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);
174 * generic_permission - check for access rights on a Posix-like filesystem
175 * @inode: inode to check access rights for
176 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
177 * @check_acl: optional callback to check for Posix ACLs
179 * Used to check for read/write/execute permissions on a file.
180 * We use "fsuid" for this, letting us set arbitrary permissions
181 * for filesystem access without changing the "normal" uids which
182 * are used for other things..
184 int generic_permission(struct inode *inode, int mask,
185 int (*check_acl)(struct inode *inode, int mask))
187 umode_t mode = inode->i_mode;
189 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
191 if (current_fsuid() == inode->i_uid)
194 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
195 int error = check_acl(inode, mask);
196 if (error == -EACCES)
197 goto check_capabilities;
198 else if (error != -EAGAIN)
202 if (in_group_p(inode->i_gid))
207 * If the DACs are ok we don't need any capability check.
209 if ((mask & ~mode) == 0)
214 * Read/write DACs are always overridable.
215 * Executable DACs are overridable if at least one exec bit is set.
217 if (!(mask & MAY_EXEC) || execute_ok(inode))
218 if (capable(CAP_DAC_OVERRIDE))
222 * Searching includes executable on directories, else just read.
224 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
225 if (capable(CAP_DAC_READ_SEARCH))
232 * inode_permission - check for access rights to a given inode
233 * @inode: inode to check permission on
234 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
236 * Used to check for read/write/execute permissions on an inode.
237 * We use "fsuid" for this, letting us set arbitrary permissions
238 * for filesystem access without changing the "normal" uids which
239 * are used for other things.
241 int inode_permission(struct inode *inode, int mask)
245 if (mask & MAY_WRITE) {
246 umode_t mode = inode->i_mode;
249 * Nobody gets write access to a read-only fs.
251 if (IS_RDONLY(inode) &&
252 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
256 * Nobody gets write access to an immutable file.
258 if (IS_IMMUTABLE(inode))
262 if (inode->i_op->permission)
263 retval = inode->i_op->permission(inode, mask);
265 retval = generic_permission(inode, mask, NULL);
270 retval = devcgroup_inode_permission(inode, mask);
274 return security_inode_permission(inode,
275 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
279 * file_permission - check for additional access rights to a given file
280 * @file: file to check access rights for
281 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
283 * Used to check for read/write/execute permissions on an already opened
287 * Do not use this function in new code. All access checks should
288 * be done using inode_permission().
290 int file_permission(struct file *file, int mask)
292 return inode_permission(file->f_path.dentry->d_inode, mask);
296 * get_write_access() gets write permission for a file.
297 * put_write_access() releases this write permission.
298 * This is used for regular files.
299 * We cannot support write (and maybe mmap read-write shared) accesses and
300 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
301 * can have the following values:
302 * 0: no writers, no VM_DENYWRITE mappings
303 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
304 * > 0: (i_writecount) users are writing to the file.
306 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
307 * except for the cases where we don't hold i_writecount yet. Then we need to
308 * use {get,deny}_write_access() - these functions check the sign and refuse
309 * to do the change if sign is wrong. Exclusion between them is provided by
310 * the inode->i_lock spinlock.
313 int get_write_access(struct inode * inode)
315 spin_lock(&inode->i_lock);
316 if (atomic_read(&inode->i_writecount) < 0) {
317 spin_unlock(&inode->i_lock);
320 atomic_inc(&inode->i_writecount);
321 spin_unlock(&inode->i_lock);
326 int deny_write_access(struct file * file)
328 struct inode *inode = file->f_path.dentry->d_inode;
330 spin_lock(&inode->i_lock);
331 if (atomic_read(&inode->i_writecount) > 0) {
332 spin_unlock(&inode->i_lock);
335 atomic_dec(&inode->i_writecount);
336 spin_unlock(&inode->i_lock);
342 * path_get - get a reference to a path
343 * @path: path to get the reference to
345 * Given a path increment the reference count to the dentry and the vfsmount.
347 void path_get(struct path *path)
352 EXPORT_SYMBOL(path_get);
355 * path_put - put a reference to a path
356 * @path: path to put the reference to
358 * Given a path decrement the reference count to the dentry and the vfsmount.
