4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user *filename, char *page)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 char * getname(const char __user * filename)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask))
175 umode_t mode = inode->i_mode;
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179 if (current_fsuid() == inode->i_uid)
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask);
184 if (error != -EAGAIN)
188 if (in_group_p(inode->i_gid))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask & ~mode) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
206 * Used to check for read/write/execute permissions on a file.
207 * We use "fsuid" for this, letting us set arbitrary permissions
208 * for filesystem access without changing the "normal" uids which
209 * are used for other things..
211 int generic_permission(struct inode *inode, int mask,
212 int (*check_acl)(struct inode *inode, int mask))
217 * Do the basic POSIX ACL permission checks.
219 ret = acl_permission_check(inode, mask, check_acl);
224 * Read/write DACs are always overridable.
225 * Executable DACs are overridable if at least one exec bit is set.
227 if (!(mask & MAY_EXEC) || execute_ok(inode))
228 if (capable(CAP_DAC_OVERRIDE))
232 * Searching includes executable on directories, else just read.
234 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
281 retval = devcgroup_inode_permission(inode, mask);
285 return security_inode_permission(inode, mask);
289 * file_permission - check for additional access rights to a given file
290 * @file: file to check access rights for
291 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
293 * Used to check for read/write/execute permissions on an already opened
297 * Do not use this function in new code. All access checks should
298 * be done using inode_permission().
300 int file_permission(struct file *file, int mask)
302 return inode_permission(file->f_path.dentry->d_inode, mask);
306 * get_write_access() gets write permission for a file.
307 * put_write_access() releases this write permission.
308 * This is used for regular files.
309 * We cannot support write (and maybe mmap read-write shared) accesses and
310 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
311 * can have the following values:
312 * 0: no writers, no VM_DENYWRITE mappings
313 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
314 * > 0: (i_writecount) users are writing to the file.
316 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
317 * except for the cases where we don't hold i_writecount yet. Then we need to
318 * use {get,deny}_write_access() - these functions check the sign and refuse
319 * to do the change if sign is wrong. Exclusion between them is provided by
320 * the inode->i_lock spinlock.
323 int get_write_access(struct inode * inode)
325 spin_lock(&inode->i_lock);
326 if (atomic_read(&inode->i_writecount) < 0) {
327 spin_unlock(&inode->i_lock);
330 atomic_inc(&inode->i_writecount);
331 spin_unlock(&inode->i_lock);
336 int deny_write_access(struct file * file)
338 struct inode *inode = file->f_path.dentry->d_inode;
340 spin_lock(&inode->i_lock);
341 if (atomic_read(&inode->i_writecount) > 0) {
342 spin_unlock(&inode->i_lock);
345 atomic_dec(&inode->i_writecount);
346 spin_unlock(&inode->i_lock);
352 * path_get - get a reference to a path
353 * @path: path to get the reference to
355 * Given a path increment the reference count to the dentry and the vfsmount.
357 void path_get(struct path *path)
362 EXPORT_SYMBOL(path_get);
365 * path_put - put a reference to a path
366 * @path: path to put the reference to
368 * Given a path decrement the reference count to the dentry and the vfsmount.
370 void path_put(struct path *path)
375 EXPORT_SYMBOL(path_put);
378 * release_open_intent - free up open intent resources
379 * @nd: pointer to nameidata
381 void release_open_intent(struct nameidata *nd)
383 if (nd->intent.open.file->f_path.dentry == NULL)
384 put_filp(nd->intent.open.file);
386 fput(nd->intent.open.file);
389 static inline struct dentry *
390 do_revalidate(struct dentry *dentry, struct nameidata *nd)
392 int status = dentry->d_op->d_revalidate(dentry, nd);
393 if (unlikely(status <= 0)) {
395 * The dentry failed validation.
396 * If d_revalidate returned 0 attempt to invalidate
397 * the dentry otherwise d_revalidate is asking us
398 * to return a fail status.
401 if (!d_invalidate(dentry)) {
407 dentry = ERR_PTR(status);
414 * force_reval_path - force revalidation of a dentry
416 * In some situations the path walking code will trust dentries without
417 * revalidating them. This causes problems for filesystems that depend on
418 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
419 * (which indicates that it's possible for the dentry to go stale), force
420 * a d_revalidate call before proceeding.
422 * Returns 0 if the revalidation was successful. If the revalidation fails,
423 * either return the error returned by d_revalidate or -ESTALE if the
424 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
425 * invalidate the dentry. It's up to the caller to handle putting references
426 * to the path if necessary.
429 force_reval_path(struct path *path, struct nameidata *nd)
432 struct dentry *dentry = path->dentry;
435 * only check on filesystems where it's possible for the dentry to
436 * become stale. It's assumed that if this flag is set then the
437 * d_revalidate op will also be defined.
439 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
442 status = dentry->d_op->d_revalidate(dentry, nd);
447 d_invalidate(dentry);
454 * Short-cut version of permission(), for calling on directories
455 * during pathname resolution. Combines parts of permission()
456 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
458 * If appropriate, check DAC only. If not appropriate, or
459 * short-cut DAC fails, then call ->permission() to do more
460 * complete permission check.
462 static int exec_permission(struct inode *inode)
466 if (inode->i_op->permission) {
467 ret = inode->i_op->permission(inode, MAY_EXEC);
472 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
476 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
481 return security_inode_permission(inode, MAY_EXEC);
484 static __always_inline void set_root(struct nameidata *nd)
487 get_fs_root(current->fs, &nd->root);
490 static int link_path_walk(const char *, struct nameidata *);
492 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
504 return link_path_walk(link, nd);
507 return PTR_ERR(link);
510 static void path_put_conditional(struct path *path, struct nameidata *nd)
513 if (path->mnt != nd->path.mnt)
517 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
519 dput(nd->path.dentry);
520 if (nd->path.mnt != path->mnt) {
521 mntput(nd->path.mnt);
522 nd->path.mnt = path->mnt;
524 nd->path.dentry = path->dentry;
527 static __always_inline int
528 __do_follow_link(struct path *path, struct nameidata *nd, void **p)
531 struct dentry *dentry = path->dentry;
533 touch_atime(path->mnt, dentry);
534 nd_set_link(nd, NULL);
536 if (path->mnt != nd->path.mnt) {
537 path_to_nameidata(path, nd);
541 nd->last_type = LAST_BIND;
542 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
545 char *s = nd_get_link(nd);
548 error = __vfs_follow_link(nd, s);
549 else if (nd->last_type == LAST_BIND) {
550 error = force_reval_path(&nd->path, nd);
559 * This limits recursive symlink follows to 8, while
560 * limiting consecutive symlinks to 40.
