmount *mounted;
731 mounted = lookup_mnt(*mnt, *dentry);
736 *dentry = dget(mounted->mnt_root);
742 static __always_inline void follow_dotdot(struct nameidata *nd)
744 struct fs_struct *fs = current->fs;
747 struct vfsmount *parent;
748 struct dentry *old = nd->path.dentry;
750 read_lock(&fs->lock);
751 if (nd->path.dentry == fs->root.dentry &&
752 nd->path.mnt == fs->root.mnt) {
753 read_unlock(&fs->lock);
756 read_unlock(&fs->lock);
757 spin_lock(&dcache_lock);
758 if (nd->path.dentry != nd->path.mnt->mnt_root) {
759 nd->path.dentry = dget(nd->path.dentry->d_parent);
760 spin_unlock(&dcache_lock);
764 spin_unlock(&dcache_lock);
765 spin_lock(&vfsmount_lock);
766 parent = nd->path.mnt->mnt_parent;
767 if (parent == nd->path.mnt) {
768 spin_unlock(&vfsmount_lock);
772 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
773 spin_unlock(&vfsmount_lock);
775 mntput(nd->path.mnt);
776 nd->path.mnt = parent;
778 follow_mount(&nd->path.mnt, &nd->path.dentry);
782 * It's more convoluted than I'd like it to be, but... it's still fairly
783 * small and for now I'd prefer to have fast path as straight as possible.
784 * It _is_ time-critical.
786 static int do_lookup(struct nameidata *nd, struct qstr *name,
789 struct vfsmount *mnt = nd->path.mnt;
790 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
794 if (dentry->d_op && dentry->d_op->d_revalidate)
795 goto need_revalidate;
798 path->dentry = dentry;
799 __follow_mount(path);
803 dentry = real_lookup(nd->path.dentry, name, nd);
809 dentry = do_revalidate(dentry, nd);
817 return PTR_ERR(dentry);
822 * This is the basic name resolution function, turning a pathname into
823 * the final dentry. We expect 'base' to be positive and a directory.
825 * Returns 0 and nd will have valid dentry and mnt on success.
826 * Returns error and drops reference to input namei data on failure.
828 static int __link_path_walk(const char *name, struct nameidata *nd)
833 unsigned int lookup_flags = nd->flags;
840 inode = nd->path.dentry->d_inode;
842 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
844 /* At this point we know we have a real path component. */
850 nd->flags |= LOOKUP_CONTINUE;
851 err = exec_permission_lite(inode);
853 err = inode_permission(nd->path.dentry->d_inode,
856 err = ima_path_check(&nd->path, MAY_EXEC,
862 c = *(const unsigned char *)name;
864 hash = init_name_hash();
867 hash = partial_name_hash(c, hash);
868 c = *(const unsigned char *)name;
869 } while (c && (c != '/'));
870 this.len = name - (const char *) this.name;
871 this.hash = end_name_hash(hash);
873 /* remove trailing slashes? */
876 while (*++name == '/');
878 goto last_with_slashes;
881 * "." and ".." are special - ".." especially so because it has
882 * to be able to know about the current root directory and
883 * parent relationships.
885 if (this.name[0] == '.') switch (this.len) {
889 if (this.name[1] != '.')
892 inode = nd->path.dentry->d_inode;
898 * See if the low-level filesystem might want
899 * to use its own hash..
901 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
902 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
907 /* This does the actual lookups.. */
908 err = do_lookup(nd, &this, &next);
913 inode = next.dentry->d_inode;
917 if (inode->i_op->follow_link) {
918 err = do_follow_link(&next, nd);
922 inode = nd->path.dentry->d_inode;
926 path_to_nameidata(&next, nd);
928 if (!inode->i_op->lookup)
931 /* here ends the main loop */
934 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
936 /* Clear LOOKUP_CONTINUE iff it was previously unset */
937 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
938 if (lookup_flags & LOOKUP_PARENT)
940 if (this.name[0] == '.') switch (this.len) {
944 if (this.name[1] != '.')
947 inode = nd->path.dentry->d_inode;
952 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
953 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
958 err = do_lookup(nd, &this, &next);
961 inode = next.dentry->d_inode;
962 if ((lookup_flags & LOOKUP_FOLLOW)
963 && inode && inode->i_op->follow_link) {
964 err = do_follow_link(&next, nd);
967 inode = nd->path.dentry->d_inode;
969 path_to_nameidata(&next, nd);
973 if (lookup_flags & LOOKUP_DIRECTORY) {
975 if (!inode->i_op->lookup)
981 nd->last_type = LAST_NORM;
982 if (this.name[0] != '.')
985 nd->last_type = LAST_DOT;
986 else if (this.len == 2 && this.name[1] == '.')
987 nd->last_type = LAST_DOTDOT;
992 * We bypassed the ordinary revalidation routines.
993 * We may need to check the cached dentry for staleness.
