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);
827 * This is the basic name resolution function, turning a pathname into
828 * the final dentry. We expect 'base' to be positive and a directory.
830 * Returns 0 and nd will have valid dentry and mnt on success.
831 * Returns error and drops reference to input namei data on failure.
833 static int link_path_walk(const char *name, struct nameidata *nd)
838 unsigned int lookup_flags = nd->flags;
845 inode = nd->path.dentry->d_inode;
847 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
849 /* At this point we know we have a real path component. */
855 nd->flags |= LOOKUP_CONTINUE;
856 err = exec_permission(inode);
861 c = *(const unsigned char *)name;
863 hash = init_name_hash();
866 hash = partial_name_hash(c, hash);
867 c = *(const unsigned char *)name;
868 } while (c && (c != '/'));
869 this.len = name - (const char *) this.name;
870 this.hash = end_name_hash(hash);
872 /* remove trailing slashes? */
875 while (*++name == '/');
877 goto last_with_slashes;
880 * "." and ".." are special - ".." especially so because it has
881 * to be able to know about the current root directory and
882 * parent relationships.
884 if (this.name[0] == '.') switch (this.len) {
888 if (this.name[1] != '.')
891 inode = nd->path.dentry->d_inode;
896 /* This does the actual lookups.. */
897 err = do_lookup(nd, &this, &next);
902 inode = next.dentry->d_inode;
906 if (inode->i_op->follow_link) {
907 err = do_follow_link(&next, nd);
911 inode = nd->path.dentry->d_inode;
915 path_to_nameidata(&next, nd);
917 if (!inode->i_op->lookup)
920 /* here ends the main loop */
923 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
925 /* Clear LOOKUP_CONTINUE iff it was previously unset */
926 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
927 if (lookup_flags & LOOKUP_PARENT)
929 if (this.name[0] == '.') switch (this.len) {
933 if (this.name[1] != '.')
936 inode = nd->path.dentry->d_inode;
941 err = do_lookup(nd, &this, &next);
944 inode = next.dentry->d_inode;
945 if ((lookup_flags & LOOKUP_FOLLOW)
946 && inode && inode->i_op->follow_link) {
947 err = do_follow_link(&next, nd);
950 inode = nd->path.dentry->d_inode;
952 path_to_nameidata(&next, nd);
956 if (lookup_flags & LOOKUP_DIRECTORY) {
958 if (!inode->i_op->lookup)
964 nd->last_type = LAST_NORM;
965 if (this.name[0] != '.')
968 nd->last_type = LAST_DOT;
969 else if (this.len == 2 && this.name[1] == '.')
970 nd->last_type = LAST_DOTDOT;
975 * We bypassed the ordinary revalidation routines.
976 * We may need to check the cached dentry for staleness.
978 if (nd->path.dentry && nd->path.dentry->d_sb &&
979 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
981 /* Note: we do not d_invalidate() */
982 if (!nd->path.dentry->d_op->d_revalidate(
983 nd->path.dentry, nd))
989 path_put_conditional(&next, nd);
997 static int path_walk(const char *name, struct nameidata *nd)
999 struct path save = nd->path;
1002 current->total_link_count = 0;
1004 /* make sure the stuff we saved doesn't go away */
1007 result = link_path_walk(name, nd);
1008 if (result == -ESTALE) {
1009 /* nd->path had been dropped */
1010 current->total_link_count = 0;
1012 path_get(&nd->path);
1013 nd->flags |= LOOKUP_REVAL;
1014 result = link_path_walk(name, nd);
1022 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1028 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1031 nd->root.mnt = NULL;
1035 nd->path = nd->root;
1036 path_get(&nd->root);
1037 } else if (dfd == AT_FDCWD) {
1038 struct fs_struct *fs = current->fs;
1039 read_lock(&fs->lock);
1042 read_unlock(&fs->lock);
1044 struct dentry *dentry;
1046 file = fget_light(dfd, &fput_needed);
1051 dentry = file->f_path.dentry;
1054 if (!S_ISDIR(dentry->d_inode->i_mode))
1057 retval = file_permission(file, MAY_EXEC);
1061 nd->path = file->f_path;
1062 path_get(&file->f_path);
1064 fput_light(file, fput_needed);
1069 fput_light(file, fput_needed);
1074 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1075 static int do_path_lookup(int dfd, const char *name,
1076 unsigned int flags, struct nameidata *nd)
1078 int retval = path_init(dfd, name, flags, nd);
1080 retval = path_walk(name, nd);
1081 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1082 nd->path.dentry->d_inode))
1083 audit_inode(name, nd->path.dentry);
1085 path_put(&nd->root);
1086 nd->root.mnt = NULL;
1091 int path_lookup(const char *name, unsigned int flags,
1092 struct nameidata *nd)
1094 return do_path_lookup(AT_FDCWD, name, flags, nd);
1097 int kern_path(const char *name, unsigned int flags, struct path *path)
1099 struct nameidata nd;
1100 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1107 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1108 * @dentry: pointer to dentry of the base directory
1109 * @mnt: pointer to vfs mount of the base directory
1110 * @name: pointer to file name
1111 * @flags: lookup flags
1112 * @nd: pointer to nameidata
1114 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1115 const char *name, unsigned int flags,
1116 struct nameidata *nd)
1120 /* same as do_path_lookup */
1121 nd->last_type = LAST_ROOT;
1125 nd->path.dentry = dentry;
1127 path_get(&nd->path);
1128 nd->root = nd->path;
1129 path_get(&nd->root);
1131 retval = path_walk(name, nd);
1132 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1133 nd->path.dentry->d_inode))
1134 audit_inode(name, nd->path.dentry);
1136 path_put(&nd->root);
1137 nd->root.mnt = NULL;
1142 static struct dentry *__lookup_hash(struct qstr *name,
1143 struct dentry *base, struct nameidata *nd)
1145 struct dentry *dentry;
1146 struct inode *inode;
1149 inode = base->d_inode;
1152 * See if the low-level filesystem might want
1153 * to use its own hash..