562 * Without that kind of total limit, nasty chains of consecutive
563 * symlinks can cause almost arbitrarily long lookups.
565 static inline int do_follow_link(struct path *path, struct nameidata *nd)
569 if (current->link_count >= MAX_NESTED_LINKS)
571 if (current->total_link_count >= 40)
573 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
575 err = security_inode_follow_link(path->dentry, nd);
578 current->link_count++;
579 current->total_link_count++;
581 err = __do_follow_link(path, nd, &cookie);
582 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
583 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
585 current->link_count--;
589 path_put_conditional(path, nd);
594 int follow_up(struct path *path)
596 struct vfsmount *parent;
597 struct dentry *mountpoint;
598 spin_lock(&vfsmount_lock);
599 parent = path->mnt->mnt_parent;
600 if (parent == path->mnt) {
601 spin_unlock(&vfsmount_lock);
605 mountpoint = dget(path->mnt->mnt_mountpoint);
606 spin_unlock(&vfsmount_lock);
608 path->dentry = mountpoint;
614 /* no need for dcache_lock, as serialization is taken care in
617 static int __follow_mount(struct path *path)
620 while (d_mountpoint(path->dentry)) {
621 struct vfsmount *mounted = lookup_mnt(path);
628 path->dentry = dget(mounted->mnt_root);
634 static void follow_mount(struct path *path)
636 while (d_mountpoint(path->dentry)) {
637 struct vfsmount *mounted = lookup_mnt(path);
643 path->dentry = dget(mounted->mnt_root);
647 /* no need for dcache_lock, as serialization is taken care in
650 int follow_down(struct path *path)
652 struct vfsmount *mounted;
654 mounted = lookup_mnt(path);
659 path->dentry = dget(mounted->mnt_root);
665 static __always_inline void follow_dotdot(struct nameidata *nd)
670 struct dentry *old = nd->path.dentry;
672 if (nd->path.dentry == nd->root.dentry &&
673 nd->path.mnt == nd->root.mnt) {
676 if (nd->path.dentry != nd->path.mnt->mnt_root) {
677 /* rare case of legitimate dget_parent()... */
678 nd->path.dentry = dget_parent(nd->path.dentry);
682 if (!follow_up(&nd->path))
685 follow_mount(&nd->path);
689 * It's more convoluted than I'd like it to be, but... it's still fairly
690 * small and for now I'd prefer to have fast path as straight as possible.
691 * It _is_ time-critical.
693 static int do_lookup(struct nameidata *nd, struct qstr *name,
696 struct vfsmount *mnt = nd->path.mnt;
697 struct dentry *dentry, *parent;
700 * See if the low-level filesystem might want
701 * to use its own hash..
703 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
704 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
709 dentry = __d_lookup(nd->path.dentry, name);
713 if (dentry->d_op && dentry->d_op->d_revalidate)
714 goto need_revalidate;
717 path->dentry = dentry;
718 __follow_mount(path);
722 parent = nd->path.dentry;
723 dir = parent->d_inode;
725 mutex_lock(&dir->i_mutex);
727 * First re-do the cached lookup just in case it was created
728 * while we waited for the directory semaphore..
730 * FIXME! This could use version numbering or similar to
731 * avoid unnecessary cache lookups.
733 * The "dcache_lock" is purely to protect the RCU list walker
734 * from concurrent renames at this point (we mustn't get false
735 * negatives from the RCU list walk here, unlike the optimistic
738 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
740 dentry = d_lookup(parent, name);
744 /* Don't create child dentry for a dead directory. */
745 dentry = ERR_PTR(-ENOENT);
749 new = d_alloc(parent, name);
750 dentry = ERR_PTR(-ENOMEM);
752 dentry = dir->i_op->lookup(dir, new, nd);
759 mutex_unlock(&dir->i_mutex);
766 * Uhhuh! Nasty case: the cache was re-populated while
767 * we waited on the semaphore. Need to revalidate.
769 mutex_unlock(&dir->i_mutex);
773 dentry = do_revalidate(dentry, nd);
781 return PTR_ERR(dentry);
785 * This is a temporary kludge to deal with "automount" symlinks; proper
786 * solution is to trigger them on follow_mount(), so that do_lookup()
787 * would DTRT. To be killed before 2.6.34-final.
789 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
791 return inode && unlikely(inode->i_op->follow_link) &&
792 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
797 * This is the basic name resolution function, turning a pathname into
798 * the final dentry. We expect 'base' to be positive and a directory.
800 * Returns 0 and nd will have valid dentry and mnt on success.
801 * Returns error and drops reference to input namei data on failure.
803 static int link_path_walk(const char *name, struct nameidata *nd)
808 unsigned int lookup_flags = nd->flags;
815 inode = nd->path.dentry->d_inode;
817 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
819 /* At this point we know we have a real path component. */
825 nd->flags |= LOOKUP_CONTINUE;
826 err = exec_permission(inode);
831 c = *(const unsigned char *)name;
833 hash = init_name_hash();
836 hash = partial_name_hash(c, hash);
837 c = *(const unsigned char *)name;
838 } while (c && (c != '/'));
839 this.len = name - (const char *) this.name;
840 this.hash = end_name_hash(hash);
842 /* remove trailing slashes? */
845 while (*++name == '/');
847 goto last_with_slashes;
850 * "." and ".." are special - ".." especially so because it has
851 * to be able to know about the current root directory and
852 * parent relationships.
854 if (this.name[0] == '.') switch (this.len) {
858 if (this.name[1] != '.')
861 inode = nd->path.dentry->d_inode;
866 /* This does the actual lookups.. */
867 err = do_lookup(nd, &this, &next);
872 inode = next.dentry->d_inode;
876 if (inode->i_op->follow_link) {
877 err = do_follow_link(&next, nd);
881 inode = nd->path.dentry->d_inode;
885 path_to_nameidata(&next, nd);
887 if (!inode->i_op->lookup)
890 /* here ends the main loop */
893 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
895 /* Clear LOOKUP_CONTINUE iff it was previously unset */
896 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
897 if (lookup_flags & LOOKUP_PARENT)
899 if (this.name[0] == '.') switch (this.len) {
903 if (this.name[1] != '.')
906 inode = nd->path.dentry->d_inode;
911 err = do_lookup(nd, &this, &next);
914 inode = next.dentry->d_inode;
915 if (follow_on_final(inode, lookup_flags)) {
916 err = do_follow_link(&next, nd);
919 inode = nd->path.dentry->d_inode;
921 path_to_nameidata(&next, nd);
925 if (lookup_flags & LOOKUP_DIRECTORY) {
927 if (!inode->i_op->lookup)
933 nd->last_type = LAST_NORM;
934 if (this.name[0] != '.')
