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)
257 if (mask & MAY_WRITE) {
258 umode_t mode = inode->i_mode;
261 * Nobody gets write access to a read-only fs.
263 if (IS_RDONLY(inode) &&
264 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
268 * Nobody gets write access to an immutable file.
270 if (IS_IMMUTABLE(inode))
274 if (!IS_WITHAPPEND(inode))
275 submask &= ~MAY_APPEND;
277 if (inode->i_op->permission)
278 retval = inode->i_op->permission(inode, submask);
280 retval = generic_permission(inode, mask, inode->i_op->check_acl);
285 retval = devcgroup_inode_permission(inode, mask);
289 return security_inode_permission(inode,
290 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
294 * file_permission - check for additional access rights to a given file
295 * @file: file to check access rights for
296 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 * Used to check for read/write/execute permissions on an already opened
302 * Do not use this function in new code. All access checks should
303 * be done using inode_permission().
305 int file_permission(struct file *file, int mask)
307 return inode_permission(file->f_path.dentry->d_inode, mask);
311 * get_write_access() gets write permission for a file.
312 * put_write_access() releases this write permission.
313 * This is used for regular files.
314 * We cannot support write (and maybe mmap read-write shared) accesses and
315 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
316 * can have the following values:
317 * 0: no writers, no VM_DENYWRITE mappings
318 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
319 * > 0: (i_writecount) users are writing to the file.
321 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
322 * except for the cases where we don't hold i_writecount yet. Then we need to
323 * use {get,deny}_write_access() - these functions check the sign and refuse
324 * to do the change if sign is wrong. Exclusion between them is provided by
325 * the inode->i_lock spinlock.
328 int get_write_access(struct inode * inode)
330 spin_lock(&inode->i_lock);
331 if (atomic_read(&inode->i_writecount) < 0) {
332 spin_unlock(&inode->i_lock);
335 atomic_inc(&inode->i_writecount);
336 spin_unlock(&inode->i_lock);
341 int deny_write_access(struct file * file)
343 struct inode *inode = file->f_path.dentry->d_inode;
345 spin_lock(&inode->i_lock);
346 if (atomic_read(&inode->i_writecount) > 0) {
347 spin_unlock(&inode->i_lock);
350 atomic_dec(&inode->i_writecount);
351 spin_unlock(&inode->i_lock);
357 * path_get - get a reference to a path
358 * @path: path to get the reference to
360 * Given a path increment the reference count to the dentry and the vfsmount.
362 void path_get(struct path *path)
367 EXPORT_SYMBOL(path_get);
370 * path_put - put a reference to a path
371 * @path: path to put the reference to
373 * Given a path decrement the reference count to the dentry and the vfsmount.
375 void path_put(struct path *path)
380 EXPORT_SYMBOL(path_put);
383 * release_open_intent - free up open intent resources
384 * @nd: pointer to nameidata
386 void release_open_intent(struct nameidata *nd)
388 if (nd->intent.open.file->f_path.dentry == NULL)
389 put_filp(nd->intent.open.file);
391 fput(nd->intent.open.file);
393 EXPORT_SYMBOL_GPL(release_open_intent);
395 static inline struct dentry *
396 do_revalidate(struct dentry *dentry, struct nameidata *nd)
398 int status = dentry->d_op->d_revalidate(dentry, nd);
399 if (unlikely(status <= 0)) {
401 * The dentry failed validation.
402 * If d_revalidate returned 0 attempt to invalidate
403 * the dentry otherwise d_revalidate is asking us
404 * to return a fail status.
407 if (!d_invalidate(dentry)) {
413 dentry = ERR_PTR(status);
420 * force_reval_path - force revalidation of a dentry
422 * In some situations the path walking code will trust dentries without
423 * revalidating them. This causes problems for filesystems that depend on
424 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
425 * (which indicates that it's possible for the dentry to go stale), force
426 * a d_revalidate call before proceeding.
428 * Returns 0 if the revalidation was successful. If the revalidation fails,
429 * either return the error returned by d_revalidate or -ESTALE if the
430 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
431 * invalidate the dentry. It's up to the caller to handle putting references
432 * to the path if necessary.
435 force_reval_path(struct path *path, struct nameidata *nd)
438 struct dentry *dentry = path->dentry;
441 * only check on filesystems where it's possible for the dentry to
442 * become stale. It's assumed that if this flag is set then the
443 * d_revalidate op will also be defined.
445 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
448 status = dentry->d_op->d_revalidate(dentry, nd);
453 d_invalidate(dentry);
460 * Short-cut version of permission(), for calling on directories
461 * during pathname resolution. Combines parts of permission()
462 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
464 * If appropriate, check DAC only. If not appropriate, or
465 * short-cut DAC fails, then call ->permission() to do more
466 * complete permission check.
468 static int exec_permission(struct inode *inode)
472 if (inode->i_op->permission) {
473 ret = inode->i_op->permission(inode, MAY_EXEC);
478 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
482 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
487 return security_inode_permission(inode, MAY_EXEC);
490 static __always_inline void set_root(struct nameidata *nd)
493 struct fs_struct *fs = current->fs;
494 read_lock(&fs->lock);
497 read_unlock(&fs->lock);
501 static int link_path_walk(const char *, struct nameidata *);
503 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
515 return link_path_walk(link, nd);
518 return PTR_ERR(link);
521 static void path_put_conditional(struct path *path, struct nameidata *nd)
524 if (path->mnt != nd->path.mnt)
528 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
530 dput(nd->path.dentry);
531 if (nd->path.mnt != path->mnt) {
532 mntput(nd->path.mnt);
533 nd->path.mnt = path->mnt;
535 nd->path.dentry = path->dentry;
538 static __always_inline int
539 __do_follow_link(struct path *path, struct nameidata *nd, void **p)
542 struct dentry *dentry = path->dentry;
544 touch_atime(path->mnt, dentry);
545 nd_set_link(nd, NULL);
547 if (path->mnt != nd->path.mnt) {
548 path_to_nameidata(path, nd);
552 nd->last_type = LAST_BIND;
553 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
556 char *s = nd_get_link(nd);
559 error = __vfs_follow_link(nd, s);
560 else if (nd->last_type == LAST_BIND) {
561 error = force_reval_path(&nd->path, nd);
570 * This limits recursive symlink follows to 8, while
571 * limiting consecutive symlinks to 40.
573 * Without that kind of total limit, nasty chains of consecutive
574 * symlinks can cause almost arbitrarily long lookups.