1155 if (base->d_op && base->d_op->d_hash) {
1156 err = base->d_op->d_hash(base, name);
1157 dentry = ERR_PTR(err);
1162 dentry = __d_lookup(base, name);
1164 /* lockess __d_lookup may fail due to concurrent d_move()
1165 * in some unrelated directory, so try with d_lookup
1168 dentry = d_lookup(base, name);
1170 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1171 dentry = do_revalidate(dentry, nd);
1176 /* Don't create child dentry for a dead directory. */
1177 dentry = ERR_PTR(-ENOENT);
1178 if (IS_DEADDIR(inode))
1181 new = d_alloc(base, name);
1182 dentry = ERR_PTR(-ENOMEM);
1185 dentry = inode->i_op->lookup(inode, new, nd);
1196 * Restricted form of lookup. Doesn't follow links, single-component only,
1197 * needs parent already locked. Doesn't follow mounts.
1200 static struct dentry *lookup_hash(struct nameidata *nd)
1204 err = exec_permission(nd->path.dentry->d_inode);
1206 return ERR_PTR(err);
1207 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1210 static int __lookup_one_len(const char *name, struct qstr *this,
1211 struct dentry *base, int len)
1221 hash = init_name_hash();
1223 c = *(const unsigned char *)name++;
1224 if (c == '/' || c == '\0')
1226 hash = partial_name_hash(c, hash);
1228 this->hash = end_name_hash(hash);
1233 * lookup_one_len - filesystem helper to lookup single pathname component
1234 * @name: pathname component to lookup
1235 * @base: base directory to lookup from
1236 * @len: maximum length @len should be interpreted to
1238 * Note that this routine is purely a helper for filesystem usage and should
1239 * not be called by generic code. Also note that by using this function the
1240 * nameidata argument is passed to the filesystem methods and a filesystem
1241 * using this helper needs to be prepared for that.
1243 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1248 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1250 err = __lookup_one_len(name, &this, base, len);
1252 return ERR_PTR(err);
1254 err = exec_permission(base->d_inode);
1256 return ERR_PTR(err);
1257 return __lookup_hash(&this, base, NULL);
1260 int user_path_at(int dfd, const char __user *name, unsigned flags,
1263 struct nameidata nd;
1264 char *tmp = getname(name);
1265 int err = PTR_ERR(tmp);
1268 BUG_ON(flags & LOOKUP_PARENT);
1270 err = do_path_lookup(dfd, tmp, flags, &nd);
1278 static int user_path_parent(int dfd, const char __user *path,
1279 struct nameidata *nd, char **name)
1281 char *s = getname(path);
1287 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1297 * It's inline, so penalty for filesystems that don't use sticky bit is
1300 static inline int check_sticky(struct inode *dir, struct inode *inode)
1302 uid_t fsuid = current_fsuid();
1304 if (!(dir->i_mode & S_ISVTX))
1306 if (inode->i_uid == fsuid)
1308 if (dir->i_uid == fsuid)
1310 return !capable(CAP_FOWNER);
1314 * Check whether we can remove a link victim from directory dir, check
1315 * whether the type of victim is right.
1316 * 1. We can't do it if dir is read-only (done in permission())
1317 * 2. We should have write and exec permissions on dir
1318 * 3. We can't remove anything from append-only dir
1319 * 4. We can't do anything with immutable dir (done in permission())
1320 * 5. If the sticky bit on dir is set we should either
1321 * a. be owner of dir, or
1322 * b. be owner of victim, or
1323 * c. have CAP_FOWNER capability
1324 * 6. If the victim is append-only or immutable we can't do antyhing with
1325 * links pointing to it.
1326 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1327 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1328 * 9. We can't remove a root or mountpoint.
1329 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1330 * nfs_async_unlink().
1332 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1336 if (!victim->d_inode)
1339 BUG_ON(victim->d_parent->d_inode != dir);
1340 audit_inode_child(victim->d_name.name, victim, dir);
1342 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1347 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1348 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1351 if (!S_ISDIR(victim->d_inode->i_mode))
1353 if (IS_ROOT(victim))
1355 } else if (S_ISDIR(victim->d_inode->i_mode))
1357 if (IS_DEADDIR(dir))
1359 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1364 /* Check whether we can create an object with dentry child in directory
1366 * 1. We can't do it if child already exists (open has special treatment for
1367 * this case, but since we are inlined it's OK)
1368 * 2. We can't do it if dir is read-only (done in permission())
1369 * 3. We should have write and exec permissions on dir
1370 * 4. We can't do it if dir is immutable (done in permission())
1372 static inline int may_create(struct inode *dir, struct dentry *child)
1376 if (IS_DEADDIR(dir))
1378 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1382 * O_DIRECTORY translates into forcing a directory lookup.