937 nd->last_type = LAST_DOT;
938 else if (this.len == 2 && this.name[1] == '.')
939 nd->last_type = LAST_DOTDOT;
944 * We bypassed the ordinary revalidation routines.
945 * We may need to check the cached dentry for staleness.
947 if (nd->path.dentry && nd->path.dentry->d_sb &&
948 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
950 /* Note: we do not d_invalidate() */
951 if (!nd->path.dentry->d_op->d_revalidate(
952 nd->path.dentry, nd))
958 path_put_conditional(&next, nd);
966 static int path_walk(const char *name, struct nameidata *nd)
968 struct path save = nd->path;
971 current->total_link_count = 0;
973 /* make sure the stuff we saved doesn't go away */
976 result = link_path_walk(name, nd);
977 if (result == -ESTALE) {
978 /* nd->path had been dropped */
979 current->total_link_count = 0;
982 nd->flags |= LOOKUP_REVAL;
983 result = link_path_walk(name, nd);
991 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
997 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1000 nd->root.mnt = NULL;
1004 nd->path = nd->root;
1005 path_get(&nd->root);
1006 } else if (dfd == AT_FDCWD) {
1007 get_fs_pwd(current->fs, &nd->path);
1009 struct dentry *dentry;
1011 file = fget_light(dfd, &fput_needed);
1016 dentry = file->f_path.dentry;
1019 if (!S_ISDIR(dentry->d_inode->i_mode))
1022 retval = file_permission(file, MAY_EXEC);
1026 nd->path = file->f_path;
1027 path_get(&file->f_path);
1029 fput_light(file, fput_needed);
1034 fput_light(file, fput_needed);
1039 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1040 static int do_path_lookup(int dfd, const char *name,
1041 unsigned int flags, struct nameidata *nd)
1043 int retval = path_init(dfd, name, flags, nd);
1045 retval = path_walk(name, nd);
1046 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1047 nd->path.dentry->d_inode))
1048 audit_inode(name, nd->path.dentry);
1050 path_put(&nd->root);
1051 nd->root.mnt = NULL;
1056 int path_lookup(const char *name, unsigned int flags,
1057 struct nameidata *nd)
1059 return do_path_lookup(AT_FDCWD, name, flags, nd);
1062 int kern_path(const char *name, unsigned int flags, struct path *path)
1064 struct nameidata nd;
1065 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1072 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1073 * @dentry: pointer to dentry of the base directory
1074 * @mnt: pointer to vfs mount of the base directory
1075 * @name: pointer to file name
1076 * @flags: lookup flags
1077 * @nd: pointer to nameidata
1079 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1080 const char *name, unsigned int flags,
1081 struct nameidata *nd)
1085 /* same as do_path_lookup */
1086 nd->last_type = LAST_ROOT;
1090 nd->path.dentry = dentry;
1092 path_get(&nd->path);
1093 nd->root = nd->path;
1094 path_get(&nd->root);
1096 retval = path_walk(name, nd);
1097 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1098 nd->path.dentry->d_inode))
1099 audit_inode(name, nd->path.dentry);
1101 path_put(&nd->root);
1102 nd->root.mnt = NULL;
1107 static struct dentry *__lookup_hash(struct qstr *name,
1108 struct dentry *base, struct nameidata *nd)
1110 struct dentry *dentry;
1111 struct inode *inode;
1114 inode = base->d_inode;
1117 * See if the low-level filesystem might want
1118 * to use its own hash..
1120 if (base->d_op && base->d_op->d_hash) {
1121 err = base->d_op->d_hash(base, name);
1122 dentry = ERR_PTR(err);
1127 dentry = __d_lookup(base, name);
1129 /* lockess __d_lookup may fail due to concurrent d_move()
1130 * in some unrelated directory, so try with d_lookup
1133 dentry = d_lookup(base, name);
1135 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1136 dentry = do_revalidate(dentry, nd);
1141 /* Don't create child dentry for a dead directory. */
1142 dentry = ERR_PTR(-ENOENT);
1143 if (IS_DEADDIR(inode))
1146 new = d_alloc(base, name);
1147 dentry = ERR_PTR(-ENOMEM);
1150 dentry = inode->i_op->lookup(inode, new, nd);
1161 * Restricted form of lookup. Doesn't follow links, single-component only,
1162 * needs parent already locked. Doesn't follow mounts.
1165 static struct dentry *lookup_hash(struct nameidata *nd)
1169 err = exec_permission(nd->path.dentry->d_inode);
1171 return ERR_PTR(err);
1172 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1175 static int __lookup_one_len(const char *name, struct qstr *this,
1176 struct dentry *base, int len)
1186 hash = init_name_hash();
1188 c = *(const unsigned char *)name++;
1189 if (c == '/' || c == '\0')
1191 hash = partial_name_hash(c, hash);
1193 this->hash = end_name_hash(hash);
1198 * lookup_one_len - filesystem helper to lookup single pathname component
1199 * @name: pathname component to lookup
1200 * @base: base directory to lookup from
1201 * @len: maximum length @len should be interpreted to
1203 * Note that this routine is purely a helper for filesystem usage and should
1204 * not be called by generic code. Also note that by using this function the
1205 * nameidata argument is passed to the filesystem methods and a filesystem
1206 * using this helper needs to be prepared for that.
1208 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1213 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1215 err = __lookup_one_len(name, &this, base, len);
1217 return ERR_PTR(err);
1219 err = exec_permission(base->d_inode);
1221 return ERR_PTR(err);
1222 return __lookup_hash(&this, base, NULL);
1225 int user_path_at(int dfd, const char __user *name, unsigned flags,
1228 struct nameidata nd;
1229 char *tmp = getname(name);
1230 int err = PTR_ERR(tmp);
1233 BUG_ON(flags & LOOKUP_PARENT);
1235 err = do_path_lookup(dfd, tmp, flags, &nd);
1243 static int user_path_parent(int dfd, const char __user *path,
1244 struct nameidata *nd, char **name)
1246 char *s = getname(path);
1252 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1262 * It's inline, so penalty for filesystems that don't use sticky bit is
1265 static inline int check_sticky(struct inode *dir, struct inode *inode)
1267 uid_t fsuid = current_fsuid();
1269 if (!(dir->i_mode & S_ISVTX))
1271 if (inode->i_uid == fsuid)
1273 if (dir->i_uid == fsuid)
1275 return !capable(CAP_FOWNER);
1279 * Check whether we can remove a link victim from directory dir, check
1280 * whether the type of victim is right.