576 static inline int do_follow_link(struct path *path, struct nameidata *nd)
580 if (current->link_count >= MAX_NESTED_LINKS)
582 if (current->total_link_count >= 40)
584 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
586 err = security_inode_follow_link(path->dentry, nd);
589 current->link_count++;
590 current->total_link_count++;
592 err = __do_follow_link(path, nd, &cookie);
593 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
594 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
596 current->link_count--;
600 path_put_conditional(path, nd);
605 int follow_up(struct path *path)
607 struct vfsmount *parent;
608 struct dentry *mountpoint;
609 spin_lock(&vfsmount_lock);
610 parent = path->mnt->mnt_parent;
611 if (parent == path->mnt) {
612 spin_unlock(&vfsmount_lock);
616 mountpoint = dget(path->mnt->mnt_mountpoint);
617 spin_unlock(&vfsmount_lock);
619 path->dentry = mountpoint;
625 /* no need for dcache_lock, as serialization is taken care in
628 static int __follow_mount(struct path *path)
631 while (d_mountpoint(path->dentry)) {
632 struct vfsmount *mounted = lookup_mnt(path);
639 path->dentry = dget(mounted->mnt_root);
645 static void follow_mount(struct path *path)
647 while (d_mountpoint(path->dentry)) {
648 struct vfsmount *mounted = lookup_mnt(path);
654 path->dentry = dget(mounted->mnt_root);
658 /* no need for dcache_lock, as serialization is taken care in
661 int follow_down(struct path *path)
663 struct vfsmount *mounted;
665 mounted = lookup_mnt(path);
670 path->dentry = dget(mounted->mnt_root);
676 static __always_inline void follow_dotdot(struct nameidata *nd)
681 struct dentry *old = nd->path.dentry;
683 if (nd->path.dentry == nd->root.dentry &&
684 nd->path.mnt == nd->root.mnt) {
687 if (nd->path.dentry != nd->path.mnt->mnt_root) {
688 /* rare case of legitimate dget_parent()... */
689 nd->path.dentry = dget_parent(nd->path.dentry);
693 if (!follow_up(&nd->path))
696 follow_mount(&nd->path);
700 * It's more convoluted than I'd like it to be, but... it's still fairly
701 * small and for now I'd prefer to have fast path as straight as possible.
702 * It _is_ time-critical.
704 static int do_lookup(struct nameidata *nd, struct qstr *name,
707 struct vfsmount *mnt = nd->path.mnt;
708 struct dentry *dentry, *parent;
711 * See if the low-level filesystem might want
712 * to use its own hash..
714 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
715 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
720 dentry = __d_lookup(nd->path.dentry, name);
723 if (dentry->d_op && dentry->d_op->d_revalidate)
724 goto need_revalidate;
727 path->dentry = dentry;
728 __follow_mount(path);
732 parent = nd->path.dentry;
733 dir = parent->d_inode;
735 mutex_lock(&dir->i_mutex);
737 * First re-do the cached lookup just in case it was created
738 * while we waited for the directory semaphore..
740 * FIXME! This could use version numbering or similar to
741 * avoid unnecessary cache lookups.
743 * The "dcache_lock" is purely to protect the RCU list walker
744 * from concurrent renames at this point (we mustn't get false
745 * negatives from the RCU list walk here, unlike the optimistic
748 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
750 dentry = d_lookup(parent, name);
754 /* Don't create child dentry for a dead directory. */
755 dentry = ERR_PTR(-ENOENT);
759 new = d_alloc(parent, name);
760 dentry = ERR_PTR(-ENOMEM);
762 dentry = dir->i_op->lookup(dir, new, nd);
769 mutex_unlock(&dir->i_mutex);
776 * Uhhuh! Nasty case: the cache was re-populated while
777 * we waited on the semaphore. Need to revalidate.
779 mutex_unlock(&dir->i_mutex);
780 if (dentry->d_op && dentry->d_op->d_revalidate) {
781 dentry = do_revalidate(dentry, nd);
783 dentry = ERR_PTR(-ENOENT);
790 dentry = do_revalidate(dentry, nd);
798 return PTR_ERR(dentry);
802 * This is a temporary kludge to deal with "automount" symlinks; proper
803 * solution is to trigger them on follow_mount(), so that do_lookup()
804 * would DTRT. To be killed before 2.6.34-final.
806 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
808 return inode && unlikely(inode->i_op->follow_link) &&
809 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
814 * This is the basic name resolution function, turning a pathname into
815 * the final dentry. We expect 'base' to be positive and a directory.
817 * Returns 0 and nd will have valid dentry and mnt on success.
818 * Returns error and drops reference to input namei data on failure.
820 static int link_path_walk(const char *name, struct nameidata *nd)
825 unsigned int lookup_flags = nd->flags;
832 inode = nd->path.dentry->d_inode;
834 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
836 /* At this point we know we have a real path component. */
842 nd->flags |= LOOKUP_CONTINUE;
843 err = exec_permission(inode);
848 c = *(const unsigned char *)name;
850 hash = init_name_hash();
853 hash = partial_name_hash(c, hash);
854 c = *(const unsigned char *)name;
855 } while (c && (c != '/'));
856 this.len = name - (const char *) this.name;
857 this.hash = end_name_hash(hash);
859 /* remove trailing slashes? */
862 while (*++name == '/');
864 goto last_with_slashes;
867 * "." and ".." are special - ".." especially so because it has
868 * to be able to know about the current root directory and
869 * parent relationships.
871 if (this.name[0] == '.') switch (this.len) {
875 if (this.name[1] != '.')
878 inode = nd->path.dentry->d_inode;
883 /* This does the actual lookups.. */
884 err = do_lookup(nd, &this, &next);
889 inode = next.dentry->d_inode;
893 if (inode->i_op->follow_link) {
894 err = do_follow_link(&next, nd);
898 inode = nd->path.dentry->d_inode;
902 path_to_nameidata(&next, nd);
904 if (!inode->i_op->lookup)
907 /* here ends the main loop */
910 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
912 /* Clear LOOKUP_CONTINUE iff it was previously unset */
913 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
914 if (lookup_flags & LOOKUP_PARENT)
916 if (this.name[0] == '.') switch (this.len) {
920 if (this.name[1] != '.')
923 inode = nd->path.dentry->d_inode;
928 err = do_lookup(nd, &this, &next);
931 inode = next.dentry->d_inode;
932 if (follow_on_final(inode, lookup_flags)) {
933 err = do_follow_link(&next, nd);
936 inode = nd->path.dentry->d_inode;
938 path_to_nameidata(&next, nd);
942 if (lookup_flags & LOOKUP_DIRECTORY) {
944 if (!inode->i_op->lookup)
950 nd->last_type = LAST_NORM;
951 if (this.name[0] != '.')
954 nd->last_type = LAST_DOT;
955 else if (this.len == 2 && this.name[1] == '.')