1384 static inline int lookup_flags(unsigned int f)
1386 unsigned long retval = LOOKUP_FOLLOW;
1389 retval &= ~LOOKUP_FOLLOW;
1391 if (f & O_DIRECTORY)
1392 retval |= LOOKUP_DIRECTORY;
1398 * p1 and p2 should be directories on the same fs.
1400 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1405 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1409 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1411 p = d_ancestor(p2, p1);
1413 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1414 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1418 p = d_ancestor(p1, p2);
1420 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1421 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1425 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1426 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1430 void unlock_rename(struct dentry *p1, struct dentry *p2)
1432 mutex_unlock(&p1->d_inode->i_mutex);
1434 mutex_unlock(&p2->d_inode->i_mutex);
1435 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1439 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1440 struct nameidata *nd)
1442 int error = may_create(dir, dentry);
1447 if (!dir->i_op->create)
1448 return -EACCES; /* shouldn't it be ENOSYS? */
1451 error = security_inode_create(dir, dentry, mode);
1455 error = dir->i_op->create(dir, dentry, mode, nd);
1457 fsnotify_create(dir, dentry);
1461 int may_open(struct path *path, int acc_mode, int flag)
1463 struct dentry *dentry = path->dentry;
1464 struct inode *inode = dentry->d_inode;
1470 switch (inode->i_mode & S_IFMT) {
1474 if (acc_mode & MAY_WRITE)
1479 if (path->mnt->mnt_flags & MNT_NODEV)
1488 error = inode_permission(inode, acc_mode);
1493 * An append-only file must be opened in append mode for writing.
1495 if (IS_APPEND(inode)) {
1496 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1502 /* O_NOATIME can only be set by the owner or superuser */
1503 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1507 * Ensure there are no outstanding leases on the file.
1509 return break_lease(inode, flag);
1512 static int handle_truncate(struct path *path)
1514 struct inode *inode = path->dentry->d_inode;
1515 int error = get_write_access(inode);
1519 * Refuse to truncate files with mandatory locks held on them.
1521 error = locks_verify_locked(inode);
1523 error = security_path_truncate(path, 0,
1524 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1526 error = do_truncate(path->dentry, 0,
1527 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1530 put_write_access(inode);
1535 * Be careful about ever adding any more callers of this
1536 * function. Its flags must be in the namei format, not
1537 * what get passed to sys_open().
1539 static int __open_namei_create(struct nameidata *nd, struct path *path,
1543 struct dentry *dir = nd->path.dentry;
1545 if (!IS_POSIXACL(dir->d_inode))
1546 mode &= ~current_umask();
1547 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1550 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1552 mutex_unlock(&dir->d_inode->i_mutex);
1553 dput(nd->path.dentry);
1554 nd->path.dentry = path->dentry;
1557 /* Don't check for write permission, don't truncate */
1558 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1562 * Note that while the flag value (low two bits) for sys_open means:
1567 * it is changed into
1568 * 00 - no permissions needed
1569 * 01 - read-permission
1570 * 10 - write-permission
1572 * for the internal routines (ie open_namei()/follow_link() etc)
1573 * This is more logical, and also allows the 00 "no perm needed"
1574 * to be used for symlinks (where the permissions are checked
1578 static inline int open_to_namei_flags(int flag)
1580 if ((flag+1) & O_ACCMODE)
1585 static int open_will_truncate(int flag, struct inode *inode)
1588 * We'll never write to the fs underlying
1591 if (special_file(inode->i_mode))
1593 return (flag & O_TRUNC);
1597 * Note that the low bits of the passed in "open_flag"
1598 * are not the same as in the local variable "flag". See
1599 * open_to_namei_flags() for more details.
1601 struct file *do_filp_open(int dfd, const char *pathname,
1602 int open_flag, int mode, int acc_mode)
1605 struct nameidata nd;
1611 int flag = open_to_namei_flags(open_flag);
1612 int force_reval = 0;
1615 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1616 * check for O_DSYNC if the need any syncing at all we enforce it's
1617 * always set instead of having to deal with possibly weird behaviour
1618 * for malicious applications setting only __O_SYNC.
1620 if (open_flag & __O_SYNC)
1621 open_flag |= O_DSYNC;
1624 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1626 /* O_TRUNC implies we need access checks for write permissions */
1628 acc_mode |= MAY_WRITE;
1630 /* Allow the LSM permission hook to distinguish append
1631 access from general write access. */
1632 if (flag & O_APPEND)
1633 acc_mode |= MAY_APPEND;
1636 * The simplest case - just a plain lookup.