1281 * 1. We can't do it if dir is read-only (done in permission())
1282 * 2. We should have write and exec permissions on dir
1283 * 3. We can't remove anything from append-only dir
1284 * 4. We can't do anything with immutable dir (done in permission())
1285 * 5. If the sticky bit on dir is set we should either
1286 * a. be owner of dir, or
1287 * b. be owner of victim, or
1288 * c. have CAP_FOWNER capability
1289 * 6. If the victim is append-only or immutable we can't do antyhing with
1290 * links pointing to it.
1291 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1292 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1293 * 9. We can't remove a root or mountpoint.
1294 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1295 * nfs_async_unlink().
1297 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1301 if (!victim->d_inode)
1304 BUG_ON(victim->d_parent->d_inode != dir);
1305 audit_inode_child(victim, dir);
1307 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1312 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1313 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1316 if (!S_ISDIR(victim->d_inode->i_mode))
1318 if (IS_ROOT(victim))
1320 } else if (S_ISDIR(victim->d_inode->i_mode))
1322 if (IS_DEADDIR(dir))
1324 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1329 /* Check whether we can create an object with dentry child in directory
1331 * 1. We can't do it if child already exists (open has special treatment for
1332 * this case, but since we are inlined it's OK)
1333 * 2. We can't do it if dir is read-only (done in permission())
1334 * 3. We should have write and exec permissions on dir
1335 * 4. We can't do it if dir is immutable (done in permission())
1337 static inline int may_create(struct inode *dir, struct dentry *child)
1341 if (IS_DEADDIR(dir))
1343 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1347 * p1 and p2 should be directories on the same fs.
1349 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1354 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1358 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1360 p = d_ancestor(p2, p1);
1362 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1363 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1367 p = d_ancestor(p1, p2);
1369 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1370 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1374 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1375 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1379 void unlock_rename(struct dentry *p1, struct dentry *p2)
1381 mutex_unlock(&p1->d_inode->i_mutex);
1383 mutex_unlock(&p2->d_inode->i_mutex);
1384 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1388 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1389 struct nameidata *nd)
1391 int error = may_create(dir, dentry);
1396 if (!dir->i_op->create)
1397 return -EACCES; /* shouldn't it be ENOSYS? */
1400 error = security_inode_create(dir, dentry, mode);
1403 error = dir->i_op->create(dir, dentry, mode, nd);
1405 fsnotify_create(dir, dentry);
1409 int may_open(struct path *path, int acc_mode, int flag)
1411 struct dentry *dentry = path->dentry;
1412 struct inode *inode = dentry->d_inode;
1418 switch (inode->i_mode & S_IFMT) {
1422 if (acc_mode & MAY_WRITE)
1427 if (path->mnt->mnt_flags & MNT_NODEV)
1436 error = inode_permission(inode, acc_mode);
1441 * An append-only file must be opened in append mode for writing.
1443 if (IS_APPEND(inode)) {
1444 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1450 /* O_NOATIME can only be set by the owner or superuser */
1451 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1455 * Ensure there are no outstanding leases on the file.
1457 return break_lease(inode, flag);
1460 static int handle_truncate(struct path *path)
1462 struct inode *inode = path->dentry->d_inode;
1463 int error = get_write_access(inode);
1467 * Refuse to truncate files with mandatory locks held on them.
1469 error = locks_verify_locked(inode);
1471 error = security_path_truncate(path);
1473 error = do_truncate(path->dentry, 0,
1474 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1477 put_write_access(inode);
1482 * Be careful about ever adding any more callers of this
1483 * function. Its flags must be in the namei format, not
1484 * what get passed to sys_open().
1486 static int __open_namei_create(struct nameidata *nd, struct path *path,
1487 int open_flag, int mode)
1490 struct dentry *dir = nd->path.dentry;
1492 if (!IS_POSIXACL(dir->d_inode))
1493 mode &= ~current_umask();
1494 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1497 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1499 mutex_unlock(&dir->d_inode->i_mutex);
1500 dput(nd->path.dentry);
1501 nd->path.dentry = path->dentry;
1504 /* Don't check for write permission, don't truncate */
1505 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
1509 * Note that while the flag value (low two bits) for sys_open means:
1514 * it is changed into
1515 * 00 - no permissions needed
1516 * 01 - read-permission
1517 * 10 - write-permission
1519 * for the internal routines (ie open_namei()/follow_link() etc)
1520 * This is more logical, and also allows the 00 "no perm needed"
1521 * to be used for symlinks (where the permissions are checked
1525 static inline int open_to_namei_flags(int flag)
1527 if ((flag+1) & O_ACCMODE)
1532 static int open_will_truncate(int flag, struct inode *inode)
1535 * We'll never write to the fs underlying
1538 if (special_file(inode->i_mode))
1540 return (flag & O_TRUNC);
1543 static struct file *finish_open(struct nameidata *nd,
1544 int open_flag, int acc_mode)
1550 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
1551 if (will_truncate) {
1552 error = mnt_want_write(nd->path.mnt);
1556 error = may_open(&nd->path, acc_mode, open_flag);
1559 mnt_drop_write(nd->path.mnt);
1562 filp = nameidata_to_filp(nd);
1563 if (!IS_ERR(filp)) {
1564 error = ima_file_check(filp, acc_mode);
1567 filp = ERR_PTR(error);
1570 if (!IS_ERR(filp)) {
1571 if (will_truncate) {
1572 error = handle_truncate(&nd->path);
1575 filp = ERR_PTR(error);
1580 * It is now safe to drop the mnt write
1581 * because the filp has had a write taken
1585 mnt_drop_write(nd->path.mnt);
1589 if (!IS_ERR(nd->intent.open.file))
1590 release_open_intent(nd);
1591 path_put(&nd->path);
1592 return ERR_PTR(error);
1595 static struct file *do_last(struct nameidata *nd, struct path *path,
1596 int open_flag, int acc_mode,
1597 int mode, const char *pathname)
1599 struct dentry *dir = nd->path.dentry;
1601 int error = -EISDIR;
1603 switch (nd->last_type) {
1606 dir = nd->path.dentry;
1608 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1609 if (!dir->d_op->d_revalidate(dir, nd)) {
1616 if (open_flag & O_CREAT)
1620 audit_inode(pathname, dir);
1624 /* trailing slashes? */
1625 if (nd->last.name[nd->last.len]) {
1626 if (open_flag & O_CREAT)
1628 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
1631 /* just plain open? */
1632 if (!(open_flag & O_CREAT)) {
1633 error = do_lookup(nd, &nd->last, path);
1637 if (!path->dentry->d_inode)
1639 if (path->dentry->d_inode->i_op->follow_link)
1642 if (nd->flags & LOOKUP_DIRECTORY) {
1643 if (!path->dentry->d_inode->i_op->lookup)
1646 path_to_nameidata(path, nd);
1647 audit_inode(pathname, nd->path.dentry);
1651 /* OK, it's O_CREAT */
1652 mutex_lock(&dir->d_inode->i_mutex);
1654 path->dentry = lookup_hash(nd);
1655 path->mnt = nd->path.mnt;
1657 error = PTR_ERR(path->dentry);
1658 if (IS_ERR(path->dentry)) {
1659 mutex_unlock(&dir->d_inode->i_mutex);
1663 if (IS_ERR(nd->intent.open.file)) {
1664 error = PTR_ERR(nd->intent.open.file);
1665 goto exit_mutex_unlock;
1668 /* Negative dentry, just create the file */
1669 if (!path->dentry->d_inode) {
1671 * This write is needed to ensure that a
1672 * ro->rw transition does not occur between
1673 * the time when the file is created and when
1674 * a permanent write count is taken through
1675 * the 'struct file' in nameidata_to_filp().
