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 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
171 * This does basic POSIX ACL permission checking
173 static int acl_permission_check(struct inode *inode, int mask,
174 int (*check_acl)(struct inode *inode, int mask))
176 umode_t mode = inode->i_mode;
178 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
180 if (current_fsuid() == inode->i_uid)
183 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
184 int error = check_acl(inode, mask);
185 if (error != -EAGAIN)
189 if (in_group_p(inode->i_gid))
194 * If the DACs are ok we don't need any capability check.
196 if ((mask & ~mode) == 0)
202 * generic_permission - check for access rights on a Posix-like filesystem
203 * @inode: inode to check access rights for
204 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
205 * @check_acl: optional callback to check for Posix ACLs
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things..
212 int generic_permission(struct inode *inode, int mask,
213 int (*check_acl)(struct inode *inode, int mask))
218 * Do the basic POSIX ACL permission checks.
220 ret = acl_permission_check(inode, mask, check_acl);
225 * Read/write DACs are always overridable.
226 * Executable DACs are overridable if at least one exec bit is set.
228 if (!(mask & MAY_EXEC) || execute_ok(inode))
229 if (capable(CAP_DAC_OVERRIDE))
233 * Searching includes executable on directories, else just read.
235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
281 retval = devcgroup_inode_permission(inode, mask);
285 return security_inode_permission(inode,
286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
294 * Used to check for read/write/execute permissions on an already opened
298 * Do not use this function in new code. All access checks should
299 * be done using inode_permission().
301 int file_permission(struct file *file, int mask)
303 return inode_permission(file->f_path.dentry->d_inode, mask);
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
324 int get_write_access(struct inode * inode)
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
337 int deny_write_access(struct file * file)
339 struct inode *inode = file->f_path.dentry->d_inode;
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
353 * path_get - get a reference to a path
354 * @path: path to get the reference to
356 * Given a path increment the reference count to the dentry and the vfsmount.
358 void path_get(struct path *path)
363 EXPORT_SYMBOL(path_get);
366 * path_put - put a reference to a path
367 * @path: path to put the reference to
369 * Given a path decrement the reference count to the dentry and the vfsmount.
371 void path_put(struct path *path)
376 EXPORT_SYMBOL(path_put);
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
382 void release_open_intent(struct nameidata *nd)
384 if (nd->intent.open.file->f_path.dentry == NULL)
385 put_filp(nd->intent.open.file);
387 fput(nd->intent.open.file);
390 static inline struct dentry *
391 do_revalidate(struct dentry *dentry, struct nameidata *nd)
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
402 if (!d_invalidate(dentry)) {
408 dentry = ERR_PTR(status);
415 * Internal lookup() using the new generic dcache.
418 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
420 struct dentry * dentry = __d_lookup(parent, name);
422 /* lockess __d_lookup may fail due to concurrent d_move()
423 * in some unrelated directory, so try with d_lookup
426 dentry = d_lookup(parent, name);
428 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
429 dentry = do_revalidate(dentry, nd);
435 * Short-cut version of permission(), for calling by
436 * path_walk(), when dcache lock is held. Combines parts
437 * of permission() and generic_permission(), and tests ONLY for
438 * MAY_EXEC permission.
440 * If appropriate, check DAC only. If not appropriate, or
441 * short-cut DAC fails, then call permission() to do more
442 * complete permission check.
444 static int exec_permission_lite(struct inode *inode)
448 if (inode->i_op->permission) {
449 ret = inode->i_op->permission(inode, MAY_EXEC);
454 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
458 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
463 return security_inode_permission(inode, MAY_EXEC);
467 * This is called when everything else fails, and we actually have
468 * to go to the low-level filesystem to find out what we should do..
470 * We get the directory semaphore, and after getting that we also
471 * make sure that nobody added the entry to the dcache in the meantime..
474 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
476 struct dentry * result;
477 struct inode *dir = parent->d_inode;
479 mutex_lock(&dir->i_mutex);
481 * First re-do the cached lookup just in case it was created
482 * while we waited for the directory semaphore..
484 * FIXME! This could use version numbering or similar to
485 * avoid unnecessary cache lookups.
487 * The "dcache_lock" is purely to protect the RCU list walker
488 * from concurrent renames at this point (we mustn't get false
489 * negatives from the RCU list walk here, unlike the optimistic
492 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
494 result = d_lookup(parent, name);
496 struct dentry *dentry;
498 /* Don't create child dentry for a dead directory. */
499 result = ERR_PTR(-ENOENT);
503 dentry = d_alloc(parent, name);
504 result = ERR_PTR(-ENOMEM);
506 result = dir->i_op->lookup(dir, dentry, nd);
513 mutex_unlock(&dir->i_mutex);
518 * Uhhuh! Nasty case: the cache was re-populated while
519 * we waited on the semaphore. Need to revalidate.
521 mutex_unlock(&dir->i_mutex);
522 if (result->d_op && result->d_op->d_revalidate) {
523 result = do_revalidate(result, nd);
525 result = ERR_PTR(-ENOENT);
530 static __always_inline void set_root(struct nameidata *nd)
533 struct fs_struct *fs = current->fs;
534 read_lock(&fs->lock);
537 read_unlock(&fs->lock);
541 static int link_path_walk(const char *, struct nameidata *);
543 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
557 res = link_path_walk(link, nd);
558 if (nd->depth || res || nd->last_type!=LAST_NORM)
561 * If it is an iterative symlinks resolution in open_namei() we
562 * have to copy the last component. And all that crap because of
563 * bloody create() on broken symlinks. Furrfu...
566 if (unlikely(!name)) {
570 strcpy(name, nd->last.name);
571 nd->last.name = name;
575 return PTR_ERR(link);
578 static void path_put_conditional(struct path *path, struct nameidata *nd)
581 if (path->mnt != nd->path.mnt)
585 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
587 dput(nd->path.dentry);
588 if (nd->path.mnt != path->mnt)
589 mntput(nd->path.mnt);
590 nd->path.mnt = path->mnt;
591 nd->path.dentry = path->dentry;
594 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
598 struct dentry *dentry = path->dentry;
600 touch_atime(path->mnt, dentry);
601 nd_set_link(nd, NULL);
603 if (path->mnt != nd->path.mnt) {
604 path_to_nameidata(path, nd);
608 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
609 error = PTR_ERR(cookie);
610 if (!IS_ERR(cookie)) {
611 char *s = nd_get_link(nd);
614 error = __vfs_follow_link(nd, s);
615 if (dentry->d_inode->i_op->put_link)
616 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
622 * This limits recursive symlink follows to 8, while
623 * limiting consecutive symlinks to 40.
