4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <asm/uaccess.h>
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
171 * This does basic POSIX ACL permission checking
173 static int acl_permission_check(struct inode *inode, int mask,
174 int (*check_acl)(struct inode *inode, int mask))
176 umode_t mode = inode->i_mode;
178 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
180 if (current_fsuid() == inode->i_uid)
183 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
184 int error = check_acl(inode, mask);
185 if (error != -EAGAIN)
189 if (in_group_p(inode->i_gid))
194 * If the DACs are ok we don't need any capability check.
196 if ((mask & ~mode) == 0)
202 * generic_permission - check for access rights on a Posix-like filesystem
203 * @inode: inode to check access rights for
204 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
205 * @check_acl: optional callback to check for Posix ACLs
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things..
212 int generic_permission(struct inode *inode, int mask,
213 int (*check_acl)(struct inode *inode, int mask))
218 * Do the basic POSIX ACL permission checks.
220 ret = acl_permission_check(inode, mask, check_acl);
225 * Read/write DACs are always overridable.
226 * Executable DACs are overridable if at least one exec bit is set.
228 if (!(mask & MAY_EXEC) || execute_ok(inode))
229 if (capable(CAP_DAC_OVERRIDE))
233 * Searching includes executable on directories, else just read.
235 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
236 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
237 if (capable(CAP_DAC_READ_SEARCH))
244 * inode_permission - check for access rights to a given inode
245 * @inode: inode to check permission on
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
248 * Used to check for read/write/execute permissions on an inode.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 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 (inode->i_op->permission)
275 retval = inode->i_op->permission(inode, mask);
277 retval = generic_permission(inode, mask, inode->i_op->check_acl);
282 retval = devcgroup_inode_permission(inode, mask);
286 return security_inode_permission(inode,
287 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
291 * file_permission - check for additional access rights to a given file
292 * @file: file to check access rights for
293 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
295 * Used to check for read/write/execute permissions on an already opened
299 * Do not use this function in new code. All access checks should
300 * be done using inode_permission().
302 int file_permission(struct file *file, int mask)
304 return inode_permission(file->f_path.dentry->d_inode, mask);
308 * get_write_access() gets write permission for a file.
309 * put_write_access() releases this write permission.
310 * This is used for regular files.
311 * We cannot support write (and maybe mmap read-write shared) accesses and
312 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
313 * can have the following values:
314 * 0: no writers, no VM_DENYWRITE mappings
315 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
316 * > 0: (i_writecount) users are writing to the file.
318 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
319 * except for the cases where we don't hold i_writecount yet. Then we need to
320 * use {get,deny}_write_access() - these functions check the sign and refuse
321 * to do the change if sign is wrong. Exclusion between them is provided by
322 * the inode->i_lock spinlock.
325 int get_write_access(struct inode * inode)
327 spin_lock(&inode->i_lock);
328 if (atomic_read(&inode->i_writecount) < 0) {
329 spin_unlock(&inode->i_lock);
332 atomic_inc(&inode->i_writecount);
333 spin_unlock(&inode->i_lock);
338 int deny_write_access(struct file * file)
340 struct inode *inode = file->f_path.dentry->d_inode;
342 spin_lock(&inode->i_lock);
343 if (atomic_read(&inode->i_writecount) > 0) {
344 spin_unlock(&inode->i_lock);
347 atomic_dec(&inode->i_writecount);
348 spin_unlock(&inode->i_lock);
354 * path_get - get a reference to a path
355 * @path: path to get the reference to
357 * Given a path increment the reference count to the dentry and the vfsmount.
359 void path_get(struct path *path)
364 EXPORT_SYMBOL(path_get);
367 * path_put - put a reference to a path
368 * @path: path to put the reference to
370 * Given a path decrement the reference count to the dentry and the vfsmount.
372 void path_put(struct path *path)
377 EXPORT_SYMBOL(path_put);
380 * release_open_intent - free up open intent resources
381 * @nd: pointer to nameidata
383 void release_open_intent(struct nameidata *nd)
385 if (nd->intent.open.file->f_path.dentry == NULL)
386 put_filp(nd->intent.open.file);
388 fput(nd->intent.open.file);
391 static inline struct dentry *
392 do_revalidate(struct dentry *dentry, struct nameidata *nd)
394 int status = dentry->d_op->d_revalidate(dentry, nd);
395 if (unlikely(status <= 0)) {
397 * The dentry failed validation.
398 * If d_revalidate returned 0 attempt to invalidate
399 * the dentry otherwise d_revalidate is asking us
400 * to return a fail status.
403 if (!d_invalidate(dentry)) {
409 dentry = ERR_PTR(status);
416 * force_reval_path - force revalidation of a dentry
418 * In some situations the path walking code will trust dentries without
419 * revalidating them. This causes problems for filesystems that depend on
420 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
421 * (which indicates that it's possible for the dentry to go stale), force
422 * a d_revalidate call before proceeding.
424 * Returns 0 if the revalidation was successful. If the revalidation fails,
425 * either return the error returned by d_revalidate or -ESTALE if the
426 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
427 * invalidate the dentry. It's up to the caller to handle putting references
428 * to the path if necessary.
431 force_reval_path(struct path *path, struct nameidata *nd)
434 struct dentry *dentry = path->dentry;
437 * only check on filesystems where it's possible for the dentry to
438 * become stale. It's assumed that if this flag is set then the
439 * d_revalidate op will also be defined.
441 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
444 status = dentry->d_op->d_revalidate(dentry, nd);
449 d_invalidate(dentry);
456 * Short-cut version of permission(), for calling on directories
457 * during pathname resolution. Combines parts of permission()
458 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
460 * If appropriate, check DAC only. If not appropriate, or
461 * short-cut DAC fails, then call ->permission() to do more
462 * complete permission check.
464 static int exec_permission(struct inode *inode)
468 if (inode->i_op->permission) {
469 ret = inode->i_op->permission(inode, MAY_EXEC);
474 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
478 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
483 return security_inode_permission(inode, MAY_EXEC);
486 static __always_inline void set_root(struct nameidata *nd)
489 struct fs_struct *fs = current->fs;
490 read_lock(&fs->lock);
493 read_unlock(&fs->lock);
497 static int link_path_walk(const char *, struct nameidata *);
499 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
513 res = link_path_walk(link, nd);
514 if (nd->depth || res || nd->last_type!=LAST_NORM)
517 * If it is an iterative symlinks resolution in open_namei() we
518 * have to copy the last component. And all that crap because of
519 * bloody create() on broken symlinks. Furrfu...