956 nd->last_type = LAST_DOTDOT;
961 * We bypassed the ordinary revalidation routines.
962 * We may need to check the cached dentry for staleness.
964 if (nd->path.dentry && nd->path.dentry->d_sb &&
965 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
967 /* Note: we do not d_invalidate() */
968 if (!nd->path.dentry->d_op->d_revalidate(
969 nd->path.dentry, nd))
975 path_put_conditional(&next, nd);
983 static int path_walk(const char *name, struct nameidata *nd)
985 struct path save = nd->path;
988 current->total_link_count = 0;
990 /* make sure the stuff we saved doesn't go away */
993 result = link_path_walk(name, nd);
994 if (result == -ESTALE) {
995 /* nd->path had been dropped */
996 current->total_link_count = 0;
999 nd->flags |= LOOKUP_REVAL;
1000 result = link_path_walk(name, nd);
1008 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1014 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1017 nd->root.mnt = NULL;
1021 nd->path = nd->root;
1022 path_get(&nd->root);
1023 } else if (dfd == AT_FDCWD) {
1024 struct fs_struct *fs = current->fs;
1025 read_lock(&fs->lock);
1028 read_unlock(&fs->lock);
1030 struct dentry *dentry;
1032 file = fget_light(dfd, &fput_needed);
1037 dentry = file->f_path.dentry;
1040 if (!S_ISDIR(dentry->d_inode->i_mode))
1043 retval = file_permission(file, MAY_EXEC);
1047 nd->path = file->f_path;
1048 path_get(&file->f_path);
1050 fput_light(file, fput_needed);
1055 fput_light(file, fput_needed);
1060 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1061 static int do_path_lookup(int dfd, const char *name,
1062 unsigned int flags, struct nameidata *nd)
1064 int retval = path_init(dfd, name, flags, nd);
1066 retval = path_walk(name, nd);
1067 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1068 nd->path.dentry->d_inode))
1069 audit_inode(name, nd->path.dentry);
1071 path_put(&nd->root);
1072 nd->root.mnt = NULL;
1077 int path_lookup(const char *name, unsigned int flags,
1078 struct nameidata *nd)
1080 return do_path_lookup(AT_FDCWD, name, flags, nd);
1083 int kern_path(const char *name, unsigned int flags, struct path *path)
1085 struct nameidata nd;
1086 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1093 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1094 * @dentry: pointer to dentry of the base directory
1095 * @mnt: pointer to vfs mount of the base directory
1096 * @name: pointer to file name
1097 * @flags: lookup flags
1098 * @nd: pointer to nameidata
1100 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1101 const char *name, unsigned int flags,
1102 struct nameidata *nd)
1106 /* same as do_path_lookup */
1107 nd->last_type = LAST_ROOT;
1111 nd->path.dentry = dentry;
1113 path_get(&nd->path);
1114 nd->root = nd->path;
1115 path_get(&nd->root);
1117 retval = path_walk(name, nd);
1118 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1119 nd->path.dentry->d_inode))
1120 audit_inode(name, nd->path.dentry);
1122 path_put(&nd->root);
1123 nd->root.mnt = NULL;
1128 static struct dentry *__lookup_hash(struct qstr *name,
1129 struct dentry *base, struct nameidata *nd)
1131 struct dentry *dentry;
1132 struct inode *inode;
1135 inode = base->d_inode;
1138 * See if the low-level filesystem might want
1139 * to use its own hash..
1141 if (base->d_op && base->d_op->d_hash) {
1142 err = base->d_op->d_hash(base, name);
1143 dentry = ERR_PTR(err);
1148 dentry = __d_lookup(base, name);
1150 /* lockess __d_lookup may fail due to concurrent d_move()
1151 * in some unrelated directory, so try with d_lookup
1154 dentry = d_lookup(base, name);
1156 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1157 dentry = do_revalidate(dentry, nd);
1162 /* Don't create child dentry for a dead directory. */
1163 dentry = ERR_PTR(-ENOENT);
1164 if (IS_DEADDIR(inode))
1167 new = d_alloc(base, name);
1168 dentry = ERR_PTR(-ENOMEM);
1171 dentry = inode->i_op->lookup(inode, new, nd);
1182 * Restricted form of lookup. Doesn't follow links, single-component only,
1183 * needs parent already locked. Doesn't follow mounts.
1186 static struct dentry *lookup_hash(struct nameidata *nd)
1190 err = exec_permission(nd->path.dentry->d_inode);
1192 return ERR_PTR(err);
1193 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1196 static int __lookup_one_len(const char *name, struct qstr *this,
1197 struct dentry *base, int len)
1207 hash = init_name_hash();
1209 c = *(const unsigned char *)name++;
1210 if (c == '/' || c == '\0')
1212 hash = partial_name_hash(c, hash);
1214 this->hash = end_name_hash(hash);
1219 * lookup_one_len - filesystem helper to lookup single pathname component
1220 * @name: pathname component to lookup
1221 * @base: base directory to lookup from
1222 * @len: maximum length @len should be interpreted to
1224 * Note that this routine is purely a helper for filesystem usage and should
1225 * not be called by generic code. Also note that by using this function the
1226 * nameidata argument is passed to the filesystem methods and a filesystem
1227 * using this helper needs to be prepared for that.
1229 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1234 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1236 err = __lookup_one_len(name, &this, base, len);
1238 return ERR_PTR(err);
1240 err = exec_permission(base->d_inode);
1242 return ERR_PTR(err);
1243 return __lookup_hash(&this, base, NULL);
1246 int user_path_at(int dfd, const char __user *name, unsigned flags,
1249 struct nameidata nd;
1250 char *tmp = getname(name);
1251 int err = PTR_ERR(tmp);
1254 BUG_ON(flags & LOOKUP_PARENT);
1256 err = do_path_lookup(dfd, tmp, flags, &nd);
1264 static int user_path_parent(int dfd, const char __user *path,
1265 struct nameidata *nd, char **name)
1267 char *s = getname(path);
1273 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1283 * It's inline, so penalty for filesystems that don't use sticky bit is
1286 static inline int check_sticky(struct inode *dir, struct inode *inode)
1288 uid_t fsuid = current_fsuid();
1290 if (!(dir->i_mode & S_ISVTX))
1292 if (inode->i_uid == fsuid)
1294 if (dir->i_uid == fsuid)
1296 return !capable(CAP_FOWNER);
1300 * Check whether we can remove a link victim from directory dir, check
1301 * whether the type of victim is right.
