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 static int __link_path_walk(const char *name, struct nameidata *nd);
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static int do_getname(const char __user *filename, char *page)
123 unsigned long len = PATH_MAX;
125 if (!segment_eq(get_fs(), KERNEL_DS)) {
126 if ((unsigned long) filename >= TASK_SIZE)
128 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
129 len = TASK_SIZE - (unsigned long) filename;
132 retval = strncpy_from_user(page, filename, len);
136 return -ENAMETOOLONG;
142 char * getname(const char __user * filename)
146 result = ERR_PTR(-ENOMEM);
149 int retval = do_getname(filename, tmp);
154 result = ERR_PTR(retval);
157 audit_getname(result);
161 #ifdef CONFIG_AUDITSYSCALL
162 void putname(const char *name)
164 if (unlikely(!audit_dummy_context()))
169 EXPORT_SYMBOL(putname);
173 * This does basic POSIX ACL permission checking
175 static int acl_permission_check(struct inode *inode, int mask,
176 int (*check_acl)(struct inode *inode, int mask))
178 umode_t mode = inode->i_mode;
180 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
182 if (current_fsuid() == inode->i_uid)
185 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
191 if (in_group_p(inode->i_gid))
196 * If the DACs are ok we don't need any capability check.
198 if ((mask & ~mode) == 0)
204 * generic_permission - check for access rights on a Posix-like filesystem
205 * @inode: inode to check access rights for
206 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
207 * @check_acl: optional callback to check for Posix ACLs
209 * Used to check for read/write/execute permissions on a file.
210 * We use "fsuid" for this, letting us set arbitrary permissions
211 * for filesystem access without changing the "normal" uids which
212 * are used for other things..
214 int generic_permission(struct inode *inode, int mask,
215 int (*check_acl)(struct inode *inode, int mask))
220 * Do the basic POSIX ACL permission checks.
222 ret = acl_permission_check(inode, mask, check_acl);
227 * Read/write DACs are always overridable.
228 * Executable DACs are overridable if at least one exec bit is set.
230 if (!(mask & MAY_EXEC) || execute_ok(inode))
231 if (capable(CAP_DAC_OVERRIDE))
235 * Searching includes executable on directories, else just read.
237 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
238 if (capable(CAP_DAC_READ_SEARCH))
245 * inode_permission - check for access rights to a given inode
246 * @inode: inode to check permission on
247 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
249 * Used to check for read/write/execute permissions on an inode.
250 * We use "fsuid" for this, letting us set arbitrary permissions
251 * for filesystem access without changing the "normal" uids which
252 * are used for other things.
254 int inode_permission(struct inode *inode, int mask)
258 if (mask & MAY_WRITE) {
259 umode_t mode = inode->i_mode;
262 * Nobody gets write access to a read-only fs.
264 if (IS_RDONLY(inode) &&
265 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
269 * Nobody gets write access to an immutable file.
271 if (IS_IMMUTABLE(inode))
275 if (inode->i_op->permission)
276 retval = inode->i_op->permission(inode, mask);
278 retval = generic_permission(inode, mask, inode->i_op->check_acl);
283 retval = devcgroup_inode_permission(inode, mask);
287 return security_inode_permission(inode,
288 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
292 * file_permission - check for additional access rights to a given file
293 * @file: file to check access rights for
294 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
296 * Used to check for read/write/execute permissions on an already opened
300 * Do not use this function in new code. All access checks should
301 * be done using inode_permission().
303 int file_permission(struct file *file, int mask)
305 return inode_permission(file->f_path.dentry->d_inode, mask);
309 * get_write_access() gets write permission for a file.
310 * put_write_access() releases this write permission.
311 * This is used for regular files.
312 * We cannot support write (and maybe mmap read-write shared) accesses and
313 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
314 * can have the following values:
315 * 0: no writers, no VM_DENYWRITE mappings
316 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
317 * > 0: (i_writecount) users are writing to the file.
319 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
320 * except for the cases where we don't hold i_writecount yet. Then we need to
321 * use {get,deny}_write_access() - these functions check the sign and refuse
322 * to do the change if sign is wrong. Exclusion between them is provided by
323 * the inode->i_lock spinlock.
326 int get_write_access(struct inode * inode)
328 spin_lock(&inode->i_lock);
329 if (atomic_read(&inode->i_writecount) < 0) {
330 spin_unlock(&inode->i_lock);
333 atomic_inc(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
339 int deny_write_access(struct file * file)
341 struct inode *inode = file->f_path.dentry->d_inode;
343 spin_lock(&inode->i_lock);
344 if (atomic_read(&inode->i_writecount) > 0) {
345 spin_unlock(&inode->i_lock);
348 atomic_dec(&inode->i_writecount);
349 spin_unlock(&inode->i_lock);
355 * path_get - get a reference to a path
356 * @path: path to get the reference to
358 * Given a path increment the reference count to the dentry and the vfsmount.
360 void path_get(struct path *path)
365 EXPORT_SYMBOL(path_get);
368 * path_put - put a reference to a path
369 * @path: path to put the reference to
371 * Given a path decrement the reference count to the dentry and the vfsmount.
373 void path_put(struct path *path)
378 EXPORT_SYMBOL(path_put);
381 * release_open_intent - free up open intent resources
382 * @nd: pointer to nameidata
384 void release_open_intent(struct nameidata *nd)
386 if (nd->intent.open.file->f_path.dentry == NULL)
387 put_filp(nd->intent.open.file);
389 fput(nd->intent.open.file);
392 static inline struct dentry *
393 do_revalidate(struct dentry *dentry, struct nameidata *nd)
395 int status = dentry->d_op->d_revalidate(dentry, nd);
396 if (unlikely(status <= 0)) {
398 * The dentry failed validation.
399 * If d_revalidate returned 0 attempt to invalidate
400 * the dentry otherwise d_revalidate is asking us
401 * to return a fail status.
404 if (!d_invalidate(dentry)) {
410 dentry = ERR_PTR(status);
417 * Internal lookup() using the new generic dcache.
420 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
422 struct dentry * dentry = __d_lookup(parent, name);
424 /* lockess __d_lookup may fail due to concurrent d_move()
425 * in some unrelated directory, so try with d_lookup
428 dentry = d_lookup(parent, name);
430 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
431 dentry = do_revalidate(dentry, nd);
437 * Short-cut version of permission(), for calling by
438 * path_walk(), when dcache lock is held. Combines parts
439 * of permission() and generic_permission(), and tests ONLY for
440 * MAY_EXEC permission.
442 * If appropriate, check DAC only. If not appropriate, or
443 * short-cut DAC fails, then call permission() to do more
444 * complete permission check.
446 static int exec_permission_lite(struct inode *inode)
450 if (inode->i_op->permission) {
451 ret = inode->i_op->permission(inode, MAY_EXEC);
456 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
460 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
465 return security_inode_permission(inode, MAY_EXEC);
469 * This is called when everything else fails, and we actually have
470 * to go to the low-level filesystem to find out what we should do..