625 * Without that kind of total limit, nasty chains of consecutive
626 * symlinks can cause almost arbitrarily long lookups.
628 static inline int do_follow_link(struct path *path, struct nameidata *nd)
631 if (current->link_count >= MAX_NESTED_LINKS)
633 if (current->total_link_count >= 40)
635 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
637 err = security_inode_follow_link(path->dentry, nd);
640 current->link_count++;
641 current->total_link_count++;
643 err = __do_follow_link(path, nd);
645 current->link_count--;
649 path_put_conditional(path, nd);
654 int follow_up(struct path *path)
656 struct vfsmount *parent;
657 struct dentry *mountpoint;
658 spin_lock(&vfsmount_lock);
659 parent = path->mnt->mnt_parent;
660 if (parent == path->mnt) {
661 spin_unlock(&vfsmount_lock);
665 mountpoint = dget(path->mnt->mnt_mountpoint);
666 spin_unlock(&vfsmount_lock);
668 path->dentry = mountpoint;
674 /* no need for dcache_lock, as serialization is taken care in
677 static int __follow_mount(struct path *path)
680 while (d_mountpoint(path->dentry)) {
681 struct vfsmount *mounted = lookup_mnt(path);
688 path->dentry = dget(mounted->mnt_root);
694 static void follow_mount(struct path *path)
696 while (d_mountpoint(path->dentry)) {
697 struct vfsmount *mounted = lookup_mnt(path);
703 path->dentry = dget(mounted->mnt_root);
707 /* no need for dcache_lock, as serialization is taken care in
710 int follow_down(struct path *path)
712 struct vfsmount *mounted;
714 mounted = lookup_mnt(path);
719 path->dentry = dget(mounted->mnt_root);
725 static __always_inline void follow_dotdot(struct nameidata *nd)
730 struct vfsmount *parent;
731 struct dentry *old = nd->path.dentry;
733 if (nd->path.dentry == nd->root.dentry &&
734 nd->path.mnt == nd->root.mnt) {
737 spin_lock(&dcache_lock);
738 if (nd->path.dentry != nd->path.mnt->mnt_root) {
739 nd->path.dentry = dget(nd->path.dentry->d_parent);
740 spin_unlock(&dcache_lock);
744 spin_unlock(&dcache_lock);
745 spin_lock(&vfsmount_lock);
746 parent = nd->path.mnt->mnt_parent;
747 if (parent == nd->path.mnt) {
748 spin_unlock(&vfsmount_lock);
752 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
753 spin_unlock(&vfsmount_lock);
755 mntput(nd->path.mnt);
756 nd->path.mnt = parent;
758 follow_mount(&nd->path);
762 * It's more convoluted than I'd like it to be, but... it's still fairly
763 * small and for now I'd prefer to have fast path as straight as possible.
764 * It _is_ time-critical.
766 static int do_lookup(struct nameidata *nd, struct qstr *name,
769 struct vfsmount *mnt = nd->path.mnt;
770 struct dentry *dentry;
772 * See if the low-level filesystem might want
773 * to use its own hash..
775 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
776 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
781 dentry = __d_lookup(nd->path.dentry, name);
784 if (dentry->d_op && dentry->d_op->d_revalidate)
785 goto need_revalidate;
788 path->dentry = dentry;
789 __follow_mount(path);
793 dentry = real_lookup(nd->path.dentry, name, nd);
799 dentry = do_revalidate(dentry, nd);
807 return PTR_ERR(dentry);
812 * This is the basic name resolution function, turning a pathname into
813 * the final dentry. We expect 'base' to be positive and a directory.
815 * Returns 0 and nd will have valid dentry and mnt on success.
816 * Returns error and drops reference to input namei data on failure.
818 static int link_path_walk(const char *name, struct nameidata *nd)
823 unsigned int lookup_flags = nd->flags;
830 inode = nd->path.dentry->d_inode;
832 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
834 /* At this point we know we have a real path component. */
840 nd->flags |= LOOKUP_CONTINUE;
841 err = exec_permission_lite(inode);
846 c = *(const unsigned char *)name;
848 hash = init_name_hash();
851 hash = partial_name_hash(c, hash);
852 c = *(const unsigned char *)name;
853 } while (c && (c != '/'));
854 this.len = name - (const char *) this.name;
855 this.hash = end_name_hash(hash);
857 /* remove trailing slashes? */
860 while (*++name == '/');
862 goto last_with_slashes;
865 * "." and ".." are special - ".." especially so because it has
866 * to be able to know about the current root directory and
867 * parent relationships.
869 if (this.name[0] == '.') switch (this.len) {
873 if (this.name[1] != '.')
876 inode = nd->path.dentry->d_inode;
881 /* This does the actual lookups.. */
882 err = do_lookup(nd, &this, &next);
887 inode = next.dentry->d_inode;
891 if (inode->i_op->follow_link) {
892 err = do_follow_link(&next, nd);
896 inode = nd->path.dentry->d_inode;
900 path_to_nameidata(&next, nd);
902 if (!inode->i_op->lookup)
905 /* here ends the main loop */
908 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
910 /* Clear LOOKUP_CONTINUE iff it was previously unset */
911 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
912 if (lookup_flags & LOOKUP_PARENT)
914 if (this.name[0] == '.') switch (this.len) {
918 if (this.name[1] != '.')
921 inode = nd->path.dentry->d_inode;
926 err = do_lookup(nd, &this, &next);
929 inode = next.dentry->d_inode;
930 if ((lookup_flags & LOOKUP_FOLLOW)
931 && inode && inode->i_op->follow_link) {
932 err = do_follow_link(&next, nd);
935 inode = nd->path.dentry->d_inode;
937 path_to_nameidata(&next, nd);
941 if (lookup_flags & LOOKUP_DIRECTORY) {
943 if (!inode->i_op->lookup)
949 nd->last_type = LAST_NORM;
950 if (this.name[0] != '.')
953 nd->last_type = LAST_DOT;
954 else if (this.len == 2 && this.name[1] == '.')
955 nd->last_type = LAST_DOTDOT;
960 * We bypassed the ordinary revalidation routines.