1302 * 1. We can't do it if dir is read-only (done in permission())
1303 * 2. We should have write and exec permissions on dir
1304 * 3. We can't remove anything from append-only dir
1305 * 4. We can't do anything with immutable dir (done in permission())
1306 * 5. If the sticky bit on dir is set we should either
1307 * a. be owner of dir, or
1308 * b. be owner of victim, or
1309 * c. have CAP_FOWNER capability
1310 * 6. If the victim is append-only or immutable we can't do antyhing with
1311 * links pointing to it.
1312 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1313 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1314 * 9. We can't remove a root or mountpoint.
1315 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1316 * nfs_async_unlink().
1318 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1322 if (!victim->d_inode)
1325 BUG_ON(victim->d_parent->d_inode != dir);
1326 audit_inode_child(victim, dir);
1328 if (dir->i_op->may_delete) {
1331 if (IS_IMMUTABLE(dir))
1333 error = dir->i_op->may_delete(dir, victim->d_inode);
1335 error = security_inode_permission(dir, MAY_WRITE | MAY_EXEC);
1337 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1338 if (!error && check_sticky(dir, victim->d_inode))
1345 if (IS_APPEND(victim->d_inode) || IS_IMMUTABLE(victim->d_inode))
1348 if (!S_ISDIR(victim->d_inode->i_mode))
1350 if (IS_ROOT(victim))
1352 } else if (S_ISDIR(victim->d_inode->i_mode))
1354 if (IS_DEADDIR(dir))
1356 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1361 /* Check whether we can create an object with dentry child in directory
1363 * 1. We can't do it if child already exists (open has special treatment for
1364 * this case, but since we are inlined it's OK)
1365 * 2. We can't do it if dir is read-only (done in permission())
1366 * 3. We should have write and exec permissions on dir
1367 * 4. We can't do it if dir is immutable (done in permission())
1369 static inline int may_create(struct inode *dir, struct dentry *child,
1376 if (IS_DEADDIR(dir))
1379 if (dir->i_op->may_create) {
1382 if (IS_IMMUTABLE(dir))
1384 error = dir->i_op->may_create(dir, isdir);
1386 error = security_inode_permission(dir, MAY_WRITE | MAY_EXEC);
1388 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1394 * p1 and p2 should be directories on the same fs.
1396 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1401 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1405 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1407 p = d_ancestor(p2, p1);
1409 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1410 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1414 p = d_ancestor(p1, p2);
1416 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1417 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1421 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1422 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1426 void unlock_rename(struct dentry *p1, struct dentry *p2)
1428 mutex_unlock(&p1->d_inode->i_mutex);
1430 mutex_unlock(&p2->d_inode->i_mutex);
1431 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1435 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1436 struct nameidata *nd)
1438 int error = may_create(dir, dentry, 0);
1443 if (!dir->i_op->create)
1444 return -EACCES; /* shouldn't it be ENOSYS? */
1447 error = security_inode_create(dir, dentry, mode);
1450 error = dir->i_op->create(dir, dentry, mode, nd);
1452 fsnotify_create(dir, dentry);
1456 int may_open(struct path *path, int acc_mode, int flag)
1458 struct dentry *dentry = path->dentry;
1459 struct inode *inode = dentry->d_inode;
1465 switch (inode->i_mode & S_IFMT) {
1469 if (acc_mode & MAY_WRITE)
1474 if (path->mnt->mnt_flags & MNT_NODEV)
1483 error = inode_permission(inode, acc_mode);
1488 * An append-only file must be opened in append mode for writing.
1490 if (IS_APPEND(inode)) {
1491 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1497 /* O_NOATIME can only be set by the owner or superuser */
1498 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1502 * Ensure there are no outstanding leases on the file.
1504 return break_lease(inode, flag);
1507 static int handle_truncate(struct path *path)
1509 struct inode *inode = path->dentry->d_inode;
1510 int error = get_write_access(inode);
1514 * Refuse to truncate files with mandatory locks held on them.
1516 error = locks_verify_locked(inode);
1518 error = security_path_truncate(path, 0,
1519 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1521 error = do_truncate(path->dentry, 0,
1522 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1525 put_write_access(inode);
1530 * Be careful about ever adding any more callers of this
1531 * function. Its flags must be in the namei format, not
1532 * what get passed to sys_open().
1534 static int __open_namei_create(struct nameidata *nd, struct path *path,
1535 int open_flag, int mode)
1538 struct dentry *dir = nd->path.dentry;
1540 if (!IS_POSIXACL(dir->d_inode))
1541 mode &= ~current_umask();
1542 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1545 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1547 mutex_unlock(&dir->d_inode->i_mutex);
1548 dput(nd->path.dentry);
1549 nd->path.dentry = path->dentry;
1552 /* Don't check for write permission, don't truncate */
1553 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
1557 * Note that while the flag value (low two bits) for sys_open means:
1562 * it is changed into
1563 * 00 - no permissions needed
1564 * 01 - read-permission
1565 * 10 - write-permission
1567 * for the internal routines (ie open_namei()/follow_link() etc)
1568 * This is more logical, and also allows the 00 "no perm needed"
1569 * to be used for symlinks (where the permissions are checked
1573 static inline int open_to_namei_flags(int flag)
1575 if ((flag+1) & O_ACCMODE)
1580 static int open_will_truncate(int flag, struct inode *inode)
1583 * We'll never write to the fs underlying
1586 if (special_file(inode->i_mode))
1588 return (flag & O_TRUNC);
1591 static struct file *finish_open(struct nameidata *nd,
1592 int open_flag, int acc_mode)
1598 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
1599 if (will_truncate) {
1600 error = mnt_want_write(nd->path.mnt);
1604 error = may_open(&nd->path, acc_mode, open_flag);
1607 mnt_drop_write(nd->path.mnt);
1610 filp = nameidata_to_filp(nd);
1611 if (!IS_ERR(filp)) {
1612 error = ima_file_check(filp, acc_mode);
1615 filp = ERR_PTR(error);
1618 if (!IS_ERR(filp)) {
1619 if (will_truncate) {
1620 error = handle_truncate(&nd->path);
1623 filp = ERR_PTR(error);
1628 * It is now safe to drop the mnt write
1629 * because the filp has had a write taken
1633 mnt_drop_write(nd->path.mnt);
1637 if (!IS_ERR(nd->intent.open.file))
1638 release_open_intent(nd);
1639 path_put(&nd->path);
1640 return ERR_PTR(error);
1643 static struct file *do_last(struct nameidata *nd, struct path *path,
1644 int open_flag, int acc_mode,
1645 int mode, const char *pathname)
1647 struct dentry *dir = nd->path.dentry;
1649 int error = -EISDIR;
1651 switch (nd->last_type) {
1654 dir = nd->path.dentry;
1656 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1657 if (!dir->d_op->d_revalidate(dir, nd)) {
1664 if (open_flag & O_CREAT)
1668 audit_inode(pathname, dir);
1672 /* trailing slashes? */
1673 if (nd->last.name[nd->last.len]) {
1674 if (open_flag & O_CREAT)
1676 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
1679 /* just plain open? */
1680 if (!(open_flag & O_CREAT)) {
1681 error = do_lookup(nd, &nd->last, path);
1685 if (!path->dentry->d_inode)
1687 if (path->dentry->d_inode->i_op->follow_link)
1690 if (nd->flags & LOOKUP_DIRECTORY) {
1691 if (!path->dentry->d_inode->i_op->lookup)
1694 path_to_nameidata(path, nd);
1695 audit_inode(pathname, nd->path.dentry);
1699 /* OK, it's O_CREAT */
1700 mutex_lock(&dir->d_inode->i_mutex);
1702 path->dentry = lookup_hash(nd);
1703 path->mnt = nd->path.mnt;
1705 error = PTR_ERR(path->dentry);
1706 if (IS_ERR(path->dentry)) {
1707 mutex_unlock(&dir->d_inode->i_mutex);
1711 if (IS_ERR(nd->intent.open.file)) {
1712 error = PTR_ERR(nd->intent.open.file);
1713 goto exit_mutex_unlock;
1716 /* Negative dentry, just create the file */
1717 if (!path->dentry->d_inode) {
1719 * This write is needed to ensure that a
1720 * ro->rw transition does not occur between
1721 * the time when the file is created and when
1722 * a permanent write count is taken through
1723 * the 'struct file' in nameidata_to_filp().