472 * We get the directory semaphore, and after getting that we also
473 * make sure that nobody added the entry to the dcache in the meantime..
476 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
478 struct dentry * result;
479 struct inode *dir = parent->d_inode;
481 mutex_lock(&dir->i_mutex);
483 * First re-do the cached lookup just in case it was created
484 * while we waited for the directory semaphore..
486 * FIXME! This could use version numbering or similar to
487 * avoid unnecessary cache lookups.
489 * The "dcache_lock" is purely to protect the RCU list walker
490 * from concurrent renames at this point (we mustn't get false
491 * negatives from the RCU list walk here, unlike the optimistic
494 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
496 result = d_lookup(parent, name);
498 struct dentry *dentry;
500 /* Don't create child dentry for a dead directory. */
501 result = ERR_PTR(-ENOENT);
505 dentry = d_alloc(parent, name);
506 result = ERR_PTR(-ENOMEM);
508 result = dir->i_op->lookup(dir, dentry, nd);
515 mutex_unlock(&dir->i_mutex);
520 * Uhhuh! Nasty case: the cache was re-populated while
521 * we waited on the semaphore. Need to revalidate.
523 mutex_unlock(&dir->i_mutex);
524 if (result->d_op && result->d_op->d_revalidate) {
525 result = do_revalidate(result, nd);
527 result = ERR_PTR(-ENOENT);
533 * Wrapper to retry pathname resolution whenever the underlying
534 * file system returns an ESTALE.
536 * Retry the whole path once, forcing real lookup requests
537 * instead of relying on the dcache.
539 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
541 struct path save = nd->path;
544 /* make sure the stuff we saved doesn't go away */
547 result = __link_path_walk(name, nd);
548 if (result == -ESTALE) {
549 /* nd->path had been dropped */
552 nd->flags |= LOOKUP_REVAL;
553 result = __link_path_walk(name, nd);
561 static __always_inline void set_root(struct nameidata *nd)
564 struct fs_struct *fs = current->fs;
565 read_lock(&fs->lock);
568 read_unlock(&fs->lock);
572 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
586 res = link_path_walk(link, nd);
587 if (nd->depth || res || nd->last_type!=LAST_NORM)
590 * If it is an iterative symlinks resolution in open_namei() we
591 * have to copy the last component. And all that crap because of
592 * bloody create() on broken symlinks. Furrfu...
595 if (unlikely(!name)) {
599 strcpy(name, nd->last.name);
600 nd->last.name = name;
604 return PTR_ERR(link);
607 static void path_put_conditional(struct path *path, struct nameidata *nd)
610 if (path->mnt != nd->path.mnt)
614 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
616 dput(nd->path.dentry);
617 if (nd->path.mnt != path->mnt)
618 mntput(nd->path.mnt);
619 nd->path.mnt = path->mnt;
620 nd->path.dentry = path->dentry;
623 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
627 struct dentry *dentry = path->dentry;
629 touch_atime(path->mnt, dentry);
630 nd_set_link(nd, NULL);
632 if (path->mnt != nd->path.mnt) {
633 path_to_nameidata(path, nd);
637 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
638 error = PTR_ERR(cookie);
639 if (!IS_ERR(cookie)) {
640 char *s = nd_get_link(nd);
643 error = __vfs_follow_link(nd, s);
644 if (dentry->d_inode->i_op->put_link)
645 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
653 * This limits recursive symlink follows to 8, while
654 * limiting consecutive symlinks to 40.
656 * Without that kind of total limit, nasty chains of consecutive
657 * symlinks can cause almost arbitrarily long lookups.
659 static inline int do_follow_link(struct path *path, struct nameidata *nd)
662 if (current->link_count >= MAX_NESTED_LINKS)
664 if (current->total_link_count >= 40)
666 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
668 err = security_inode_follow_link(path->dentry, nd);
671 current->link_count++;
672 current->total_link_count++;
674 err = __do_follow_link(path, nd);
675 current->link_count--;
679 path_put_conditional(path, nd);
684 int follow_up(struct path *path)
686 struct vfsmount *parent;
687 struct dentry *mountpoint;
688 spin_lock(&vfsmount_lock);
689 parent = path->mnt->mnt_parent;
690 if (parent == path->mnt) {
691 spin_unlock(&vfsmount_lock);
695 mountpoint = dget(path->mnt->mnt_mountpoint);
696 spin_unlock(&vfsmount_lock);
698 path->dentry = mountpoint;
704 /* no need for dcache_lock, as serialization is taken care in
707 static int __follow_mount(struct path *path)
710 while (d_mountpoint(path->dentry)) {
711 struct vfsmount *mounted = lookup_mnt(path);
718 path->dentry = dget(mounted->mnt_root);
724 static void follow_mount(struct path *path)
726 while (d_mountpoint(path->dentry)) {
727 struct vfsmount *mounted = lookup_mnt(path);
733 path->dentry = dget(mounted->mnt_root);
737 /* no need for dcache_lock, as serialization is taken care in
740 int follow_down(struct path *path)
742 struct vfsmount *mounted;
744 mounted = lookup_mnt(path);
749 path->dentry = dget(mounted->mnt_root);
755 static __always_inline void follow_dotdot(struct nameidata *nd)
760 struct vfsmount *parent;
761 struct dentry *old = nd->path.dentry;
763 if (nd->path.dentry == nd->root.dentry &&
764 nd->path.mnt == nd->root.mnt) {
767 spin_lock(&dcache_lock);
768 if (nd->path.dentry != nd->path.mnt->mnt_root) {
769 nd->path.dentry = dget(nd->path.dentry->d_parent);
770 spin_unlock(&dcache_lock);
774 spin_unlock(&dcache_lock);
775 spin_lock(&vfsmount_lock);
776 parent = nd->path.mnt->mnt_parent;
777 if (parent == nd->path.mnt) {
778 spin_unlock(&vfsmount_lock);
782 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
783 spin_unlock(&vfsmount_lock);
785 mntput(nd->path.mnt);
786 nd->path.mnt = parent;
788 follow_mount(&nd->path);
792 * It's more convoluted than I'd like it to be, but... it's still fairly
793 * small and for now I'd prefer to have fast path as straight as possible.
794 * It _is_ time-critical.
796 static int do_lookup(struct nameidata *nd, struct qstr *name,
799 struct vfsmount *mnt = nd->path.mnt;
800 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
804 if (dentry->d_op && dentry->d_op->d_revalidate)
805 goto need_revalidate;
808 path->dentry = dentry;
809 __follow_mount(path);
813 dentry = real_lookup(nd->path.dentry, name, nd);
819 dentry = do_revalidate(dentry, nd);
827 return PTR_ERR(dentry);
832 * This is the basic name resolution function, turning a pathname into
833 * the final dentry. We expect 'base' to be positive and a directory.