1638 if (!(flag & O_CREAT)) {
1639 filp = get_empty_filp();
1642 return ERR_PTR(-ENFILE);
1643 nd.intent.open.file = filp;
1644 filp->f_flags = open_flag;
1645 nd.intent.open.flags = flag;
1646 nd.intent.open.create_mode = 0;
1647 error = do_path_lookup(dfd, pathname,
1648 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1649 if (IS_ERR(nd.intent.open.file)) {
1651 error = PTR_ERR(nd.intent.open.file);
1655 release_open_intent(&nd);
1657 return ERR_PTR(error);
1662 * Create - we need to know the parent.
1665 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1667 return ERR_PTR(error);
1669 nd.flags |= LOOKUP_REVAL;
1670 error = path_walk(pathname, &nd);
1674 return ERR_PTR(error);
1676 if (unlikely(!audit_dummy_context()))
1677 audit_inode(pathname, nd.path.dentry);
1680 * We have the parent and last component. First of all, check
1681 * that we are not asked to creat(2) an obvious directory - that
1685 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1689 filp = get_empty_filp();
1692 nd.intent.open.file = filp;
1693 filp->f_flags = open_flag;
1694 nd.intent.open.flags = flag;
1695 nd.intent.open.create_mode = mode;
1696 dir = nd.path.dentry;
1697 nd.flags &= ~LOOKUP_PARENT;
1698 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1700 nd.flags |= LOOKUP_EXCL;
1701 mutex_lock(&dir->d_inode->i_mutex);
1702 path.dentry = lookup_hash(&nd);
1703 path.mnt = nd.path.mnt;
1706 error = PTR_ERR(path.dentry);
1707 if (IS_ERR(path.dentry)) {
1708 mutex_unlock(&dir->d_inode->i_mutex);
1712 if (IS_ERR(nd.intent.open.file)) {
1713 error = PTR_ERR(nd.intent.open.file);
1714 goto exit_mutex_unlock;
1717 /* Negative dentry, just create the file */
1718 if (!path.dentry->d_inode) {
1720 * This write is needed to ensure that a
1721 * ro->rw transition does not occur between
1722 * the time when the file is created and when
1723 * a permanent write count is taken through
1724 * the 'struct file' in nameidata_to_filp().
1726 error = mnt_want_write(nd.path.mnt);
1728 goto exit_mutex_unlock;
1729 error = __open_namei_create(&nd, &path, flag, mode);
1731 mnt_drop_write(nd.path.mnt);
1734 filp = nameidata_to_filp(&nd);
1735 mnt_drop_write(nd.path.mnt);
1738 if (!IS_ERR(filp)) {
1739 error = ima_path_check(filp, acc_mode);
1742 filp = ERR_PTR(error);
1749 * It already exists.
1751 mutex_unlock(&dir->d_inode->i_mutex);
1752 audit_inode(pathname, path.dentry);
1758 if (__follow_mount(&path)) {
1760 if (flag & O_NOFOLLOW)
1765 if (!path.dentry->d_inode)
1767 if (path.dentry->d_inode->i_op->follow_link)
1770 path_to_nameidata(&path, &nd);
1772 if (S_ISDIR(path.dentry->d_inode->i_mode))
1777 * 1. may_open() truncates a file
1778 * 2. a rw->ro mount transition occurs
1779 * 3. nameidata_to_filp() fails due to
1781 * That would be inconsistent, and should
1782 * be avoided. Taking this mnt write here
1783 * ensures that (2) can not occur.
1785 will_truncate = open_will_truncate(flag, nd.path.dentry->d_inode);
1786 if (will_truncate) {
1787 error = mnt_want_write(nd.path.mnt);
1791 error = may_open(&nd.path, acc_mode, flag);
1794 mnt_drop_write(nd.path.mnt);
1797 filp = nameidata_to_filp(&nd);
1798 if (!IS_ERR(filp)) {
1799 error = ima_path_check(filp, acc_mode);
1802 filp = ERR_PTR(error);
1805 if (!IS_ERR(filp)) {
1806 if (acc_mode & MAY_WRITE)
1807 vfs_dq_init(nd.path.dentry->d_inode);
1809 if (will_truncate) {
1810 error = handle_truncate(&nd.path);
1813 filp = ERR_PTR(error);
1818 * It is now safe to drop the mnt write
1819 * because the filp has had a write taken
1823 mnt_drop_write(nd.path.mnt);
1829 mutex_unlock(&dir->d_inode->i_mutex);
1831 path_put_conditional(&path, &nd);
1833 if (!IS_ERR(nd.intent.open.file))
1834 release_open_intent(&nd);
1839 return ERR_PTR(error);
1843 if (flag & O_NOFOLLOW)
1846 * This is subtle. Instead of calling do_follow_link() we do the
1847 * thing by hands. The reason is that this way we have zero link_count
1848 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1849 * After that we have the parent and last component, i.e.
1850 * we are in the same situation as after the first path_walk().
1851 * Well, almost - if the last component is normal we get its copy
1852 * stored in nd->last.name and we will have to putname() it when we
1853 * are done. Procfs-like symlinks just set LAST_BIND.
1855 nd.flags |= LOOKUP_PARENT;
1856 error = security_inode_follow_link(path.dentry, &nd);
1859 error = __do_follow_link(&path, &nd);
1862 /* Does someone understand code flow here? Or it is only
1863 * me so stupid? Anathema to whoever designed this non-sense
1864 * with "intent.open".