1725 error = mnt_want_write(nd->path.mnt);
1727 goto exit_mutex_unlock;
1728 error = __open_namei_create(nd, path, open_flag, mode);
1730 mnt_drop_write(nd->path.mnt);
1733 filp = nameidata_to_filp(nd);
1734 mnt_drop_write(nd->path.mnt);
1735 if (!IS_ERR(filp)) {
1736 error = ima_file_check(filp, acc_mode);
1739 filp = ERR_PTR(error);
1746 * It already exists.
1748 mutex_unlock(&dir->d_inode->i_mutex);
1749 audit_inode(pathname, path->dentry);
1752 if (open_flag & O_EXCL)
1755 if (__follow_mount(path)) {
1757 if (open_flag & O_NOFOLLOW)
1762 if (!path->dentry->d_inode)
1765 if (path->dentry->d_inode->i_op->follow_link)
1768 path_to_nameidata(path, nd);
1770 if (S_ISDIR(path->dentry->d_inode->i_mode))
1773 filp = finish_open(nd, open_flag, acc_mode);
1777 mutex_unlock(&dir->d_inode->i_mutex);
1779 path_put_conditional(path, nd);
1781 if (!IS_ERR(nd->intent.open.file))
1782 release_open_intent(nd);
1783 path_put(&nd->path);
1784 return ERR_PTR(error);
1788 * Note that the low bits of the passed in "open_flag"
1789 * are not the same as in the local variable "flag". See
1790 * open_to_namei_flags() for more details.
1792 struct file *do_filp_open(int dfd, const char *pathname,
1793 int open_flag, int mode, int acc_mode)
1796 struct nameidata nd;
1800 int flag = open_to_namei_flags(open_flag);
1801 int force_reval = 0;
1803 if (!(open_flag & O_CREAT))
1807 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1808 * check for O_DSYNC if the need any syncing at all we enforce it's
1809 * always set instead of having to deal with possibly weird behaviour
1810 * for malicious applications setting only __O_SYNC.
1812 if (open_flag & __O_SYNC)
1813 open_flag |= O_DSYNC;
1816 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1818 /* O_TRUNC implies we need access checks for write permissions */
1819 if (open_flag & O_TRUNC)
1820 acc_mode |= MAY_WRITE;
1822 /* Allow the LSM permission hook to distinguish append
1823 access from general write access. */
1824 if (open_flag & O_APPEND)
1825 acc_mode |= MAY_APPEND;
1827 /* find the parent */
1829 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1831 return ERR_PTR(error);
1833 nd.flags |= LOOKUP_REVAL;
1835 current->total_link_count = 0;
1836 error = link_path_walk(pathname, &nd);
1838 filp = ERR_PTR(error);
1841 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
1842 audit_inode(pathname, nd.path.dentry);
1845 * We have the parent and last component.
1849 filp = get_empty_filp();
1852 nd.intent.open.file = filp;
1853 filp->f_flags = open_flag;
1854 nd.intent.open.flags = flag;
1855 nd.intent.open.create_mode = mode;
1856 nd.flags &= ~LOOKUP_PARENT;
1857 nd.flags |= LOOKUP_OPEN;
1858 if (open_flag & O_CREAT) {
1859 nd.flags |= LOOKUP_CREATE;
1860 if (open_flag & O_EXCL)
1861 nd.flags |= LOOKUP_EXCL;
1863 if (open_flag & O_DIRECTORY)
1864 nd.flags |= LOOKUP_DIRECTORY;
1865 if (!(open_flag & O_NOFOLLOW))
1866 nd.flags |= LOOKUP_FOLLOW;
1867 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1868 while (unlikely(!filp)) { /* trailing symlink */
1870 struct inode *inode = path.dentry->d_inode;
1873 /* S_ISDIR part is a temporary automount kludge */
1874 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
1879 * This is subtle. Instead of calling do_follow_link() we do
1880 * the thing by hands. The reason is that this way we have zero
1881 * link_count and path_walk() (called from ->follow_link)
1882 * honoring LOOKUP_PARENT. After that we have the parent and
1883 * last component, i.e. we are in the same situation as after
1884 * the first path_walk(). Well, almost - if the last component
1885 * is normal we get its copy stored in nd->last.name and we will
1886 * have to putname() it when we are done. Procfs-like symlinks
1887 * just set LAST_BIND.
1889 nd.flags |= LOOKUP_PARENT;
1890 error = security_inode_follow_link(path.dentry, &nd);
1893 error = __do_follow_link(&path, &nd, &cookie);
1894 if (unlikely(error)) {
1895 /* nd.path had been dropped */
1896 if (!IS_ERR(cookie) && inode->i_op->put_link)
1897 inode->i_op->put_link(path.dentry, &nd, cookie);
1899 release_open_intent(&nd);
1900 filp = ERR_PTR(error);
1904 nd.flags &= ~LOOKUP_PARENT;
1905 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1906 if (inode->i_op->put_link)
1907 inode->i_op->put_link(holder.dentry, &nd, cookie);
1913 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1920 path_put_conditional(&path, &nd);
1921 if (!IS_ERR(nd.intent.open.file))
1922 release_open_intent(&nd);
1925 filp = ERR_PTR(error);
1930 * filp_open - open file and return file pointer
1932 * @filename: path to open
1933 * @flags: open flags as per the open(2) second argument
1934 * @mode: mode for the new file if O_CREAT is set, else ignored
1936 * This is the helper to open a file from kernelspace if you really
1937 * have to. But in generally you should not do this, so please move
1938 * along, nothing to see here..
