4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 char * getname(const char __user * filename)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(!audit_dummy_context()))
167 EXPORT_SYMBOL(putname);
171 * This does basic POSIX ACL permission checking
173 static int acl_permission_check(struct inode *inode, int mask,
174 int (*check_acl)(struct inode *inode, int mask))
176 umode_t mode = inode->i_mode;
178 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
180 if (current_fsuid() == inode->i_uid)
183 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
184 int error = check_acl(inode, mask);
185 if (error != -EAGAIN)
189 if (in_group_p(inode->i_gid))
194 * If the DACs are ok we don't need any capability check.
196 if ((mask & ~mode) == 0)
202 * generic_permission - check for access rights on a Posix-like filesystem
203 * @inode: inode to check access rights for
204 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
205 * @check_acl: optional callback to check for Posix ACLs
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things..
212 int generic_permission(struct inode *inode, int mask,
213 int (*check_acl)(struct inode *inode, int mask))
218 * Do the basic POSIX ACL permission checks.
220 ret = acl_permission_check(inode, mask, check_acl);
225 * Read/write DACs are always overridable.
226 * Executable DACs are overridable if at least one exec bit is set.
228 if (!(mask & MAY_EXEC) || execute_ok(inode))
229 if (capable(CAP_DAC_OVERRIDE))
233 * Searching includes executable on directories, else just read.
235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
281 retval = devcgroup_inode_permission(inode, mask);
285 return security_inode_permission(inode,
286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
294 * Used to check for read/write/execute permissions on an already opened
298 * Do not use this function in new code. All access checks should
299 * be done using inode_permission().
301 int file_permission(struct file *file, int mask)
303 return inode_permission(file->f_path.dentry->d_inode, mask);
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
324 int get_write_access(struct inode * inode)
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
337 int deny_write_access(struct file * file)
339 struct inode *inode = file->f_path.dentry->d_inode;
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
353 * path_get - get a reference to a path
354 * @path: path to get the reference to
356 * Given a path increment the reference count to the dentry and the vfsmount.
358 void path_get(struct path *path)
363 EXPORT_SYMBOL(path_get);
366 * path_put - put a reference to a path
367 * @path: path to put the reference to
369 * Given a path decrement the reference count to the dentry and the vfsmount.
371 void path_put(struct path *path)
376 EXPORT_SYMBOL(path_put);
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
382 void release_open_intent(struct nameidata *nd)
384 if (nd->intent.open.file->f_path.dentry == NULL)
385 put_filp(nd->intent.open.file);
387 fput(nd->intent.open.file);
390 static inline struct dentry *
391 do_revalidate(struct dentry *dentry, struct nameidata *nd)
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
402 if (!d_invalidate(dentry)) {
408 dentry = ERR_PTR(status);
415 * Short-cut version of permission(), for calling on directories
416 * during pathname resolution. Combines parts of permission()
417 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
419 * If appropriate, check DAC only. If not appropriate, or
420 * short-cut DAC fails, then call ->permission() to do more
421 * complete permission check.
423 static int exec_permission(struct inode *inode)
427 if (inode->i_op->permission) {
428 ret = inode->i_op->permission(inode, MAY_EXEC);
433 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
437 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
442 return security_inode_permission(inode, MAY_EXEC);
445 static __always_inline void set_root(struct nameidata *nd)
448 struct fs_struct *fs = current->fs;
449 read_lock(&fs->lock);
452 read_unlock(&fs->lock);
456 static int link_path_walk(const char *, struct nameidata *);
458 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
472 res = link_path_walk(link, nd);
473 if (nd->depth || res || nd->last_type!=LAST_NORM)
476 * If it is an iterative symlinks resolution in open_namei() we
477 * have to copy the last component. And all that crap because of
478 * bloody create() on broken symlinks. Furrfu...
481 if (unlikely(!name)) {
485 strcpy(name, nd->last.name);
486 nd->last.name = name;
490 return PTR_ERR(link);
493 static void path_put_conditional(struct path *path, struct nameidata *nd)
496 if (path->mnt != nd->path.mnt)
500 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
502 dput(nd->path.dentry);
503 if (nd->path.mnt != path->mnt)
504 mntput(nd->path.mnt);
505 nd->path.mnt = path->mnt;
506 nd->path.dentry = path->dentry;
509 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
513 struct dentry *dentry = path->dentry;
515 touch_atime(path->mnt, dentry);
516 nd_set_link(nd, NULL);
518 if (path->mnt != nd->path.mnt) {
519 path_to_nameidata(path, nd);
523 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
524 error = PTR_ERR(cookie);
525 if (!IS_ERR(cookie)) {
526 char *s = nd_get_link(nd);
529 error = __vfs_follow_link(nd, s);
530 if (dentry->d_inode->i_op->put_link)
531 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
537 * This limits recursive symlink follows to 8, while
538 * limiting consecutive symlinks to 40.
540 * Without that kind of total limit, nasty chains of consecutive
541 * symlinks can cause almost arbitrarily long lookups.