522 if (unlikely(!name)) {
526 strcpy(name, nd->last.name);
527 nd->last.name = name;
531 return PTR_ERR(link);
534 static void path_put_conditional(struct path *path, struct nameidata *nd)
537 if (path->mnt != nd->path.mnt)
541 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
543 dput(nd->path.dentry);
544 if (nd->path.mnt != path->mnt)
545 mntput(nd->path.mnt);
546 nd->path.mnt = path->mnt;
547 nd->path.dentry = path->dentry;
550 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
554 struct dentry *dentry = path->dentry;
556 touch_atime(path->mnt, dentry);
557 nd_set_link(nd, NULL);
559 if (path->mnt != nd->path.mnt) {
560 path_to_nameidata(path, nd);
564 nd->last_type = LAST_BIND;
565 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
566 error = PTR_ERR(cookie);
567 if (!IS_ERR(cookie)) {
568 char *s = nd_get_link(nd);
571 error = __vfs_follow_link(nd, s);
572 else if (nd->last_type == LAST_BIND) {
573 error = force_reval_path(&nd->path, nd);
577 if (dentry->d_inode->i_op->put_link)
578 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
584 * This limits recursive symlink follows to 8, while
585 * limiting consecutive symlinks to 40.
587 * Without that kind of total limit, nasty chains of consecutive
588 * symlinks can cause almost arbitrarily long lookups.
590 static inline int do_follow_link(struct path *path, struct nameidata *nd)
593 if (current->link_count >= MAX_NESTED_LINKS)
595 if (current->total_link_count >= 40)
597 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
599 err = security_inode_follow_link(path->dentry, nd);
602 current->link_count++;
603 current->total_link_count++;
605 err = __do_follow_link(path, nd);
607 current->link_count--;
611 path_put_conditional(path, nd);
616 int follow_up(struct path *path)
618 struct vfsmount *parent;
619 struct dentry *mountpoint;
620 spin_lock(&vfsmount_lock);
621 parent = path->mnt->mnt_parent;
622 if (parent == path->mnt) {
623 spin_unlock(&vfsmount_lock);
627 mountpoint = dget(path->mnt->mnt_mountpoint);
628 spin_unlock(&vfsmount_lock);
630 path->dentry = mountpoint;
636 /* no need for dcache_lock, as serialization is taken care in
639 static int __follow_mount(struct path *path)
642 while (d_mountpoint(path->dentry)) {
643 struct vfsmount *mounted = lookup_mnt(path);
650 path->dentry = dget(mounted->mnt_root);
656 static void follow_mount(struct path *path)
658 while (d_mountpoint(path->dentry)) {
659 struct vfsmount *mounted = lookup_mnt(path);
665 path->dentry = dget(mounted->mnt_root);
669 /* no need for dcache_lock, as serialization is taken care in
672 int follow_down(struct path *path)
674 struct vfsmount *mounted;
676 mounted = lookup_mnt(path);
681 path->dentry = dget(mounted->mnt_root);
687 static __always_inline void follow_dotdot(struct nameidata *nd)
692 struct vfsmount *parent;
693 struct dentry *old = nd->path.dentry;
695 if (nd->path.dentry == nd->root.dentry &&
696 nd->path.mnt == nd->root.mnt) {
699 spin_lock(&dcache_lock);
700 if (nd->path.dentry != nd->path.mnt->mnt_root) {
701 nd->path.dentry = dget(nd->path.dentry->d_parent);
702 spin_unlock(&dcache_lock);
706 spin_unlock(&dcache_lock);
707 spin_lock(&vfsmount_lock);
708 parent = nd->path.mnt->mnt_parent;
709 if (parent == nd->path.mnt) {
710 spin_unlock(&vfsmount_lock);
714 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
715 spin_unlock(&vfsmount_lock);
717 mntput(nd->path.mnt);
718 nd->path.mnt = parent;
720 follow_mount(&nd->path);
724 * It's more convoluted than I'd like it to be, but... it's still fairly
725 * small and for now I'd prefer to have fast path as straight as possible.
726 * It _is_ time-critical.
728 static int do_lookup(struct nameidata *nd, struct qstr *name,
731 struct vfsmount *mnt = nd->path.mnt;
732 struct dentry *dentry, *parent;
735 * See if the low-level filesystem might want
736 * to use its own hash..
738 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
739 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
744 dentry = __d_lookup(nd->path.dentry, name);
747 if (dentry->d_op && dentry->d_op->d_revalidate)
748 goto need_revalidate;
751 path->dentry = dentry;
752 __follow_mount(path);
756 parent = nd->path.dentry;
757 dir = parent->d_inode;
759 mutex_lock(&dir->i_mutex);
761 * First re-do the cached lookup just in case it was created
762 * while we waited for the directory semaphore..
764 * FIXME! This could use version numbering or similar to
765 * avoid unnecessary cache lookups.
767 * The "dcache_lock" is purely to protect the RCU list walker
768 * from concurrent renames at this point (we mustn't get false
769 * negatives from the RCU list walk here, unlike the optimistic
772 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
774 dentry = d_lookup(parent, name);
778 /* Don't create child dentry for a dead directory. */
779 dentry = ERR_PTR(-ENOENT);
783 new = d_alloc(parent, name);
784 dentry = ERR_PTR(-ENOMEM);
786 dentry = dir->i_op->lookup(dir, new, nd);
793 mutex_unlock(&dir->i_mutex);
800 * Uhhuh! Nasty case: the cache was re-populated while
801 * we waited on the semaphore. Need to revalidate.
803 mutex_unlock(&dir->i_mutex);
804 if (dentry->d_op && dentry->d_op->d_revalidate) {
805 dentry = do_revalidate(dentry, nd);
807 dentry = ERR_PTR(-ENOENT);
814 dentry = do_revalidate(dentry, nd);
822 return PTR_ERR(dentry);
826 * This is a temporary kludge to deal with "automount" symlinks; proper
827 * solution is to trigger them on follow_mount(), so that do_lookup()
828 * would DTRT. To be killed before 2.6.34-final.
830 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
832 return inode && unlikely(inode->i_op->follow_link) &&
833 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
838 * This is the basic name resolution function, turning a pathname into
839 * the final dentry. We expect 'base' to be positive and a directory.
841 * Returns 0 and nd will have valid dentry and mnt on success.
842 * Returns error and drops reference to input namei data on failure.
844 static int link_path_walk(const char *name, struct nameidata *nd)
849 unsigned int lookup_flags = nd->flags;
856 inode = nd->path.dentry->d_inode;
858 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
860 /* At this point we know we have a real path component. */
866 nd->flags |= LOOKUP_CONTINUE;
867 err = exec_permission(inode);
872 c = *(const unsigned char *)name;
874 hash = init_name_hash();
877 hash = partial_name_hash(c, hash);
878 c = *(const unsigned char *)name;
879 } while (c && (c != '/'));
880 this.len = name - (const char *) this.name;
881 this.hash = end_name_hash(hash);
883 /* remove trailing slashes? */
886 while (*++name == '/');
888 goto last_with_slashes;
891 * "." and ".." are special - ".." especially so because it has
892 * to be able to know about the current root directory and
893 * parent relationships.