1866 release_open_intent(&nd);
1869 if (error == -ESTALE && !force_reval) {
1873 return ERR_PTR(error);
1875 nd.flags &= ~LOOKUP_PARENT;
1876 if (nd.last_type == LAST_BIND)
1879 if (nd.last_type != LAST_NORM)
1881 if (nd.last.name[nd.last.len]) {
1882 __putname(nd.last.name);
1887 __putname(nd.last.name);
1890 dir = nd.path.dentry;
1891 mutex_lock(&dir->d_inode->i_mutex);
1892 path.dentry = lookup_hash(&nd);
1893 path.mnt = nd.path.mnt;
1894 __putname(nd.last.name);
1899 * filp_open - open file and return file pointer
1901 * @filename: path to open
1902 * @flags: open flags as per the open(2) second argument
1903 * @mode: mode for the new file if O_CREAT is set, else ignored
1905 * This is the helper to open a file from kernelspace if you really
1906 * have to. But in generally you should not do this, so please move
1907 * along, nothing to see here..
1909 struct file *filp_open(const char *filename, int flags, int mode)
1911 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1913 EXPORT_SYMBOL(filp_open);
1916 * lookup_create - lookup a dentry, creating it if it doesn't exist
1917 * @nd: nameidata info
1918 * @is_dir: directory flag
1920 * Simple function to lookup and return a dentry and create it
1921 * if it doesn't exist. Is SMP-safe.
1923 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1925 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1927 struct dentry *dentry = ERR_PTR(-EEXIST);
1929 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1931 * Yucky last component or no last component at all?
1932 * (foo/., foo/.., /////)
1934 if (nd->last_type != LAST_NORM)
1936 nd->flags &= ~LOOKUP_PARENT;
1937 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1938 nd->intent.open.flags = O_EXCL;
1941 * Do the final lookup.
1943 dentry = lookup_hash(nd);
1947 if (dentry->d_inode)
1950 * Special case - lookup gave negative, but... we had foo/bar/
1951 * From the vfs_mknod() POV we just have a negative dentry -
1952 * all is fine. Let's be bastards - you had / on the end, you've
1953 * been asking for (non-existent) directory. -ENOENT for you.
1955 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1957 dentry = ERR_PTR(-ENOENT);
1962 dentry = ERR_PTR(-EEXIST);
1966 EXPORT_SYMBOL_GPL(lookup_create);
1968 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1970 int error = may_create(dir, dentry);
1975 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1978 if (!dir->i_op->mknod)
1981 error = devcgroup_inode_mknod(mode, dev);
1985 error = security_inode_mknod(dir, dentry, mode, dev);
1990 error = dir->i_op->mknod(dir, dentry, mode, dev);
1992 fsnotify_create(dir, dentry);
1996 static int may_mknod(mode_t mode)
1998 switch (mode & S_IFMT) {
2004 case 0: /* zero mode translates to S_IFREG */
2013 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2018 struct dentry *dentry;
2019 struct nameidata nd;
2024 error = user_path_parent(dfd, filename, &nd, &tmp);
2028 dentry = lookup_create(&nd, 0);
2029 if (IS_ERR(dentry)) {
2030 error = PTR_ERR(dentry);
2033 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2034 mode &= ~current_umask();
2035 error = may_mknod(mode);
2038 error = mnt_want_write(nd.path.mnt);
2041 error = security_path_mknod(&nd.path, dentry, mode, dev);
2043 goto out_drop_write;
2044 switch (mode & S_IFMT) {
2045 case 0: case S_IFREG:
2046 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2048 case S_IFCHR: case S_IFBLK:
2049 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2050 new_decode_dev(dev));
2052 case S_IFIFO: case S_IFSOCK:
2053 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2057 mnt_drop_write(nd.path.mnt);
2061 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2068 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2070 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2073 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2075 int error = may_create(dir, dentry);
2080 if (!dir->i_op->mkdir)
2083 mode &= (S_IRWXUGO|S_ISVTX);
2084 error = security_inode_mkdir(dir, dentry, mode);
2089 error = dir->i_op->mkdir(dir, dentry, mode);
2091 fsnotify_mkdir(dir, dentry);
2095 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2099 struct dentry *dentry;
2100 struct nameidata nd;
2102 error = user_path_parent(dfd, pathname, &nd, &tmp);
2106 dentry = lookup_create(&nd, 1);
2107 error = PTR_ERR(dentry);
2111 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2112 mode &= ~current_umask();
2113 error = mnt_want_write(nd.path.mnt);
2116 error = security_path_mkdir(&nd.path, dentry, mode);
2118 goto out_drop_write;
2119 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2121 mnt_drop_write(nd.path.mnt);
2125 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2132 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2134 return sys_mkdirat(AT_FDCWD, pathname, mode);
2138 * We try to drop the dentry early: we should have
2139 * a usage count of 2 if we're the only user of this
2140 * dentry, and if that is true (possibly after pruning
2141 * the dcache), then we drop the dentry now.