835 * Returns 0 and nd will have valid dentry and mnt on success.
836 * Returns error and drops reference to input namei data on failure.
838 static int __link_path_walk(const char *name, struct nameidata *nd)
843 unsigned int lookup_flags = nd->flags;
850 inode = nd->path.dentry->d_inode;
852 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
854 /* At this point we know we have a real path component. */
860 nd->flags |= LOOKUP_CONTINUE;
861 err = exec_permission_lite(inode);
866 c = *(const unsigned char *)name;
868 hash = init_name_hash();
871 hash = partial_name_hash(c, hash);
872 c = *(const unsigned char *)name;
873 } while (c && (c != '/'));
874 this.len = name - (const char *) this.name;
875 this.hash = end_name_hash(hash);
877 /* remove trailing slashes? */
880 while (*++name == '/');
882 goto last_with_slashes;
885 * "." and ".." are special - ".." especially so because it has
886 * to be able to know about the current root directory and
887 * parent relationships.
889 if (this.name[0] == '.') switch (this.len) {
893 if (this.name[1] != '.')
896 inode = nd->path.dentry->d_inode;
902 * See if the low-level filesystem might want
903 * to use its own hash..
905 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
906 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
911 /* This does the actual lookups.. */
912 err = do_lookup(nd, &this, &next);
917 inode = next.dentry->d_inode;
921 if (inode->i_op->follow_link) {
922 err = do_follow_link(&next, nd);
926 inode = nd->path.dentry->d_inode;
930 path_to_nameidata(&next, nd);
932 if (!inode->i_op->lookup)
935 /* here ends the main loop */
938 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
940 /* Clear LOOKUP_CONTINUE iff it was previously unset */
941 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
942 if (lookup_flags & LOOKUP_PARENT)
944 if (this.name[0] == '.') switch (this.len) {
948 if (this.name[1] != '.')
951 inode = nd->path.dentry->d_inode;
956 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
957 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
962 err = do_lookup(nd, &this, &next);
965 inode = next.dentry->d_inode;
966 if ((lookup_flags & LOOKUP_FOLLOW)
967 && inode && inode->i_op->follow_link) {
968 err = do_follow_link(&next, nd);
971 inode = nd->path.dentry->d_inode;
973 path_to_nameidata(&next, nd);
977 if (lookup_flags & LOOKUP_DIRECTORY) {
979 if (!inode->i_op->lookup)
985 nd->last_type = LAST_NORM;
986 if (this.name[0] != '.')
989 nd->last_type = LAST_DOT;
990 else if (this.len == 2 && this.name[1] == '.')
991 nd->last_type = LAST_DOTDOT;
996 * We bypassed the ordinary revalidation routines.
997 * We may need to check the cached dentry for staleness.
999 if (nd->path.dentry && nd->path.dentry->d_sb &&
1000 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1002 /* Note: we do not d_invalidate() */
1003 if (!nd->path.dentry->d_op->d_revalidate(
1004 nd->path.dentry, nd))
1010 path_put_conditional(&next, nd);
1013 path_put(&nd->path);
1018 static int path_walk(const char *name, struct nameidata *nd)
1020 current->total_link_count = 0;
1021 return link_path_walk(name, nd);
1024 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1030 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1033 nd->root.mnt = NULL;
1037 nd->path = nd->root;
1038 path_get(&nd->root);
1039 } else if (dfd == AT_FDCWD) {
1040 struct fs_struct *fs = current->fs;
1041 read_lock(&fs->lock);
1044 read_unlock(&fs->lock);
1046 struct dentry *dentry;
1048 file = fget_light(dfd, &fput_needed);
1053 dentry = file->f_path.dentry;
1056 if (!S_ISDIR(dentry->d_inode->i_mode))
1059 retval = file_permission(file, MAY_EXEC);
1063 nd->path = file->f_path;
1064 path_get(&file->f_path);
1066 fput_light(file, fput_needed);
1071 fput_light(file, fput_needed);
1076 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1077 static int do_path_lookup(int dfd, const char *name,
1078 unsigned int flags, struct nameidata *nd)
1080 int retval = path_init(dfd, name, flags, nd);
1082 retval = path_walk(name, nd);
1083 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1084 nd->path.dentry->d_inode))
1085 audit_inode(name, nd->path.dentry);
1087 path_put(&nd->root);
1088 nd->root.mnt = NULL;
1093 int path_lookup(const char *name, unsigned int flags,
1094 struct nameidata *nd)
1096 return do_path_lookup(AT_FDCWD, name, flags, nd);
1099 int kern_path(const char *name, unsigned int flags, struct path *path)
1101 struct nameidata nd;
1102 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1109 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1110 * @dentry: pointer to dentry of the base directory
1111 * @mnt: pointer to vfs mount of the base directory
1112 * @name: pointer to file name
1113 * @flags: lookup flags
1114 * @nd: pointer to nameidata
1116 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1117 const char *name, unsigned int flags,
1118 struct nameidata *nd)
1122 /* same as do_path_lookup */
1123 nd->last_type = LAST_ROOT;
1127 nd->path.dentry = dentry;
1129 path_get(&nd->path);
1130 nd->root = nd->path;
1131 path_get(&nd->root);
1133 retval = path_walk(name, nd);
1134 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1135 nd->path.dentry->d_inode))
1136 audit_inode(name, nd->path.dentry);
1138 path_put(&nd->root);
1139 nd->root.mnt = NULL;
1145 * path_lookup_open - lookup a file path with open intent
1146 * @dfd: the directory to use as base, or AT_FDCWD
1147 * @name: pointer to file name
1148 * @lookup_flags: lookup intent flags
1149 * @nd: pointer to nameidata
1150 * @open_flags: open intent flags
1152 static int path_lookup_open(int dfd, const char *name,
1153 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1155 struct file *filp = get_empty_filp();
1160 nd->intent.open.file = filp;
1161 nd->intent.open.flags = open_flags;
1162 nd->intent.open.create_mode = 0;
1163 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1164 if (IS_ERR(nd->intent.open.file)) {
1166 err = PTR_ERR(nd->intent.open.file);
1167 path_put(&nd->path);
1169 } else if (err != 0)
1170 release_open_intent(nd);
1174 static struct dentry *__lookup_hash(struct qstr *name,
1175 struct dentry *base, struct nameidata *nd)
1177 struct dentry *dentry;
1178 struct inode *inode;
1181 inode = base->d_inode;
1184 * See if the low-level filesystem might want
1185 * to use its own hash..