1940 struct file *filp_open(const char *filename, int flags, int mode)
1942 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1944 EXPORT_SYMBOL(filp_open);
1947 * lookup_create - lookup a dentry, creating it if it doesn't exist
1948 * @nd: nameidata info
1949 * @is_dir: directory flag
1951 * Simple function to lookup and return a dentry and create it
1952 * if it doesn't exist. Is SMP-safe.
1954 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1956 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1958 struct dentry *dentry = ERR_PTR(-EEXIST);
1960 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1962 * Yucky last component or no last component at all?
1963 * (foo/., foo/.., /////)
1965 if (nd->last_type != LAST_NORM)
1967 nd->flags &= ~LOOKUP_PARENT;
1968 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1969 nd->intent.open.flags = O_EXCL;
1972 * Do the final lookup.
1974 dentry = lookup_hash(nd);
1978 if (dentry->d_inode)
1981 * Special case - lookup gave negative, but... we had foo/bar/
1982 * From the vfs_mknod() POV we just have a negative dentry -
1983 * all is fine. Let's be bastards - you had / on the end, you've
1984 * been asking for (non-existent) directory. -ENOENT for you.
1986 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1988 dentry = ERR_PTR(-ENOENT);
1993 dentry = ERR_PTR(-EEXIST);
1997 EXPORT_SYMBOL_GPL(lookup_create);
1999 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2001 int error = may_create(dir, dentry, 0);
2006 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2009 if (!dir->i_op->mknod)
2012 error = devcgroup_inode_mknod(mode, dev);
2016 error = security_inode_mknod(dir, dentry, mode, dev);
2020 error = dir->i_op->mknod(dir, dentry, mode, dev);
2022 fsnotify_create(dir, dentry);
2026 static int may_mknod(mode_t mode)
2028 switch (mode & S_IFMT) {
2034 case 0: /* zero mode translates to S_IFREG */
2043 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2048 struct dentry *dentry;
2049 struct nameidata nd;
2054 error = user_path_parent(dfd, filename, &nd, &tmp);
2058 dentry = lookup_create(&nd, 0);
2059 if (IS_ERR(dentry)) {
2060 error = PTR_ERR(dentry);
2063 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2064 mode &= ~current_umask();
2065 error = may_mknod(mode);
2068 error = mnt_want_write(nd.path.mnt);
2071 error = security_path_mknod(&nd.path, dentry, mode, dev);
2073 goto out_drop_write;
2074 switch (mode & S_IFMT) {
2075 case 0: case S_IFREG:
2076 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2078 case S_IFCHR: case S_IFBLK:
2079 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2080 new_decode_dev(dev));
2082 case S_IFIFO: case S_IFSOCK:
2083 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2087 mnt_drop_write(nd.path.mnt);
2091 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2098 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2100 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2103 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2105 int error = may_create(dir, dentry, 1);
2110 if (!dir->i_op->mkdir)
2113 mode &= (S_IRWXUGO|S_ISVTX);
2114 error = security_inode_mkdir(dir, dentry, mode);
2118 error = dir->i_op->mkdir(dir, dentry, mode);
2120 fsnotify_mkdir(dir, dentry);
2124 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2128 struct dentry *dentry;
2129 struct nameidata nd;
2131 error = user_path_parent(dfd, pathname, &nd, &tmp);
2135 dentry = lookup_create(&nd, 1);
2136 error = PTR_ERR(dentry);
2140 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2141 mode &= ~current_umask();
2142 error = mnt_want_write(nd.path.mnt);
2145 error = security_path_mkdir(&nd.path, dentry, mode);
2147 goto out_drop_write;
2148 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2150 mnt_drop_write(nd.path.mnt);
2154 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2161 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2163 return sys_mkdirat(AT_FDCWD, pathname, mode);
2167 * We try to drop the dentry early: we should have
2168 * a usage count of 2 if we're the only user of this
2169 * dentry, and if that is true (possibly after pruning
2170 * the dcache), then we drop the dentry now.
2172 * A low-level filesystem can, if it choses, legally
2175 * if (!d_unhashed(dentry))
2178 * if it cannot handle the case of removing a directory
2179 * that is still in use by something else..
2181 void dentry_unhash(struct dentry *dentry)
2184 shrink_dcache_parent(dentry);
2185 spin_lock(&dcache_lock);
2186 spin_lock(&dentry->d_lock);
2187 if (atomic_read(&dentry->d_count) == 2)
2189 spin_unlock(&dentry->d_lock);
2190 spin_unlock(&dcache_lock);
2193 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2195 int error = may_delete(dir, dentry, 1);
2200 if (!dir->i_op->rmdir)
2203 mutex_lock(&dentry->d_inode->i_mutex);
2204 dentry_unhash(dentry);
2205 if (d_mountpoint(dentry))
2208 error = security_inode_rmdir(dir, dentry);
2210 error = dir->i_op->rmdir(dir, dentry);
2212 dentry->d_inode->i_flags |= S_DEAD;
2217 mutex_unlock(&dentry->d_inode->i_mutex);
2226 static long do_rmdir(int dfd, const char __user *pathname)
2230 struct dentry *dentry;
2231 struct nameidata nd;
2233 error = user_path_parent(dfd, pathname, &nd, &name);
2237 switch(nd.last_type) {
2249 nd.flags &= ~LOOKUP_PARENT;
2251 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2252 dentry = lookup_hash(&nd);
2253 error = PTR_ERR(dentry);
2256 error = mnt_want_write(nd.path.mnt);
2259 error = security_path_rmdir(&nd.path, dentry);
2262 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2264 mnt_drop_write(nd.path.mnt);
2268 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2275 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2277 return do_rmdir(AT_FDCWD, pathname);
2280 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2282 int error = may_delete(dir, dentry, 0);
2287 if (!dir->i_op->unlink)
2290 mutex_lock(&dentry->d_inode->i_mutex);
2291 if (d_mountpoint(dentry))
2294 error = security_inode_unlink(dir, dentry);
2296 error = dir->i_op->unlink(dir, dentry);
2301 mutex_unlock(&dentry->d_inode->i_mutex);
2303 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2304 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2305 fsnotify_link_count(dentry->d_inode);
2313 * Make sure that the actual truncation of the file will occur outside its
2314 * directory's i_mutex. Truncate can take a long time if there is a lot of
2315 * writeout happening, and we don't want to prevent access to the directory
2316 * while waiting on the I/O.