2143 * A low-level filesystem can, if it choses, legally
2146 * if (!d_unhashed(dentry))
2149 * if it cannot handle the case of removing a directory
2150 * that is still in use by something else..
2152 void dentry_unhash(struct dentry *dentry)
2155 shrink_dcache_parent(dentry);
2156 spin_lock(&dcache_lock);
2157 spin_lock(&dentry->d_lock);
2158 if (atomic_read(&dentry->d_count) == 2)
2160 spin_unlock(&dentry->d_lock);
2161 spin_unlock(&dcache_lock);
2164 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2166 int error = may_delete(dir, dentry, 1);
2171 if (!dir->i_op->rmdir)
2176 mutex_lock(&dentry->d_inode->i_mutex);
2177 dentry_unhash(dentry);
2178 if (d_mountpoint(dentry))
2181 error = security_inode_rmdir(dir, dentry);
2183 error = dir->i_op->rmdir(dir, dentry);
2185 dentry->d_inode->i_flags |= S_DEAD;
2188 mutex_unlock(&dentry->d_inode->i_mutex);
2197 static long do_rmdir(int dfd, const char __user *pathname)
2201 struct dentry *dentry;
2202 struct nameidata nd;
2204 error = user_path_parent(dfd, pathname, &nd, &name);
2208 switch(nd.last_type) {
2220 nd.flags &= ~LOOKUP_PARENT;
2222 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2223 dentry = lookup_hash(&nd);
2224 error = PTR_ERR(dentry);
2227 error = mnt_want_write(nd.path.mnt);
2230 error = security_path_rmdir(&nd.path, dentry);
2233 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2235 mnt_drop_write(nd.path.mnt);
2239 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2246 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2248 return do_rmdir(AT_FDCWD, pathname);
2251 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2253 int error = may_delete(dir, dentry, 0);
2258 if (!dir->i_op->unlink)
2263 mutex_lock(&dentry->d_inode->i_mutex);
2264 if (d_mountpoint(dentry))
2267 error = security_inode_unlink(dir, dentry);
2269 error = dir->i_op->unlink(dir, dentry);
2271 mutex_unlock(&dentry->d_inode->i_mutex);
2273 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2274 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2275 fsnotify_link_count(dentry->d_inode);
2283 * Make sure that the actual truncation of the file will occur outside its
2284 * directory's i_mutex. Truncate can take a long time if there is a lot of
2285 * writeout happening, and we don't want to prevent access to the directory
2286 * while waiting on the I/O.
2288 static long do_unlinkat(int dfd, const char __user *pathname)
2292 struct dentry *dentry;
2293 struct nameidata nd;
2294 struct inode *inode = NULL;
2296 error = user_path_parent(dfd, pathname, &nd, &name);
2301 if (nd.last_type != LAST_NORM)
2304 nd.flags &= ~LOOKUP_PARENT;
2306 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2307 dentry = lookup_hash(&nd);
2308 error = PTR_ERR(dentry);
2309 if (!IS_ERR(dentry)) {
2310 /* Why not before? Because we want correct error value */
2311 if (nd.last.name[nd.last.len])
2313 inode = dentry->d_inode;
2315 atomic_inc(&inode->i_count);
2316 error = mnt_want_write(nd.path.mnt);
2319 error = security_path_unlink(&nd.path, dentry);
2322 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2324 mnt_drop_write(nd.path.mnt);
2328 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2330 iput(inode); /* truncate the inode here */
2337 error = !dentry->d_inode ? -ENOENT :
2338 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2342 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2344 if ((flag & ~AT_REMOVEDIR) != 0)
2347 if (flag & AT_REMOVEDIR)
2348 return do_rmdir(dfd, pathname);
2350 return do_unlinkat(dfd, pathname);
2353 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2355 return do_unlinkat(AT_FDCWD, pathname);
2358 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2360 int error = may_create(dir, dentry);
2365 if (!dir->i_op->symlink)
2368 error = security_inode_symlink(dir, dentry, oldname);
2373 error = dir->i_op->symlink(dir, dentry, oldname);
2375 fsnotify_create(dir, dentry);
2379 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2380 int, newdfd, const char __user *, newname)
2385 struct dentry *dentry;
2386 struct nameidata nd;
2388 from = getname(oldname);
2390 return PTR_ERR(from);
2392 error = user_path_parent(newdfd, newname, &nd, &to);
2396 dentry = lookup_create(&nd, 0);
2397 error = PTR_ERR(dentry);
2401 error = mnt_want_write(nd.path.mnt);
2404 error = security_path_symlink(&nd.path, dentry, from);
2406 goto out_drop_write;
2407 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2409 mnt_drop_write(nd.path.mnt);
2413 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2421 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2423 return sys_symlinkat(oldname, AT_FDCWD, newname);
2426 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2428 struct inode *inode = old_dentry->d_inode;
2434 error = may_create(dir, new_dentry);
2438 if (dir->i_sb != inode->i_sb)
2442 * A link to an append-only or immutable file cannot be created.