895 if (this.name[0] == '.') switch (this.len) {
899 if (this.name[1] != '.')
902 inode = nd->path.dentry->d_inode;
907 /* This does the actual lookups.. */
908 err = do_lookup(nd, &this, &next);
913 inode = next.dentry->d_inode;
917 if (inode->i_op->follow_link) {
918 err = do_follow_link(&next, nd);
922 inode = nd->path.dentry->d_inode;
926 path_to_nameidata(&next, nd);
928 if (!inode->i_op->lookup)
931 /* here ends the main loop */
934 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
936 /* Clear LOOKUP_CONTINUE iff it was previously unset */
937 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
938 if (lookup_flags & LOOKUP_PARENT)
940 if (this.name[0] == '.') switch (this.len) {
944 if (this.name[1] != '.')
947 inode = nd->path.dentry->d_inode;
952 err = do_lookup(nd, &this, &next);
955 inode = next.dentry->d_inode;
956 if (follow_on_final(inode, lookup_flags)) {
957 err = do_follow_link(&next, nd);
960 inode = nd->path.dentry->d_inode;
962 path_to_nameidata(&next, nd);
966 if (lookup_flags & LOOKUP_DIRECTORY) {
968 if (!inode->i_op->lookup)
974 nd->last_type = LAST_NORM;
975 if (this.name[0] != '.')
978 nd->last_type = LAST_DOT;
979 else if (this.len == 2 && this.name[1] == '.')
980 nd->last_type = LAST_DOTDOT;
985 * We bypassed the ordinary revalidation routines.
986 * We may need to check the cached dentry for staleness.
988 if (nd->path.dentry && nd->path.dentry->d_sb &&
989 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
991 /* Note: we do not d_invalidate() */
992 if (!nd->path.dentry->d_op->d_revalidate(
993 nd->path.dentry, nd))
999 path_put_conditional(&next, nd);
1002 path_put(&nd->path);
1007 static int path_walk(const char *name, struct nameidata *nd)
1009 struct path save = nd->path;
1012 current->total_link_count = 0;
1014 /* make sure the stuff we saved doesn't go away */
1017 result = link_path_walk(name, nd);
1018 if (result == -ESTALE) {
1019 /* nd->path had been dropped */
1020 current->total_link_count = 0;
1022 path_get(&nd->path);
1023 nd->flags |= LOOKUP_REVAL;
1024 result = link_path_walk(name, nd);
1032 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1038 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1041 nd->root.mnt = NULL;
1045 nd->path = nd->root;
1046 path_get(&nd->root);
1047 } else if (dfd == AT_FDCWD) {
1048 struct fs_struct *fs = current->fs;
1049 read_lock(&fs->lock);
1052 read_unlock(&fs->lock);
1054 struct dentry *dentry;
1056 file = fget_light(dfd, &fput_needed);
1061 dentry = file->f_path.dentry;
1064 if (!S_ISDIR(dentry->d_inode->i_mode))
1067 retval = file_permission(file, MAY_EXEC);
1071 nd->path = file->f_path;
1072 path_get(&file->f_path);
1074 fput_light(file, fput_needed);
1079 fput_light(file, fput_needed);
1084 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1085 static int do_path_lookup(int dfd, const char *name,
1086 unsigned int flags, struct nameidata *nd)
1088 int retval = path_init(dfd, name, flags, nd);
1090 retval = path_walk(name, nd);
1091 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1092 nd->path.dentry->d_inode))
1093 audit_inode(name, nd->path.dentry);
1095 path_put(&nd->root);
1096 nd->root.mnt = NULL;
1101 int path_lookup(const char *name, unsigned int flags,
1102 struct nameidata *nd)
1104 return do_path_lookup(AT_FDCWD, name, flags, nd);
1107 int kern_path(const char *name, unsigned int flags, struct path *path)
1109 struct nameidata nd;
1110 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1117 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1118 * @dentry: pointer to dentry of the base directory
1119 * @mnt: pointer to vfs mount of the base directory
1120 * @name: pointer to file name
1121 * @flags: lookup flags
1122 * @nd: pointer to nameidata
1124 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1125 const char *name, unsigned int flags,
1126 struct nameidata *nd)
1130 /* same as do_path_lookup */
1131 nd->last_type = LAST_ROOT;
1135 nd->path.dentry = dentry;
1137 path_get(&nd->path);
1138 nd->root = nd->path;
1139 path_get(&nd->root);
1141 retval = path_walk(name, nd);
1142 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1143 nd->path.dentry->d_inode))
1144 audit_inode(name, nd->path.dentry);
1146 path_put(&nd->root);
1147 nd->root.mnt = NULL;
1152 static struct dentry *__lookup_hash(struct qstr *name,
1153 struct dentry *base, struct nameidata *nd)
1155 struct dentry *dentry;
1156 struct inode *inode;
1159 inode = base->d_inode;
1162 * See if the low-level filesystem might want
1163 * to use its own hash..
1165 if (base->d_op && base->d_op->d_hash) {
1166 err = base->d_op->d_hash(base, name);
1167 dentry = ERR_PTR(err);
1172 dentry = __d_lookup(base, name);
1174 /* lockess __d_lookup may fail due to concurrent d_move()
1175 * in some unrelated directory, so try with d_lookup
1178 dentry = d_lookup(base, name);
1180 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1181 dentry = do_revalidate(dentry, nd);
1186 /* Don't create child dentry for a dead directory. */
1187 dentry = ERR_PTR(-ENOENT);
1188 if (IS_DEADDIR(inode))
1191 new = d_alloc(base, name);
1192 dentry = ERR_PTR(-ENOMEM);
1195 dentry = inode->i_op->lookup(inode, new, nd);
1206 * Restricted form of lookup. Doesn't follow links, single-component only,
1207 * needs parent already locked. Doesn't follow mounts.
1210 static struct dentry *lookup_hash(struct nameidata *nd)
1214 err = exec_permission(nd->path.dentry->d_inode);
1216 return ERR_PTR(err);
1217 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1220 static int __lookup_one_len(const char *name, struct qstr *this,
1221 struct dentry *base, int len)
1231 hash = init_name_hash();
1233 c = *(const unsigned char *)name++;
1234 if (c == '/' || c == '\0')
1236 hash = partial_name_hash(c, hash);
1238 this->hash = end_name_hash(hash);
1243 * lookup_one_len - filesystem helper to lookup single pathname component
1244 * @name: pathname component to lookup
1245 * @base: base directory to lookup from
1246 * @len: maximum length @len should be interpreted to
1248 * Note that this routine is purely a helper for filesystem usage and should
1249 * not be called by generic code. Also note that by using this function the
1250 * nameidata argument is passed to the filesystem methods and a filesystem
1251 * using this helper needs to be prepared for that.