543 static inline int do_follow_link(struct path *path, struct nameidata *nd)
546 if (current->link_count >= MAX_NESTED_LINKS)
548 if (current->total_link_count >= 40)
550 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
552 err = security_inode_follow_link(path->dentry, nd);
555 current->link_count++;
556 current->total_link_count++;
558 err = __do_follow_link(path, nd);
560 current->link_count--;
564 path_put_conditional(path, nd);
569 int follow_up(struct path *path)
571 struct vfsmount *parent;
572 struct dentry *mountpoint;
573 spin_lock(&vfsmount_lock);
574 parent = path->mnt->mnt_parent;
575 if (parent == path->mnt) {
576 spin_unlock(&vfsmount_lock);
580 mountpoint = dget(path->mnt->mnt_mountpoint);
581 spin_unlock(&vfsmount_lock);
583 path->dentry = mountpoint;
589 /* no need for dcache_lock, as serialization is taken care in
592 static int __follow_mount(struct path *path)
595 while (d_mountpoint(path->dentry)) {
596 struct vfsmount *mounted = lookup_mnt(path);
603 path->dentry = dget(mounted->mnt_root);
609 static void follow_mount(struct path *path)
611 while (d_mountpoint(path->dentry)) {
612 struct vfsmount *mounted = lookup_mnt(path);
618 path->dentry = dget(mounted->mnt_root);
622 /* no need for dcache_lock, as serialization is taken care in
625 int follow_down(struct path *path)
627 struct vfsmount *mounted;
629 mounted = lookup_mnt(path);
634 path->dentry = dget(mounted->mnt_root);
640 static __always_inline void follow_dotdot(struct nameidata *nd)
645 struct vfsmount *parent;
646 struct dentry *old = nd->path.dentry;
648 if (nd->path.dentry == nd->root.dentry &&
649 nd->path.mnt == nd->root.mnt) {
652 spin_lock(&dcache_lock);
653 if (nd->path.dentry != nd->path.mnt->mnt_root) {
654 nd->path.dentry = dget(nd->path.dentry->d_parent);
655 spin_unlock(&dcache_lock);
659 spin_unlock(&dcache_lock);
660 spin_lock(&vfsmount_lock);
661 parent = nd->path.mnt->mnt_parent;
662 if (parent == nd->path.mnt) {
663 spin_unlock(&vfsmount_lock);
667 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
668 spin_unlock(&vfsmount_lock);
670 mntput(nd->path.mnt);
671 nd->path.mnt = parent;
673 follow_mount(&nd->path);
677 * It's more convoluted than I'd like it to be, but... it's still fairly
678 * small and for now I'd prefer to have fast path as straight as possible.
679 * It _is_ time-critical.
681 static int do_lookup(struct nameidata *nd, struct qstr *name,
684 struct vfsmount *mnt = nd->path.mnt;
685 struct dentry *dentry, *parent;
688 * See if the low-level filesystem might want
689 * to use its own hash..
691 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
692 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
697 dentry = __d_lookup(nd->path.dentry, name);
700 if (dentry->d_op && dentry->d_op->d_revalidate)
701 goto need_revalidate;
704 path->dentry = dentry;
705 __follow_mount(path);
709 parent = nd->path.dentry;
710 dir = parent->d_inode;
712 mutex_lock(&dir->i_mutex);
714 * First re-do the cached lookup just in case it was created
715 * while we waited for the directory semaphore..
717 * FIXME! This could use version numbering or similar to
718 * avoid unnecessary cache lookups.
720 * The "dcache_lock" is purely to protect the RCU list walker
721 * from concurrent renames at this point (we mustn't get false
722 * negatives from the RCU list walk here, unlike the optimistic
725 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
727 dentry = d_lookup(parent, name);
731 /* Don't create child dentry for a dead directory. */
732 dentry = ERR_PTR(-ENOENT);
736 new = d_alloc(parent, name);
737 dentry = ERR_PTR(-ENOMEM);
739 dentry = dir->i_op->lookup(dir, new, nd);
746 mutex_unlock(&dir->i_mutex);
753 * Uhhuh! Nasty case: the cache was re-populated while
754 * we waited on the semaphore. Need to revalidate.
756 mutex_unlock(&dir->i_mutex);
757 if (dentry->d_op && dentry->d_op->d_revalidate) {
758 dentry = do_revalidate(dentry, nd);
760 dentry = ERR_PTR(-ENOENT);
767 dentry = do_revalidate(dentry, nd);
775 return PTR_ERR(dentry);
780 * This is the basic name resolution function, turning a pathname into
781 * the final dentry. We expect 'base' to be positive and a directory.
783 * Returns 0 and nd will have valid dentry and mnt on success.
784 * Returns error and drops reference to input namei data on failure.
786 static int link_path_walk(const char *name, struct nameidata *nd)
791 unsigned int lookup_flags = nd->flags;
798 inode = nd->path.dentry->d_inode;
800 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
802 /* At this point we know we have a real path component. */
808 nd->flags |= LOOKUP_CONTINUE;
809 err = exec_permission(inode);
814 c = *(const unsigned char *)name;
816 hash = init_name_hash();
819 hash = partial_name_hash(c, hash);
820 c = *(const unsigned char *)name;
821 } while (c && (c != '/'));
822 this.len = name - (const char *) this.name;
823 this.hash = end_name_hash(hash);
825 /* remove trailing slashes? */
828 while (*++name == '/');
830 goto last_with_slashes;
833 * "." and ".." are special - ".." especially so because it has
834 * to be able to know about the current root directory and
835 * parent relationships.
837 if (this.name[0] == '.') switch (this.len) {
841 if (this.name[1] != '.')
844 inode = nd->path.dentry->d_inode;
849 /* This does the actual lookups.. */
850 err = do_lookup(nd, &this, &next);
855 inode = next.dentry->d_inode;
859 if (inode->i_op->follow_link) {
860 err = do_follow_link(&next, nd);
864 inode = nd->path.dentry->d_inode;
868 path_to_nameidata(&next, nd);
870 if (!inode->i_op->lookup)
873 /* here ends the main loop */
876 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
878 /* Clear LOOKUP_CONTINUE iff it was previously unset */
879 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
880 if (lookup_flags & LOOKUP_PARENT)
882 if (this.name[0] == '.') switch (this.len) {
886 if (this.name[1] != '.')
889 inode = nd->path.dentry->d_inode;
894 err = do_lookup(nd, &this, &next);
897 inode = next.dentry->d_inode;
898 if ((lookup_flags & LOOKUP_FOLLOW)
899 && inode && inode->i_op->follow_link) {
900 err = do_follow_link(&next, nd);
903 inode = nd->path.dentry->d_inode;
905 path_to_nameidata(&next, nd);
909 if (lookup_flags & LOOKUP_DIRECTORY) {
911 if (!inode->i_op->lookup)
917 nd->last_type = LAST_NORM;
918 if (this.name[0] != '.')
921 nd->last_type = LAST_DOT;
922 else if (this.len == 2 && this.name[1] == '.')
923 nd->last_type = LAST_DOTDOT;
928 * We bypassed the ordinary revalidation routines.
929 * We may need to check the cached dentry for staleness.