995 if (nd->path.dentry && nd->path.dentry->d_sb &&
996 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
998 /* Note: we do not d_invalidate() */
999 if (!nd->path.dentry->d_op->d_revalidate(
1000 nd->path.dentry, nd))
1006 path_put_conditional(&next, nd);
1009 path_put(&nd->path);
1014 static int path_walk(const char *name, struct nameidata *nd)
1016 current->total_link_count = 0;
1017 return link_path_walk(name, nd);
1020 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1025 struct fs_struct *fs = current->fs;
1027 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1032 read_lock(&fs->lock);
1033 nd->path = fs->root;
1034 path_get(&fs->root);
1035 read_unlock(&fs->lock);
1036 } else if (dfd == AT_FDCWD) {
1037 read_lock(&fs->lock);
1040 read_unlock(&fs->lock);
1042 struct dentry *dentry;
1044 file = fget_light(dfd, &fput_needed);
1049 dentry = file->f_path.dentry;
1052 if (!S_ISDIR(dentry->d_inode->i_mode))
1055 retval = file_permission(file, MAY_EXEC);
1059 nd->path = file->f_path;
1060 path_get(&file->f_path);
1062 fput_light(file, fput_needed);
1067 fput_light(file, fput_needed);
1072 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1073 static int do_path_lookup(int dfd, const char *name,
1074 unsigned int flags, struct nameidata *nd)
1076 int retval = path_init(dfd, name, flags, nd);
1078 retval = path_walk(name, nd);
1079 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1080 nd->path.dentry->d_inode))
1081 audit_inode(name, nd->path.dentry);
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);
1123 retval = path_walk(name, nd);
1124 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1125 nd->path.dentry->d_inode))
1126 audit_inode(name, nd->path.dentry);
1133 * path_lookup_open - lookup a file path with open intent
1134 * @dfd: the directory to use as base, or AT_FDCWD
1135 * @name: pointer to file name
1136 * @lookup_flags: lookup intent flags
1137 * @nd: pointer to nameidata
1138 * @open_flags: open intent flags
1140 static int path_lookup_open(int dfd, const char *name,
1141 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1143 struct file *filp = get_empty_filp();
1148 nd->intent.open.file = filp;
1149 nd->intent.open.flags = open_flags;
1150 nd->intent.open.create_mode = 0;
1151 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1152 if (IS_ERR(nd->intent.open.file)) {
1154 err = PTR_ERR(nd->intent.open.file);
1155 path_put(&nd->path);
1157 } else if (err != 0)
1158 release_open_intent(nd);
1162 static struct dentry *__lookup_hash(struct qstr *name,
1163 struct dentry *base, struct nameidata *nd)
1165 struct dentry *dentry;
1166 struct inode *inode;
1169 inode = base->d_inode;
1172 * See if the low-level filesystem might want
1173 * to use its own hash..
1175 if (base->d_op && base->d_op->d_hash) {
1176 err = base->d_op->d_hash(base, name);
1177 dentry = ERR_PTR(err);
1182 dentry = cached_lookup(base, name, nd);
1186 /* Don't create child dentry for a dead directory. */
1187 dentry = ERR_PTR(-ENOENT);
1188 if (IS_DEADDIR(inode))
1191 new = d_alloc(base, name);
1192 dentry = ERR_PTR(-ENOMEM);
1195 dentry = inode->i_op->lookup(inode, new, nd);
1206 * Restricted form of lookup. Doesn't follow links, single-component only,
1207 * needs parent already locked. Doesn't follow mounts.
1210 static struct dentry *lookup_hash(struct nameidata *nd)
1214 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1216 return ERR_PTR(err);
1217 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1220 static int __lookup_one_len(const char *name, struct qstr *this,
1221 struct dentry *base, int len)
1231 hash = init_name_hash();
1233 c = *(const unsigned char *)name++;
1234 if (c == '/' || c == '\0')
1236 hash = partial_name_hash(c, hash);
1238 this->hash = end_name_hash(hash);
1243 * lookup_one_len - filesystem helper to lookup single pathname component
1244 * @name: pathname component to lookup
1245 * @base: base directory to lookup from
1246 * @len: maximum length @len should be interpreted to
1248 * Note that this routine is purely a helper for filesystem usage and should
1249 * not be called by generic code. Also note that by using this function the
1250 * nameidata argument is passed to the filesystem methods and a filesystem
1251 * using this helper needs to be prepared for that.
1253 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1258 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1260 err = __lookup_one_len(name, &this, base, len);
1262 return ERR_PTR(err);
1264 err = inode_permission(base->d_inode, MAY_EXEC);
1266 return ERR_PTR(err);
1267 return __lookup_hash(&this, base, NULL);
1271 * lookup_one_noperm - bad hack for sysfs
1272 * @name: pathname component to lookup
1273 * @base: base directory to lookup from
1275 * This is a variant of lookup_one_len that doesn't perform any permission
1276 * checks. It's a horrible hack to work around the braindead sysfs
1277 * architecture and should not be used anywhere else.
1279 * DON'T USE THIS FUNCTION EVER, thanks.