1253 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1258 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1260 err = __lookup_one_len(name, &this, base, len);
1262 return ERR_PTR(err);
1264 err = exec_permission(base->d_inode);
1266 return ERR_PTR(err);
1267 return __lookup_hash(&this, base, NULL);
1270 int user_path_at(int dfd, const char __user *name, unsigned flags,
1273 struct nameidata nd;
1274 char *tmp = getname(name);
1275 int err = PTR_ERR(tmp);
1278 BUG_ON(flags & LOOKUP_PARENT);
1280 err = do_path_lookup(dfd, tmp, flags, &nd);
1288 static int user_path_parent(int dfd, const char __user *path,
1289 struct nameidata *nd, char **name)
1291 char *s = getname(path);
1297 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1307 * It's inline, so penalty for filesystems that don't use sticky bit is
1310 static inline int check_sticky(struct inode *dir, struct inode *inode)
1312 uid_t fsuid = current_fsuid();
1314 if (!(dir->i_mode & S_ISVTX))
1316 if (inode->i_uid == fsuid)
1318 if (dir->i_uid == fsuid)
1320 return !capable(CAP_FOWNER);
1324 * Check whether we can remove a link victim from directory dir, check
1325 * whether the type of victim is right.
1326 * 1. We can't do it if dir is read-only (done in permission())
1327 * 2. We should have write and exec permissions on dir
1328 * 3. We can't remove anything from append-only dir
1329 * 4. We can't do anything with immutable dir (done in permission())
1330 * 5. If the sticky bit on dir is set we should either
1331 * a. be owner of dir, or
1332 * b. be owner of victim, or
1333 * c. have CAP_FOWNER capability
1334 * 6. If the victim is append-only or immutable we can't do antyhing with
1335 * links pointing to it.
1336 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1337 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1338 * 9. We can't remove a root or mountpoint.
1339 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1340 * nfs_async_unlink().
1342 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1346 if (!victim->d_inode)
1349 BUG_ON(victim->d_parent->d_inode != dir);
1350 audit_inode_child(victim->d_name.name, victim, dir);
1352 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1357 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1358 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1361 if (!S_ISDIR(victim->d_inode->i_mode))
1363 if (IS_ROOT(victim))
1365 } else if (S_ISDIR(victim->d_inode->i_mode))
1367 if (IS_DEADDIR(dir))
1369 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1374 /* Check whether we can create an object with dentry child in directory
1376 * 1. We can't do it if child already exists (open has special treatment for
1377 * this case, but since we are inlined it's OK)
1378 * 2. We can't do it if dir is read-only (done in permission())
1379 * 3. We should have write and exec permissions on dir
1380 * 4. We can't do it if dir is immutable (done in permission())
1382 static inline int may_create(struct inode *dir, struct dentry *child)
1386 if (IS_DEADDIR(dir))
1388 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1392 * O_DIRECTORY translates into forcing a directory lookup.
1394 static inline int lookup_flags(unsigned int f)
1396 unsigned long retval = LOOKUP_FOLLOW;
1399 retval &= ~LOOKUP_FOLLOW;
1401 if (f & O_DIRECTORY)
1402 retval |= LOOKUP_DIRECTORY;
1408 * p1 and p2 should be directories on the same fs.
1410 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1415 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1419 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1421 p = d_ancestor(p2, p1);
1423 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1424 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1428 p = d_ancestor(p1, p2);
1430 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1431 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1435 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1436 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1440 void unlock_rename(struct dentry *p1, struct dentry *p2)
1442 mutex_unlock(&p1->d_inode->i_mutex);
1444 mutex_unlock(&p2->d_inode->i_mutex);
1445 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1449 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1450 struct nameidata *nd)
1452 int error = may_create(dir, dentry);
1457 if (!dir->i_op->create)
1458 return -EACCES; /* shouldn't it be ENOSYS? */
1461 error = security_inode_create(dir, dentry, mode);
1465 error = dir->i_op->create(dir, dentry, mode, nd);
1467 fsnotify_create(dir, dentry);
1471 int may_open(struct path *path, int acc_mode, int flag)
1473 struct dentry *dentry = path->dentry;
1474 struct inode *inode = dentry->d_inode;
1480 switch (inode->i_mode & S_IFMT) {
1484 if (acc_mode & MAY_WRITE)
1489 if (path->mnt->mnt_flags & MNT_NODEV)
1498 error = inode_permission(inode, acc_mode);
1503 * An append-only file must be opened in append mode for writing.
1505 if (IS_APPEND(inode)) {
1506 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1512 /* O_NOATIME can only be set by the owner or superuser */
1513 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1517 * Ensure there are no outstanding leases on the file.
1519 return break_lease(inode, flag);
1522 static int handle_truncate(struct path *path)
1524 struct inode *inode = path->dentry->d_inode;
1525 int error = get_write_access(inode);
1529 * Refuse to truncate files with mandatory locks held on them.
1531 error = locks_verify_locked(inode);
1533 error = security_path_truncate(path, 0,
1534 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1536 error = do_truncate(path->dentry, 0,
1537 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1540 put_write_access(inode);
1545 * Be careful about ever adding any more callers of this
1546 * function. Its flags must be in the namei format, not
1547 * what get passed to sys_open().
1549 static int __open_namei_create(struct nameidata *nd, struct path *path,
1553 struct dentry *dir = nd->path.dentry;
1555 if (!IS_POSIXACL(dir->d_inode))
1556 mode &= ~current_umask();
1557 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1560 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1562 mutex_unlock(&dir->d_inode->i_mutex);
1563 dput(nd->path.dentry);
1564 nd->path.dentry = path->dentry;
1567 /* Don't check for write permission, don't truncate */
1568 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1572 * Note that while the flag value (low two bits) for sys_open means:
1577 * it is changed into
1578 * 00 - no permissions needed
1579 * 01 - read-permission
1580 * 10 - write-permission
1582 * for the internal routines (ie open_namei()/follow_link() etc)
1583 * This is more logical, and also allows the 00 "no perm needed"
1584 * to be used for symlinks (where the permissions are checked
1588 static inline int open_to_namei_flags(int flag)
1590 if ((flag+1) & O_ACCMODE)
1595 static int open_will_truncate(int flag, struct inode *inode)
1598 * We'll never write to the fs underlying
1601 if (special_file(inode->i_mode))
1603 return (flag & O_TRUNC);
1607 * Note that the low bits of the passed in "open_flag"
1608 * are not the same as in the local variable "flag". See
1609 * open_to_namei_flags() for more details.