1677 error = mnt_want_write(nd->path.mnt);
1679 goto exit_mutex_unlock;
1680 error = __open_namei_create(nd, path, open_flag, mode);
1682 mnt_drop_write(nd->path.mnt);
1685 filp = nameidata_to_filp(nd);
1686 mnt_drop_write(nd->path.mnt);
1687 if (!IS_ERR(filp)) {
1688 error = ima_file_check(filp, acc_mode);
1691 filp = ERR_PTR(error);
1698 * It already exists.
1700 mutex_unlock(&dir->d_inode->i_mutex);
1701 audit_inode(pathname, path->dentry);
1704 if (open_flag & O_EXCL)
1707 if (__follow_mount(path)) {
1709 if (open_flag & O_NOFOLLOW)
1714 if (!path->dentry->d_inode)
1717 if (path->dentry->d_inode->i_op->follow_link)
1720 path_to_nameidata(path, nd);
1722 if (S_ISDIR(path->dentry->d_inode->i_mode))
1725 filp = finish_open(nd, open_flag, acc_mode);
1729 mutex_unlock(&dir->d_inode->i_mutex);
1731 path_put_conditional(path, nd);
1733 if (!IS_ERR(nd->intent.open.file))
1734 release_open_intent(nd);
1735 path_put(&nd->path);
1736 return ERR_PTR(error);
1740 * Note that the low bits of the passed in "open_flag"
1741 * are not the same as in the local variable "flag". See
1742 * open_to_namei_flags() for more details.
1744 struct file *do_filp_open(int dfd, const char *pathname,
1745 int open_flag, int mode, int acc_mode)
1748 struct nameidata nd;
1752 int flag = open_to_namei_flags(open_flag);
1753 int force_reval = 0;
1755 if (!(open_flag & O_CREAT))
1759 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1760 * check for O_DSYNC if the need any syncing at all we enforce it's
1761 * always set instead of having to deal with possibly weird behaviour
1762 * for malicious applications setting only __O_SYNC.
1764 if (open_flag & __O_SYNC)
1765 open_flag |= O_DSYNC;
1768 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1770 /* O_TRUNC implies we need access checks for write permissions */
1771 if (open_flag & O_TRUNC)
1772 acc_mode |= MAY_WRITE;
1774 /* Allow the LSM permission hook to distinguish append
1775 access from general write access. */
1776 if (open_flag & O_APPEND)
1777 acc_mode |= MAY_APPEND;
1779 /* find the parent */
1781 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1783 return ERR_PTR(error);
1785 nd.flags |= LOOKUP_REVAL;
1787 current->total_link_count = 0;
1788 error = link_path_walk(pathname, &nd);
1790 filp = ERR_PTR(error);
1793 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
1794 audit_inode(pathname, nd.path.dentry);
1797 * We have the parent and last component.
1801 filp = get_empty_filp();
1804 nd.intent.open.file = filp;
1805 filp->f_flags = open_flag;
1806 nd.intent.open.flags = flag;
1807 nd.intent.open.create_mode = mode;
1808 nd.flags &= ~LOOKUP_PARENT;
1809 nd.flags |= LOOKUP_OPEN;
1810 if (open_flag & O_CREAT) {
1811 nd.flags |= LOOKUP_CREATE;
1812 if (open_flag & O_EXCL)
1813 nd.flags |= LOOKUP_EXCL;
1815 if (open_flag & O_DIRECTORY)
1816 nd.flags |= LOOKUP_DIRECTORY;
1817 if (!(open_flag & O_NOFOLLOW))
1818 nd.flags |= LOOKUP_FOLLOW;
1819 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1820 while (unlikely(!filp)) { /* trailing symlink */
1822 struct inode *inode = path.dentry->d_inode;
1825 /* S_ISDIR part is a temporary automount kludge */
1826 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
1831 * This is subtle. Instead of calling do_follow_link() we do
1832 * the thing by hands. The reason is that this way we have zero
1833 * link_count and path_walk() (called from ->follow_link)
1834 * honoring LOOKUP_PARENT. After that we have the parent and
1835 * last component, i.e. we are in the same situation as after
1836 * the first path_walk(). Well, almost - if the last component
1837 * is normal we get its copy stored in nd->last.name and we will
1838 * have to putname() it when we are done. Procfs-like symlinks
1839 * just set LAST_BIND.
1841 nd.flags |= LOOKUP_PARENT;
1842 error = security_inode_follow_link(path.dentry, &nd);
1845 error = __do_follow_link(&path, &nd, &cookie);
1846 if (unlikely(error)) {
1847 /* nd.path had been dropped */
1848 if (!IS_ERR(cookie) && inode->i_op->put_link)
1849 inode->i_op->put_link(path.dentry, &nd, cookie);
1851 release_open_intent(&nd);
1852 filp = ERR_PTR(error);
1856 nd.flags &= ~LOOKUP_PARENT;
1857 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1858 if (inode->i_op->put_link)
1859 inode->i_op->put_link(holder.dentry, &nd, cookie);
1865 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1872 path_put_conditional(&path, &nd);
1873 if (!IS_ERR(nd.intent.open.file))
1874 release_open_intent(&nd);
1877 filp = ERR_PTR(error);
1882 * filp_open - open file and return file pointer
1884 * @filename: path to open
1885 * @flags: open flags as per the open(2) second argument
1886 * @mode: mode for the new file if O_CREAT is set, else ignored
1888 * This is the helper to open a file from kernelspace if you really
1889 * have to. But in generally you should not do this, so please move
1890 * along, nothing to see here..