931 if (nd->path.dentry && nd->path.dentry->d_sb &&
932 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
934 /* Note: we do not d_invalidate() */
935 if (!nd->path.dentry->d_op->d_revalidate(
936 nd->path.dentry, nd))
942 path_put_conditional(&next, nd);
950 static int path_walk(const char *name, struct nameidata *nd)
952 struct path save = nd->path;
955 current->total_link_count = 0;
957 /* make sure the stuff we saved doesn't go away */
960 result = link_path_walk(name, nd);
961 if (result == -ESTALE) {
962 /* nd->path had been dropped */
963 current->total_link_count = 0;
966 nd->flags |= LOOKUP_REVAL;
967 result = link_path_walk(name, nd);
975 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
981 nd->last_type = LAST_ROOT; /* if there are only slashes... */
990 } else if (dfd == AT_FDCWD) {
991 struct fs_struct *fs = current->fs;
992 read_lock(&fs->lock);
995 read_unlock(&fs->lock);
997 struct dentry *dentry;
999 file = fget_light(dfd, &fput_needed);
1004 dentry = file->f_path.dentry;
1007 if (!S_ISDIR(dentry->d_inode->i_mode))
1010 retval = file_permission(file, MAY_EXEC);
1014 nd->path = file->f_path;
1015 path_get(&file->f_path);
1017 fput_light(file, fput_needed);
1022 fput_light(file, fput_needed);
1027 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1028 static int do_path_lookup(int dfd, const char *name,
1029 unsigned int flags, struct nameidata *nd)
1031 int retval = path_init(dfd, name, flags, nd);
1033 retval = path_walk(name, nd);
1034 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1035 nd->path.dentry->d_inode))
1036 audit_inode(name, nd->path.dentry);
1038 path_put(&nd->root);
1039 nd->root.mnt = NULL;
1044 int path_lookup(const char *name, unsigned int flags,
1045 struct nameidata *nd)
1047 return do_path_lookup(AT_FDCWD, name, flags, nd);
1050 int kern_path(const char *name, unsigned int flags, struct path *path)
1052 struct nameidata nd;
1053 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1060 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1061 * @dentry: pointer to dentry of the base directory
1062 * @mnt: pointer to vfs mount of the base directory
1063 * @name: pointer to file name
1064 * @flags: lookup flags
1065 * @nd: pointer to nameidata
1067 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1068 const char *name, unsigned int flags,
1069 struct nameidata *nd)
1073 /* same as do_path_lookup */
1074 nd->last_type = LAST_ROOT;
1078 nd->path.dentry = dentry;
1080 path_get(&nd->path);
1081 nd->root = nd->path;
1082 path_get(&nd->root);
1084 retval = path_walk(name, nd);
1085 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1086 nd->path.dentry->d_inode))
1087 audit_inode(name, nd->path.dentry);
1089 path_put(&nd->root);
1090 nd->root.mnt = NULL;
1095 static struct dentry *__lookup_hash(struct qstr *name,
1096 struct dentry *base, struct nameidata *nd)
1098 struct dentry *dentry;
1099 struct inode *inode;
1102 inode = base->d_inode;
1105 * See if the low-level filesystem might want
1106 * to use its own hash..
1108 if (base->d_op && base->d_op->d_hash) {
1109 err = base->d_op->d_hash(base, name);
1110 dentry = ERR_PTR(err);
1115 dentry = __d_lookup(base, name);
1117 /* lockess __d_lookup may fail due to concurrent d_move()
1118 * in some unrelated directory, so try with d_lookup
1121 dentry = d_lookup(base, name);
1123 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1124 dentry = do_revalidate(dentry, nd);
1129 /* Don't create child dentry for a dead directory. */
1130 dentry = ERR_PTR(-ENOENT);
1131 if (IS_DEADDIR(inode))
1134 new = d_alloc(base, name);
1135 dentry = ERR_PTR(-ENOMEM);
1138 dentry = inode->i_op->lookup(inode, new, nd);
1149 * Restricted form of lookup. Doesn't follow links, single-component only,
1150 * needs parent already locked. Doesn't follow mounts.
1153 static struct dentry *lookup_hash(struct nameidata *nd)
1157 err = exec_permission(nd->path.dentry->d_inode);
1159 return ERR_PTR(err);
1160 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1163 static int __lookup_one_len(const char *name, struct qstr *this,
1164 struct dentry *base, int len)
1174 hash = init_name_hash();
1176 c = *(const unsigned char *)name++;
1177 if (c == '/' || c == '\0')
1179 hash = partial_name_hash(c, hash);
1181 this->hash = end_name_hash(hash);
1186 * lookup_one_len - filesystem helper to lookup single pathname component
1187 * @name: pathname component to lookup
1188 * @base: base directory to lookup from
1189 * @len: maximum length @len should be interpreted to
1191 * Note that this routine is purely a helper for filesystem usage and should
1192 * not be called by generic code. Also note that by using this function the
1193 * nameidata argument is passed to the filesystem methods and a filesystem
1194 * using this helper needs to be prepared for that.
1196 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1201 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1203 err = __lookup_one_len(name, &this, base, len);
1205 return ERR_PTR(err);
1207 err = exec_permission(base->d_inode);
1209 return ERR_PTR(err);
1210 return __lookup_hash(&this, base, NULL);
1213 int user_path_at(int dfd, const char __user *name, unsigned flags,
1216 struct nameidata nd;
1217 char *tmp = getname(name);
1218 int err = PTR_ERR(tmp);
1221 BUG_ON(flags & LOOKUP_PARENT);
1223 err = do_path_lookup(dfd, tmp, flags, &nd);
1231 static int user_path_parent(int dfd, const char __user *path,
1232 struct nameidata *nd, char **name)
1234 char *s = getname(path);
1240 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1250 * It's inline, so penalty for filesystems that don't use sticky bit is
1253 static inline int check_sticky(struct inode *dir, struct inode *inode)
1255 uid_t fsuid = current_fsuid();
1257 if (!(dir->i_mode & S_ISVTX))
1259 if (inode->i_uid == fsuid)
1261 if (dir->i_uid == fsuid)
1263 return !capable(CAP_FOWNER);
1267 * Check whether we can remove a link victim from directory dir, check
1268 * whether the type of victim is right.
1269 * 1. We can't do it if dir is read-only (done in permission())
1270 * 2. We should have write and exec permissions on dir
1271 * 3. We can't remove anything from append-only dir
1272 * 4. We can't do anything with immutable dir (done in permission())
1273 * 5. If the sticky bit on dir is set we should either
1274 * a. be owner of dir, or
1275 * b. be owner of victim, or
1276 * c. have CAP_FOWNER capability
1277 * 6. If the victim is append-only or immutable we can't do antyhing with
1278 * links pointing to it.