961 * We may need to check the cached dentry for staleness.
963 if (nd->path.dentry && nd->path.dentry->d_sb &&
964 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
966 /* Note: we do not d_invalidate() */
967 if (!nd->path.dentry->d_op->d_revalidate(
968 nd->path.dentry, nd))
974 path_put_conditional(&next, nd);
982 static int path_walk(const char *name, struct nameidata *nd)
984 struct path save = nd->path;
987 current->total_link_count = 0;
989 /* make sure the stuff we saved doesn't go away */
992 result = link_path_walk(name, nd);
993 if (result == -ESTALE) {
994 /* nd->path had been dropped */
995 current->total_link_count = 0;
998 nd->flags |= LOOKUP_REVAL;
999 result = link_path_walk(name, nd);
1007 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1013 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1016 nd->root.mnt = NULL;
1020 nd->path = nd->root;
1021 path_get(&nd->root);
1022 } else if (dfd == AT_FDCWD) {
1023 struct fs_struct *fs = current->fs;
1024 read_lock(&fs->lock);
1027 read_unlock(&fs->lock);
1029 struct dentry *dentry;
1031 file = fget_light(dfd, &fput_needed);
1036 dentry = file->f_path.dentry;
1039 if (!S_ISDIR(dentry->d_inode->i_mode))
1042 retval = file_permission(file, MAY_EXEC);
1046 nd->path = file->f_path;
1047 path_get(&file->f_path);
1049 fput_light(file, fput_needed);
1054 fput_light(file, fput_needed);
1059 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1060 static int do_path_lookup(int dfd, const char *name,
1061 unsigned int flags, struct nameidata *nd)
1063 int retval = path_init(dfd, name, flags, nd);
1065 retval = path_walk(name, nd);
1066 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1067 nd->path.dentry->d_inode))
1068 audit_inode(name, nd->path.dentry);
1070 path_put(&nd->root);
1071 nd->root.mnt = NULL;
1076 int path_lookup(const char *name, unsigned int flags,
1077 struct nameidata *nd)
1079 return do_path_lookup(AT_FDCWD, name, flags, nd);
1082 int kern_path(const char *name, unsigned int flags, struct path *path)
1084 struct nameidata nd;
1085 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1092 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1093 * @dentry: pointer to dentry of the base directory
1094 * @mnt: pointer to vfs mount of the base directory
1095 * @name: pointer to file name
1096 * @flags: lookup flags
1097 * @nd: pointer to nameidata
1099 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1100 const char *name, unsigned int flags,
1101 struct nameidata *nd)
1105 /* same as do_path_lookup */
1106 nd->last_type = LAST_ROOT;
1110 nd->path.dentry = dentry;
1112 path_get(&nd->path);
1113 nd->root = nd->path;
1114 path_get(&nd->root);
1116 retval = path_walk(name, nd);
1117 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1118 nd->path.dentry->d_inode))
1119 audit_inode(name, nd->path.dentry);
1121 path_put(&nd->root);
1122 nd->root.mnt = NULL;
1128 * path_lookup_open - lookup a file path with open intent
1129 * @dfd: the directory to use as base, or AT_FDCWD
1130 * @name: pointer to file name
1131 * @lookup_flags: lookup intent flags
1132 * @nd: pointer to nameidata
1133 * @open_flags: open intent flags
1135 static int path_lookup_open(int dfd, const char *name,
1136 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1138 struct file *filp = get_empty_filp();
1143 nd->intent.open.file = filp;
1144 nd->intent.open.flags = open_flags;
1145 nd->intent.open.create_mode = 0;
1146 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1147 if (IS_ERR(nd->intent.open.file)) {
1149 err = PTR_ERR(nd->intent.open.file);
1150 path_put(&nd->path);
1152 } else if (err != 0)
1153 release_open_intent(nd);
1157 static struct dentry *__lookup_hash(struct qstr *name,
1158 struct dentry *base, struct nameidata *nd)
1160 struct dentry *dentry;
1161 struct inode *inode;
1164 inode = base->d_inode;
1167 * See if the low-level filesystem might want
1168 * to use its own hash..
1170 if (base->d_op && base->d_op->d_hash) {
1171 err = base->d_op->d_hash(base, name);
1172 dentry = ERR_PTR(err);
1177 dentry = cached_lookup(base, name, nd);
1181 /* Don't create child dentry for a dead directory. */
1182 dentry = ERR_PTR(-ENOENT);
1183 if (IS_DEADDIR(inode))
1186 new = d_alloc(base, name);
1187 dentry = ERR_PTR(-ENOMEM);
1190 dentry = inode->i_op->lookup(inode, new, nd);
1201 * Restricted form of lookup. Doesn't follow links, single-component only,
1202 * needs parent already locked. Doesn't follow mounts.
1205 static struct dentry *lookup_hash(struct nameidata *nd)
1209 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1211 return ERR_PTR(err);
1212 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1215 static int __lookup_one_len(const char *name, struct qstr *this,
1216 struct dentry *base, int len)
1226 hash = init_name_hash();
1228 c = *(const unsigned char *)name++;
1229 if (c == '/' || c == '\0')
1231 hash = partial_name_hash(c, hash);
1233 this->hash = end_name_hash(hash);
1238 * lookup_one_len - filesystem helper to lookup single pathname component
1239 * @name: pathname component to lookup
1240 * @base: base directory to lookup from
1241 * @len: maximum length @len should be interpreted to
1243 * Note that this routine is purely a helper for filesystem usage and should
1244 * not be called by generic code. Also note that by using this function the
1245 * nameidata argument is passed to the filesystem methods and a filesystem
1246 * using this helper needs to be prepared for that.
1248 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1253 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1255 err = __lookup_one_len(name, &this, base, len);
1257 return ERR_PTR(err);
1259 err = inode_permission(base->d_inode, MAY_EXEC);
1261 return ERR_PTR(err);
1262 return __lookup_hash(&this, base, NULL);
1265 int user_path_at(int dfd, const char __user *name, unsigned flags,
1268 struct nameidata nd;
1269 char *tmp = getname(name);
1270 int err = PTR_ERR(tmp);
1273 BUG_ON(flags & LOOKUP_PARENT);
1275 err = do_path_lookup(dfd, tmp, flags, &nd);
1283 static int user_path_parent(int dfd, const char __user *path,
1284 struct nameidata *nd, char **name)
1286 char *s = getname(path);
1292 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1302 * It's inline, so penalty for filesystems that don't use sticky bit is
1305 static inline int check_sticky(struct inode *dir, struct inode *inode)
1307 uid_t fsuid = current_fsuid();
1309 if (!(dir->i_mode & S_ISVTX))
1311 if (inode->i_uid == fsuid)
1313 if (dir->i_uid == fsuid)
1315 return !capable(CAP_FOWNER);
1319 * Check whether we can remove a link victim from directory dir, check
1320 * whether the type of victim is right.