1892 struct file *filp_open(const char *filename, int flags, int mode)
1894 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1896 EXPORT_SYMBOL(filp_open);
1899 * lookup_create - lookup a dentry, creating it if it doesn't exist
1900 * @nd: nameidata info
1901 * @is_dir: directory flag
1903 * Simple function to lookup and return a dentry and create it
1904 * if it doesn't exist. Is SMP-safe.
1906 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1908 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1910 struct dentry *dentry = ERR_PTR(-EEXIST);
1912 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1914 * Yucky last component or no last component at all?
1915 * (foo/., foo/.., /////)
1917 if (nd->last_type != LAST_NORM)
1919 nd->flags &= ~LOOKUP_PARENT;
1920 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1921 nd->intent.open.flags = O_EXCL;
1924 * Do the final lookup.
1926 dentry = lookup_hash(nd);
1930 if (dentry->d_inode)
1933 * Special case - lookup gave negative, but... we had foo/bar/
1934 * From the vfs_mknod() POV we just have a negative dentry -
1935 * all is fine. Let's be bastards - you had / on the end, you've
1936 * been asking for (non-existent) directory. -ENOENT for you.
1938 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1940 dentry = ERR_PTR(-ENOENT);
1945 dentry = ERR_PTR(-EEXIST);
1949 EXPORT_SYMBOL_GPL(lookup_create);
1951 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1953 int error = may_create(dir, dentry);
1958 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1961 if (!dir->i_op->mknod)
1964 error = devcgroup_inode_mknod(mode, dev);
1968 error = security_inode_mknod(dir, dentry, mode, dev);
1972 error = dir->i_op->mknod(dir, dentry, mode, dev);
1974 fsnotify_create(dir, dentry);
1978 static int may_mknod(mode_t mode)
1980 switch (mode & S_IFMT) {
1986 case 0: /* zero mode translates to S_IFREG */
1995 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2000 struct dentry *dentry;
2001 struct nameidata nd;
2006 error = user_path_parent(dfd, filename, &nd, &tmp);
2010 dentry = lookup_create(&nd, 0);
2011 if (IS_ERR(dentry)) {
2012 error = PTR_ERR(dentry);
2015 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2016 mode &= ~current_umask();
2017 error = may_mknod(mode);
2020 error = mnt_want_write(nd.path.mnt);
2023 error = security_path_mknod(&nd.path, dentry, mode, dev);
2025 goto out_drop_write;
2026 switch (mode & S_IFMT) {
2027 case 0: case S_IFREG:
2028 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2030 case S_IFCHR: case S_IFBLK:
2031 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2032 new_decode_dev(dev));
2034 case S_IFIFO: case S_IFSOCK:
2035 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2039 mnt_drop_write(nd.path.mnt);
2043 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2050 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2052 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2055 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2057 int error = may_create(dir, dentry);
2062 if (!dir->i_op->mkdir)
2065 mode &= (S_IRWXUGO|S_ISVTX);
2066 error = security_inode_mkdir(dir, dentry, mode);
2070 error = dir->i_op->mkdir(dir, dentry, mode);
2072 fsnotify_mkdir(dir, dentry);
2076 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2080 struct dentry *dentry;
2081 struct nameidata nd;
2083 error = user_path_parent(dfd, pathname, &nd, &tmp);
2087 dentry = lookup_create(&nd, 1);
2088 error = PTR_ERR(dentry);
2092 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2093 mode &= ~current_umask();
2094 error = mnt_want_write(nd.path.mnt);
2097 error = security_path_mkdir(&nd.path, dentry, mode);
2099 goto out_drop_write;
2100 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2102 mnt_drop_write(nd.path.mnt);
2106 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2113 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2115 return sys_mkdirat(AT_FDCWD, pathname, mode);
2119 * We try to drop the dentry early: we should have
2120 * a usage count of 2 if we're the only user of this
2121 * dentry, and if that is true (possibly after pruning
2122 * the dcache), then we drop the dentry now.
2124 * A low-level filesystem can, if it choses, legally
2127 * if (!d_unhashed(dentry))
2130 * if it cannot handle the case of removing a directory
2131 * that is still in use by something else..
2133 void dentry_unhash(struct dentry *dentry)
2136 shrink_dcache_parent(dentry);
2137 spin_lock(&dcache_lock);
2138 spin_lock(&dentry->d_lock);
2139 if (atomic_read(&dentry->d_count) == 2)
2141 spin_unlock(&dentry->d_lock);
2142 spin_unlock(&dcache_lock);
2145 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2147 int error = may_delete(dir, dentry, 1);
2152 if (!dir->i_op->rmdir)
2155 mutex_lock(&dentry->d_inode->i_mutex);
2156 dentry_unhash(dentry);
2157 if (d_mountpoint(dentry))
2160 error = security_inode_rmdir(dir, dentry);
2162 error = dir->i_op->rmdir(dir, dentry);
2164 dentry->d_inode->i_flags |= S_DEAD;
2169 mutex_unlock(&dentry->d_inode->i_mutex);
2178 static long do_rmdir(int dfd, const char __user *pathname)
2182 struct dentry *dentry;
2183 struct nameidata nd;
2185 error = user_path_parent(dfd, pathname, &nd, &name);
2189 switch(nd.last_type) {
2201 nd.flags &= ~LOOKUP_PARENT;
2203 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2204 dentry = lookup_hash(&nd);
2205 error = PTR_ERR(dentry);
2208 error = mnt_want_write(nd.path.mnt);
2211 error = security_path_rmdir(&nd.path, dentry);
2214 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2216 mnt_drop_write(nd.path.mnt);
2220 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2227 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2229 return do_rmdir(AT_FDCWD, pathname);
2232 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2234 int error = may_delete(dir, dentry, 0);
2239 if (!dir->i_op->unlink)
2242 mutex_lock(&dentry->d_inode->i_mutex);
2243 if (d_mountpoint(dentry))
2246 error = security_inode_unlink(dir, dentry);
2248 error = dir->i_op->unlink(dir, dentry);
2253 mutex_unlock(&dentry->d_inode->i_mutex);
2255 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2256 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2257 fsnotify_link_count(dentry->d_inode);
2265 * Make sure that the actual truncation of the file will occur outside its
2266 * directory's i_mutex. Truncate can take a long time if there is a lot of
2267 * writeout happening, and we don't want to prevent access to the directory
2268 * while waiting on the I/O.