1281 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1286 err = __lookup_one_len(name, &this, base, strlen(name));
1288 return ERR_PTR(err);
1289 return __lookup_hash(&this, base, NULL);
1292 int user_path_at(int dfd, const char __user *name, unsigned flags,
1295 struct nameidata nd;
1296 char *tmp = getname(name);
1297 int err = PTR_ERR(tmp);
1300 BUG_ON(flags & LOOKUP_PARENT);
1302 err = do_path_lookup(dfd, tmp, flags, &nd);
1310 static int user_path_parent(int dfd, const char __user *path,
1311 struct nameidata *nd, char **name)
1313 char *s = getname(path);
1319 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1329 * It's inline, so penalty for filesystems that don't use sticky bit is
1332 static inline int check_sticky(struct inode *dir, struct inode *inode)
1334 uid_t fsuid = current_fsuid();
1336 if (!(dir->i_mode & S_ISVTX))
1338 if (inode->i_uid == fsuid)
1340 if (dir->i_uid == fsuid)
1342 return !capable(CAP_FOWNER);
1346 * Check whether we can remove a link victim from directory dir, check
1347 * whether the type of victim is right.
1348 * 1. We can't do it if dir is read-only (done in permission())
1349 * 2. We should have write and exec permissions on dir
1350 * 3. We can't remove anything from append-only dir
1351 * 4. We can't do anything with immutable dir (done in permission())
1352 * 5. If the sticky bit on dir is set we should either
1353 * a. be owner of dir, or
1354 * b. be owner of victim, or
1355 * c. have CAP_FOWNER capability
1356 * 6. If the victim is append-only or immutable we can't do antyhing with
1357 * links pointing to it.
1358 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1359 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1360 * 9. We can't remove a root or mountpoint.
1361 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1362 * nfs_async_unlink().
1364 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1368 if (!victim->d_inode)
1371 BUG_ON(victim->d_parent->d_inode != dir);
1372 audit_inode_child(victim->d_name.name, victim, dir);
1374 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1379 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1380 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1383 if (!S_ISDIR(victim->d_inode->i_mode))
1385 if (IS_ROOT(victim))
1387 } else if (S_ISDIR(victim->d_inode->i_mode))
1389 if (IS_DEADDIR(dir))
1391 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1396 /* Check whether we can create an object with dentry child in directory
1398 * 1. We can't do it if child already exists (open has special treatment for
1399 * this case, but since we are inlined it's OK)
1400 * 2. We can't do it if dir is read-only (done in permission())
1401 * 3. We should have write and exec permissions on dir
1402 * 4. We can't do it if dir is immutable (done in permission())
1404 static inline int may_create(struct inode *dir, struct dentry *child)
1408 if (IS_DEADDIR(dir))
1410 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1414 * O_DIRECTORY translates into forcing a directory lookup.
1416 static inline int lookup_flags(unsigned int f)
1418 unsigned long retval = LOOKUP_FOLLOW;
1421 retval &= ~LOOKUP_FOLLOW;
1423 if (f & O_DIRECTORY)
1424 retval |= LOOKUP_DIRECTORY;
1430 * p1 and p2 should be directories on the same fs.
1432 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1437 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1441 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1443 p = d_ancestor(p2, p1);
1445 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1446 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1450 p = d_ancestor(p1, p2);
1452 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1453 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1457 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1458 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1462 void unlock_rename(struct dentry *p1, struct dentry *p2)
1464 mutex_unlock(&p1->d_inode->i_mutex);
1466 mutex_unlock(&p2->d_inode->i_mutex);
1467 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1471 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1472 struct nameidata *nd)
1474 int error = may_create(dir, dentry);
1479 if (!dir->i_op->create)
1480 return -EACCES; /* shouldn't it be ENOSYS? */
1483 error = security_inode_create(dir, dentry, mode);
1487 error = dir->i_op->create(dir, dentry, mode, nd);
1489 fsnotify_create(dir, dentry);
1493 int may_open(struct path *path, int acc_mode, int flag)
1495 struct dentry *dentry = path->dentry;
1496 struct inode *inode = dentry->d_inode;
1502 switch (inode->i_mode & S_IFMT) {
1506 if (acc_mode & MAY_WRITE)
1511 if (path->mnt->mnt_flags & MNT_NODEV)
1520 error = inode_permission(inode, acc_mode);
1524 error = ima_path_check(path,
1525 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC),
1530 * An append-only file must be opened in append mode for writing.
1532 if (IS_APPEND(inode)) {
1533 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1539 /* O_NOATIME can only be set by the owner or superuser */
1540 if (flag & O_NOATIME)
1541 if (!is_owner_or_cap(inode))
1545 * Ensure there are no outstanding leases on the file.
1547 error = break_lease(inode, flag);
1551 if (flag & O_TRUNC) {
1552 error = get_write_access(inode);
1557 * Refuse to truncate files with mandatory locks held on them.