1321 * 1. We can't do it if dir is read-only (done in permission())
1322 * 2. We should have write and exec permissions on dir
1323 * 3. We can't remove anything from append-only dir
1324 * 4. We can't do anything with immutable dir (done in permission())
1325 * 5. If the sticky bit on dir is set we should either
1326 * a. be owner of dir, or
1327 * b. be owner of victim, or
1328 * c. have CAP_FOWNER capability
1329 * 6. If the victim is append-only or immutable we can't do antyhing with
1330 * links pointing to it.
1331 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1332 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1333 * 9. We can't remove a root or mountpoint.
1334 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1335 * nfs_async_unlink().
1337 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1341 if (!victim->d_inode)
1344 BUG_ON(victim->d_parent->d_inode != dir);
1345 audit_inode_child(victim->d_name.name, victim, dir);
1347 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1352 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1353 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1356 if (!S_ISDIR(victim->d_inode->i_mode))
1358 if (IS_ROOT(victim))
1360 } else if (S_ISDIR(victim->d_inode->i_mode))
1362 if (IS_DEADDIR(dir))
1364 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1369 /* Check whether we can create an object with dentry child in directory
1371 * 1. We can't do it if child already exists (open has special treatment for
1372 * this case, but since we are inlined it's OK)
1373 * 2. We can't do it if dir is read-only (done in permission())
1374 * 3. We should have write and exec permissions on dir
1375 * 4. We can't do it if dir is immutable (done in permission())
1377 static inline int may_create(struct inode *dir, struct dentry *child)
1381 if (IS_DEADDIR(dir))
1383 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1387 * O_DIRECTORY translates into forcing a directory lookup.
1389 static inline int lookup_flags(unsigned int f)
1391 unsigned long retval = LOOKUP_FOLLOW;
1394 retval &= ~LOOKUP_FOLLOW;
1396 if (f & O_DIRECTORY)
1397 retval |= LOOKUP_DIRECTORY;
1403 * p1 and p2 should be directories on the same fs.
1405 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1410 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1414 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1416 p = d_ancestor(p2, p1);
1418 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1419 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1423 p = d_ancestor(p1, p2);
1425 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1426 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1430 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1431 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1435 void unlock_rename(struct dentry *p1, struct dentry *p2)
1437 mutex_unlock(&p1->d_inode->i_mutex);
1439 mutex_unlock(&p2->d_inode->i_mutex);
1440 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1444 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1445 struct nameidata *nd)
1447 int error = may_create(dir, dentry);
1452 if (!dir->i_op->create)
1453 return -EACCES; /* shouldn't it be ENOSYS? */
1456 error = security_inode_create(dir, dentry, mode);
1460 error = dir->i_op->create(dir, dentry, mode, nd);
1462 fsnotify_create(dir, dentry);
1466 int may_open(struct path *path, int acc_mode, int flag)
1468 struct dentry *dentry = path->dentry;
1469 struct inode *inode = dentry->d_inode;
1475 switch (inode->i_mode & S_IFMT) {
1479 if (acc_mode & MAY_WRITE)
1484 if (path->mnt->mnt_flags & MNT_NODEV)
1493 error = inode_permission(inode, acc_mode);
1497 error = ima_path_check(path, acc_mode ?
1498 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1499 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1505 * An append-only file must be opened in append mode for writing.
1507 if (IS_APPEND(inode)) {
1509 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1515 /* O_NOATIME can only be set by the owner or superuser */
1516 if (flag & O_NOATIME)
1517 if (!is_owner_or_cap(inode)) {
1523 * Ensure there are no outstanding leases on the file.
1525 error = break_lease(inode, flag);
1529 if (flag & O_TRUNC) {
1530 error = get_write_access(inode);
1535 * Refuse to truncate files with mandatory locks held on them.
1537 error = locks_verify_locked(inode);
1539 error = security_path_truncate(path, 0,
1540 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1544 error = do_truncate(dentry, 0,
1545 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1548 put_write_access(inode);
1552 if (flag & FMODE_WRITE)
1557 ima_counts_put(path, acc_mode ?
1558 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1559 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1564 * Be careful about ever adding any more callers of this
1565 * function. Its flags must be in the namei format, not
1566 * what get passed to sys_open().
1568 static int __open_namei_create(struct nameidata *nd, struct path *path,
1572 struct dentry *dir = nd->path.dentry;
1574 if (!IS_POSIXACL(dir->d_inode))
1575 mode &= ~current_umask();
1576 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1579 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1581 mutex_unlock(&dir->d_inode->i_mutex);
1582 dput(nd->path.dentry);
1583 nd->path.dentry = path->dentry;
1586 /* Don't check for write permission, don't truncate */
1587 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1591 * Note that while the flag value (low two bits) for sys_open means:
1596 * it is changed into
1597 * 00 - no permissions needed
1598 * 01 - read-permission
1599 * 10 - write-permission
1601 * for the internal routines (ie open_namei()/follow_link() etc)
1602 * This is more logical, and also allows the 00 "no perm needed"
1603 * to be used for symlinks (where the permissions are checked
1607 static inline int open_to_namei_flags(int flag)
1609 if ((flag+1) & O_ACCMODE)
1614 static int open_will_write_to_fs(int flag, struct inode *inode)
1617 * We'll never write to the fs underlying
1620 if (special_file(inode->i_mode))
1622 return (flag & O_TRUNC);
1626 * Note that the low bits of the passed in "open_flag"
1627 * are not the same as in the local variable "flag". See
1628 * open_to_namei_flags() for more details.
1630 struct file *do_filp_open(int dfd, const char *pathname,
1631 int open_flag, int mode, int acc_mode)
1634 struct nameidata nd;
1636 struct path path, save;
1640 int flag = open_to_namei_flags(open_flag);
1643 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1644 * check for O_DSYNC if the need any syncing at all we enforce it's
1645 * always set instead of having to deal with possibly weird behaviour
1646 * for malicious applications setting only __O_SYNC.