1279 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1280 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1281 * 9. We can't remove a root or mountpoint.
1282 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1283 * nfs_async_unlink().
1285 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1289 if (!victim->d_inode)
1292 BUG_ON(victim->d_parent->d_inode != dir);
1293 audit_inode_child(victim->d_name.name, victim, dir);
1295 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1300 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1301 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1304 if (!S_ISDIR(victim->d_inode->i_mode))
1306 if (IS_ROOT(victim))
1308 } else if (S_ISDIR(victim->d_inode->i_mode))
1310 if (IS_DEADDIR(dir))
1312 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1317 /* Check whether we can create an object with dentry child in directory
1319 * 1. We can't do it if child already exists (open has special treatment for
1320 * this case, but since we are inlined it's OK)
1321 * 2. We can't do it if dir is read-only (done in permission())
1322 * 3. We should have write and exec permissions on dir
1323 * 4. We can't do it if dir is immutable (done in permission())
1325 static inline int may_create(struct inode *dir, struct dentry *child)
1329 if (IS_DEADDIR(dir))
1331 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1335 * O_DIRECTORY translates into forcing a directory lookup.
1337 static inline int lookup_flags(unsigned int f)
1339 unsigned long retval = LOOKUP_FOLLOW;
1342 retval &= ~LOOKUP_FOLLOW;
1344 if (f & O_DIRECTORY)
1345 retval |= LOOKUP_DIRECTORY;
1351 * p1 and p2 should be directories on the same fs.
1353 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1358 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1362 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1364 p = d_ancestor(p2, p1);
1366 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1367 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1371 p = d_ancestor(p1, p2);
1373 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1374 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1378 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1379 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1383 void unlock_rename(struct dentry *p1, struct dentry *p2)
1385 mutex_unlock(&p1->d_inode->i_mutex);
1387 mutex_unlock(&p2->d_inode->i_mutex);
1388 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1392 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1393 struct nameidata *nd)
1395 int error = may_create(dir, dentry);
1400 if (!dir->i_op->create)
1401 return -EACCES; /* shouldn't it be ENOSYS? */
1404 error = security_inode_create(dir, dentry, mode);
1408 error = dir->i_op->create(dir, dentry, mode, nd);
1410 fsnotify_create(dir, dentry);
1414 int may_open(struct path *path, int acc_mode, int flag)
1416 struct dentry *dentry = path->dentry;
1417 struct inode *inode = dentry->d_inode;
1423 switch (inode->i_mode & S_IFMT) {
1427 if (acc_mode & MAY_WRITE)
1432 if (path->mnt->mnt_flags & MNT_NODEV)
1441 error = inode_permission(inode, acc_mode);
1445 error = ima_path_check(path, acc_mode ?
1446 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1447 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1453 * An append-only file must be opened in append mode for writing.
1455 if (IS_APPEND(inode)) {
1457 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1463 /* O_NOATIME can only be set by the owner or superuser */
1464 if (flag & O_NOATIME)
1465 if (!is_owner_or_cap(inode)) {
1471 * Ensure there are no outstanding leases on the file.
1473 error = break_lease(inode, flag);
1477 if (flag & O_TRUNC) {
1478 error = get_write_access(inode);
1483 * Refuse to truncate files with mandatory locks held on them.
1485 error = locks_verify_locked(inode);
1487 error = security_path_truncate(path, 0,
1488 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1492 error = do_truncate(dentry, 0,
1493 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1496 put_write_access(inode);
1500 if (flag & FMODE_WRITE)
1505 ima_counts_put(path, acc_mode ?
1506 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1507 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1512 * Be careful about ever adding any more callers of this
1513 * function. Its flags must be in the namei format, not
1514 * what get passed to sys_open().
1516 static int __open_namei_create(struct nameidata *nd, struct path *path,
1520 struct dentry *dir = nd->path.dentry;
1522 if (!IS_POSIXACL(dir->d_inode))
1523 mode &= ~current_umask();
1524 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1527 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1529 mutex_unlock(&dir->d_inode->i_mutex);
1530 dput(nd->path.dentry);
1531 nd->path.dentry = path->dentry;
1534 /* Don't check for write permission, don't truncate */
1535 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1539 * Note that while the flag value (low two bits) for sys_open means:
1544 * it is changed into
1545 * 00 - no permissions needed
1546 * 01 - read-permission
1547 * 10 - write-permission
1549 * for the internal routines (ie open_namei()/follow_link() etc)
1550 * This is more logical, and also allows the 00 "no perm needed"
1551 * to be used for symlinks (where the permissions are checked
1555 static inline int open_to_namei_flags(int flag)
1557 if ((flag+1) & O_ACCMODE)
1562 static int open_will_write_to_fs(int flag, struct inode *inode)
1565 * We'll never write to the fs underlying
1568 if (special_file(inode->i_mode))
1570 return (flag & O_TRUNC);
1574 * Note that the low bits of the passed in "open_flag"
1575 * are not the same as in the local variable "flag". See
1576 * open_to_namei_flags() for more details.
1578 struct file *do_filp_open(int dfd, const char *pathname,
1579 int open_flag, int mode, int acc_mode)
1582 struct nameidata nd;
1584 struct path path, save;
1588 int flag = open_to_namei_flags(open_flag);
1591 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1592 * check for O_DSYNC if the need any syncing at all we enforce it's
1593 * always set instead of having to deal with possibly weird behaviour
1594 * for malicious applications setting only __O_SYNC.
1596 if (open_flag & __O_SYNC)
1597 open_flag |= O_DSYNC;
1600 acc_mode = MAY_OPEN | ACC_MODE(flag);
1602 /* O_TRUNC implies we need access checks for write permissions */
1604 acc_mode |= MAY_WRITE;
1606 /* Allow the LSM permission hook to distinguish append
1607 access from general write access. */
1608 if (flag & O_APPEND)
1609 acc_mode |= MAY_APPEND;
1612 * The simplest case - just a plain lookup.