2318 static long do_unlinkat(int dfd, const char __user *pathname)
2322 struct dentry *dentry;
2323 struct nameidata nd;
2324 struct inode *inode = NULL;
2326 error = user_path_parent(dfd, pathname, &nd, &name);
2331 if (nd.last_type != LAST_NORM)
2334 nd.flags &= ~LOOKUP_PARENT;
2336 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2337 dentry = lookup_hash(&nd);
2338 error = PTR_ERR(dentry);
2339 if (!IS_ERR(dentry)) {
2340 /* Why not before? Because we want correct error value */
2341 if (nd.last.name[nd.last.len])
2343 inode = dentry->d_inode;
2345 atomic_inc(&inode->i_count);
2346 error = mnt_want_write(nd.path.mnt);
2349 error = security_path_unlink(&nd.path, dentry);
2352 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2354 mnt_drop_write(nd.path.mnt);
2358 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2360 iput(inode); /* truncate the inode here */
2367 error = !dentry->d_inode ? -ENOENT :
2368 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2372 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2374 if ((flag & ~AT_REMOVEDIR) != 0)
2377 if (flag & AT_REMOVEDIR)
2378 return do_rmdir(dfd, pathname);
2380 return do_unlinkat(dfd, pathname);
2383 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2385 return do_unlinkat(AT_FDCWD, pathname);
2388 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2390 int error = may_create(dir, dentry, 0);
2395 if (!dir->i_op->symlink)
2398 error = security_inode_symlink(dir, dentry, oldname);
2402 error = dir->i_op->symlink(dir, dentry, oldname);
2404 fsnotify_create(dir, dentry);
2408 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2409 int, newdfd, const char __user *, newname)
2414 struct dentry *dentry;
2415 struct nameidata nd;
2417 from = getname(oldname);
2419 return PTR_ERR(from);
2421 error = user_path_parent(newdfd, newname, &nd, &to);
2425 dentry = lookup_create(&nd, 0);
2426 error = PTR_ERR(dentry);
2430 error = mnt_want_write(nd.path.mnt);
2433 error = security_path_symlink(&nd.path, dentry, from);
2435 goto out_drop_write;
2436 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2438 mnt_drop_write(nd.path.mnt);
2442 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2450 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2452 return sys_symlinkat(oldname, AT_FDCWD, newname);
2455 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2457 struct inode *inode = old_dentry->d_inode;
2463 error = may_create(dir, new_dentry, S_ISDIR(inode->i_mode));
2467 if (dir->i_sb != inode->i_sb)
2471 * A link to an append-only or immutable file cannot be created.
2473 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2475 if (!dir->i_op->link)
2477 if (S_ISDIR(inode->i_mode))
2480 error = security_inode_link(old_dentry, dir, new_dentry);
2484 mutex_lock(&inode->i_mutex);
2485 error = dir->i_op->link(old_dentry, dir, new_dentry);
2486 mutex_unlock(&inode->i_mutex);
2488 fsnotify_link(dir, inode, new_dentry);
2493 * Hardlinks are often used in delicate situations. We avoid
2494 * security-related surprises by not following symlinks on the
2497 * We don't follow them on the oldname either to be compatible
2498 * with linux 2.0, and to avoid hard-linking to directories
2499 * and other special files. --ADM
2501 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2502 int, newdfd, const char __user *, newname, int, flags)
2504 struct dentry *new_dentry;
2505 struct nameidata nd;
2506 struct path old_path;
2510 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2513 error = user_path_at(olddfd, oldname,
2514 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2519 error = user_path_parent(newdfd, newname, &nd, &to);
2523 if (old_path.mnt != nd.path.mnt)
2525 new_dentry = lookup_create(&nd, 0);
2526 error = PTR_ERR(new_dentry);
2527 if (IS_ERR(new_dentry))
2529 error = mnt_want_write(nd.path.mnt);
2532 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2534 goto out_drop_write;
2535 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2537 mnt_drop_write(nd.path.mnt);
2541 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2546 path_put(&old_path);
2551 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2553 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2557 * The worst of all namespace operations - renaming directory. "Perverted"
2558 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2560 * a) we can get into loop creation. Check is done in is_subdir().
2561 * b) race potential - two innocent renames can create a loop together.
2562 * That's where 4.4 screws up. Current fix: serialization on
2563 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2565 * c) we have to lock _three_ objects - parents and victim (if it exists).
2566 * And that - after we got ->i_mutex on parents (until then we don't know
2567 * whether the target exists). Solution: try to be smart with locking
2568 * order for inodes. We rely on the fact that tree topology may change
2569 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2570 * move will be locked. Thus we can rank directories by the tree
2571 * (ancestors first) and rank all non-directories after them.
2572 * That works since everybody except rename does "lock parent, lookup,
2573 * lock child" and rename is under ->s_vfs_rename_mutex.
2574 * HOWEVER, it relies on the assumption that any object with ->lookup()
2575 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2576 * we'd better make sure that there's no link(2) for them.
2577 * d) some filesystems don't support opened-but-unlinked directories,
2578 * either because of layout or because they are not ready to deal with
2579 * all cases correctly. The latter will be fixed (taking this sort of
2580 * stuff into VFS), but the former is not going away. Solution: the same
2581 * trick as in rmdir().
2582 * e) conversion from fhandle to dentry may come in the wrong moment - when
2583 * we are removing the target. Solution: we will have to grab ->i_mutex
2584 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2585 * ->i_mutex on parents, which works but leads to some truly excessive
2588 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2589 struct inode *new_dir, struct dentry *new_dentry)
2592 struct inode *target;
2595 * If we are going to change the parent - check write permissions,
2596 * we'll need to flip '..'.