1648 if (open_flag & __O_SYNC)
1649 open_flag |= O_DSYNC;
1652 acc_mode = MAY_OPEN | ACC_MODE(flag);
1654 /* O_TRUNC implies we need access checks for write permissions */
1656 acc_mode |= MAY_WRITE;
1658 /* Allow the LSM permission hook to distinguish append
1659 access from general write access. */
1660 if (flag & O_APPEND)
1661 acc_mode |= MAY_APPEND;
1664 * The simplest case - just a plain lookup.
1666 if (!(flag & O_CREAT)) {
1667 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1670 return ERR_PTR(error);
1675 * Create - we need to know the parent.
1677 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1679 return ERR_PTR(error);
1680 error = path_walk(pathname, &nd);
1684 return ERR_PTR(error);
1686 if (unlikely(!audit_dummy_context()))
1687 audit_inode(pathname, nd.path.dentry);
1690 * We have the parent and last component. First of all, check
1691 * that we are not asked to creat(2) an obvious directory - that
1695 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1699 filp = get_empty_filp();
1702 nd.intent.open.file = filp;
1703 nd.intent.open.flags = flag;
1704 nd.intent.open.create_mode = mode;
1705 dir = nd.path.dentry;
1706 nd.flags &= ~LOOKUP_PARENT;
1707 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1709 nd.flags |= LOOKUP_EXCL;
1710 mutex_lock(&dir->d_inode->i_mutex);
1711 path.dentry = lookup_hash(&nd);
1712 path.mnt = nd.path.mnt;
1715 error = PTR_ERR(path.dentry);
1716 if (IS_ERR(path.dentry)) {
1717 mutex_unlock(&dir->d_inode->i_mutex);
1721 if (IS_ERR(nd.intent.open.file)) {
1722 error = PTR_ERR(nd.intent.open.file);
1723 goto exit_mutex_unlock;
1726 /* Negative dentry, just create the file */
1727 if (!path.dentry->d_inode) {
1729 * This write is needed to ensure that a
1730 * ro->rw transition does not occur between
1731 * the time when the file is created and when
1732 * a permanent write count is taken through
1733 * the 'struct file' in nameidata_to_filp().
1735 error = mnt_want_write(nd.path.mnt);
1737 goto exit_mutex_unlock;
1738 error = __open_namei_create(&nd, &path, flag, mode);
1740 mnt_drop_write(nd.path.mnt);
1743 filp = nameidata_to_filp(&nd, open_flag);
1745 ima_counts_put(&nd.path,
1746 acc_mode & (MAY_READ | MAY_WRITE |
1748 mnt_drop_write(nd.path.mnt);
1755 * It already exists.
1757 mutex_unlock(&dir->d_inode->i_mutex);
1758 audit_inode(pathname, path.dentry);
1764 if (__follow_mount(&path)) {
1766 if (flag & O_NOFOLLOW)
1771 if (!path.dentry->d_inode)
1773 if (path.dentry->d_inode->i_op->follow_link)
1776 path_to_nameidata(&path, &nd);
1778 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1783 * 1. may_open() truncates a file
1784 * 2. a rw->ro mount transition occurs
1785 * 3. nameidata_to_filp() fails due to
1787 * That would be inconsistent, and should
1788 * be avoided. Taking this mnt write here
1789 * ensures that (2) can not occur.
1791 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1793 error = mnt_want_write(nd.path.mnt);
1797 error = may_open(&nd.path, acc_mode, flag);
1800 mnt_drop_write(nd.path.mnt);
1803 filp = nameidata_to_filp(&nd, open_flag);
1805 ima_counts_put(&nd.path,
1806 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1808 * It is now safe to drop the mnt write
1809 * because the filp has had a write taken
1813 mnt_drop_write(nd.path.mnt);
1819 mutex_unlock(&dir->d_inode->i_mutex);
1821 path_put_conditional(&path, &nd);
1823 if (!IS_ERR(nd.intent.open.file))
1824 release_open_intent(&nd);
1829 return ERR_PTR(error);
1833 if (flag & O_NOFOLLOW)
1836 * This is subtle. Instead of calling do_follow_link() we do the
1837 * thing by hands. The reason is that this way we have zero link_count
1838 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1839 * After that we have the parent and last component, i.e.
1840 * we are in the same situation as after the first path_walk().
1841 * Well, almost - if the last component is normal we get its copy
1842 * stored in nd->last.name and we will have to putname() it when we
1843 * are done. Procfs-like symlinks just set LAST_BIND.
1845 nd.flags |= LOOKUP_PARENT;
1846 error = security_inode_follow_link(path.dentry, &nd);
1851 error = __do_follow_link(&path, &nd);
1852 if (error == -ESTALE) {
1853 /* nd.path had been dropped */
1856 nd.flags |= LOOKUP_REVAL;
1857 error = __do_follow_link(&path, &nd);
1862 /* Does someone understand code flow here? Or it is only
1863 * me so stupid? Anathema to whoever designed this non-sense
1864 * with "intent.open".
1866 release_open_intent(&nd);
1869 return ERR_PTR(error);
1871 nd.flags &= ~LOOKUP_PARENT;
1872 if (nd.last_type == LAST_BIND)
1875 if (nd.last_type != LAST_NORM)
1877 if (nd.last.name[nd.last.len]) {
1878 __putname(nd.last.name);
1883 __putname(nd.last.name);
1886 dir = nd.path.dentry;
1887 mutex_lock(&dir->d_inode->i_mutex);
1888 path.dentry = lookup_hash(&nd);
1889 path.mnt = nd.path.mnt;
1890 __putname(nd.last.name);
1895 * filp_open - open file and return file pointer
1897 * @filename: path to open
1898 * @flags: open flags as per the open(2) second argument
1899 * @mode: mode for the new file if O_CREAT is set, else ignored
1901 * This is the helper to open a file from kernelspace if you really
1902 * have to. But in generally you should not do this, so please move
1903 * along, nothing to see here..
1905 struct file *filp_open(const char *filename, int flags, int mode)
1907 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1909 EXPORT_SYMBOL(filp_open);
1912 * lookup_create - lookup a dentry, creating it if it doesn't exist
1913 * @nd: nameidata info
1914 * @is_dir: directory flag
1916 * Simple function to lookup and return a dentry and create it
1917 * if it doesn't exist. Is SMP-safe.