1559 error = locks_verify_locked(inode);
1561 error = security_path_truncate(path, 0,
1562 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1566 error = do_truncate(dentry, 0,
1567 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1570 put_write_access(inode);
1574 if (flag & FMODE_WRITE)
1581 * Be careful about ever adding any more callers of this
1582 * function. Its flags must be in the namei format, not
1583 * what get passed to sys_open().
1585 static int __open_namei_create(struct nameidata *nd, struct path *path,
1589 struct dentry *dir = nd->path.dentry;
1591 if (!IS_POSIXACL(dir->d_inode))
1592 mode &= ~current_umask();
1593 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1596 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1598 mutex_unlock(&dir->d_inode->i_mutex);
1599 dput(nd->path.dentry);
1600 nd->path.dentry = path->dentry;
1603 /* Don't check for write permission, don't truncate */
1604 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1608 * Note that while the flag value (low two bits) for sys_open means:
1613 * it is changed into
1614 * 00 - no permissions needed
1615 * 01 - read-permission
1616 * 10 - write-permission
1618 * for the internal routines (ie open_namei()/follow_link() etc)
1619 * This is more logical, and also allows the 00 "no perm needed"
1620 * to be used for symlinks (where the permissions are checked
1624 static inline int open_to_namei_flags(int flag)
1626 if ((flag+1) & O_ACCMODE)
1631 static int open_will_write_to_fs(int flag, struct inode *inode)
1634 * We'll never write to the fs underlying
1637 if (special_file(inode->i_mode))
1639 return (flag & O_TRUNC);
1643 * Note that the low bits of the passed in "open_flag"
1644 * are not the same as in the local variable "flag". See
1645 * open_to_namei_flags() for more details.
1647 struct file *do_filp_open(int dfd, const char *pathname,
1648 int open_flag, int mode, int acc_mode)
1651 struct nameidata nd;
1657 int flag = open_to_namei_flags(open_flag);
1660 acc_mode = MAY_OPEN | ACC_MODE(flag);
1662 /* O_TRUNC implies we need access checks for write permissions */
1664 acc_mode |= MAY_WRITE;
1666 /* Allow the LSM permission hook to distinguish append
1667 access from general write access. */
1668 if (flag & O_APPEND)
1669 acc_mode |= MAY_APPEND;
1672 * The simplest case - just a plain lookup.
1674 if (!(flag & O_CREAT)) {
1675 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1678 return ERR_PTR(error);
1683 * Create - we need to know the parent.
1685 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1687 return ERR_PTR(error);
1688 error = path_walk(pathname, &nd);
1690 return ERR_PTR(error);
1691 if (unlikely(!audit_dummy_context()))
1692 audit_inode(pathname, nd.path.dentry);
1695 * We have the parent and last component. First of all, check
1696 * that we are not asked to creat(2) an obvious directory - that
1700 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1704 filp = get_empty_filp();
1707 nd.intent.open.file = filp;
1708 nd.intent.open.flags = flag;
1709 nd.intent.open.create_mode = mode;
1710 dir = nd.path.dentry;
1711 nd.flags &= ~LOOKUP_PARENT;
1712 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1714 nd.flags |= LOOKUP_EXCL;
1715 mutex_lock(&dir->d_inode->i_mutex);
1716 path.dentry = lookup_hash(&nd);
1717 path.mnt = nd.path.mnt;
1720 error = PTR_ERR(path.dentry);
1721 if (IS_ERR(path.dentry)) {
1722 mutex_unlock(&dir->d_inode->i_mutex);
1726 if (IS_ERR(nd.intent.open.file)) {
1727 error = PTR_ERR(nd.intent.open.file);
1728 goto exit_mutex_unlock;
1731 /* Negative dentry, just create the file */
1732 if (!path.dentry->d_inode) {
1734 * This write is needed to ensure that a
1735 * ro->rw transition does not occur between
1736 * the time when the file is created and when
1737 * a permanent write count is taken through
1738 * the 'struct file' in nameidata_to_filp().
1740 error = mnt_want_write(nd.path.mnt);
1742 goto exit_mutex_unlock;
1743 error = __open_namei_create(&nd, &path, flag, mode);
1745 mnt_drop_write(nd.path.mnt);
1748 filp = nameidata_to_filp(&nd, open_flag);
1749 mnt_drop_write(nd.path.mnt);
1754 * It already exists.
1756 mutex_unlock(&dir->d_inode->i_mutex);
1757 audit_inode(pathname, path.dentry);
1763 if (__follow_mount(&path)) {
1765 if (flag & O_NOFOLLOW)
1770 if (!path.dentry->d_inode)
1772 if (path.dentry->d_inode->i_op->follow_link)
1775 path_to_nameidata(&path, &nd);
1777 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1782 * 1. may_open() truncates a file
1783 * 2. a rw->ro mount transition occurs
1784 * 3. nameidata_to_filp() fails due to
1786 * That would be inconsistent, and should
1787 * be avoided. Taking this mnt write here
1788 * ensures that (2) can not occur.