1614 if (!(flag & O_CREAT)) {
1615 filp = get_empty_filp();
1618 return ERR_PTR(-ENFILE);
1619 nd.intent.open.file = filp;
1620 nd.intent.open.flags = flag;
1621 nd.intent.open.create_mode = 0;
1622 error = do_path_lookup(dfd, pathname,
1623 lookup_flags(flag)|LOOKUP_OPEN, &nd);
1624 if (IS_ERR(nd.intent.open.file)) {
1626 error = PTR_ERR(nd.intent.open.file);
1630 release_open_intent(&nd);
1632 return ERR_PTR(error);
1637 * Create - we need to know the parent.
1639 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1641 return ERR_PTR(error);
1642 error = path_walk(pathname, &nd);
1646 return ERR_PTR(error);
1648 if (unlikely(!audit_dummy_context()))
1649 audit_inode(pathname, nd.path.dentry);
1652 * We have the parent and last component. First of all, check
1653 * that we are not asked to creat(2) an obvious directory - that
1657 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1661 filp = get_empty_filp();
1664 nd.intent.open.file = filp;
1665 nd.intent.open.flags = flag;
1666 nd.intent.open.create_mode = mode;
1667 dir = nd.path.dentry;
1668 nd.flags &= ~LOOKUP_PARENT;
1669 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1671 nd.flags |= LOOKUP_EXCL;
1672 mutex_lock(&dir->d_inode->i_mutex);
1673 path.dentry = lookup_hash(&nd);
1674 path.mnt = nd.path.mnt;
1677 error = PTR_ERR(path.dentry);
1678 if (IS_ERR(path.dentry)) {
1679 mutex_unlock(&dir->d_inode->i_mutex);
1683 if (IS_ERR(nd.intent.open.file)) {
1684 error = PTR_ERR(nd.intent.open.file);
1685 goto exit_mutex_unlock;
1688 /* Negative dentry, just create the file */
1689 if (!path.dentry->d_inode) {
1691 * This write is needed to ensure that a
1692 * ro->rw transition does not occur between
1693 * the time when the file is created and when
1694 * a permanent write count is taken through
1695 * the 'struct file' in nameidata_to_filp().
1697 error = mnt_want_write(nd.path.mnt);
1699 goto exit_mutex_unlock;
1700 error = __open_namei_create(&nd, &path, flag, mode);
1702 mnt_drop_write(nd.path.mnt);
1705 filp = nameidata_to_filp(&nd, open_flag);
1707 ima_counts_put(&nd.path,
1708 acc_mode & (MAY_READ | MAY_WRITE |
1710 mnt_drop_write(nd.path.mnt);
1717 * It already exists.
1719 mutex_unlock(&dir->d_inode->i_mutex);
1720 audit_inode(pathname, path.dentry);
1726 if (__follow_mount(&path)) {
1728 if (flag & O_NOFOLLOW)
1733 if (!path.dentry->d_inode)
1735 if (path.dentry->d_inode->i_op->follow_link)
1738 path_to_nameidata(&path, &nd);
1740 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1745 * 1. may_open() truncates a file
1746 * 2. a rw->ro mount transition occurs
1747 * 3. nameidata_to_filp() fails due to
1749 * That would be inconsistent, and should
1750 * be avoided. Taking this mnt write here
1751 * ensures that (2) can not occur.
1753 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1755 error = mnt_want_write(nd.path.mnt);
1759 error = may_open(&nd.path, acc_mode, flag);
1762 mnt_drop_write(nd.path.mnt);
1765 filp = nameidata_to_filp(&nd, open_flag);
1767 ima_counts_put(&nd.path,
1768 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1770 * It is now safe to drop the mnt write
1771 * because the filp has had a write taken
1775 mnt_drop_write(nd.path.mnt);
1781 mutex_unlock(&dir->d_inode->i_mutex);
1783 path_put_conditional(&path, &nd);
1785 if (!IS_ERR(nd.intent.open.file))
1786 release_open_intent(&nd);
1791 return ERR_PTR(error);
1795 if (flag & O_NOFOLLOW)
1798 * This is subtle. Instead of calling do_follow_link() we do the
1799 * thing by hands. The reason is that this way we have zero link_count
1800 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1801 * After that we have the parent and last component, i.e.
1802 * we are in the same situation as after the first path_walk().
1803 * Well, almost - if the last component is normal we get its copy
1804 * stored in nd->last.name and we will have to putname() it when we
1805 * are done. Procfs-like symlinks just set LAST_BIND.
1807 nd.flags |= LOOKUP_PARENT;
1808 error = security_inode_follow_link(path.dentry, &nd);
1813 error = __do_follow_link(&path, &nd);
1814 if (error == -ESTALE) {
1815 /* nd.path had been dropped */
1818 nd.flags |= LOOKUP_REVAL;
1819 error = __do_follow_link(&path, &nd);
1824 /* Does someone understand code flow here? Or it is only
1825 * me so stupid? Anathema to whoever designed this non-sense
1826 * with "intent.open".
1828 release_open_intent(&nd);
1831 return ERR_PTR(error);
1833 nd.flags &= ~LOOKUP_PARENT;
1834 if (nd.last_type == LAST_BIND)
1837 if (nd.last_type != LAST_NORM)
1839 if (nd.last.name[nd.last.len]) {
1840 __putname(nd.last.name);
1845 __putname(nd.last.name);
1848 dir = nd.path.dentry;
1849 mutex_lock(&dir->d_inode->i_mutex);
1850 path.dentry = lookup_hash(&nd);
1851 path.mnt = nd.path.mnt;
1852 __putname(nd.last.name);
1857 * filp_open - open file and return file pointer
1859 * @filename: path to open
1860 * @flags: open flags as per the open(2) second argument
1861 * @mode: mode for the new file if O_CREAT is set, else ignored
1863 * This is the helper to open a file from kernelspace if you really
1864 * have to. But in generally you should not do this, so please move
1865 * along, nothing to see here..
1867 struct file *filp_open(const char *filename, int flags, int mode)
1869 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1871 EXPORT_SYMBOL(filp_open);
1874 * lookup_create - lookup a dentry, creating it if it doesn't exist
1875 * @nd: nameidata info
1876 * @is_dir: directory flag
1878 * Simple function to lookup and return a dentry and create it
1879 * if it doesn't exist. Is SMP-safe.
1881 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1883 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1885 struct dentry *dentry = ERR_PTR(-EEXIST);
1887 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1889 * Yucky last component or no last component at all?