2270 static long do_unlinkat(int dfd, const char __user *pathname)
2274 struct dentry *dentry;
2275 struct nameidata nd;
2276 struct inode *inode = NULL;
2278 error = user_path_parent(dfd, pathname, &nd, &name);
2283 if (nd.last_type != LAST_NORM)
2286 nd.flags &= ~LOOKUP_PARENT;
2288 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2289 dentry = lookup_hash(&nd);
2290 error = PTR_ERR(dentry);
2291 if (!IS_ERR(dentry)) {
2292 /* Why not before? Because we want correct error value */
2293 if (nd.last.name[nd.last.len])
2295 inode = dentry->d_inode;
2297 atomic_inc(&inode->i_count);
2298 error = mnt_want_write(nd.path.mnt);
2301 error = security_path_unlink(&nd.path, dentry);
2304 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2306 mnt_drop_write(nd.path.mnt);
2310 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2312 iput(inode); /* truncate the inode here */
2319 error = !dentry->d_inode ? -ENOENT :
2320 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2324 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2326 if ((flag & ~AT_REMOVEDIR) != 0)
2329 if (flag & AT_REMOVEDIR)
2330 return do_rmdir(dfd, pathname);
2332 return do_unlinkat(dfd, pathname);
2335 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2337 return do_unlinkat(AT_FDCWD, pathname);
2340 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2342 int error = may_create(dir, dentry);
2347 if (!dir->i_op->symlink)
2350 error = security_inode_symlink(dir, dentry, oldname);
2354 error = dir->i_op->symlink(dir, dentry, oldname);
2356 fsnotify_create(dir, dentry);
2360 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2361 int, newdfd, const char __user *, newname)
2366 struct dentry *dentry;
2367 struct nameidata nd;
2369 from = getname(oldname);
2371 return PTR_ERR(from);
2373 error = user_path_parent(newdfd, newname, &nd, &to);
2377 dentry = lookup_create(&nd, 0);
2378 error = PTR_ERR(dentry);
2382 error = mnt_want_write(nd.path.mnt);
2385 error = security_path_symlink(&nd.path, dentry, from);
2387 goto out_drop_write;
2388 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2390 mnt_drop_write(nd.path.mnt);
2394 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2402 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2404 return sys_symlinkat(oldname, AT_FDCWD, newname);
2407 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2409 struct inode *inode = old_dentry->d_inode;
2415 error = may_create(dir, new_dentry);
2419 if (dir->i_sb != inode->i_sb)
2423 * A link to an append-only or immutable file cannot be created.
2425 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2427 if (!dir->i_op->link)
2429 if (S_ISDIR(inode->i_mode))
2432 error = security_inode_link(old_dentry, dir, new_dentry);
2436 mutex_lock(&inode->i_mutex);
2437 error = dir->i_op->link(old_dentry, dir, new_dentry);
2438 mutex_unlock(&inode->i_mutex);
2440 fsnotify_link(dir, inode, new_dentry);
2445 * Hardlinks are often used in delicate situations. We avoid
2446 * security-related surprises by not following symlinks on the
2449 * We don't follow them on the oldname either to be compatible
2450 * with linux 2.0, and to avoid hard-linking to directories
2451 * and other special files. --ADM
2453 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2454 int, newdfd, const char __user *, newname, int, flags)
2456 struct dentry *new_dentry;
2457 struct nameidata nd;
2458 struct path old_path;
2462 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2465 error = user_path_at(olddfd, oldname,
2466 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2471 error = user_path_parent(newdfd, newname, &nd, &to);
2475 if (old_path.mnt != nd.path.mnt)
2477 new_dentry = lookup_create(&nd, 0);
2478 error = PTR_ERR(new_dentry);
2479 if (IS_ERR(new_dentry))
2481 error = mnt_want_write(nd.path.mnt);
2484 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2486 goto out_drop_write;
2487 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2489 mnt_drop_write(nd.path.mnt);
2493 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2498 path_put(&old_path);
2503 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2505 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2509 * The worst of all namespace operations - renaming directory. "Perverted"
2510 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2512 * a) we can get into loop creation. Check is done in is_subdir().
2513 * b) race potential - two innocent renames can create a loop together.
2514 * That's where 4.4 screws up. Current fix: serialization on
2515 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2517 * c) we have to lock _three_ objects - parents and victim (if it exists).
2518 * And that - after we got ->i_mutex on parents (until then we don't know
2519 * whether the target exists). Solution: try to be smart with locking
2520 * order for inodes. We rely on the fact that tree topology may change
2521 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2522 * move will be locked. Thus we can rank directories by the tree
2523 * (ancestors first) and rank all non-directories after them.
2524 * That works since everybody except rename does "lock parent, lookup,
2525 * lock child" and rename is under ->s_vfs_rename_mutex.
2526 * HOWEVER, it relies on the assumption that any object with ->lookup()
2527 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2528 * we'd better make sure that there's no link(2) for them.
2529 * d) some filesystems don't support opened-but-unlinked directories,
2530 * either because of layout or because they are not ready to deal with
2531 * all cases correctly. The latter will be fixed (taking this sort of
2532 * stuff into VFS), but the former is not going away. Solution: the same
2533 * trick as in rmdir().
2534 * e) conversion from fhandle to dentry may come in the wrong moment - when
2535 * we are removing the target. Solution: we will have to grab ->i_mutex
2536 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2537 * ->i_mutex on parents, which works but leads to some truly excessive
2540 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2541 struct inode *new_dir, struct dentry *new_dentry)
2544 struct inode *target;
2547 * If we are going to change the parent - check write permissions,
2548 * we'll need to flip '..'.