1611 struct file *do_filp_open(int dfd, const char *pathname,
1612 int open_flag, int mode, int acc_mode)
1615 struct nameidata nd;
1621 int flag = open_to_namei_flags(open_flag);
1622 int force_reval = 0;
1625 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1626 * check for O_DSYNC if the need any syncing at all we enforce it's
1627 * always set instead of having to deal with possibly weird behaviour
1628 * for malicious applications setting only __O_SYNC.
1630 if (open_flag & __O_SYNC)
1631 open_flag |= O_DSYNC;
1634 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1636 /* O_TRUNC implies we need access checks for write permissions */
1638 acc_mode |= MAY_WRITE;
1640 /* Allow the LSM permission hook to distinguish append
1641 access from general write access. */
1642 if (flag & O_APPEND)
1643 acc_mode |= MAY_APPEND;
1646 * The simplest case - just a plain lookup.
1648 if (!(flag & O_CREAT)) {
1649 filp = get_empty_filp();
1652 return ERR_PTR(-ENFILE);
1653 nd.intent.open.file = filp;
1654 filp->f_flags = open_flag;
1655 nd.intent.open.flags = flag;
1656 nd.intent.open.create_mode = 0;
1657 error = do_path_lookup(dfd, pathname,
1658 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1659 if (IS_ERR(nd.intent.open.file)) {
1661 error = PTR_ERR(nd.intent.open.file);
1665 release_open_intent(&nd);
1667 return ERR_PTR(error);
1672 * Create - we need to know the parent.
1675 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1677 return ERR_PTR(error);
1679 nd.flags |= LOOKUP_REVAL;
1680 error = path_walk(pathname, &nd);
1684 return ERR_PTR(error);
1686 if (unlikely(!audit_dummy_context()))
1687 audit_inode(pathname, nd.path.dentry);
1690 * We have the parent and last component. First of all, check
1691 * that we are not asked to creat(2) an obvious directory - that
1695 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1699 filp = get_empty_filp();
1702 nd.intent.open.file = filp;
1703 filp->f_flags = open_flag;
1704 nd.intent.open.flags = flag;
1705 nd.intent.open.create_mode = mode;
1706 dir = nd.path.dentry;
1707 nd.flags &= ~LOOKUP_PARENT;
1708 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1710 nd.flags |= LOOKUP_EXCL;
1711 mutex_lock(&dir->d_inode->i_mutex);
1712 path.dentry = lookup_hash(&nd);
1713 path.mnt = nd.path.mnt;
1716 error = PTR_ERR(path.dentry);
1717 if (IS_ERR(path.dentry)) {
1718 mutex_unlock(&dir->d_inode->i_mutex);
1722 if (IS_ERR(nd.intent.open.file)) {
1723 error = PTR_ERR(nd.intent.open.file);
1724 goto exit_mutex_unlock;
1727 /* Negative dentry, just create the file */
1728 if (!path.dentry->d_inode) {
1730 * This write is needed to ensure that a
1731 * ro->rw transition does not occur between
1732 * the time when the file is created and when
1733 * a permanent write count is taken through
1734 * the 'struct file' in nameidata_to_filp().
1736 error = mnt_want_write(nd.path.mnt);
1738 goto exit_mutex_unlock;
1739 error = __open_namei_create(&nd, &path, flag, mode);
1741 mnt_drop_write(nd.path.mnt);
1744 filp = nameidata_to_filp(&nd);
1745 mnt_drop_write(nd.path.mnt);
1748 if (!IS_ERR(filp)) {
1749 error = ima_file_check(filp, acc_mode);
1752 filp = ERR_PTR(error);
1759 * It already exists.
1761 mutex_unlock(&dir->d_inode->i_mutex);
1762 audit_inode(pathname, path.dentry);
1768 if (__follow_mount(&path)) {
1770 if (flag & O_NOFOLLOW)
1775 if (!path.dentry->d_inode)
1777 if (path.dentry->d_inode->i_op->follow_link)
1780 path_to_nameidata(&path, &nd);
1782 if (S_ISDIR(path.dentry->d_inode->i_mode))
1787 * 1. may_open() truncates a file
1788 * 2. a rw->ro mount transition occurs
1789 * 3. nameidata_to_filp() fails due to
1791 * That would be inconsistent, and should
1792 * be avoided. Taking this mnt write here
1793 * ensures that (2) can not occur.
1795 will_truncate = open_will_truncate(flag, nd.path.dentry->d_inode);
1796 if (will_truncate) {
1797 error = mnt_want_write(nd.path.mnt);
1801 error = may_open(&nd.path, acc_mode, flag);
1804 mnt_drop_write(nd.path.mnt);
1807 filp = nameidata_to_filp(&nd);
1808 if (!IS_ERR(filp)) {
1809 error = ima_file_check(filp, acc_mode);
1812 filp = ERR_PTR(error);
1815 if (!IS_ERR(filp)) {
1816 if (acc_mode & MAY_WRITE)
1817 vfs_dq_init(nd.path.dentry->d_inode);
1819 if (will_truncate) {
1820 error = handle_truncate(&nd.path);
1823 filp = ERR_PTR(error);
1828 * It is now safe to drop the mnt write
1829 * because the filp has had a write taken
1833 mnt_drop_write(nd.path.mnt);
1839 mutex_unlock(&dir->d_inode->i_mutex);
1841 path_put_conditional(&path, &nd);
1843 if (!IS_ERR(nd.intent.open.file))
1844 release_open_intent(&nd);
1849 return ERR_PTR(error);
1853 if (flag & O_NOFOLLOW)
1856 * This is subtle. Instead of calling do_follow_link() we do the
1857 * thing by hands. The reason is that this way we have zero link_count
1858 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1859 * After that we have the parent and last component, i.e.
1860 * we are in the same situation as after the first path_walk().
1861 * Well, almost - if the last component is normal we get its copy
1862 * stored in nd->last.name and we will have to putname() it when we
1863 * are done. Procfs-like symlinks just set LAST_BIND.
1865 nd.flags |= LOOKUP_PARENT;
1866 error = security_inode_follow_link(path.dentry, &nd);
1869 error = __do_follow_link(&path, &nd);
1872 /* Does someone understand code flow here? Or it is only
1873 * me so stupid? Anathema to whoever designed this non-sense
1874 * with "intent.open".