1890 * (foo/., foo/.., /////)
1892 if (nd->last_type != LAST_NORM)
1894 nd->flags &= ~LOOKUP_PARENT;
1895 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1896 nd->intent.open.flags = O_EXCL;
1899 * Do the final lookup.
1901 dentry = lookup_hash(nd);
1905 if (dentry->d_inode)
1908 * Special case - lookup gave negative, but... we had foo/bar/
1909 * From the vfs_mknod() POV we just have a negative dentry -
1910 * all is fine. Let's be bastards - you had / on the end, you've
1911 * been asking for (non-existent) directory. -ENOENT for you.
1913 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1915 dentry = ERR_PTR(-ENOENT);
1920 dentry = ERR_PTR(-EEXIST);
1924 EXPORT_SYMBOL_GPL(lookup_create);
1926 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1928 int error = may_create(dir, dentry);
1933 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1936 if (!dir->i_op->mknod)
1939 error = devcgroup_inode_mknod(mode, dev);
1943 error = security_inode_mknod(dir, dentry, mode, dev);
1948 error = dir->i_op->mknod(dir, dentry, mode, dev);
1950 fsnotify_create(dir, dentry);
1954 static int may_mknod(mode_t mode)
1956 switch (mode & S_IFMT) {
1962 case 0: /* zero mode translates to S_IFREG */
1971 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
1976 struct dentry *dentry;
1977 struct nameidata nd;
1982 error = user_path_parent(dfd, filename, &nd, &tmp);
1986 dentry = lookup_create(&nd, 0);
1987 if (IS_ERR(dentry)) {
1988 error = PTR_ERR(dentry);
1991 if (!IS_POSIXACL(nd.path.dentry->d_inode))
1992 mode &= ~current_umask();
1993 error = may_mknod(mode);
1996 error = mnt_want_write(nd.path.mnt);
1999 error = security_path_mknod(&nd.path, dentry, mode, dev);
2001 goto out_drop_write;
2002 switch (mode & S_IFMT) {
2003 case 0: case S_IFREG:
2004 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2006 case S_IFCHR: case S_IFBLK:
2007 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2008 new_decode_dev(dev));
2010 case S_IFIFO: case S_IFSOCK:
2011 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2015 mnt_drop_write(nd.path.mnt);
2019 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2026 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2028 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2031 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2033 int error = may_create(dir, dentry);
2038 if (!dir->i_op->mkdir)
2041 mode &= (S_IRWXUGO|S_ISVTX);
2042 error = security_inode_mkdir(dir, dentry, mode);
2047 error = dir->i_op->mkdir(dir, dentry, mode);
2049 fsnotify_mkdir(dir, dentry);
2053 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2057 struct dentry *dentry;
2058 struct nameidata nd;
2060 error = user_path_parent(dfd, pathname, &nd, &tmp);
2064 dentry = lookup_create(&nd, 1);
2065 error = PTR_ERR(dentry);
2069 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2070 mode &= ~current_umask();
2071 error = mnt_want_write(nd.path.mnt);
2074 error = security_path_mkdir(&nd.path, dentry, mode);
2076 goto out_drop_write;
2077 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2079 mnt_drop_write(nd.path.mnt);
2083 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2090 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2092 return sys_mkdirat(AT_FDCWD, pathname, mode);
2096 * We try to drop the dentry early: we should have
2097 * a usage count of 2 if we're the only user of this
2098 * dentry, and if that is true (possibly after pruning
2099 * the dcache), then we drop the dentry now.
2101 * A low-level filesystem can, if it choses, legally
2104 * if (!d_unhashed(dentry))
2107 * if it cannot handle the case of removing a directory
2108 * that is still in use by something else..
2110 void dentry_unhash(struct dentry *dentry)
2113 shrink_dcache_parent(dentry);
2114 spin_lock(&dcache_lock);
2115 spin_lock(&dentry->d_lock);
2116 if (atomic_read(&dentry->d_count) == 2)
2118 spin_unlock(&dentry->d_lock);
2119 spin_unlock(&dcache_lock);
2122 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2124 int error = may_delete(dir, dentry, 1);
2129 if (!dir->i_op->rmdir)
2134 mutex_lock(&dentry->d_inode->i_mutex);
2135 dentry_unhash(dentry);
2136 if (d_mountpoint(dentry))
2139 error = security_inode_rmdir(dir, dentry);
2141 error = dir->i_op->rmdir(dir, dentry);
2143 dentry->d_inode->i_flags |= S_DEAD;
2146 mutex_unlock(&dentry->d_inode->i_mutex);
2155 static long do_rmdir(int dfd, const char __user *pathname)
2159 struct dentry *dentry;
2160 struct nameidata nd;
2162 error = user_path_parent(dfd, pathname, &nd, &name);
2166 switch(nd.last_type) {
2178 nd.flags &= ~LOOKUP_PARENT;
2180 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2181 dentry = lookup_hash(&nd);
2182 error = PTR_ERR(dentry);
2185 error = mnt_want_write(nd.path.mnt);
2188 error = security_path_rmdir(&nd.path, dentry);
2191 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2193 mnt_drop_write(nd.path.mnt);
2197 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2204 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2206 return do_rmdir(AT_FDCWD, pathname);
2209 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2211 int error = may_delete(dir, dentry, 0);
2216 if (!dir->i_op->unlink)
2221 mutex_lock(&dentry->d_inode->i_mutex);
2222 if (d_mountpoint(dentry))
2225 error = security_inode_unlink(dir, dentry);
2227 error = dir->i_op->unlink(dir, dentry);
2229 mutex_unlock(&dentry->d_inode->i_mutex);
2231 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2232 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2233 fsnotify_link_count(dentry->d_inode);
2241 * Make sure that the actual truncation of the file will occur outside its
2242 * directory's i_mutex. Truncate can take a long time if there is a lot of
2243 * writeout happening, and we don't want to prevent access to the directory
2244 * while waiting on the I/O.