1187 if (base->d_op && base->d_op->d_hash) {
1188 err = base->d_op->d_hash(base, name);
1189 dentry = ERR_PTR(err);
1194 dentry = cached_lookup(base, name, nd);
1198 /* Don't create child dentry for a dead directory. */
1199 dentry = ERR_PTR(-ENOENT);
1200 if (IS_DEADDIR(inode))
1203 new = d_alloc(base, name);
1204 dentry = ERR_PTR(-ENOMEM);
1207 dentry = inode->i_op->lookup(inode, new, nd);
1218 * Restricted form of lookup. Doesn't follow links, single-component only,
1219 * needs parent already locked. Doesn't follow mounts.
1222 static struct dentry *lookup_hash(struct nameidata *nd)
1226 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1228 return ERR_PTR(err);
1229 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1232 static int __lookup_one_len(const char *name, struct qstr *this,
1233 struct dentry *base, int len)
1243 hash = init_name_hash();
1245 c = *(const unsigned char *)name++;
1246 if (c == '/' || c == '\0')
1248 hash = partial_name_hash(c, hash);
1250 this->hash = end_name_hash(hash);
1255 * lookup_one_len - filesystem helper to lookup single pathname component
1256 * @name: pathname component to lookup
1257 * @base: base directory to lookup from
1258 * @len: maximum length @len should be interpreted to
1260 * Note that this routine is purely a helper for filesystem usage and should
1261 * not be called by generic code. Also note that by using this function the
1262 * nameidata argument is passed to the filesystem methods and a filesystem
1263 * using this helper needs to be prepared for that.
1265 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1270 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1272 err = __lookup_one_len(name, &this, base, len);
1274 return ERR_PTR(err);
1276 err = inode_permission(base->d_inode, MAY_EXEC);
1278 return ERR_PTR(err);
1279 return __lookup_hash(&this, base, NULL);
1283 * lookup_one_noperm - bad hack for sysfs
1284 * @name: pathname component to lookup
1285 * @base: base directory to lookup from
1287 * This is a variant of lookup_one_len that doesn't perform any permission
1288 * checks. It's a horrible hack to work around the braindead sysfs
1289 * architecture and should not be used anywhere else.
1291 * DON'T USE THIS FUNCTION EVER, thanks.
1293 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1298 err = __lookup_one_len(name, &this, base, strlen(name));
1300 return ERR_PTR(err);
1301 return __lookup_hash(&this, base, NULL);
1304 int user_path_at(int dfd, const char __user *name, unsigned flags,
1307 struct nameidata nd;
1308 char *tmp = getname(name);
1309 int err = PTR_ERR(tmp);
1312 BUG_ON(flags & LOOKUP_PARENT);
1314 err = do_path_lookup(dfd, tmp, flags, &nd);
1322 static int user_path_parent(int dfd, const char __user *path,
1323 struct nameidata *nd, char **name)
1325 char *s = getname(path);
1331 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1341 * It's inline, so penalty for filesystems that don't use sticky bit is
1344 static inline int check_sticky(struct inode *dir, struct inode *inode)
1346 uid_t fsuid = current_fsuid();
1348 if (!(dir->i_mode & S_ISVTX))
1350 if (inode->i_uid == fsuid)
1352 if (dir->i_uid == fsuid)
1354 return !capable(CAP_FOWNER);
1358 * Check whether we can remove a link victim from directory dir, check
1359 * whether the type of victim is right.
1360 * 1. We can't do it if dir is read-only (done in permission())
1361 * 2. We should have write and exec permissions on dir
1362 * 3. We can't remove anything from append-only dir
1363 * 4. We can't do anything with immutable dir (done in permission())
1364 * 5. If the sticky bit on dir is set we should either
1365 * a. be owner of dir, or
1366 * b. be owner of victim, or
1367 * c. have CAP_FOWNER capability
1368 * 6. If the victim is append-only or immutable we can't do antyhing with
1369 * links pointing to it.
1370 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1371 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1372 * 9. We can't remove a root or mountpoint.
1373 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1374 * nfs_async_unlink().
1376 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1380 if (!victim->d_inode)
1383 BUG_ON(victim->d_parent->d_inode != dir);
1384 audit_inode_child(victim->d_name.name, victim, dir);
1386 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1391 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1392 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1395 if (!S_ISDIR(victim->d_inode->i_mode))
1397 if (IS_ROOT(victim))
1399 } else if (S_ISDIR(victim->d_inode->i_mode))
1401 if (IS_DEADDIR(dir))
1403 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1408 /* Check whether we can create an object with dentry child in directory
1410 * 1. We can't do it if child already exists (open has special treatment for
1411 * this case, but since we are inlined it's OK)
1412 * 2. We can't do it if dir is read-only (done in permission())
1413 * 3. We should have write and exec permissions on dir
1414 * 4. We can't do it if dir is immutable (done in permission())
1416 static inline int may_create(struct inode *dir, struct dentry *child)
1420 if (IS_DEADDIR(dir))
1422 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1426 * O_DIRECTORY translates into forcing a directory lookup.
1428 static inline int lookup_flags(unsigned int f)
1430 unsigned long retval = LOOKUP_FOLLOW;
1433 retval &= ~LOOKUP_FOLLOW;
1435 if (f & O_DIRECTORY)
1436 retval |= LOOKUP_DIRECTORY;
1442 * p1 and p2 should be directories on the same fs.
1444 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1449 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1453 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1455 p = d_ancestor(p2, p1);
1457 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1458 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1462 p = d_ancestor(p1, p2);
1464 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1465 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1469 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1470 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1474 void unlock_rename(struct dentry *p1, struct dentry *p2)
1476 mutex_unlock(&p1->d_inode->i_mutex);
1478 mutex_unlock(&p2->d_inode->i_mutex);
1479 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1483 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1484 struct nameidata *nd)
1486 int error = may_create(dir, dentry);
1491 if (!dir->i_op->create)
1492 return -EACCES; /* shouldn't it be ENOSYS? */
1495 error = security_inode_create(dir, dentry, mode);
1499 error = dir->i_op->create(dir, dentry, mode, nd);
1501 fsnotify_create(dir, dentry);
1505 int may_open(struct path *path, int acc_mode, int flag)
1507 struct dentry *dentry = path->dentry;
1508 struct inode *inode = dentry->d_inode;
1514 switch (inode->i_mode & S_IFMT) {
1518 if (acc_mode & MAY_WRITE)
1523 if (path->mnt->mnt_flags & MNT_NODEV)
1532 error = inode_permission(inode, acc_mode);
1536 error = ima_path_check(path, acc_mode ?
1537 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1538 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1544 * An append-only file must be opened in append mode for writing.
1546 if (IS_APPEND(inode)) {
1548 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1554 /* O_NOATIME can only be set by the owner or superuser */
1555 if (flag & O_NOATIME)
1556 if (!is_owner_or_cap(inode)) {
1562 * Ensure there are no outstanding leases on the file.
1564 error = break_lease(inode, flag);
1568 if (flag & O_TRUNC) {
1569 error = get_write_access(inode);
1574 * Refuse to truncate files with mandatory locks held on them.