2598 if (new_dir != old_dir) {
2599 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2604 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2608 target = new_dentry->d_inode;
2610 mutex_lock(&target->i_mutex);
2611 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2615 dentry_unhash(new_dentry);
2616 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2620 target->i_flags |= S_DEAD;
2621 dont_mount(new_dentry);
2623 mutex_unlock(&target->i_mutex);
2624 if (d_unhashed(new_dentry))
2625 d_rehash(new_dentry);
2629 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2630 d_move(old_dentry,new_dentry);
2634 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2635 struct inode *new_dir, struct dentry *new_dentry)
2637 struct inode *target;
2640 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2645 target = new_dentry->d_inode;
2647 mutex_lock(&target->i_mutex);
2648 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2651 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2654 dont_mount(new_dentry);
2655 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2656 d_move(old_dentry, new_dentry);
2659 mutex_unlock(&target->i_mutex);
2664 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2665 struct inode *new_dir, struct dentry *new_dentry)
2668 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2669 const char *old_name;
2671 if (old_dentry->d_inode == new_dentry->d_inode)
2674 error = may_delete(old_dir, old_dentry, is_dir);
2678 if (!new_dentry->d_inode)
2679 error = may_create(new_dir, new_dentry, is_dir);
2681 error = may_delete(new_dir, new_dentry, is_dir);
2685 if (!old_dir->i_op->rename)
2688 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2691 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2693 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2695 fsnotify_move(old_dir, new_dir, old_name, is_dir,
2696 new_dentry->d_inode, old_dentry);
2697 fsnotify_oldname_free(old_name);
2702 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2703 int, newdfd, const char __user *, newname)
2705 struct dentry *old_dir, *new_dir;
2706 struct dentry *old_dentry, *new_dentry;
2707 struct dentry *trap;
2708 struct nameidata oldnd, newnd;
2713 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2717 error = user_path_parent(newdfd, newname, &newnd, &to);
2722 if (oldnd.path.mnt != newnd.path.mnt)
2725 old_dir = oldnd.path.dentry;
2727 if (oldnd.last_type != LAST_NORM)
2730 new_dir = newnd.path.dentry;
2731 if (newnd.last_type != LAST_NORM)
2734 oldnd.flags &= ~LOOKUP_PARENT;
2735 newnd.flags &= ~LOOKUP_PARENT;
2736 newnd.flags |= LOOKUP_RENAME_TARGET;
2738 trap = lock_rename(new_dir, old_dir);
2740 old_dentry = lookup_hash(&oldnd);
2741 error = PTR_ERR(old_dentry);
2742 if (IS_ERR(old_dentry))
2744 /* source must exist */
2746 if (!old_dentry->d_inode)
2748 /* unless the source is a directory trailing slashes give -ENOTDIR */
2749 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2751 if (oldnd.last.name[oldnd.last.len])
2753 if (newnd.last.name[newnd.last.len])
2756 /* source should not be ancestor of target */
2758 if (old_dentry == trap)
2760 new_dentry = lookup_hash(&newnd);
2761 error = PTR_ERR(new_dentry);
2762 if (IS_ERR(new_dentry))
2764 /* target should not be an ancestor of source */
2766 if (new_dentry == trap)
2769 error = mnt_want_write(oldnd.path.mnt);
2772 error = security_path_rename(&oldnd.path, old_dentry,
2773 &newnd.path, new_dentry);
2776 error = vfs_rename(old_dir->d_inode, old_dentry,
2777 new_dir->d_inode, new_dentry);
2779 mnt_drop_write(oldnd.path.mnt);
2785 unlock_rename(new_dir, old_dir);
2787 path_put(&newnd.path);
2790 path_put(&oldnd.path);
2796 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2798 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2801 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2805 len = PTR_ERR(link);
2810 if (len > (unsigned) buflen)
2812 if (copy_to_user(buffer, link, len))
2819 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2820 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2821 * using) it for any given inode is up to filesystem.
2823 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2825 struct nameidata nd;
2830 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2832 return PTR_ERR(cookie);
2834 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2835 if (dentry->d_inode->i_op->put_link)
2836 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2840 int vfs_follow_link(struct nameidata *nd, const char *link)
2842 return __vfs_follow_link(nd, link);
2845 /* get the link contents into pagecache */
2846 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2850 struct address_space *mapping = dentry->d_inode->i_mapping;
2851 page = read_mapping_page(mapping, 0, NULL);
2856 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2860 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2862 struct page *page = NULL;
2863 char *s = page_getlink(dentry, &page);
2864 int res = vfs_readlink(dentry,buffer,buflen,s);
2867 page_cache_release(page);
2872 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2874 struct page *page = NULL;
2875 nd_set_link(nd, page_getlink(dentry, &page));
2879 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2881 struct page *page = cookie;
2885 page_cache_release(page);
2890 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2892 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2894 struct address_space *mapping = inode->i_mapping;
2899 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2901 flags |= AOP_FLAG_NOFS;
2904 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2905 flags, &page, &fsdata);
2909 kaddr = kmap_atomic(page, KM_USER0);
2910 memcpy(kaddr, symname, len-1);
2911 kunmap_atomic(kaddr, KM_USER0);
2913 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2920 mark_inode_dirty(inode);
2926 int page_symlink(struct inode *inode, const char *symname, int len)
2928 return __page_symlink(inode, symname, len,
2929 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2932 const struct inode_operations page_symlink_inode_operations = {
2933 .readlink = generic_readlink,
2934 .follow_link = page_follow_link_light,
2935 .put_link = page_put_link,
2938 EXPORT_SYMBOL(user_path_at);
2939 EXPORT_SYMBOL(follow_down);
2940 EXPORT_SYMBOL(follow_up);
2941 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2942 EXPORT_SYMBOL(getname);
2943 EXPORT_SYMBOL(lock_rename);
2944 EXPORT_SYMBOL(lookup_one_len);
2945 EXPORT_SYMBOL(page_follow_link_light);
2946 EXPORT_SYMBOL(page_put_link);
2947 EXPORT_SYMBOL(page_readlink);
2948 EXPORT_SYMBOL(__page_symlink);
2949 EXPORT_SYMBOL(page_symlink);
2950 EXPORT_SYMBOL(page_symlink_inode_operations);
2951 EXPORT_SYMBOL(path_lookup);
2952 EXPORT_SYMBOL(kern_path);
2953 EXPORT_SYMBOL(vfs_path_lookup);
2954 EXPORT_SYMBOL(inode_permission);
2955 EXPORT_SYMBOL(file_permission);
2956 EXPORT_SYMBOL(unlock_rename);
2957 EXPORT_SYMBOL(vfs_create);
2958 EXPORT_SYMBOL(vfs_follow_link);
2959 EXPORT_SYMBOL(vfs_link);
2960 EXPORT_SYMBOL(vfs_mkdir);
2961 EXPORT_SYMBOL(vfs_mknod);
2962 EXPORT_SYMBOL(generic_permission);
2963 EXPORT_SYMBOL(vfs_readlink);
2964 EXPORT_SYMBOL(vfs_rename);
2965 EXPORT_SYMBOL(vfs_rmdir);
2966 EXPORT_SYMBOL(vfs_symlink);
2967 EXPORT_SYMBOL(vfs_unlink);
2968 EXPORT_SYMBOL(dentry_unhash);
2969 EXPORT_SYMBOL(generic_readlink);