2550 if (new_dir != old_dir) {
2551 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2556 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2560 target = new_dentry->d_inode;
2562 mutex_lock(&target->i_mutex);
2563 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2567 dentry_unhash(new_dentry);
2568 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2572 target->i_flags |= S_DEAD;
2573 dont_mount(new_dentry);
2575 mutex_unlock(&target->i_mutex);
2576 if (d_unhashed(new_dentry))
2577 d_rehash(new_dentry);
2581 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2582 d_move(old_dentry,new_dentry);
2586 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2587 struct inode *new_dir, struct dentry *new_dentry)
2589 struct inode *target;
2592 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2597 target = new_dentry->d_inode;
2599 mutex_lock(&target->i_mutex);
2600 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2603 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2606 dont_mount(new_dentry);
2607 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2608 d_move(old_dentry, new_dentry);
2611 mutex_unlock(&target->i_mutex);
2616 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2617 struct inode *new_dir, struct dentry *new_dentry)
2620 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2621 const unsigned char *old_name;
2623 if (old_dentry->d_inode == new_dentry->d_inode)
2626 error = may_delete(old_dir, old_dentry, is_dir);
2630 if (!new_dentry->d_inode)
2631 error = may_create(new_dir, new_dentry);
2633 error = may_delete(new_dir, new_dentry, is_dir);
2637 if (!old_dir->i_op->rename)
2640 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2643 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2645 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2647 fsnotify_move(old_dir, new_dir, old_name, is_dir,
2648 new_dentry->d_inode, old_dentry);
2649 fsnotify_oldname_free(old_name);
2654 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2655 int, newdfd, const char __user *, newname)
2657 struct dentry *old_dir, *new_dir;
2658 struct dentry *old_dentry, *new_dentry;
2659 struct dentry *trap;
2660 struct nameidata oldnd, newnd;
2665 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2669 error = user_path_parent(newdfd, newname, &newnd, &to);
2674 if (oldnd.path.mnt != newnd.path.mnt)
2677 old_dir = oldnd.path.dentry;
2679 if (oldnd.last_type != LAST_NORM)
2682 new_dir = newnd.path.dentry;
2683 if (newnd.last_type != LAST_NORM)
2686 oldnd.flags &= ~LOOKUP_PARENT;
2687 newnd.flags &= ~LOOKUP_PARENT;
2688 newnd.flags |= LOOKUP_RENAME_TARGET;
2690 trap = lock_rename(new_dir, old_dir);
2692 old_dentry = lookup_hash(&oldnd);
2693 error = PTR_ERR(old_dentry);
2694 if (IS_ERR(old_dentry))
2696 /* source must exist */
2698 if (!old_dentry->d_inode)
2700 /* unless the source is a directory trailing slashes give -ENOTDIR */
2701 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2703 if (oldnd.last.name[oldnd.last.len])
2705 if (newnd.last.name[newnd.last.len])
2708 /* source should not be ancestor of target */
2710 if (old_dentry == trap)
2712 new_dentry = lookup_hash(&newnd);
2713 error = PTR_ERR(new_dentry);
2714 if (IS_ERR(new_dentry))
2716 /* target should not be an ancestor of source */
2718 if (new_dentry == trap)
2721 error = mnt_want_write(oldnd.path.mnt);
2724 error = security_path_rename(&oldnd.path, old_dentry,
2725 &newnd.path, new_dentry);
2728 error = vfs_rename(old_dir->d_inode, old_dentry,
2729 new_dir->d_inode, new_dentry);
2731 mnt_drop_write(oldnd.path.mnt);
2737 unlock_rename(new_dir, old_dir);
2739 path_put(&newnd.path);
2742 path_put(&oldnd.path);
2748 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2750 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2753 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2757 len = PTR_ERR(link);
2762 if (len > (unsigned) buflen)
2764 if (copy_to_user(buffer, link, len))
2771 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2772 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2773 * using) it for any given inode is up to filesystem.
2775 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2777 struct nameidata nd;
2782 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2784 return PTR_ERR(cookie);
2786 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2787 if (dentry->d_inode->i_op->put_link)
2788 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2792 int vfs_follow_link(struct nameidata *nd, const char *link)
2794 return __vfs_follow_link(nd, link);
2797 /* get the link contents into pagecache */
2798 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2802 struct address_space *mapping = dentry->d_inode->i_mapping;
2803 page = read_mapping_page(mapping, 0, NULL);
2808 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2812 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2814 struct page *page = NULL;
2815 char *s = page_getlink(dentry, &page);
2816 int res = vfs_readlink(dentry,buffer,buflen,s);
2819 page_cache_release(page);
2824 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2826 struct page *page = NULL;
2827 nd_set_link(nd, page_getlink(dentry, &page));
2831 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2833 struct page *page = cookie;
2837 page_cache_release(page);
2842 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2844 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2846 struct address_space *mapping = inode->i_mapping;
2851 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2853 flags |= AOP_FLAG_NOFS;
2856 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2857 flags, &page, &fsdata);
2861 kaddr = kmap_atomic(page, KM_USER0);
2862 memcpy(kaddr, symname, len-1);
2863 kunmap_atomic(kaddr, KM_USER0);
2865 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2872 mark_inode_dirty(inode);
2878 int page_symlink(struct inode *inode, const char *symname, int len)
2880 return __page_symlink(inode, symname, len,
2881 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2884 const struct inode_operations page_symlink_inode_operations = {
2885 .readlink = generic_readlink,
2886 .follow_link = page_follow_link_light,
2887 .put_link = page_put_link,
2890 EXPORT_SYMBOL(user_path_at);
2891 EXPORT_SYMBOL(follow_down);
2892 EXPORT_SYMBOL(follow_up);
2893 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2894 EXPORT_SYMBOL(getname);
2895 EXPORT_SYMBOL(lock_rename);
2896 EXPORT_SYMBOL(lookup_one_len);
2897 EXPORT_SYMBOL(page_follow_link_light);
2898 EXPORT_SYMBOL(page_put_link);
2899 EXPORT_SYMBOL(page_readlink);
2900 EXPORT_SYMBOL(__page_symlink);
2901 EXPORT_SYMBOL(page_symlink);
2902 EXPORT_SYMBOL(page_symlink_inode_operations);
2903 EXPORT_SYMBOL(path_lookup);
2904 EXPORT_SYMBOL(kern_path);
2905 EXPORT_SYMBOL(vfs_path_lookup);
2906 EXPORT_SYMBOL(inode_permission);
2907 EXPORT_SYMBOL(file_permission);
2908 EXPORT_SYMBOL(unlock_rename);
2909 EXPORT_SYMBOL(vfs_create);
2910 EXPORT_SYMBOL(vfs_follow_link);
2911 EXPORT_SYMBOL(vfs_link);
2912 EXPORT_SYMBOL(vfs_mkdir);
2913 EXPORT_SYMBOL(vfs_mknod);
2914 EXPORT_SYMBOL(generic_permission);
2915 EXPORT_SYMBOL(vfs_readlink);
2916 EXPORT_SYMBOL(vfs_rename);
2917 EXPORT_SYMBOL(vfs_rmdir);
2918 EXPORT_SYMBOL(vfs_symlink);
2919 EXPORT_SYMBOL(vfs_unlink);
2920 EXPORT_SYMBOL(dentry_unhash);
2921 EXPORT_SYMBOL(generic_readlink);