1919 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1921 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1923 struct dentry *dentry = ERR_PTR(-EEXIST);
1925 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1927 * Yucky last component or no last component at all?
1928 * (foo/., foo/.., /////)
1930 if (nd->last_type != LAST_NORM)
1932 nd->flags &= ~LOOKUP_PARENT;
1933 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1934 nd->intent.open.flags = O_EXCL;
1937 * Do the final lookup.
1939 dentry = lookup_hash(nd);
1943 if (dentry->d_inode)
1946 * Special case - lookup gave negative, but... we had foo/bar/
1947 * From the vfs_mknod() POV we just have a negative dentry -
1948 * all is fine. Let's be bastards - you had / on the end, you've
1949 * been asking for (non-existent) directory. -ENOENT for you.
1951 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1953 dentry = ERR_PTR(-ENOENT);
1958 dentry = ERR_PTR(-EEXIST);
1962 EXPORT_SYMBOL_GPL(lookup_create);
1964 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1966 int error = may_create(dir, dentry);
1971 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1974 if (!dir->i_op->mknod)
1977 error = devcgroup_inode_mknod(mode, dev);
1981 error = security_inode_mknod(dir, dentry, mode, dev);
1986 error = dir->i_op->mknod(dir, dentry, mode, dev);
1988 fsnotify_create(dir, dentry);
1992 static int may_mknod(mode_t mode)
1994 switch (mode & S_IFMT) {
2000 case 0: /* zero mode translates to S_IFREG */
2009 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2014 struct dentry *dentry;
2015 struct nameidata nd;
2020 error = user_path_parent(dfd, filename, &nd, &tmp);
2024 dentry = lookup_create(&nd, 0);
2025 if (IS_ERR(dentry)) {
2026 error = PTR_ERR(dentry);
2029 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2030 mode &= ~current_umask();
2031 error = may_mknod(mode);
2034 error = mnt_want_write(nd.path.mnt);
2037 error = security_path_mknod(&nd.path, dentry, mode, dev);
2039 goto out_drop_write;
2040 switch (mode & S_IFMT) {
2041 case 0: case S_IFREG:
2042 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2044 case S_IFCHR: case S_IFBLK:
2045 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2046 new_decode_dev(dev));
2048 case S_IFIFO: case S_IFSOCK:
2049 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2053 mnt_drop_write(nd.path.mnt);
2057 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2064 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2066 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2069 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2071 int error = may_create(dir, dentry);
2076 if (!dir->i_op->mkdir)
2079 mode &= (S_IRWXUGO|S_ISVTX);
2080 error = security_inode_mkdir(dir, dentry, mode);
2085 error = dir->i_op->mkdir(dir, dentry, mode);
2087 fsnotify_mkdir(dir, dentry);
2091 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2095 struct dentry *dentry;
2096 struct nameidata nd;
2098 error = user_path_parent(dfd, pathname, &nd, &tmp);
2102 dentry = lookup_create(&nd, 1);
2103 error = PTR_ERR(dentry);
2107 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2108 mode &= ~current_umask();
2109 error = mnt_want_write(nd.path.mnt);
2112 error = security_path_mkdir(&nd.path, dentry, mode);
2114 goto out_drop_write;
2115 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2117 mnt_drop_write(nd.path.mnt);
2121 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2128 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2130 return sys_mkdirat(AT_FDCWD, pathname, mode);
2134 * We try to drop the dentry early: we should have
2135 * a usage count of 2 if we're the only user of this
2136 * dentry, and if that is true (possibly after pruning
2137 * the dcache), then we drop the dentry now.
2139 * A low-level filesystem can, if it choses, legally
2142 * if (!d_unhashed(dentry))
2145 * if it cannot handle the case of removing a directory
2146 * that is still in use by something else..
2148 void dentry_unhash(struct dentry *dentry)
2151 shrink_dcache_parent(dentry);
2152 spin_lock(&dcache_lock);
2153 spin_lock(&dentry->d_lock);
2154 if (atomic_read(&dentry->d_count) == 2)
2156 spin_unlock(&dentry->d_lock);
2157 spin_unlock(&dcache_lock);
2160 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2162 int error = may_delete(dir, dentry, 1);
2167 if (!dir->i_op->rmdir)
2172 mutex_lock(&dentry->d_inode->i_mutex);
2173 dentry_unhash(dentry);
2174 if (d_mountpoint(dentry))
2177 error = security_inode_rmdir(dir, dentry);
2179 error = dir->i_op->rmdir(dir, dentry);
2181 dentry->d_inode->i_flags |= S_DEAD;
2184 mutex_unlock(&dentry->d_inode->i_mutex);
2193 static long do_rmdir(int dfd, const char __user *pathname)
2197 struct dentry *dentry;
2198 struct nameidata nd;
2200 error = user_path_parent(dfd, pathname, &nd, &name);
2204 switch(nd.last_type) {
2216 nd.flags &= ~LOOKUP_PARENT;
2218 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2219 dentry = lookup_hash(&nd);
2220 error = PTR_ERR(dentry);
2223 error = mnt_want_write(nd.path.mnt);
2226 error = security_path_rmdir(&nd.path, dentry);
2229 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2231 mnt_drop_write(nd.path.mnt);
2235 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2242 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2244 return do_rmdir(AT_FDCWD, pathname);
2247 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2249 int error = may_delete(dir, dentry, 0);
2254 if (!dir->i_op->unlink)
2259 mutex_lock(&dentry->d_inode->i_mutex);
2260 if (d_mountpoint(dentry))
2263 error = security_inode_unlink(dir, dentry);
2265 error = dir->i_op->unlink(dir, dentry);
2267 mutex_unlock(&dentry->d_inode->i_mutex);
2269 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2270 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2271 fsnotify_link_count(dentry->d_inode);
2279 * Make sure that the actual truncation of the file will occur outside its
2280 * directory's i_mutex. Truncate can take a long time if there is a lot of
2281 * writeout happening, and we don't want to prevent access to the directory
2282 * while waiting on the I/O.