1576 error = locks_verify_locked(inode);
1578 error = security_path_truncate(path, 0,
1579 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1583 error = do_truncate(dentry, 0,
1584 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1587 put_write_access(inode);
1591 if (flag & FMODE_WRITE)
1596 ima_counts_put(path, acc_mode ?
1597 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1598 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1603 * Be careful about ever adding any more callers of this
1604 * function. Its flags must be in the namei format, not
1605 * what get passed to sys_open().
1607 static int __open_namei_create(struct nameidata *nd, struct path *path,
1611 struct dentry *dir = nd->path.dentry;
1613 if (!IS_POSIXACL(dir->d_inode))
1614 mode &= ~current_umask();
1615 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1618 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1620 mutex_unlock(&dir->d_inode->i_mutex);
1621 dput(nd->path.dentry);
1622 nd->path.dentry = path->dentry;
1625 /* Don't check for write permission, don't truncate */
1626 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1630 * Note that while the flag value (low two bits) for sys_open means:
1635 * it is changed into
1636 * 00 - no permissions needed
1637 * 01 - read-permission
1638 * 10 - write-permission
1640 * for the internal routines (ie open_namei()/follow_link() etc)
1641 * This is more logical, and also allows the 00 "no perm needed"
1642 * to be used for symlinks (where the permissions are checked
1646 static inline int open_to_namei_flags(int flag)
1648 if ((flag+1) & O_ACCMODE)
1653 static int open_will_write_to_fs(int flag, struct inode *inode)
1656 * We'll never write to the fs underlying
1659 if (special_file(inode->i_mode))
1661 return (flag & O_TRUNC);
1665 * Note that the low bits of the passed in "open_flag"
1666 * are not the same as in the local variable "flag". See
1667 * open_to_namei_flags() for more details.
1669 struct file *do_filp_open(int dfd, const char *pathname,
1670 int open_flag, int mode, int acc_mode)
1673 struct nameidata nd;
1679 int flag = open_to_namei_flags(open_flag);
1682 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1683 * check for O_DSYNC if the need any syncing at all we enforce it's
1684 * always set instead of having to deal with possibly weird behaviour
1685 * for malicious applications setting only __O_SYNC.
1687 if (open_flag & __O_SYNC)
1688 open_flag |= O_DSYNC;
1691 acc_mode = MAY_OPEN | ACC_MODE(flag);
1693 /* O_TRUNC implies we need access checks for write permissions */
1695 acc_mode |= MAY_WRITE;
1697 /* Allow the LSM permission hook to distinguish append
1698 access from general write access. */
1699 if (flag & O_APPEND)
1700 acc_mode |= MAY_APPEND;
1703 * The simplest case - just a plain lookup.
1705 if (!(flag & O_CREAT)) {
1706 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1709 return ERR_PTR(error);
1714 * Create - we need to know the parent.
1716 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1718 return ERR_PTR(error);
1719 error = path_walk(pathname, &nd);
1723 return ERR_PTR(error);
1725 if (unlikely(!audit_dummy_context()))
1726 audit_inode(pathname, nd.path.dentry);
1729 * We have the parent and last component. First of all, check
1730 * that we are not asked to creat(2) an obvious directory - that
1734 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1738 filp = get_empty_filp();
1741 nd.intent.open.file = filp;
1742 nd.intent.open.flags = flag;
1743 nd.intent.open.create_mode = mode;
1744 dir = nd.path.dentry;
1745 nd.flags &= ~LOOKUP_PARENT;
1746 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1748 nd.flags |= LOOKUP_EXCL;
1749 mutex_lock(&dir->d_inode->i_mutex);
1750 path.dentry = lookup_hash(&nd);
1751 path.mnt = nd.path.mnt;
1754 error = PTR_ERR(path.dentry);
1755 if (IS_ERR(path.dentry)) {
1756 mutex_unlock(&dir->d_inode->i_mutex);
1760 if (IS_ERR(nd.intent.open.file)) {
1761 error = PTR_ERR(nd.intent.open.file);
1762 goto exit_mutex_unlock;
1765 /* Negative dentry, just create the file */
1766 if (!path.dentry->d_inode) {
1768 * This write is needed to ensure that a
1769 * ro->rw transition does not occur between
1770 * the time when the file is created and when
1771 * a permanent write count is taken through
1772 * the 'struct file' in nameidata_to_filp().
1774 error = mnt_want_write(nd.path.mnt);
1776 goto exit_mutex_unlock;
1777 error = __open_namei_create(&nd, &path, flag, mode);
1779 mnt_drop_write(nd.path.mnt);
1782 filp = nameidata_to_filp(&nd, open_flag);
1784 ima_counts_put(&nd.path,
1785 acc_mode & (MAY_READ | MAY_WRITE |
1787 mnt_drop_write(nd.path.mnt);
1794 * It already exists.
1796 mutex_unlock(&dir->d_inode->i_mutex);
1797 audit_inode(pathname, path.dentry);
1803 if (__follow_mount(&path)) {
1805 if (flag & O_NOFOLLOW)
1810 if (!path.dentry->d_inode)
1812 if (path.dentry->d_inode->i_op->follow_link)
1815 path_to_nameidata(&path, &nd);
1817 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1822 * 1. may_open() truncates a file
1823 * 2. a rw->ro mount transition occurs
1824 * 3. nameidata_to_filp() fails due to
1826 * That would be inconsistent, and should
1827 * be avoided. Taking this mnt write here
1828 * ensures that (2) can not occur.
1830 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1832 error = mnt_want_write(nd.path.mnt);
1836 error = may_open(&nd.path, acc_mode, flag);
1839 mnt_drop_write(nd.path.mnt);
1842 filp = nameidata_to_filp(&nd, open_flag);
1844 ima_counts_put(&nd.path,
1845 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1847 * It is now safe to drop the mnt write
1848 * because the filp has had a write taken
1852 mnt_drop_write(nd.path.mnt);
1858 mutex_unlock(&dir->d_inode->i_mutex);
1860 path_put_conditional(&path, &nd);
1862 if (!IS_ERR(nd.intent.open.file))
1863 release_open_intent(&nd);
1868 return ERR_PTR(error);
1872 if (flag & O_NOFOLLOW)
1875 * This is subtle. Instead of calling do_follow_link() we do the
1876 * thing by hands. The reason is that this way we have zero link_count
1877 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1878 * After that we have the parent and last component, i.e.
1879 * we are in the same situation as after the first path_walk().
1880 * Well, almost - if the last component is normal we get its copy
1881 * stored in nd->last.name and we will have to putname() it when we
1882 * are done. Procfs-like symlinks just set LAST_BIND.