1790 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1792 error = mnt_want_write(nd.path.mnt);
1796 error = may_open(&nd.path, acc_mode, flag);
1799 mnt_drop_write(nd.path.mnt);
1802 filp = nameidata_to_filp(&nd, open_flag);
1804 * It is now safe to drop the mnt write
1805 * because the filp has had a write taken
1809 mnt_drop_write(nd.path.mnt);
1813 mutex_unlock(&dir->d_inode->i_mutex);
1815 path_put_conditional(&path, &nd);
1817 if (!IS_ERR(nd.intent.open.file))
1818 release_open_intent(&nd);
1821 return ERR_PTR(error);
1825 if (flag & O_NOFOLLOW)
1828 * This is subtle. Instead of calling do_follow_link() we do the
1829 * thing by hands. The reason is that this way we have zero link_count
1830 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1831 * After that we have the parent and last component, i.e.
1832 * we are in the same situation as after the first path_walk().
1833 * Well, almost - if the last component is normal we get its copy
1834 * stored in nd->last.name and we will have to putname() it when we
1835 * are done. Procfs-like symlinks just set LAST_BIND.
1837 nd.flags |= LOOKUP_PARENT;
1838 error = security_inode_follow_link(path.dentry, &nd);
1841 error = __do_follow_link(&path, &nd);
1843 /* Does someone understand code flow here? Or it is only
1844 * me so stupid? Anathema to whoever designed this non-sense
1845 * with "intent.open".
1847 release_open_intent(&nd);
1848 return ERR_PTR(error);
1850 nd.flags &= ~LOOKUP_PARENT;
1851 if (nd.last_type == LAST_BIND)
1854 if (nd.last_type != LAST_NORM)
1856 if (nd.last.name[nd.last.len]) {
1857 __putname(nd.last.name);
1862 __putname(nd.last.name);
1865 dir = nd.path.dentry;
1866 mutex_lock(&dir->d_inode->i_mutex);
1867 path.dentry = lookup_hash(&nd);
1868 path.mnt = nd.path.mnt;
1869 __putname(nd.last.name);
1874 * filp_open - open file and return file pointer
1876 * @filename: path to open
1877 * @flags: open flags as per the open(2) second argument
1878 * @mode: mode for the new file if O_CREAT is set, else ignored
1880 * This is the helper to open a file from kernelspace if you really
1881 * have to. But in generally you should not do this, so please move
1882 * along, nothing to see here..
1884 struct file *filp_open(const char *filename, int flags, int mode)
1886 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1888 EXPORT_SYMBOL(filp_open);
1891 * lookup_create - lookup a dentry, creating it if it doesn't exist
1892 * @nd: nameidata info
1893 * @is_dir: directory flag
1895 * Simple function to lookup and return a dentry and create it
1896 * if it doesn't exist. Is SMP-safe.
1898 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1900 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1902 struct dentry *dentry = ERR_PTR(-EEXIST);
1904 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1906 * Yucky last component or no last component at all?
1907 * (foo/., foo/.., /////)
1909 if (nd->last_type != LAST_NORM)
1911 nd->flags &= ~LOOKUP_PARENT;
1912 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1913 nd->intent.open.flags = O_EXCL;
1916 * Do the final lookup.
1918 dentry = lookup_hash(nd);
1922 if (dentry->d_inode)
1925 * Special case - lookup gave negative, but... we had foo/bar/
1926 * From the vfs_mknod() POV we just have a negative dentry -
1927 * all is fine. Let's be bastards - you had / on the end, you've
1928 * been asking for (non-existent) directory. -ENOENT for you.
1930 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1932 dentry = ERR_PTR(-ENOENT);
1937 dentry = ERR_PTR(-EEXIST);
1941 EXPORT_SYMBOL_GPL(lookup_create);
1943 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1945 int error = may_create(dir, dentry);
1950 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1953 if (!dir->i_op->mknod)
1956 error = devcgroup_inode_mknod(mode, dev);
1960 error = security_inode_mknod(dir, dentry, mode, dev);
1965 error = dir->i_op->mknod(dir, dentry, mode, dev);
1967 fsnotify_create(dir, dentry);
1971 static int may_mknod(mode_t mode)
1973 switch (mode & S_IFMT) {
1979 case 0: /* zero mode translates to S_IFREG */
1988 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1993 struct dentry *dentry;
1994 struct nameidata nd;
1999 error = user_path_parent(dfd, filename, &nd, &tmp);
2003 dentry = lookup_create(&nd, 0);
2004 if (IS_ERR(dentry)) {
2005 error = PTR_ERR(dentry);
2008 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2009 mode &= ~current_umask();
2010 error = may_mknod(mode);
2013 error = mnt_want_write(nd.path.mnt);
2016 error = security_path_mknod(&nd.path, dentry, mode, dev);
2018 goto out_drop_write;
2019 switch (mode & S_IFMT) {
2020 case 0: case S_IFREG:
2021 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2023 case S_IFCHR: case S_IFBLK:
2024 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2025 new_decode_dev(dev));
2027 case S_IFIFO: case S_IFSOCK:
2028 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2032 mnt_drop_write(nd.path.mnt);