2444 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2446 if (!dir->i_op->link)
2448 if (S_ISDIR(inode->i_mode))
2451 error = security_inode_link(old_dentry, dir, new_dentry);
2455 mutex_lock(&inode->i_mutex);
2457 error = dir->i_op->link(old_dentry, dir, new_dentry);
2458 mutex_unlock(&inode->i_mutex);
2460 fsnotify_link(dir, inode, new_dentry);
2465 * Hardlinks are often used in delicate situations. We avoid
2466 * security-related surprises by not following symlinks on the
2469 * We don't follow them on the oldname either to be compatible
2470 * with linux 2.0, and to avoid hard-linking to directories
2471 * and other special files. --ADM
2473 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2474 int, newdfd, const char __user *, newname, int, flags)
2476 struct dentry *new_dentry;
2477 struct nameidata nd;
2478 struct path old_path;
2482 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2485 error = user_path_at(olddfd, oldname,
2486 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2491 error = user_path_parent(newdfd, newname, &nd, &to);
2495 if (old_path.mnt != nd.path.mnt)
2497 new_dentry = lookup_create(&nd, 0);
2498 error = PTR_ERR(new_dentry);
2499 if (IS_ERR(new_dentry))
2501 error = mnt_want_write(nd.path.mnt);
2504 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2506 goto out_drop_write;
2507 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2509 mnt_drop_write(nd.path.mnt);
2513 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2518 path_put(&old_path);
2523 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2525 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2529 * The worst of all namespace operations - renaming directory. "Perverted"
2530 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2532 * a) we can get into loop creation. Check is done in is_subdir().
2533 * b) race potential - two innocent renames can create a loop together.
2534 * That's where 4.4 screws up. Current fix: serialization on
2535 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2537 * c) we have to lock _three_ objects - parents and victim (if it exists).
2538 * And that - after we got ->i_mutex on parents (until then we don't know
2539 * whether the target exists). Solution: try to be smart with locking
2540 * order for inodes. We rely on the fact that tree topology may change
2541 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2542 * move will be locked. Thus we can rank directories by the tree
2543 * (ancestors first) and rank all non-directories after them.
2544 * That works since everybody except rename does "lock parent, lookup,
2545 * lock child" and rename is under ->s_vfs_rename_mutex.
2546 * HOWEVER, it relies on the assumption that any object with ->lookup()
2547 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2548 * we'd better make sure that there's no link(2) for them.
2549 * d) some filesystems don't support opened-but-unlinked directories,
2550 * either because of layout or because they are not ready to deal with
2551 * all cases correctly. The latter will be fixed (taking this sort of
2552 * stuff into VFS), but the former is not going away. Solution: the same
2553 * trick as in rmdir().
2554 * e) conversion from fhandle to dentry may come in the wrong moment - when
2555 * we are removing the target. Solution: we will have to grab ->i_mutex
2556 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2557 * ->i_mutex on parents, which works but leads to some truely excessive
2560 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2561 struct inode *new_dir, struct dentry *new_dentry)
2564 struct inode *target;
2567 * If we are going to change the parent - check write permissions,
2568 * we'll need to flip '..'.
2570 if (new_dir != old_dir) {
2571 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2576 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2580 target = new_dentry->d_inode;
2582 mutex_lock(&target->i_mutex);
2583 dentry_unhash(new_dentry);
2585 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2588 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2591 target->i_flags |= S_DEAD;
2592 mutex_unlock(&target->i_mutex);
2593 if (d_unhashed(new_dentry))
2594 d_rehash(new_dentry);
2598 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2599 d_move(old_dentry,new_dentry);
2603 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2604 struct inode *new_dir, struct dentry *new_dentry)
2606 struct inode *target;
2609 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2614 target = new_dentry->d_inode;
2616 mutex_lock(&target->i_mutex);
2617 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2620 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2622 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2623 d_move(old_dentry, new_dentry);
2626 mutex_unlock(&target->i_mutex);
2631 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2632 struct inode *new_dir, struct dentry *new_dentry)
2635 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2636 const char *old_name;
2638 if (old_dentry->d_inode == new_dentry->d_inode)
2641 error = may_delete(old_dir, old_dentry, is_dir);
2645 if (!new_dentry->d_inode)
2646 error = may_create(new_dir, new_dentry);
2648 error = may_delete(new_dir, new_dentry, is_dir);
2652 if (!old_dir->i_op->rename)
2655 vfs_dq_init(old_dir);
2656 vfs_dq_init(new_dir);
2658 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2661 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2663 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2665 const char *new_name = old_dentry->d_name.name;