1884 nd.flags |= LOOKUP_PARENT;
1885 error = security_inode_follow_link(path.dentry, &nd);
1888 error = __do_follow_link(&path, &nd);
1890 /* Does someone understand code flow here? Or it is only
1891 * me so stupid? Anathema to whoever designed this non-sense
1892 * with "intent.open".
1894 release_open_intent(&nd);
1897 return ERR_PTR(error);
1899 nd.flags &= ~LOOKUP_PARENT;
1900 if (nd.last_type == LAST_BIND)
1903 if (nd.last_type != LAST_NORM)
1905 if (nd.last.name[nd.last.len]) {
1906 __putname(nd.last.name);
1911 __putname(nd.last.name);
1914 dir = nd.path.dentry;
1915 mutex_lock(&dir->d_inode->i_mutex);
1916 path.dentry = lookup_hash(&nd);
1917 path.mnt = nd.path.mnt;
1918 __putname(nd.last.name);
1923 * filp_open - open file and return file pointer
1925 * @filename: path to open
1926 * @flags: open flags as per the open(2) second argument
1927 * @mode: mode for the new file if O_CREAT is set, else ignored
1929 * This is the helper to open a file from kernelspace if you really
1930 * have to. But in generally you should not do this, so please move
1931 * along, nothing to see here..
1933 struct file *filp_open(const char *filename, int flags, int mode)
1935 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1937 EXPORT_SYMBOL(filp_open);
1940 * lookup_create - lookup a dentry, creating it if it doesn't exist
1941 * @nd: nameidata info
1942 * @is_dir: directory flag
1944 * Simple function to lookup and return a dentry and create it
1945 * if it doesn't exist. Is SMP-safe.
1947 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1949 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1951 struct dentry *dentry = ERR_PTR(-EEXIST);
1953 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1955 * Yucky last component or no last component at all?
1956 * (foo/., foo/.., /////)
1958 if (nd->last_type != LAST_NORM)
1960 nd->flags &= ~LOOKUP_PARENT;
1961 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1962 nd->intent.open.flags = O_EXCL;
1965 * Do the final lookup.
1967 dentry = lookup_hash(nd);
1971 if (dentry->d_inode)
1974 * Special case - lookup gave negative, but... we had foo/bar/
1975 * From the vfs_mknod() POV we just have a negative dentry -
1976 * all is fine. Let's be bastards - you had / on the end, you've
1977 * been asking for (non-existent) directory. -ENOENT for you.
1979 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1981 dentry = ERR_PTR(-ENOENT);
1986 dentry = ERR_PTR(-EEXIST);
1990 EXPORT_SYMBOL_GPL(lookup_create);
1992 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1994 int error = may_create(dir, dentry);
1999 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2002 if (!dir->i_op->mknod)
2005 error = devcgroup_inode_mknod(mode, dev);
2009 error = security_inode_mknod(dir, dentry, mode, dev);
2014 error = dir->i_op->mknod(dir, dentry, mode, dev);
2016 fsnotify_create(dir, dentry);
2020 static int may_mknod(mode_t mode)
2022 switch (mode & S_IFMT) {
2028 case 0: /* zero mode translates to S_IFREG */
2037 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2042 struct dentry *dentry;
2043 struct nameidata nd;
2048 error = user_path_parent(dfd, filename, &nd, &tmp);
2052 dentry = lookup_create(&nd, 0);
2053 if (IS_ERR(dentry)) {
2054 error = PTR_ERR(dentry);
2057 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2058 mode &= ~current_umask();
2059 error = may_mknod(mode);
2062 error = mnt_want_write(nd.path.mnt);
2065 error = security_path_mknod(&nd.path, dentry, mode, dev);
2067 goto out_drop_write;
2068 switch (mode & S_IFMT) {
2069 case 0: case S_IFREG:
2070 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2072 case S_IFCHR: case S_IFBLK:
2073 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2074 new_decode_dev(dev));
2076 case S_IFIFO: case S_IFSOCK:
2077 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2081 mnt_drop_write(nd.path.mnt);
2085 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2092 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2094 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2097 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2099 int error = may_create(dir, dentry);
2104 if (!dir->i_op->mkdir)
2107 mode &= (S_IRWXUGO|S_ISVTX);
2108 error = security_inode_mkdir(dir, dentry, mode);
2113 error = dir->i_op->mkdir(dir, dentry, mode);
2115 fsnotify_mkdir(dir, dentry);
2119 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2123 struct dentry *dentry;
2124 struct nameidata nd;
2126 error = user_path_parent(dfd, pathname, &nd, &tmp);
2130 dentry = lookup_create(&nd, 1);
2131 error = PTR_ERR(dentry);
2135 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2136 mode &= ~current_umask();
2137 error = mnt_want_write(nd.path.mnt);
2140 error = security_path_mkdir(&nd.path, dentry, mode);
2142 goto out_drop_write;
2143 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2145 mnt_drop_write(nd.path.mnt);
2149 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2156 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2158 return sys_mkdirat(AT_FDCWD, pathname, mode);
2162 * We try to drop the dentry early: we should have
2163 * a usage count of 2 if we're the only user of this
2164 * dentry, and if that is true (possibly after pruning
2165 * the dcache), then we drop the dentry now.
2167 * A low-level filesystem can, if it choses, legally
2170 * if (!d_unhashed(dentry))
2173 * if it cannot handle the case of removing a directory
2174 * that is still in use by something else..
2176 void dentry_unhash(struct dentry *dentry)
2179 shrink_dcache_parent(dentry);
2180 spin_lock(&dcache_lock);
2181 spin_lock(&dentry->d_lock);
2182 if (atomic_read(&dentry->d_count) == 2)
2184 spin_unlock(&dentry->d_lock);
2185 spin_unlock(&dcache_lock);
2188 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2190 int error = may_delete(dir, dentry, 1);
2195 if (!dir->i_op->rmdir)
2200 mutex_lock(&dentry->d_inode->i_mutex);
2201 dentry_unhash(dentry);
2202 if (d_mountpoint(dentry))
2205 error = security_inode_rmdir(dir, dentry);
2207 error = dir->i_op->rmdir(dir, dentry);
2209 dentry->d_inode->i_flags |= S_DEAD;
2212 mutex_unlock(&dentry->d_inode->i_mutex);
2221 static long do_rmdir(int dfd, const char __user *pathname)
2225 struct dentry *dentry;
2226 struct nameidata nd;
2228 error = user_path_parent(dfd, pathname, &nd, &name);
2232 switch(nd.last_type) {
2244 nd.flags &= ~LOOKUP_PARENT;
2246 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2247 dentry = lookup_hash(&nd);
2248 error = PTR_ERR(dentry);
2251 error = mnt_want_write(nd.path.mnt);
2254 error = security_path_rmdir(&nd.path, dentry);
2257 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2259 mnt_drop_write(nd.path.mnt);
2263 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2270 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2272 return do_rmdir(AT_FDCWD, pathname);
2275 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2277 int error = may_delete(dir, dentry, 0);
2282 if (!dir->i_op->unlink)
2287 mutex_lock(&dentry->d_inode->i_mutex);
2288 if (d_mountpoint(dentry))
2291 error = security_inode_unlink(dir, dentry);
2293 error = dir->i_op->unlink(dir, dentry);
2295 mutex_unlock(&dentry->d_inode->i_mutex);
2297 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2298 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2299 fsnotify_link_count(dentry->d_inode);
2307 * Make sure that the actual truncation of the file will occur outside its
2308 * directory's i_mutex. Truncate can take a long time if there is a lot of
2309 * writeout happening, and we don't want to prevent access to the directory
2310 * while waiting on the I/O.