2036 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2043 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2045 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2048 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2050 int error = may_create(dir, dentry);
2055 if (!dir->i_op->mkdir)
2058 mode &= (S_IRWXUGO|S_ISVTX);
2059 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(&dcache_lock);
2132 spin_lock(&dentry->d_lock);
2133 if (atomic_read(&dentry->d_count) == 2)
2135 spin_unlock(&dentry->d_lock);
2136 spin_unlock(&dcache_lock);
2139 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2141 int error = may_delete(dir, dentry, 1);
2146 if (!dir->i_op->rmdir)
2151 mutex_lock(&dentry->d_inode->i_mutex);
2152 dentry_unhash(dentry);
2153 if (d_mountpoint(dentry))
2156 error = security_inode_rmdir(dir, dentry);
2158 error = dir->i_op->rmdir(dir, dentry);
2160 dentry->d_inode->i_flags |= S_DEAD;
2163 mutex_unlock(&dentry->d_inode->i_mutex);
2172 static long do_rmdir(int dfd, const char __user *pathname)
2176 struct dentry *dentry;
2177 struct nameidata nd;
2179 error = user_path_parent(dfd, pathname, &nd, &name);
2183 switch(nd.last_type) {
2195 nd.flags &= ~LOOKUP_PARENT;
2197 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2198 dentry = lookup_hash(&nd);
2199 error = PTR_ERR(dentry);
2202 error = mnt_want_write(nd.path.mnt);
2205 error = security_path_rmdir(&nd.path, dentry);
2208 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2210 mnt_drop_write(nd.path.mnt);
2214 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2221 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2223 return do_rmdir(AT_FDCWD, pathname);
2226 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2228 int error = may_delete(dir, dentry, 0);
2233 if (!dir->i_op->unlink)
2238 mutex_lock(&dentry->d_inode->i_mutex);
2239 if (d_mountpoint(dentry))
2242 error = security_inode_unlink(dir, dentry);
2244 error = dir->i_op->unlink(dir, dentry);
2246 mutex_unlock(&dentry->d_inode->i_mutex);
2248 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2249 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2250 fsnotify_link_count(dentry->d_inode);
2258 * Make sure that the actual truncation of the file will occur outside its
2259 * directory's i_mutex. Truncate can take a long time if there is a lot of
2260 * writeout happening, and we don't want to prevent access to the directory
2261 * while waiting on the I/O.
2263 static long do_unlinkat(int dfd, const char __user *pathname)
2267 struct dentry *dentry;
2268 struct nameidata nd;
2269 struct inode *inode = NULL;
2271 error = user_path_parent(dfd, pathname, &nd, &name);
2276 if (nd.last_type != LAST_NORM)
2279 nd.flags &= ~LOOKUP_PARENT;
2281 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2282 dentry = lookup_hash(&nd);
2283 error = PTR_ERR(dentry);
2284 if (!IS_ERR(dentry)) {
2285 /* Why not before? Because we want correct error value */
2286 if (nd.last.name[nd.last.len])
2288 inode = dentry->d_inode;
2290 atomic_inc(&inode->i_count);
2291 error = mnt_want_write(nd.path.mnt);
2294 error = security_path_unlink(&nd.path, dentry);
2297 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2299 mnt_drop_write(nd.path.mnt);
2303 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2305 iput(inode); /* truncate the inode here */
2312 error = !dentry->d_inode ? -ENOENT :
2313 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2317 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2319 if ((flag & ~AT_REMOVEDIR) != 0)
2322 if (flag & AT_REMOVEDIR)
2323 return do_rmdir(dfd, pathname);
2325 return do_unlinkat(dfd, pathname);
2328 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2330 return do_unlinkat(AT_FDCWD, pathname);
2333 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2335 int error = may_create(dir, dentry);
2340 if (!dir->i_op->symlink)
2343 error = security_inode_symlink(dir, dentry, oldname);
2348 error = dir->i_op->symlink(dir, dentry, oldname);
2350 fsnotify_create(dir, dentry);
2354 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2355 int, newdfd, const char __user *, newname)
2360 struct dentry *dentry;
2361 struct nameidata nd;
2363 from = getname(oldname);
2365 return PTR_ERR(from);
2367 error = user_path_parent(newdfd, newname, &nd, &to);
2371 dentry = lookup_create(&nd, 0);
2372 error = PTR_ERR(dentry);
2376 error = mnt_want_write(nd.path.mnt);
2379 error = security_path_symlink(&nd.path, dentry, from);
2381 goto out_drop_write;
2382 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2384 mnt_drop_write(nd.path.mnt);
2388 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2396 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2398 return sys_symlinkat(oldname, AT_FDCWD, newname);
2401 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2403 struct inode *inode = old_dentry->d_inode;
2409 error = may_create(dir, new_dentry);
2413 if (dir->i_sb != inode->i_sb)
2417 * A link to an append-only or immutable file cannot be created.