1876 release_open_intent(&nd);
1879 if (error == -ESTALE && !force_reval) {
1883 return ERR_PTR(error);
1885 nd.flags &= ~LOOKUP_PARENT;
1886 if (nd.last_type == LAST_BIND)
1889 if (nd.last_type != LAST_NORM)
1891 if (nd.last.name[nd.last.len]) {
1892 __putname(nd.last.name);
1897 __putname(nd.last.name);
1900 dir = nd.path.dentry;
1901 mutex_lock(&dir->d_inode->i_mutex);
1902 path.dentry = lookup_hash(&nd);
1903 path.mnt = nd.path.mnt;
1904 __putname(nd.last.name);
1909 * filp_open - open file and return file pointer
1911 * @filename: path to open
1912 * @flags: open flags as per the open(2) second argument
1913 * @mode: mode for the new file if O_CREAT is set, else ignored
1915 * This is the helper to open a file from kernelspace if you really
1916 * have to. But in generally you should not do this, so please move
1917 * along, nothing to see here..
1919 struct file *filp_open(const char *filename, int flags, int mode)
1921 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1923 EXPORT_SYMBOL(filp_open);
1926 * lookup_create - lookup a dentry, creating it if it doesn't exist
1927 * @nd: nameidata info
1928 * @is_dir: directory flag
1930 * Simple function to lookup and return a dentry and create it
1931 * if it doesn't exist. Is SMP-safe.
1933 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1935 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1937 struct dentry *dentry = ERR_PTR(-EEXIST);
1939 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1941 * Yucky last component or no last component at all?
1942 * (foo/., foo/.., /////)
1944 if (nd->last_type != LAST_NORM)
1946 nd->flags &= ~LOOKUP_PARENT;
1947 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1948 nd->intent.open.flags = O_EXCL;
1951 * Do the final lookup.
1953 dentry = lookup_hash(nd);
1957 if (dentry->d_inode)
1960 * Special case - lookup gave negative, but... we had foo/bar/
1961 * From the vfs_mknod() POV we just have a negative dentry -
1962 * all is fine. Let's be bastards - you had / on the end, you've
1963 * been asking for (non-existent) directory. -ENOENT for you.
1965 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1967 dentry = ERR_PTR(-ENOENT);
1972 dentry = ERR_PTR(-EEXIST);
1976 EXPORT_SYMBOL_GPL(lookup_create);
1978 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1980 int error = may_create(dir, dentry);
1985 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1988 if (!dir->i_op->mknod)
1991 error = devcgroup_inode_mknod(mode, dev);
1995 error = security_inode_mknod(dir, dentry, mode, dev);
2000 error = dir->i_op->mknod(dir, dentry, mode, dev);
2002 fsnotify_create(dir, dentry);
2006 static int may_mknod(mode_t mode)
2008 switch (mode & S_IFMT) {
2014 case 0: /* zero mode translates to S_IFREG */
2023 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2028 struct dentry *dentry;
2029 struct nameidata nd;
2034 error = user_path_parent(dfd, filename, &nd, &tmp);
2038 dentry = lookup_create(&nd, 0);
2039 if (IS_ERR(dentry)) {
2040 error = PTR_ERR(dentry);
2043 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2044 mode &= ~current_umask();
2045 error = may_mknod(mode);
2048 error = mnt_want_write(nd.path.mnt);
2051 error = security_path_mknod(&nd.path, dentry, mode, dev);
2053 goto out_drop_write;
2054 switch (mode & S_IFMT) {
2055 case 0: case S_IFREG:
2056 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2058 case S_IFCHR: case S_IFBLK:
2059 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2060 new_decode_dev(dev));
2062 case S_IFIFO: case S_IFSOCK:
2063 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2067 mnt_drop_write(nd.path.mnt);
2071 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2078 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2080 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2083 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2085 int error = may_create(dir, dentry);
2090 if (!dir->i_op->mkdir)
2093 mode &= (S_IRWXUGO|S_ISVTX);
2094 error = security_inode_mkdir(dir, dentry, mode);
2099 error = dir->i_op->mkdir(dir, dentry, mode);
2101 fsnotify_mkdir(dir, dentry);
2105 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2109 struct dentry *dentry;
2110 struct nameidata nd;
2112 error = user_path_parent(dfd, pathname, &nd, &tmp);
2116 dentry = lookup_create(&nd, 1);
2117 error = PTR_ERR(dentry);
2121 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2122 mode &= ~current_umask();
2123 error = mnt_want_write(nd.path.mnt);
2126 error = security_path_mkdir(&nd.path, dentry, mode);
2128 goto out_drop_write;
2129 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2131 mnt_drop_write(nd.path.mnt);
2135 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2142 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2144 return sys_mkdirat(AT_FDCWD, pathname, mode);
2148 * We try to drop the dentry early: we should have
2149 * a usage count of 2 if we're the only user of this
2150 * dentry, and if that is true (possibly after pruning
2151 * the dcache), then we drop the dentry now.
2153 * A low-level filesystem can, if it choses, legally
2156 * if (!d_unhashed(dentry))
2159 * if it cannot handle the case of removing a directory
2160 * that is still in use by something else..
2162 void dentry_unhash(struct dentry *dentry)
2165 shrink_dcache_parent(dentry);
2166 spin_lock(&dcache_lock);
2167 spin_lock(&dentry->d_lock);
2168 if (atomic_read(&dentry->d_count) == 2)
2170 spin_unlock(&dentry->d_lock);
2171 spin_unlock(&dcache_lock);
2174 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2176 int error = may_delete(dir, dentry, 1);
2181 if (!dir->i_op->rmdir)
2186 mutex_lock(&dentry->d_inode->i_mutex);
2187 dentry_unhash(dentry);
2188 if (d_mountpoint(dentry))
2191 error = security_inode_rmdir(dir, dentry);
2193 error = dir->i_op->rmdir(dir, dentry);
2195 dentry->d_inode->i_flags |= S_DEAD;
2198 mutex_unlock(&dentry->d_inode->i_mutex);
2207 static long do_rmdir(int dfd, const char __user *pathname)
2211 struct dentry *dentry;
2212 struct nameidata nd;
2214 error = user_path_parent(dfd, pathname, &nd, &name);
2218 switch(nd.last_type) {
2230 nd.flags &= ~LOOKUP_PARENT;
2232 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2233 dentry = lookup_hash(&nd);
2234 error = PTR_ERR(dentry);
2237 error = mnt_want_write(nd.path.mnt);
2240 error = security_path_rmdir(&nd.path, dentry);
2243 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2245 mnt_drop_write(nd.path.mnt);
2249 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2256 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2258 return do_rmdir(AT_FDCWD, pathname);
2261 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2263 int error = may_delete(dir, dentry, 0);
2268 if (!dir->i_op->unlink)
2273 mutex_lock(&dentry->d_inode->i_mutex);
2274 if (d_mountpoint(dentry))
2277 error = security_inode_unlink(dir, dentry);
2279 error = dir->i_op->unlink(dir, dentry);
2281 mutex_unlock(&dentry->d_inode->i_mutex);
2283 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2284 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2285 fsnotify_link_count(dentry->d_inode);
2293 * Make sure that the actual truncation of the file will occur outside its
2294 * directory's i_mutex. Truncate can take a long time if there is a lot of
2295 * writeout happening, and we don't want to prevent access to the directory
2296 * while waiting on the I/O.