2312 static long do_unlinkat(int dfd, const char __user *pathname)
2316 struct dentry *dentry;
2317 struct nameidata nd;
2318 struct inode *inode = NULL;
2320 error = user_path_parent(dfd, pathname, &nd, &name);
2325 if (nd.last_type != LAST_NORM)
2328 nd.flags &= ~LOOKUP_PARENT;
2330 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2331 dentry = lookup_hash(&nd);
2332 error = PTR_ERR(dentry);
2333 if (!IS_ERR(dentry)) {
2334 /* Why not before? Because we want correct error value */
2335 if (nd.last.name[nd.last.len])
2337 inode = dentry->d_inode;
2339 atomic_inc(&inode->i_count);
2340 error = mnt_want_write(nd.path.mnt);
2343 error = security_path_unlink(&nd.path, dentry);
2346 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2348 mnt_drop_write(nd.path.mnt);
2352 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2354 iput(inode); /* truncate the inode here */
2361 error = !dentry->d_inode ? -ENOENT :
2362 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2366 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2368 if ((flag & ~AT_REMOVEDIR) != 0)
2371 if (flag & AT_REMOVEDIR)
2372 return do_rmdir(dfd, pathname);
2374 return do_unlinkat(dfd, pathname);
2377 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2379 return do_unlinkat(AT_FDCWD, pathname);
2382 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2384 int error = may_create(dir, dentry);
2389 if (!dir->i_op->symlink)
2392 error = security_inode_symlink(dir, dentry, oldname);
2397 error = dir->i_op->symlink(dir, dentry, oldname);
2399 fsnotify_create(dir, dentry);
2403 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2404 int, newdfd, const char __user *, newname)
2409 struct dentry *dentry;
2410 struct nameidata nd;
2412 from = getname(oldname);
2414 return PTR_ERR(from);
2416 error = user_path_parent(newdfd, newname, &nd, &to);
2420 dentry = lookup_create(&nd, 0);
2421 error = PTR_ERR(dentry);
2425 error = mnt_want_write(nd.path.mnt);
2428 error = security_path_symlink(&nd.path, dentry, from);
2430 goto out_drop_write;
2431 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2433 mnt_drop_write(nd.path.mnt);
2437 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2445 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2447 return sys_symlinkat(oldname, AT_FDCWD, newname);
2450 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2452 struct inode *inode = old_dentry->d_inode;
2458 error = may_create(dir, new_dentry);
2462 if (dir->i_sb != inode->i_sb)
2466 * A link to an append-only or immutable file cannot be created.
2468 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2470 if (!dir->i_op->link)
2472 if (S_ISDIR(inode->i_mode))
2475 error = security_inode_link(old_dentry, dir, new_dentry);
2479 mutex_lock(&inode->i_mutex);
2481 error = dir->i_op->link(old_dentry, dir, new_dentry);
2482 mutex_unlock(&inode->i_mutex);
2484 fsnotify_link(dir, inode, new_dentry);
2489 * Hardlinks are often used in delicate situations. We avoid
2490 * security-related surprises by not following symlinks on the
2493 * We don't follow them on the oldname either to be compatible
2494 * with linux 2.0, and to avoid hard-linking to directories
2495 * and other special files. --ADM
2497 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2498 int, newdfd, const char __user *, newname, int, flags)
2500 struct dentry *new_dentry;
2501 struct nameidata nd;
2502 struct path old_path;
2506 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2509 error = user_path_at(olddfd, oldname,
2510 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2515 error = user_path_parent(newdfd, newname, &nd, &to);
2519 if (old_path.mnt != nd.path.mnt)
2521 new_dentry = lookup_create(&nd, 0);
2522 error = PTR_ERR(new_dentry);
2523 if (IS_ERR(new_dentry))
2525 error = mnt_want_write(nd.path.mnt);
2528 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2530 goto out_drop_write;
2531 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2533 mnt_drop_write(nd.path.mnt);
2537 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2542 path_put(&old_path);
2547 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2549 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2553 * The worst of all namespace operations - renaming directory. "Perverted"
2554 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2556 * a) we can get into loop creation. Check is done in is_subdir().
2557 * b) race potential - two innocent renames can create a loop together.
2558 * That's where 4.4 screws up. Current fix: serialization on
2559 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2561 * c) we have to lock _three_ objects - parents and victim (if it exists).
2562 * And that - after we got ->i_mutex on parents (until then we don't know
2563 * whether the target exists). Solution: try to be smart with locking
2564 * order for inodes. We rely on the fact that tree topology may change
2565 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2566 * move will be locked. Thus we can rank directories by the tree
2567 * (ancestors first) and rank all non-directories after them.
2568 * That works since everybody except rename does "lock parent, lookup,
2569 * lock child" and rename is under ->s_vfs_rename_mutex.
2570 * HOWEVER, it relies on the assumption that any object with ->lookup()
2571 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2572 * we'd better make sure that there's no link(2) for them.
2573 * d) some filesystems don't support opened-but-unlinked directories,
2574 * either because of layout or because they are not ready to deal with
2575 * all cases correctly. The latter will be fixed (taking this sort of
2576 * stuff into VFS), but the former is not going away. Solution: the same
2577 * trick as in rmdir().
2578 * e) conversion from fhandle to dentry may come in the wrong moment - when
2579 * we are removing the target. Solution: we will have to grab ->i_mutex
2580 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2581 * ->i_mutex on parents, which works but leads to some truely excessive
2584 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2585 struct inode *new_dir, struct dentry *new_dentry)
2588 struct inode *target;
2591 * If we are going to change the parent - check write permissions,
2592 * we'll need to flip '..'.