2419 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2421 if (!dir->i_op->link)
2423 if (S_ISDIR(inode->i_mode))
2426 error = security_inode_link(old_dentry, dir, new_dentry);
2430 mutex_lock(&inode->i_mutex);
2432 error = dir->i_op->link(old_dentry, dir, new_dentry);
2433 mutex_unlock(&inode->i_mutex);
2435 fsnotify_link(dir, inode, new_dentry);
2440 * Hardlinks are often used in delicate situations. We avoid
2441 * security-related surprises by not following symlinks on the
2444 * We don't follow them on the oldname either to be compatible
2445 * with linux 2.0, and to avoid hard-linking to directories
2446 * and other special files. --ADM
2448 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2449 int, newdfd, const char __user *, newname, int, flags)
2451 struct dentry *new_dentry;
2452 struct nameidata nd;
2453 struct path old_path;
2457 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2460 error = user_path_at(olddfd, oldname,
2461 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2466 error = user_path_parent(newdfd, newname, &nd, &to);
2470 if (old_path.mnt != nd.path.mnt)
2472 new_dentry = lookup_create(&nd, 0);
2473 error = PTR_ERR(new_dentry);
2474 if (IS_ERR(new_dentry))
2476 error = mnt_want_write(nd.path.mnt);
2479 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2481 goto out_drop_write;
2482 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2484 mnt_drop_write(nd.path.mnt);
2488 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2493 path_put(&old_path);
2498 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2500 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2504 * The worst of all namespace operations - renaming directory. "Perverted"
2505 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2507 * a) we can get into loop creation. Check is done in is_subdir().
2508 * b) race potential - two innocent renames can create a loop together.
2509 * That's where 4.4 screws up. Current fix: serialization on
2510 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2512 * c) we have to lock _three_ objects - parents and victim (if it exists).
2513 * And that - after we got ->i_mutex on parents (until then we don't know
2514 * whether the target exists). Solution: try to be smart with locking
2515 * order for inodes. We rely on the fact that tree topology may change
2516 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2517 * move will be locked. Thus we can rank directories by the tree
2518 * (ancestors first) and rank all non-directories after them.
2519 * That works since everybody except rename does "lock parent, lookup,
2520 * lock child" and rename is under ->s_vfs_rename_mutex.
2521 * HOWEVER, it relies on the assumption that any object with ->lookup()
2522 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2523 * we'd better make sure that there's no link(2) for them.
2524 * d) some filesystems don't support opened-but-unlinked directories,
2525 * either because of layout or because they are not ready to deal with
2526 * all cases correctly. The latter will be fixed (taking this sort of
2527 * stuff into VFS), but the former is not going away. Solution: the same
2528 * trick as in rmdir().
2529 * e) conversion from fhandle to dentry may come in the wrong moment - when
2530 * we are removing the target. Solution: we will have to grab ->i_mutex
2531 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2532 * ->i_mutex on parents, which works but leads to some truely excessive
2535 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2536 struct inode *new_dir, struct dentry *new_dentry)
2539 struct inode *target;
2542 * If we are going to change the parent - check write permissions,
2543 * we'll need to flip '..'.
2545 if (new_dir != old_dir) {
2546 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2551 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2555 target = new_dentry->d_inode;
2557 mutex_lock(&target->i_mutex);
2558 dentry_unhash(new_dentry);
2560 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2563 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2566 target->i_flags |= S_DEAD;
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);
2597 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2598 d_move(old_dentry, new_dentry);
2601 mutex_unlock(&target->i_mutex);
2606 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2607 struct inode *new_dir, struct dentry *new_dentry)
2610 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2611 const char *old_name;
2613 if (old_dentry->d_inode == new_dentry->d_inode)
2616 error = may_delete(old_dir, old_dentry, is_dir);
2620 if (!new_dentry->d_inode)
2621 error = may_create(new_dir, new_dentry);
2623 error = may_delete(new_dir, new_dentry, is_dir);
2627 if (!old_dir->i_op->rename)
2630 vfs_dq_init(old_dir);
2631 vfs_dq_init(new_dir);
2633 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2636 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2638 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2640 const char *new_name = old_dentry->d_name.name;
2641 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2642 new_dentry->d_inode, old_dentry);
2644 fsnotify_oldname_free(old_name);
2649 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2650 int, newdfd, const char __user *, newname)
2652 struct dentry *old_dir, *new_dir;
2653 struct dentry *old_dentry, *new_dentry;
2654 struct dentry *trap;
2655 struct nameidata oldnd, newnd;
2660 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2664 error = user_path_parent(newdfd, newname, &newnd, &to);
2669 if (oldnd.path.mnt != newnd.path.mnt)
2672 old_dir = oldnd.path.dentry;
2674 if (oldnd.last_type != LAST_NORM)
2677 new_dir = newnd.path.dentry;
2678 if (newnd.last_type != LAST_NORM)
2681 oldnd.flags &= ~LOOKUP_PARENT;
2682 newnd.flags &= ~LOOKUP_PARENT;
2683 newnd.flags |= LOOKUP_RENAME_TARGET;
2685 trap = lock_rename(new_dir, old_dir);
2687 old_dentry = lookup_hash(&oldnd);
2688 error = PTR_ERR(old_dentry);
2689 if (IS_ERR(old_dentry))
2691 /* source must exist */
2693 if (!old_dentry->d_inode)
2695 /* unless the source is a directory trailing slashes give -ENOTDIR */
2696 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2698 if (oldnd.last.name[oldnd.last.len])
2700 if (newnd.last.name[newnd.last.len])
2703 /* source should not be ancestor of target */
2705 if (old_dentry == trap)
2707 new_dentry = lookup_hash(&newnd);
2708 error = PTR_ERR(new_dentry);
2709 if (IS_ERR(new_dentry))
2711 /* target should not be an ancestor of source */
2713 if (new_dentry == trap)
2716 error = mnt_want_write(oldnd.path.mnt);
2719 error = security_path_rename(&oldnd.path, old_dentry,
2720 &newnd.path, new_dentry);
2723 error = vfs_rename(old_dir->d_inode, old_dentry,
2724 new_dir->d_inode, new_dentry);
2726 mnt_drop_write(oldnd.path.mnt);
2732 unlock_rename(new_dir, old_dir);
2734 path_put(&newnd.path);
2737 path_put(&oldnd.path);
2743 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2745 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2748 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2752 len = PTR_ERR(link);
2757 if (len > (unsigned) buflen)
2759 if (copy_to_user(buffer, link, len))
2766 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2767 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2768 * using) it for any given inode is up to filesystem.