360 void path_put(struct path *path)
365 EXPORT_SYMBOL(path_put);
368 * release_open_intent - free up open intent resources
369 * @nd: pointer to nameidata
371 void release_open_intent(struct nameidata *nd)
373 if (nd->intent.open.file->f_path.dentry == NULL)
374 put_filp(nd->intent.open.file);
376 fput(nd->intent.open.file);
379 static inline struct dentry *
380 do_revalidate(struct dentry *dentry, struct nameidata *nd)
382 int status = dentry->d_op->d_revalidate(dentry, nd);
383 if (unlikely(status <= 0)) {
385 * The dentry failed validation.
386 * If d_revalidate returned 0 attempt to invalidate
387 * the dentry otherwise d_revalidate is asking us
388 * to return a fail status.
391 if (!d_invalidate(dentry)) {
397 dentry = ERR_PTR(status);
404 * Internal lookup() using the new generic dcache.
407 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
409 struct dentry * dentry = __d_lookup(parent, name);
411 /* lockess __d_lookup may fail due to concurrent d_move()
412 * in some unrelated directory, so try with d_lookup
415 dentry = d_lookup(parent, name);
417 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
418 dentry = do_revalidate(dentry, nd);
424 * Short-cut version of permission(), for calling by
425 * path_walk(), when dcache lock is held. Combines parts
426 * of permission() and generic_permission(), and tests ONLY for
427 * MAY_EXEC permission.
429 * If appropriate, check DAC only. If not appropriate, or
430 * short-cut DAC fails, then call permission() to do more
431 * complete permission check.
433 static int exec_permission_lite(struct inode *inode)
435 umode_t mode = inode->i_mode;
437 if (inode->i_op->permission)
438 return inode_permission(inode, MAY_EXEC);
440 if (current_fsuid() == inode->i_uid)
442 else if (in_group_p(inode->i_gid))
448 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
453 return security_inode_permission(inode, MAY_EXEC);
457 * This is called when everything else fails, and we actually have
458 * to go to the low-level filesystem to find out what we should do..
460 * We get the directory semaphore, and after getting that we also
461 * make sure that nobody added the entry to the dcache in the meantime..
464 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
466 struct dentry * result;
467 struct inode *dir = parent->d_inode;
469 mutex_lock(&dir->i_mutex);
471 * First re-do the cached lookup just in case it was created
472 * while we waited for the directory semaphore..
474 * FIXME! This could use version numbering or similar to
475 * avoid unnecessary cache lookups.
477 * The "dcache_lock" is purely to protect the RCU list walker
478 * from concurrent renames at this point (we mustn't get false
479 * negatives from the RCU list walk here, unlike the optimistic
482 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
484 result = d_lookup(parent, name);
486 struct dentry *dentry;
488 /* Don't create child dentry for a dead directory. */
489 result = ERR_PTR(-ENOENT);
493 dentry = d_alloc(parent, name);
494 result = ERR_PTR(-ENOMEM);
496 result = dir->i_op->lookup(dir, dentry, nd);
503 mutex_unlock(&dir->i_mutex);
508 * Uhhuh! Nasty case: the cache was re-populated while
509 * we waited on the semaphore. Need to revalidate.
511 mutex_unlock(&dir->i_mutex);
512 if (result->d_op && result->d_op->d_revalidate) {
513 result = do_revalidate(result, nd);
515 result = ERR_PTR(-ENOENT);
521 * Wrapper to retry pathname resolution whenever the underlying
522 * file system returns an ESTALE.
524 * Retry the whole path once, forcing real lookup requests
525 * instead of relying on the dcache.
527 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
529 struct path save = nd->path;
532 /* make sure the stuff we saved doesn't go away */
535 result = __link_path_walk(name, nd);
536 if (result == -ESTALE) {
537 /* nd->path had been dropped */
540 nd->flags |= LOOKUP_REVAL;
541 result = __link_path_walk(name, nd);
549 static __always_inline void set_root(struct nameidata *nd)
552 struct fs_struct *fs = current->fs;
553 read_lock(&fs->lock);
556 read_unlock(&fs->lock);
560 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
574 res = link_path_walk(link, nd);
575 if (nd->depth || res || nd->last_type!=LAST_NORM)
578 * If it is an iterative symlinks resolution in open_namei() we
579 * have to copy the last component. And all that crap because of
580 * bloody create() on broken symlinks. Furrfu...