2666 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2667 new_dentry->d_inode, old_dentry);
2669 fsnotify_oldname_free(old_name);
2674 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2675 int, newdfd, const char __user *, newname)
2677 struct dentry *old_dir, *new_dir;
2678 struct dentry *old_dentry, *new_dentry;
2679 struct dentry *trap;
2680 struct nameidata oldnd, newnd;
2685 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2689 error = user_path_parent(newdfd, newname, &newnd, &to);
2694 if (oldnd.path.mnt != newnd.path.mnt)
2697 old_dir = oldnd.path.dentry;
2699 if (oldnd.last_type != LAST_NORM)
2702 new_dir = newnd.path.dentry;
2703 if (newnd.last_type != LAST_NORM)
2706 oldnd.flags &= ~LOOKUP_PARENT;
2707 newnd.flags &= ~LOOKUP_PARENT;
2708 newnd.flags |= LOOKUP_RENAME_TARGET;
2710 trap = lock_rename(new_dir, old_dir);
2712 old_dentry = lookup_hash(&oldnd);
2713 error = PTR_ERR(old_dentry);
2714 if (IS_ERR(old_dentry))
2716 /* source must exist */
2718 if (!old_dentry->d_inode)
2720 /* unless the source is a directory trailing slashes give -ENOTDIR */
2721 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2723 if (oldnd.last.name[oldnd.last.len])
2725 if (newnd.last.name[newnd.last.len])
2728 /* source should not be ancestor of target */
2730 if (old_dentry == trap)
2732 new_dentry = lookup_hash(&newnd);
2733 error = PTR_ERR(new_dentry);
2734 if (IS_ERR(new_dentry))
2736 /* target should not be an ancestor of source */
2738 if (new_dentry == trap)
2741 error = mnt_want_write(oldnd.path.mnt);
2744 error = security_path_rename(&oldnd.path, old_dentry,
2745 &newnd.path, new_dentry);
2748 error = vfs_rename(old_dir->d_inode, old_dentry,
2749 new_dir->d_inode, new_dentry);
2751 mnt_drop_write(oldnd.path.mnt);
2757 unlock_rename(new_dir, old_dir);
2759 path_put(&newnd.path);
2762 path_put(&oldnd.path);
2768 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2770 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2773 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2777 len = PTR_ERR(link);
2782 if (len > (unsigned) buflen)
2784 if (copy_to_user(buffer, link, len))
2791 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2792 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2793 * using) it for any given inode is up to filesystem.
2795 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2797 struct nameidata nd;
2802 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2804 return PTR_ERR(cookie);
2806 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2807 if (dentry->d_inode->i_op->put_link)
2808 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2812 int vfs_follow_link(struct nameidata *nd, const char *link)
2814 return __vfs_follow_link(nd, link);
2817 /* get the link contents into pagecache */
2818 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2822 struct address_space *mapping = dentry->d_inode->i_mapping;
2823 page = read_mapping_page(mapping, 0, NULL);
2828 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2832 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2834 struct page *page = NULL;
2835 char *s = page_getlink(dentry, &page);
2836 int res = vfs_readlink(dentry,buffer,buflen,s);
2839 page_cache_release(page);
2844 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2846 struct page *page = NULL;
2847 nd_set_link(nd, page_getlink(dentry, &page));
2851 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2853 struct page *page = cookie;
2857 page_cache_release(page);
2862 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2864 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2866 struct address_space *mapping = inode->i_mapping;
2871 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2873 flags |= AOP_FLAG_NOFS;
2876 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2877 flags, &page, &fsdata);
2881 kaddr = kmap_atomic(page, KM_USER0);
2882 memcpy(kaddr, symname, len-1);
2883 kunmap_atomic(kaddr, KM_USER0);
2885 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2892 mark_inode_dirty(inode);
2898 int page_symlink(struct inode *inode, const char *symname, int len)
2900 return __page_symlink(inode, symname, len,
2901 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2904 const struct inode_operations page_symlink_inode_operations = {
2905 .readlink = generic_readlink,
2906 .follow_link = page_follow_link_light,
2907 .put_link = page_put_link,
2910 EXPORT_SYMBOL(user_path_at);
2911 EXPORT_SYMBOL(follow_down);
2912 EXPORT_SYMBOL(follow_up);
2913 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2914 EXPORT_SYMBOL(getname);
2915 EXPORT_SYMBOL(lock_rename);
2916 EXPORT_SYMBOL(lookup_one_len);
2917 EXPORT_SYMBOL(page_follow_link_light);
2918 EXPORT_SYMBOL(page_put_link);
2919 EXPORT_SYMBOL(page_readlink);
2920 EXPORT_SYMBOL(__page_symlink);
2921 EXPORT_SYMBOL(page_symlink);
2922 EXPORT_SYMBOL(page_symlink_inode_operations);
2923 EXPORT_SYMBOL(path_lookup);
2924 EXPORT_SYMBOL(kern_path);
2925 EXPORT_SYMBOL(vfs_path_lookup);
2926 EXPORT_SYMBOL(inode_permission);
2927 EXPORT_SYMBOL(file_permission);
2928 EXPORT_SYMBOL(unlock_rename);
2929 EXPORT_SYMBOL(vfs_create);
2930 EXPORT_SYMBOL(vfs_follow_link);
2931 EXPORT_SYMBOL(vfs_link);
2932 EXPORT_SYMBOL(vfs_mkdir);
2933 EXPORT_SYMBOL(vfs_mknod);
2934 EXPORT_SYMBOL(generic_permission);
2935 EXPORT_SYMBOL(vfs_readlink);
2936 EXPORT_SYMBOL(vfs_rename);
2937 EXPORT_SYMBOL(vfs_rmdir);
2938 EXPORT_SYMBOL(vfs_symlink);
2939 EXPORT_SYMBOL(vfs_unlink);
2940 EXPORT_SYMBOL(dentry_unhash);
2941 EXPORT_SYMBOL(generic_readlink);