2284 static long do_unlinkat(int dfd, const char __user *pathname)
2288 struct dentry *dentry;
2289 struct nameidata nd;
2290 struct inode *inode = NULL;
2292 error = user_path_parent(dfd, pathname, &nd, &name);
2297 if (nd.last_type != LAST_NORM)
2300 nd.flags &= ~LOOKUP_PARENT;
2302 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2303 dentry = lookup_hash(&nd);
2304 error = PTR_ERR(dentry);
2305 if (!IS_ERR(dentry)) {
2306 /* Why not before? Because we want correct error value */
2307 if (nd.last.name[nd.last.len])
2309 inode = dentry->d_inode;
2311 atomic_inc(&inode->i_count);
2312 error = mnt_want_write(nd.path.mnt);
2315 error = security_path_unlink(&nd.path, dentry);
2318 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2320 mnt_drop_write(nd.path.mnt);
2324 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2326 iput(inode); /* truncate the inode here */
2333 error = !dentry->d_inode ? -ENOENT :
2334 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2338 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2340 if ((flag & ~AT_REMOVEDIR) != 0)
2343 if (flag & AT_REMOVEDIR)
2344 return do_rmdir(dfd, pathname);
2346 return do_unlinkat(dfd, pathname);
2349 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2351 return do_unlinkat(AT_FDCWD, pathname);
2354 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2356 int error = may_create(dir, dentry);
2361 if (!dir->i_op->symlink)
2364 error = security_inode_symlink(dir, dentry, oldname);
2369 error = dir->i_op->symlink(dir, dentry, oldname);
2371 fsnotify_create(dir, dentry);
2375 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2376 int, newdfd, const char __user *, newname)
2381 struct dentry *dentry;
2382 struct nameidata nd;
2384 from = getname(oldname);
2386 return PTR_ERR(from);
2388 error = user_path_parent(newdfd, newname, &nd, &to);
2392 dentry = lookup_create(&nd, 0);
2393 error = PTR_ERR(dentry);
2397 error = mnt_want_write(nd.path.mnt);
2400 error = security_path_symlink(&nd.path, dentry, from);
2402 goto out_drop_write;
2403 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2405 mnt_drop_write(nd.path.mnt);
2409 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2417 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2419 return sys_symlinkat(oldname, AT_FDCWD, newname);
2422 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2424 struct inode *inode = old_dentry->d_inode;
2430 error = may_create(dir, new_dentry);
2434 if (dir->i_sb != inode->i_sb)
2438 * A link to an append-only or immutable file cannot be created.
2440 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2442 if (!dir->i_op->link)
2444 if (S_ISDIR(inode->i_mode))
2447 error = security_inode_link(old_dentry, dir, new_dentry);
2451 mutex_lock(&inode->i_mutex);
2453 error = dir->i_op->link(old_dentry, dir, new_dentry);
2454 mutex_unlock(&inode->i_mutex);
2456 fsnotify_link(dir, inode, new_dentry);
2461 * Hardlinks are often used in delicate situations. We avoid
2462 * security-related surprises by not following symlinks on the
2465 * We don't follow them on the oldname either to be compatible
2466 * with linux 2.0, and to avoid hard-linking to directories
2467 * and other special files. --ADM
2469 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2470 int, newdfd, const char __user *, newname, int, flags)
2472 struct dentry *new_dentry;
2473 struct nameidata nd;
2474 struct path old_path;
2478 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2481 error = user_path_at(olddfd, oldname,
2482 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2487 error = user_path_parent(newdfd, newname, &nd, &to);
2491 if (old_path.mnt != nd.path.mnt)
2493 new_dentry = lookup_create(&nd, 0);
2494 error = PTR_ERR(new_dentry);
2495 if (IS_ERR(new_dentry))
2497 error = mnt_want_write(nd.path.mnt);
2500 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2502 goto out_drop_write;
2503 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2505 mnt_drop_write(nd.path.mnt);
2509 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2514 path_put(&old_path);
2519 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2521 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2525 * The worst of all namespace operations - renaming directory. "Perverted"
2526 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2528 * a) we can get into loop creation. Check is done in is_subdir().
2529 * b) race potential - two innocent renames can create a loop together.
2530 * That's where 4.4 screws up. Current fix: serialization on
2531 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2533 * c) we have to lock _three_ objects - parents and victim (if it exists).
2534 * And that - after we got ->i_mutex on parents (until then we don't know
2535 * whether the target exists). Solution: try to be smart with locking
2536 * order for inodes. We rely on the fact that tree topology may change
2537 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2538 * move will be locked. Thus we can rank directories by the tree
2539 * (ancestors first) and rank all non-directories after them.
2540 * That works since everybody except rename does "lock parent, lookup,
2541 * lock child" and rename is under ->s_vfs_rename_mutex.
2542 * HOWEVER, it relies on the assumption that any object with ->lookup()
2543 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2544 * we'd better make sure that there's no link(2) for them.
2545 * d) some filesystems don't support opened-but-unlinked directories,
2546 * either because of layout or because they are not ready to deal with
2547 * all cases correctly. The latter will be fixed (taking this sort of
2548 * stuff into VFS), but the former is not going away. Solution: the same
2549 * trick as in rmdir().
2550 * e) conversion from fhandle to dentry may come in the wrong moment - when
2551 * we are removing the target. Solution: we will have to grab ->i_mutex
2552 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2553 * ->i_mutex on parents, which works but leads to some truely excessive
2556 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2557 struct inode *new_dir, struct dentry *new_dentry)
2560 struct inode *target;
2563 * If we are going to change the parent - check write permissions,
2564 * we'll need to flip '..'.