2298 static long do_unlinkat(int dfd, const char __user *pathname)
2302 struct dentry *dentry;
2303 struct nameidata nd;
2304 struct inode *inode = NULL;
2306 error = user_path_parent(dfd, pathname, &nd, &name);
2311 if (nd.last_type != LAST_NORM)
2314 nd.flags &= ~LOOKUP_PARENT;
2316 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2317 dentry = lookup_hash(&nd);
2318 error = PTR_ERR(dentry);
2319 if (!IS_ERR(dentry)) {
2320 /* Why not before? Because we want correct error value */
2321 if (nd.last.name[nd.last.len])
2323 inode = dentry->d_inode;
2325 atomic_inc(&inode->i_count);
2326 error = mnt_want_write(nd.path.mnt);
2329 error = security_path_unlink(&nd.path, dentry);
2332 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2334 mnt_drop_write(nd.path.mnt);
2338 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2340 iput(inode); /* truncate the inode here */
2347 error = !dentry->d_inode ? -ENOENT :
2348 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2352 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2354 if ((flag & ~AT_REMOVEDIR) != 0)
2357 if (flag & AT_REMOVEDIR)
2358 return do_rmdir(dfd, pathname);
2360 return do_unlinkat(dfd, pathname);
2363 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2365 return do_unlinkat(AT_FDCWD, pathname);
2368 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2370 int error = may_create(dir, dentry);
2375 if (!dir->i_op->symlink)
2378 error = security_inode_symlink(dir, dentry, oldname);
2383 error = dir->i_op->symlink(dir, dentry, oldname);
2385 fsnotify_create(dir, dentry);
2389 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2390 int, newdfd, const char __user *, newname)
2395 struct dentry *dentry;
2396 struct nameidata nd;
2398 from = getname(oldname);
2400 return PTR_ERR(from);
2402 error = user_path_parent(newdfd, newname, &nd, &to);
2406 dentry = lookup_create(&nd, 0);
2407 error = PTR_ERR(dentry);
2411 error = mnt_want_write(nd.path.mnt);
2414 error = security_path_symlink(&nd.path, dentry, from);
2416 goto out_drop_write;
2417 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2419 mnt_drop_write(nd.path.mnt);
2423 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2431 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2433 return sys_symlinkat(oldname, AT_FDCWD, newname);
2436 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2438 struct inode *inode = old_dentry->d_inode;
2444 error = may_create(dir, new_dentry);
2448 if (dir->i_sb != inode->i_sb)
2452 * A link to an append-only or immutable file cannot be created.
2454 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2456 if (!dir->i_op->link)
2458 if (S_ISDIR(inode->i_mode))
2461 error = security_inode_link(old_dentry, dir, new_dentry);
2465 mutex_lock(&inode->i_mutex);
2467 error = dir->i_op->link(old_dentry, dir, new_dentry);
2468 mutex_unlock(&inode->i_mutex);
2470 fsnotify_link(dir, inode, new_dentry);
2475 * Hardlinks are often used in delicate situations. We avoid
2476 * security-related surprises by not following symlinks on the
2479 * We don't follow them on the oldname either to be compatible
2480 * with linux 2.0, and to avoid hard-linking to directories
2481 * and other special files. --ADM
2483 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2484 int, newdfd, const char __user *, newname, int, flags)
2486 struct dentry *new_dentry;
2487 struct nameidata nd;
2488 struct path old_path;
2492 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2495 error = user_path_at(olddfd, oldname,
2496 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2501 error = user_path_parent(newdfd, newname, &nd, &to);
2505 if (old_path.mnt != nd.path.mnt)
2507 new_dentry = lookup_create(&nd, 0);
2508 error = PTR_ERR(new_dentry);
2509 if (IS_ERR(new_dentry))
2511 error = mnt_want_write(nd.path.mnt);
2514 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2516 goto out_drop_write;
2517 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2519 mnt_drop_write(nd.path.mnt);
2523 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2528 path_put(&old_path);
2533 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2535 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2539 * The worst of all namespace operations - renaming directory. "Perverted"
2540 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2542 * a) we can get into loop creation. Check is done in is_subdir().
2543 * b) race potential - two innocent renames can create a loop together.
2544 * That's where 4.4 screws up. Current fix: serialization on
2545 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2547 * c) we have to lock _three_ objects - parents and victim (if it exists).
2548 * And that - after we got ->i_mutex on parents (until then we don't know
2549 * whether the target exists). Solution: try to be smart with locking
2550 * order for inodes. We rely on the fact that tree topology may change
2551 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2552 * move will be locked. Thus we can rank directories by the tree
2553 * (ancestors first) and rank all non-directories after them.
2554 * That works since everybody except rename does "lock parent, lookup,
2555 * lock child" and rename is under ->s_vfs_rename_mutex.
2556 * HOWEVER, it relies on the assumption that any object with ->lookup()
2557 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2558 * we'd better make sure that there's no link(2) for them.
2559 * d) some filesystems don't support opened-but-unlinked directories,
2560 * either because of layout or because they are not ready to deal with
2561 * all cases correctly. The latter will be fixed (taking this sort of
2562 * stuff into VFS), but the former is not going away. Solution: the same
2563 * trick as in rmdir().
2564 * e) conversion from fhandle to dentry may come in the wrong moment - when
2565 * we are removing the target. Solution: we will have to grab ->i_mutex
2566 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2567 * ->i_mutex on parents, which works but leads to some truely excessive
2570 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2571 struct inode *new_dir, struct dentry *new_dentry)
2574 struct inode *target;
2577 * If we are going to change the parent - check write permissions,
2578 * we'll need to flip '..'.