2594 if (new_dir != old_dir) {
2595 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2600 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2604 target = new_dentry->d_inode;
2606 mutex_lock(&target->i_mutex);
2607 dentry_unhash(new_dentry);
2609 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2612 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2615 target->i_flags |= S_DEAD;
2616 mutex_unlock(&target->i_mutex);
2617 if (d_unhashed(new_dentry))
2618 d_rehash(new_dentry);
2622 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2623 d_move(old_dentry,new_dentry);
2627 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2628 struct inode *new_dir, struct dentry *new_dentry)
2630 struct inode *target;
2633 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2638 target = new_dentry->d_inode;
2640 mutex_lock(&target->i_mutex);
2641 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2644 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2646 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2647 d_move(old_dentry, new_dentry);
2650 mutex_unlock(&target->i_mutex);
2655 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2656 struct inode *new_dir, struct dentry *new_dentry)
2659 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2660 const char *old_name;
2662 if (old_dentry->d_inode == new_dentry->d_inode)
2665 error = may_delete(old_dir, old_dentry, is_dir);
2669 if (!new_dentry->d_inode)
2670 error = may_create(new_dir, new_dentry);
2672 error = may_delete(new_dir, new_dentry, is_dir);
2676 if (!old_dir->i_op->rename)
2679 vfs_dq_init(old_dir);
2680 vfs_dq_init(new_dir);
2682 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2685 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2687 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2689 const char *new_name = old_dentry->d_name.name;
2690 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2691 new_dentry->d_inode, old_dentry);
2693 fsnotify_oldname_free(old_name);
2698 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2699 int, newdfd, const char __user *, newname)
2701 struct dentry *old_dir, *new_dir;
2702 struct dentry *old_dentry, *new_dentry;
2703 struct dentry *trap;
2704 struct nameidata oldnd, newnd;
2709 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2713 error = user_path_parent(newdfd, newname, &newnd, &to);
2718 if (oldnd.path.mnt != newnd.path.mnt)
2721 old_dir = oldnd.path.dentry;
2723 if (oldnd.last_type != LAST_NORM)
2726 new_dir = newnd.path.dentry;
2727 if (newnd.last_type != LAST_NORM)
2730 oldnd.flags &= ~LOOKUP_PARENT;
2731 newnd.flags &= ~LOOKUP_PARENT;
2732 newnd.flags |= LOOKUP_RENAME_TARGET;
2734 trap = lock_rename(new_dir, old_dir);
2736 old_dentry = lookup_hash(&oldnd);
2737 error = PTR_ERR(old_dentry);
2738 if (IS_ERR(old_dentry))
2740 /* source must exist */
2742 if (!old_dentry->d_inode)
2744 /* unless the source is a directory trailing slashes give -ENOTDIR */
2745 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2747 if (oldnd.last.name[oldnd.last.len])
2749 if (newnd.last.name[newnd.last.len])
2752 /* source should not be ancestor of target */
2754 if (old_dentry == trap)
2756 new_dentry = lookup_hash(&newnd);
2757 error = PTR_ERR(new_dentry);
2758 if (IS_ERR(new_dentry))
2760 /* target should not be an ancestor of source */
2762 if (new_dentry == trap)
2765 error = mnt_want_write(oldnd.path.mnt);
2768 error = security_path_rename(&oldnd.path, old_dentry,
2769 &newnd.path, new_dentry);
2772 error = vfs_rename(old_dir->d_inode, old_dentry,
2773 new_dir->d_inode, new_dentry);
2775 mnt_drop_write(oldnd.path.mnt);
2781 unlock_rename(new_dir, old_dir);
2783 path_put(&newnd.path);
2786 path_put(&oldnd.path);
2792 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2794 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2797 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2801 len = PTR_ERR(link);
2806 if (len > (unsigned) buflen)
2808 if (copy_to_user(buffer, link, len))
2815 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2816 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2817 * using) it for any given inode is up to filesystem.
2819 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2821 struct nameidata nd;
2826 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2828 return PTR_ERR(cookie);
2830 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2831 if (dentry->d_inode->i_op->put_link)
2832 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2836 int vfs_follow_link(struct nameidata *nd, const char *link)
2838 return __vfs_follow_link(nd, link);
2841 /* get the link contents into pagecache */
2842 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2846 struct address_space *mapping = dentry->d_inode->i_mapping;
2847 page = read_mapping_page(mapping, 0, NULL);
2852 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2856 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2858 struct page *page = NULL;
2859 char *s = page_getlink(dentry, &page);
2860 int res = vfs_readlink(dentry,buffer,buflen,s);
2863 page_cache_release(page);
2868 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2870 struct page *page = NULL;
2871 nd_set_link(nd, page_getlink(dentry, &page));
2875 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2877 struct page *page = cookie;
2881 page_cache_release(page);
2886 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2888 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2890 struct address_space *mapping = inode->i_mapping;
2895 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2897 flags |= AOP_FLAG_NOFS;
2900 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2901 flags, &page, &fsdata);
2905 kaddr = kmap_atomic(page, KM_USER0);
2906 memcpy(kaddr, symname, len-1);
2907 kunmap_atomic(kaddr, KM_USER0);
2909 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2916 mark_inode_dirty(inode);
2922 int page_symlink(struct inode *inode, const char *symname, int len)
2924 return __page_symlink(inode, symname, len,
2925 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2928 const struct inode_operations page_symlink_inode_operations = {
2929 .readlink = generic_readlink,
2930 .follow_link = page_follow_link_light,
2931 .put_link = page_put_link,
2934 EXPORT_SYMBOL(user_path_at);
2935 EXPORT_SYMBOL(follow_down);
2936 EXPORT_SYMBOL(follow_up);
2937 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2938 EXPORT_SYMBOL(getname);
2939 EXPORT_SYMBOL(lock_rename);
2940 EXPORT_SYMBOL(lookup_one_len);
2941 EXPORT_SYMBOL(page_follow_link_light);
2942 EXPORT_SYMBOL(page_put_link);
2943 EXPORT_SYMBOL(page_readlink);
2944 EXPORT_SYMBOL(__page_symlink);
2945 EXPORT_SYMBOL(page_symlink);
2946 EXPORT_SYMBOL(page_symlink_inode_operations);
2947 EXPORT_SYMBOL(path_lookup);
2948 EXPORT_SYMBOL(kern_path);
2949 EXPORT_SYMBOL(vfs_path_lookup);
2950 EXPORT_SYMBOL(inode_permission);
2951 EXPORT_SYMBOL(file_permission);
2952 EXPORT_SYMBOL(unlock_rename);
2953 EXPORT_SYMBOL(vfs_create);
2954 EXPORT_SYMBOL(vfs_follow_link);
2955 EXPORT_SYMBOL(vfs_link);
2956 EXPORT_SYMBOL(vfs_mkdir);
2957 EXPORT_SYMBOL(vfs_mknod);
2958 EXPORT_SYMBOL(generic_permission);
2959 EXPORT_SYMBOL(vfs_readlink);
2960 EXPORT_SYMBOL(vfs_rename);
2961 EXPORT_SYMBOL(vfs_rmdir);
2962 EXPORT_SYMBOL(vfs_symlink);
2963 EXPORT_SYMBOL(vfs_unlink);
2964 EXPORT_SYMBOL(dentry_unhash);
2965 EXPORT_SYMBOL(generic_readlink);