2566 if (new_dir != old_dir) {
2567 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2572 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2576 target = new_dentry->d_inode;
2578 mutex_lock(&target->i_mutex);
2579 dentry_unhash(new_dentry);
2581 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2584 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2587 target->i_flags |= S_DEAD;
2588 mutex_unlock(&target->i_mutex);
2589 if (d_unhashed(new_dentry))
2590 d_rehash(new_dentry);
2594 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2595 d_move(old_dentry,new_dentry);
2599 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2600 struct inode *new_dir, struct dentry *new_dentry)
2602 struct inode *target;
2605 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2610 target = new_dentry->d_inode;
2612 mutex_lock(&target->i_mutex);
2613 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2616 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2618 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2619 d_move(old_dentry, new_dentry);
2622 mutex_unlock(&target->i_mutex);
2627 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2628 struct inode *new_dir, struct dentry *new_dentry)
2631 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2632 const char *old_name;
2634 if (old_dentry->d_inode == new_dentry->d_inode)
2637 error = may_delete(old_dir, old_dentry, is_dir);
2641 if (!new_dentry->d_inode)
2642 error = may_create(new_dir, new_dentry);
2644 error = may_delete(new_dir, new_dentry, is_dir);
2648 if (!old_dir->i_op->rename)
2651 vfs_dq_init(old_dir);
2652 vfs_dq_init(new_dir);
2654 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2657 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2659 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2661 const char *new_name = old_dentry->d_name.name;
2662 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2663 new_dentry->d_inode, old_dentry);
2665 fsnotify_oldname_free(old_name);
2670 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2671 int, newdfd, const char __user *, newname)
2673 struct dentry *old_dir, *new_dir;
2674 struct dentry *old_dentry, *new_dentry;
2675 struct dentry *trap;
2676 struct nameidata oldnd, newnd;
2681 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2685 error = user_path_parent(newdfd, newname, &newnd, &to);
2690 if (oldnd.path.mnt != newnd.path.mnt)
2693 old_dir = oldnd.path.dentry;
2695 if (oldnd.last_type != LAST_NORM)
2698 new_dir = newnd.path.dentry;
2699 if (newnd.last_type != LAST_NORM)
2702 oldnd.flags &= ~LOOKUP_PARENT;
2703 newnd.flags &= ~LOOKUP_PARENT;
2704 newnd.flags |= LOOKUP_RENAME_TARGET;
2706 trap = lock_rename(new_dir, old_dir);
2708 old_dentry = lookup_hash(&oldnd);
2709 error = PTR_ERR(old_dentry);
2710 if (IS_ERR(old_dentry))
2712 /* source must exist */
2714 if (!old_dentry->d_inode)
2716 /* unless the source is a directory trailing slashes give -ENOTDIR */
2717 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2719 if (oldnd.last.name[oldnd.last.len])
2721 if (newnd.last.name[newnd.last.len])
2724 /* source should not be ancestor of target */
2726 if (old_dentry == trap)
2728 new_dentry = lookup_hash(&newnd);
2729 error = PTR_ERR(new_dentry);
2730 if (IS_ERR(new_dentry))
2732 /* target should not be an ancestor of source */
2734 if (new_dentry == trap)
2737 error = mnt_want_write(oldnd.path.mnt);
2740 error = security_path_rename(&oldnd.path, old_dentry,
2741 &newnd.path, new_dentry);
2744 error = vfs_rename(old_dir->d_inode, old_dentry,
2745 new_dir->d_inode, new_dentry);
2747 mnt_drop_write(oldnd.path.mnt);
2753 unlock_rename(new_dir, old_dir);
2755 path_put(&newnd.path);
2758 path_put(&oldnd.path);
2764 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2766 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2769 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2773 len = PTR_ERR(link);
2778 if (len > (unsigned) buflen)
2780 if (copy_to_user(buffer, link, len))
2787 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2788 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2789 * using) it for any given inode is up to filesystem.
2791 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2793 struct nameidata nd;
2798 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2800 return PTR_ERR(cookie);
2802 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2803 if (dentry->d_inode->i_op->put_link)
2804 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2808 int vfs_follow_link(struct nameidata *nd, const char *link)
2810 return __vfs_follow_link(nd, link);
2813 /* get the link contents into pagecache */
2814 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2818 struct address_space *mapping = dentry->d_inode->i_mapping;
2819 page = read_mapping_page(mapping, 0, NULL);
2824 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2828 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2830 struct page *page = NULL;
2831 char *s = page_getlink(dentry, &page);
2832 int res = vfs_readlink(dentry,buffer,buflen,s);
2835 page_cache_release(page);
2840 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2842 struct page *page = NULL;
2843 nd_set_link(nd, page_getlink(dentry, &page));
2847 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2849 struct page *page = cookie;
2853 page_cache_release(page);
2858 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2860 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2862 struct address_space *mapping = inode->i_mapping;
2867 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2869 flags |= AOP_FLAG_NOFS;
2872 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2873 flags, &page, &fsdata);
2877 kaddr = kmap_atomic(page, KM_USER0);
2878 memcpy(kaddr, symname, len-1);
2879 kunmap_atomic(kaddr, KM_USER0);
2881 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2888 mark_inode_dirty(inode);
2894 int page_symlink(struct inode *inode, const char *symname, int len)
2896 return __page_symlink(inode, symname, len,
2897 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2900 const struct inode_operations page_symlink_inode_operations = {
2901 .readlink = generic_readlink,
2902 .follow_link = page_follow_link_light,
2903 .put_link = page_put_link,
2906 EXPORT_SYMBOL(user_path_at);
2907 EXPORT_SYMBOL(follow_down);
2908 EXPORT_SYMBOL(follow_up);
2909 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2910 EXPORT_SYMBOL(getname);
2911 EXPORT_SYMBOL(lock_rename);
2912 EXPORT_SYMBOL(lookup_one_len);
2913 EXPORT_SYMBOL(page_follow_link_light);
2914 EXPORT_SYMBOL(page_put_link);
2915 EXPORT_SYMBOL(page_readlink);
2916 EXPORT_SYMBOL(__page_symlink);
2917 EXPORT_SYMBOL(page_symlink);
2918 EXPORT_SYMBOL(page_symlink_inode_operations);
2919 EXPORT_SYMBOL(path_lookup);
2920 EXPORT_SYMBOL(kern_path);
2921 EXPORT_SYMBOL(vfs_path_lookup);
2922 EXPORT_SYMBOL(inode_permission);
2923 EXPORT_SYMBOL(file_permission);
2924 EXPORT_SYMBOL(unlock_rename);
2925 EXPORT_SYMBOL(vfs_create);
2926 EXPORT_SYMBOL(vfs_follow_link);
2927 EXPORT_SYMBOL(vfs_link);
2928 EXPORT_SYMBOL(vfs_mkdir);
2929 EXPORT_SYMBOL(vfs_mknod);
2930 EXPORT_SYMBOL(generic_permission);
2931 EXPORT_SYMBOL(vfs_readlink);
2932 EXPORT_SYMBOL(vfs_rename);
2933 EXPORT_SYMBOL(vfs_rmdir);
2934 EXPORT_SYMBOL(vfs_symlink);
2935 EXPORT_SYMBOL(vfs_unlink);
2936 EXPORT_SYMBOL(dentry_unhash);
2937 EXPORT_SYMBOL(generic_readlink);