2580 if (new_dir != old_dir) {
2581 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2586 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2590 target = new_dentry->d_inode;
2592 mutex_lock(&target->i_mutex);
2593 dentry_unhash(new_dentry);
2595 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2598 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2601 target->i_flags |= S_DEAD;
2602 mutex_unlock(&target->i_mutex);
2603 if (d_unhashed(new_dentry))
2604 d_rehash(new_dentry);
2608 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2609 d_move(old_dentry,new_dentry);
2613 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2614 struct inode *new_dir, struct dentry *new_dentry)
2616 struct inode *target;
2619 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2624 target = new_dentry->d_inode;
2626 mutex_lock(&target->i_mutex);
2627 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2630 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2632 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2633 d_move(old_dentry, new_dentry);
2636 mutex_unlock(&target->i_mutex);
2641 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2642 struct inode *new_dir, struct dentry *new_dentry)
2645 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2646 const char *old_name;
2648 if (old_dentry->d_inode == new_dentry->d_inode)
2651 error = may_delete(old_dir, old_dentry, is_dir);
2655 if (!new_dentry->d_inode)
2656 error = may_create(new_dir, new_dentry);
2658 error = may_delete(new_dir, new_dentry, is_dir);
2662 if (!old_dir->i_op->rename)
2665 vfs_dq_init(old_dir);
2666 vfs_dq_init(new_dir);
2668 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2671 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2673 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2675 const char *new_name = old_dentry->d_name.name;
2676 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2677 new_dentry->d_inode, old_dentry);
2679 fsnotify_oldname_free(old_name);
2684 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2685 int, newdfd, const char __user *, newname)
2687 struct dentry *old_dir, *new_dir;
2688 struct dentry *old_dentry, *new_dentry;
2689 struct dentry *trap;
2690 struct nameidata oldnd, newnd;
2695 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2699 error = user_path_parent(newdfd, newname, &newnd, &to);
2704 if (oldnd.path.mnt != newnd.path.mnt)
2707 old_dir = oldnd.path.dentry;
2709 if (oldnd.last_type != LAST_NORM)
2712 new_dir = newnd.path.dentry;
2713 if (newnd.last_type != LAST_NORM)
2716 oldnd.flags &= ~LOOKUP_PARENT;
2717 newnd.flags &= ~LOOKUP_PARENT;
2718 newnd.flags |= LOOKUP_RENAME_TARGET;
2720 trap = lock_rename(new_dir, old_dir);
2722 old_dentry = lookup_hash(&oldnd);
2723 error = PTR_ERR(old_dentry);
2724 if (IS_ERR(old_dentry))
2726 /* source must exist */
2728 if (!old_dentry->d_inode)
2730 /* unless the source is a directory trailing slashes give -ENOTDIR */
2731 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2733 if (oldnd.last.name[oldnd.last.len])
2735 if (newnd.last.name[newnd.last.len])
2738 /* source should not be ancestor of target */
2740 if (old_dentry == trap)
2742 new_dentry = lookup_hash(&newnd);
2743 error = PTR_ERR(new_dentry);
2744 if (IS_ERR(new_dentry))
2746 /* target should not be an ancestor of source */
2748 if (new_dentry == trap)
2751 error = mnt_want_write(oldnd.path.mnt);
2754 error = security_path_rename(&oldnd.path, old_dentry,
2755 &newnd.path, new_dentry);
2758 error = vfs_rename(old_dir->d_inode, old_dentry,
2759 new_dir->d_inode, new_dentry);
2761 mnt_drop_write(oldnd.path.mnt);
2767 unlock_rename(new_dir, old_dir);
2769 path_put(&newnd.path);
2772 path_put(&oldnd.path);
2778 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2780 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2783 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2787 len = PTR_ERR(link);
2792 if (len > (unsigned) buflen)
2794 if (copy_to_user(buffer, link, len))
2801 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2802 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2803 * using) it for any given inode is up to filesystem.
2805 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2807 struct nameidata nd;
2812 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2814 return PTR_ERR(cookie);
2816 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2817 if (dentry->d_inode->i_op->put_link)
2818 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2822 int vfs_follow_link(struct nameidata *nd, const char *link)
2824 return __vfs_follow_link(nd, link);
2827 /* get the link contents into pagecache */
2828 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2832 struct address_space *mapping = dentry->d_inode->i_mapping;
2833 page = read_mapping_page(mapping, 0, NULL);
2838 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2842 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2844 struct page *page = NULL;
2845 char *s = page_getlink(dentry, &page);
2846 int res = vfs_readlink(dentry,buffer,buflen,s);
2849 page_cache_release(page);
2854 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2856 struct page *page = NULL;
2857 nd_set_link(nd, page_getlink(dentry, &page));
2861 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2863 struct page *page = cookie;
2867 page_cache_release(page);
2872 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2874 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2876 struct address_space *mapping = inode->i_mapping;
2881 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2883 flags |= AOP_FLAG_NOFS;
2886 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2887 flags, &page, &fsdata);
2891 kaddr = kmap_atomic(page, KM_USER0);
2892 memcpy(kaddr, symname, len-1);
2893 kunmap_atomic(kaddr, KM_USER0);
2895 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2902 mark_inode_dirty(inode);
2908 int page_symlink(struct inode *inode, const char *symname, int len)
2910 return __page_symlink(inode, symname, len,
2911 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2914 const struct inode_operations page_symlink_inode_operations = {
2915 .readlink = generic_readlink,
2916 .follow_link = page_follow_link_light,
2917 .put_link = page_put_link,
2920 EXPORT_SYMBOL(user_path_at);
2921 EXPORT_SYMBOL(follow_down);
2922 EXPORT_SYMBOL(follow_up);
2923 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2924 EXPORT_SYMBOL(getname);
2925 EXPORT_SYMBOL(lock_rename);
2926 EXPORT_SYMBOL(lookup_one_len);
2927 EXPORT_SYMBOL(page_follow_link_light);
2928 EXPORT_SYMBOL(page_put_link);
2929 EXPORT_SYMBOL(page_readlink);
2930 EXPORT_SYMBOL(__page_symlink);
2931 EXPORT_SYMBOL(page_symlink);
2932 EXPORT_SYMBOL(page_symlink_inode_operations);
2933 EXPORT_SYMBOL(path_lookup);
2934 EXPORT_SYMBOL(kern_path);
2935 EXPORT_SYMBOL(vfs_path_lookup);
2936 EXPORT_SYMBOL(inode_permission);
2937 EXPORT_SYMBOL(file_permission);
2938 EXPORT_SYMBOL(unlock_rename);
2939 EXPORT_SYMBOL(vfs_create);
2940 EXPORT_SYMBOL(vfs_follow_link);
2941 EXPORT_SYMBOL(vfs_link);
2942 EXPORT_SYMBOL(vfs_mkdir);
2943 EXPORT_SYMBOL(vfs_mknod);
2944 EXPORT_SYMBOL(generic_permission);
2945 EXPORT_SYMBOL(vfs_readlink);
2946 EXPORT_SYMBOL(vfs_rename);
2947 EXPORT_SYMBOL(vfs_rmdir);
2948 EXPORT_SYMBOL(vfs_symlink);
2949 EXPORT_SYMBOL(vfs_unlink);
2950 EXPORT_SYMBOL(dentry_unhash);
2951 EXPORT_SYMBOL(generic_readlink);