2246 static long do_unlinkat(int dfd, const char __user *pathname)
2250 struct dentry *dentry;
2251 struct nameidata nd;
2252 struct inode *inode = NULL;
2254 error = user_path_parent(dfd, pathname, &nd, &name);
2259 if (nd.last_type != LAST_NORM)
2262 nd.flags &= ~LOOKUP_PARENT;
2264 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2265 dentry = lookup_hash(&nd);
2266 error = PTR_ERR(dentry);
2267 if (!IS_ERR(dentry)) {
2268 /* Why not before? Because we want correct error value */
2269 if (nd.last.name[nd.last.len])
2271 inode = dentry->d_inode;
2273 atomic_inc(&inode->i_count);
2274 error = mnt_want_write(nd.path.mnt);
2277 error = security_path_unlink(&nd.path, dentry);
2280 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2282 mnt_drop_write(nd.path.mnt);
2286 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2288 iput(inode); /* truncate the inode here */
2295 error = !dentry->d_inode ? -ENOENT :
2296 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2300 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2302 if ((flag & ~AT_REMOVEDIR) != 0)
2305 if (flag & AT_REMOVEDIR)
2306 return do_rmdir(dfd, pathname);
2308 return do_unlinkat(dfd, pathname);
2311 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2313 return do_unlinkat(AT_FDCWD, pathname);
2316 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2318 int error = may_create(dir, dentry);
2323 if (!dir->i_op->symlink)
2326 error = security_inode_symlink(dir, dentry, oldname);
2331 error = dir->i_op->symlink(dir, dentry, oldname);
2333 fsnotify_create(dir, dentry);
2337 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2338 int, newdfd, const char __user *, newname)
2343 struct dentry *dentry;
2344 struct nameidata nd;
2346 from = getname(oldname);
2348 return PTR_ERR(from);
2350 error = user_path_parent(newdfd, newname, &nd, &to);
2354 dentry = lookup_create(&nd, 0);
2355 error = PTR_ERR(dentry);
2359 error = mnt_want_write(nd.path.mnt);
2362 error = security_path_symlink(&nd.path, dentry, from);
2364 goto out_drop_write;
2365 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2367 mnt_drop_write(nd.path.mnt);
2371 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2379 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2381 return sys_symlinkat(oldname, AT_FDCWD, newname);
2384 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2386 struct inode *inode = old_dentry->d_inode;
2392 error = may_create(dir, new_dentry);
2396 if (dir->i_sb != inode->i_sb)
2400 * A link to an append-only or immutable file cannot be created.
2402 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2404 if (!dir->i_op->link)
2406 if (S_ISDIR(inode->i_mode))
2409 error = security_inode_link(old_dentry, dir, new_dentry);
2413 mutex_lock(&inode->i_mutex);
2415 error = dir->i_op->link(old_dentry, dir, new_dentry);
2416 mutex_unlock(&inode->i_mutex);
2418 fsnotify_link(dir, inode, new_dentry);
2423 * Hardlinks are often used in delicate situations. We avoid
2424 * security-related surprises by not following symlinks on the
2427 * We don't follow them on the oldname either to be compatible
2428 * with linux 2.0, and to avoid hard-linking to directories
2429 * and other special files. --ADM
2431 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2432 int, newdfd, const char __user *, newname, int, flags)
2434 struct dentry *new_dentry;
2435 struct nameidata nd;
2436 struct path old_path;
2440 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2443 error = user_path_at(olddfd, oldname,
2444 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2449 error = user_path_parent(newdfd, newname, &nd, &to);
2453 if (old_path.mnt != nd.path.mnt)
2455 new_dentry = lookup_create(&nd, 0);
2456 error = PTR_ERR(new_dentry);
2457 if (IS_ERR(new_dentry))
2459 error = mnt_want_write(nd.path.mnt);
2462 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2464 goto out_drop_write;
2465 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2467 mnt_drop_write(nd.path.mnt);
2471 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2476 path_put(&old_path);
2481 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2483 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2487 * The worst of all namespace operations - renaming directory. "Perverted"
2488 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2490 * a) we can get into loop creation. Check is done in is_subdir().
2491 * b) race potential - two innocent renames can create a loop together.
2492 * That's where 4.4 screws up. Current fix: serialization on
2493 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2495 * c) we have to lock _three_ objects - parents and victim (if it exists).
2496 * And that - after we got ->i_mutex on parents (until then we don't know
2497 * whether the target exists). Solution: try to be smart with locking
2498 * order for inodes. We rely on the fact that tree topology may change
2499 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2500 * move will be locked. Thus we can rank directories by the tree
2501 * (ancestors first) and rank all non-directories after them.
2502 * That works since everybody except rename does "lock parent, lookup,
2503 * lock child" and rename is under ->s_vfs_rename_mutex.
2504 * HOWEVER, it relies on the assumption that any object with ->lookup()
2505 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2506 * we'd better make sure that there's no link(2) for them.
2507 * d) some filesystems don't support opened-but-unlinked directories,
2508 * either because of layout or because they are not ready to deal with
2509 * all cases correctly. The latter will be fixed (taking this sort of
2510 * stuff into VFS), but the former is not going away. Solution: the same
2511 * trick as in rmdir().
2512 * e) conversion from fhandle to dentry may come in the wrong moment - when
2513 * we are removing the target. Solution: we will have to grab ->i_mutex
2514 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2515 * ->i_mutex on parents, which works but leads to some truely excessive
2518 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2519 struct inode *new_dir, struct dentry *new_dentry)
2522 struct inode *target;
2525 * If we are going to change the parent - check write permissions,
2526 * we'll need to flip '..'.