2770 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2772 struct nameidata nd;
2777 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2779 return PTR_ERR(cookie);
2781 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2782 if (dentry->d_inode->i_op->put_link)
2783 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2787 int vfs_follow_link(struct nameidata *nd, const char *link)
2789 return __vfs_follow_link(nd, link);
2792 /* get the link contents into pagecache */
2793 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2797 struct address_space *mapping = dentry->d_inode->i_mapping;
2798 page = read_mapping_page(mapping, 0, NULL);
2803 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2807 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2809 struct page *page = NULL;
2810 char *s = page_getlink(dentry, &page);
2811 int res = vfs_readlink(dentry,buffer,buflen,s);
2814 page_cache_release(page);
2819 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2821 struct page *page = NULL;
2822 nd_set_link(nd, page_getlink(dentry, &page));
2826 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2828 struct page *page = cookie;
2832 page_cache_release(page);
2837 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2839 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2841 struct address_space *mapping = inode->i_mapping;
2846 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2848 flags |= AOP_FLAG_NOFS;
2851 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2852 flags, &page, &fsdata);
2856 kaddr = kmap_atomic(page, KM_USER0);
2857 memcpy(kaddr, symname, len-1);
2858 kunmap_atomic(kaddr, KM_USER0);
2860 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2867 mark_inode_dirty(inode);
2873 int page_symlink(struct inode *inode, const char *symname, int len)
2875 return __page_symlink(inode, symname, len,
2876 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2879 const struct inode_operations page_symlink_inode_operations = {
2880 .readlink = generic_readlink,
2881 .follow_link = page_follow_link_light,
2882 .put_link = page_put_link,
2885 EXPORT_SYMBOL(user_path_at);
2886 EXPORT_SYMBOL(follow_down);
2887 EXPORT_SYMBOL(follow_up);
2888 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2889 EXPORT_SYMBOL(getname);
2890 EXPORT_SYMBOL(lock_rename);
2891 EXPORT_SYMBOL(lookup_one_len);
2892 EXPORT_SYMBOL(page_follow_link_light);
2893 EXPORT_SYMBOL(page_put_link);
2894 EXPORT_SYMBOL(page_readlink);
2895 EXPORT_SYMBOL(__page_symlink);
2896 EXPORT_SYMBOL(page_symlink);
2897 EXPORT_SYMBOL(page_symlink_inode_operations);
2898 EXPORT_SYMBOL(path_lookup);
2899 EXPORT_SYMBOL(kern_path);
2900 EXPORT_SYMBOL(vfs_path_lookup);
2901 EXPORT_SYMBOL(inode_permission);
2902 EXPORT_SYMBOL(file_permission);
2903 EXPORT_SYMBOL(unlock_rename);
2904 EXPORT_SYMBOL(vfs_create);
2905 EXPORT_SYMBOL(vfs_follow_link);
2906 EXPORT_SYMBOL(vfs_link);
2907 EXPORT_SYMBOL(vfs_mkdir);
2908 EXPORT_SYMBOL(vfs_mknod);
2909 EXPORT_SYMBOL(generic_permission);
2910 EXPORT_SYMBOL(vfs_readlink);
2911 EXPORT_SYMBOL(vfs_rename);
2912 EXPORT_SYMBOL(vfs_rmdir);
2913 EXPORT_SYMBOL(vfs_symlink);
2914 EXPORT_SYMBOL(vfs_unlink);
2915 EXPORT_SYMBOL(dentry_unhash);
2916 EXPORT_SYMBOL(generic_readlink);