2528 if (new_dir != old_dir) {
2529 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2534 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2538 target = new_dentry->d_inode;
2540 mutex_lock(&target->i_mutex);
2541 dentry_unhash(new_dentry);
2543 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2546 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2549 target->i_flags |= S_DEAD;
2550 mutex_unlock(&target->i_mutex);
2551 if (d_unhashed(new_dentry))
2552 d_rehash(new_dentry);
2556 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2557 d_move(old_dentry,new_dentry);
2561 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2562 struct inode *new_dir, struct dentry *new_dentry)
2564 struct inode *target;
2567 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2572 target = new_dentry->d_inode;
2574 mutex_lock(&target->i_mutex);
2575 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2578 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2580 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2581 d_move(old_dentry, new_dentry);
2584 mutex_unlock(&target->i_mutex);
2589 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2590 struct inode *new_dir, struct dentry *new_dentry)
2593 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2594 const char *old_name;
2596 if (old_dentry->d_inode == new_dentry->d_inode)
2599 error = may_delete(old_dir, old_dentry, is_dir);
2603 if (!new_dentry->d_inode)
2604 error = may_create(new_dir, new_dentry);
2606 error = may_delete(new_dir, new_dentry, is_dir);
2610 if (!old_dir->i_op->rename)
2613 vfs_dq_init(old_dir);
2614 vfs_dq_init(new_dir);
2616 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2619 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2621 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2623 const char *new_name = old_dentry->d_name.name;
2624 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2625 new_dentry->d_inode, old_dentry);
2627 fsnotify_oldname_free(old_name);
2632 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2633 int, newdfd, const char __user *, newname)
2635 struct dentry *old_dir, *new_dir;
2636 struct dentry *old_dentry, *new_dentry;
2637 struct dentry *trap;
2638 struct nameidata oldnd, newnd;
2643 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2647 error = user_path_parent(newdfd, newname, &newnd, &to);
2652 if (oldnd.path.mnt != newnd.path.mnt)
2655 old_dir = oldnd.path.dentry;
2657 if (oldnd.last_type != LAST_NORM)
2660 new_dir = newnd.path.dentry;
2661 if (newnd.last_type != LAST_NORM)
2664 oldnd.flags &= ~LOOKUP_PARENT;
2665 newnd.flags &= ~LOOKUP_PARENT;
2666 newnd.flags |= LOOKUP_RENAME_TARGET;
2668 trap = lock_rename(new_dir, old_dir);
2670 old_dentry = lookup_hash(&oldnd);
2671 error = PTR_ERR(old_dentry);
2672 if (IS_ERR(old_dentry))
2674 /* source must exist */
2676 if (!old_dentry->d_inode)
2678 /* unless the source is a directory trailing slashes give -ENOTDIR */
2679 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2681 if (oldnd.last.name[oldnd.last.len])
2683 if (newnd.last.name[newnd.last.len])
2686 /* source should not be ancestor of target */
2688 if (old_dentry == trap)
2690 new_dentry = lookup_hash(&newnd);
2691 error = PTR_ERR(new_dentry);
2692 if (IS_ERR(new_dentry))
2694 /* target should not be an ancestor of source */
2696 if (new_dentry == trap)
2699 error = mnt_want_write(oldnd.path.mnt);
2702 error = security_path_rename(&oldnd.path, old_dentry,
2703 &newnd.path, new_dentry);
2706 error = vfs_rename(old_dir->d_inode, old_dentry,
2707 new_dir->d_inode, new_dentry);
2709 mnt_drop_write(oldnd.path.mnt);
2715 unlock_rename(new_dir, old_dir);
2717 path_put(&newnd.path);
2720 path_put(&oldnd.path);
2726 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2728 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2731 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2735 len = PTR_ERR(link);
2740 if (len > (unsigned) buflen)
2742 if (copy_to_user(buffer, link, len))
2749 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2750 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2751 * using) it for any given inode is up to filesystem.
2753 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2755 struct nameidata nd;
2760 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2762 return PTR_ERR(cookie);
2764 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2765 if (dentry->d_inode->i_op->put_link)
2766 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2770 int vfs_follow_link(struct nameidata *nd, const char *link)
2772 return __vfs_follow_link(nd, link);
2775 /* get the link contents into pagecache */
2776 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2780 struct address_space *mapping = dentry->d_inode->i_mapping;
2781 page = read_mapping_page(mapping, 0, NULL);
2786 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2790 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2792 struct page *page = NULL;
2793 char *s = page_getlink(dentry, &page);
2794 int res = vfs_readlink(dentry,buffer,buflen,s);
2797 page_cache_release(page);
2802 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2804 struct page *page = NULL;
2805 nd_set_link(nd, page_getlink(dentry, &page));
2809 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2811 struct page *page = cookie;
2815 page_cache_release(page);
2820 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2822 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2824 struct address_space *mapping = inode->i_mapping;
2829 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2831 flags |= AOP_FLAG_NOFS;
2834 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2835 flags, &page, &fsdata);
2839 kaddr = kmap_atomic(page, KM_USER0);
2840 memcpy(kaddr, symname, len-1);
2841 kunmap_atomic(kaddr, KM_USER0);
2843 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2850 mark_inode_dirty(inode);
2856 int page_symlink(struct inode *inode, const char *symname, int len)
2858 return __page_symlink(inode, symname, len,
2859 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2862 const struct inode_operations page_symlink_inode_operations = {
2863 .readlink = generic_readlink,
2864 .follow_link = page_follow_link_light,
2865 .put_link = page_put_link,
2868 EXPORT_SYMBOL(user_path_at);
2869 EXPORT_SYMBOL(follow_down);
2870 EXPORT_SYMBOL(follow_up);
2871 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2872 EXPORT_SYMBOL(getname);
2873 EXPORT_SYMBOL(lock_rename);
2874 EXPORT_SYMBOL(lookup_one_len);
2875 EXPORT_SYMBOL(page_follow_link_light);
2876 EXPORT_SYMBOL(page_put_link);
2877 EXPORT_SYMBOL(page_readlink);
2878 EXPORT_SYMBOL(__page_symlink);
2879 EXPORT_SYMBOL(page_symlink);
2880 EXPORT_SYMBOL(page_symlink_inode_operations);
2881 EXPORT_SYMBOL(path_lookup);
2882 EXPORT_SYMBOL(kern_path);
2883 EXPORT_SYMBOL(vfs_path_lookup);
2884 EXPORT_SYMBOL(inode_permission);
2885 EXPORT_SYMBOL(file_permission);
2886 EXPORT_SYMBOL(unlock_rename);
2887 EXPORT_SYMBOL(vfs_create);
2888 EXPORT_SYMBOL(vfs_follow_link);
2889 EXPORT_SYMBOL(vfs_link);
2890 EXPORT_SYMBOL(vfs_mkdir);
2891 EXPORT_SYMBOL(vfs_mknod);
2892 EXPORT_SYMBOL(generic_permission);
2893 EXPORT_SYMBOL(vfs_readlink);
2894 EXPORT_SYMBOL(vfs_rename);
2895 EXPORT_SYMBOL(vfs_rmdir);
2896 EXPORT_SYMBOL(vfs_symlink);
2897 EXPORT_SYMBOL(vfs_unlink);
2898 EXPORT_SYMBOL(dentry_unhash);
2899 EXPORT_SYMBOL(generic_readlink);