583 if (unlikely(!name)) {
587 strcpy(name, nd->last.name);
588 nd->last.name = name;
592 return PTR_ERR(link);
595 static void path_put_conditional(struct path *path, struct nameidata *nd)
598 if (path->mnt != nd->path.mnt)
602 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
604 dput(nd->path.dentry);
605 if (nd->path.mnt != path->mnt)
606 mntput(nd->path.mnt);
607 nd->path.mnt = path->mnt;
608 nd->path.dentry = path->dentry;
611 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
615 struct dentry *dentry = path->dentry;
617 touch_atime(path->mnt, dentry);
618 nd_set_link(nd, NULL);
620 if (path->mnt != nd->path.mnt) {
621 path_to_nameidata(path, nd);
625 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
626 error = PTR_ERR(cookie);
627 if (!IS_ERR(cookie)) {
628 char *s = nd_get_link(nd);
631 error = __vfs_follow_link(nd, s);
632 if (dentry->d_inode->i_op->put_link)
633 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
641 * This limits recursive symlink follows to 8, while
642 * limiting consecutive symlinks to 40.
644 * Without that kind of total limit, nasty chains of consecutive
645 * symlinks can cause almost arbitrarily long lookups.
647 static inline int do_follow_link(struct path *path, struct nameidata *nd)
650 if (current->link_count >= MAX_NESTED_LINKS)
652 if (current->total_link_count >= 40)
654 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
656 err = security_inode_follow_link(path->dentry, nd);
659 current->link_count++;
660 current->total_link_count++;
662 err = __do_follow_link(path, nd);
663 current->link_count--;
667 path_put_conditional(path, nd);
672 int follow_up(struct path *path)
674 struct vfsmount *parent;
675 struct dentry *mountpoint;
676 spin_lock(&vfsmount_lock);
677 parent = path->mnt->mnt_parent;
678 if (parent == path->mnt) {
679 spin_unlock(&vfsmount_lock);
683 mountpoint = dget(path->mnt->mnt_mountpoint);
684 spin_unlock(&vfsmount_lock);
686 path->dentry = mountpoint;
692 /* no need for dcache_lock, as serialization is taken care in
695 static int __follow_mount(struct path *path)
698 while (d_mountpoint(path->dentry)) {
699 struct vfsmount *mounted = lookup_mnt(path);
706 path->dentry = dget(mounted->mnt_root);
712 static void follow_mount(struct path *path)
714 while (d_mountpoint(path->dentry)) {
715 struct vfsmount *mounted = lookup_mnt(path);
721 path->dentry = dget(mounted->mnt_root);
725 /* no need for dcache_lock, as serialization is taken care in
728 int follow_down(struct path *path)
730 struct vfsmount *mounted;
732 mounted = lookup_mnt(path);
737 path->dentry = dget(mounted->mnt_root);
743 static __always_inline void follow_dotdot(struct nameidata *nd)
748 struct vfsmount *parent;
749 struct dentry *old = nd->path.dentry;
751 if (nd->path.dentry == nd->root.dentry &&
752 nd->path.mnt == nd->root.mnt) {
755 spin_lock(&dcache_lock);
756 if (nd->path.dentry != nd->path.mnt->mnt_root) {
757 nd->path.dentry = dget(nd->path.dentry->d_parent);
758 spin_unlock(&dcache_lock);
762 spin_unlock(&dcache_lock);
763 spin_lock(&vfsmount_lock);
764 parent = nd->path.mnt->mnt_parent;
765 if (parent == nd->path.mnt) {
766 spin_unlock(&vfsmount_lock);
770 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
771 spin_unlock(&vfsmount_lock);
773 mntput(nd->path.mnt);
774 nd->path.mnt = parent;
776 follow_mount(&nd->path);
780 * It's more convoluted than I'd like it to be, but... it's still fairly
781 * small and for now I'd prefer to have fast path as straight as possible.
782 * It _is_ time-critical.
784 static int do_lookup(struct nameidata *nd, struct qstr *name,
787 struct vfsmount *mnt = nd->path.mnt;
788 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
792 if (dentry->d_op && dentry->d_op->d_revalidate)
793 goto need_revalidate;
796 path->dentry = dentry;
797 __follow_mount(path);
801 dentry = real_lookup(nd->path.dentry, name, nd);
807 dentry = do_revalidate(dentry, nd);
815 return PTR_ERR(dentry);
820 * This is the basic name resolution function, turning a pathname into
821 * the final dentry. We expect 'base' to be positive and a directory.
823 * Returns 0 and nd will have valid dentry and mnt on success.
824 * Returns error and drops reference to input namei data on failure.
826 static int __link_path_walk(const char *name, struct nameidata *nd)
831 unsigned int lookup_flags = nd->flags;
838 inode = nd->path.dentry->d_inode;
840 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
842 /* At this point we know we have a real path component. */
848 nd->flags |= LOOKUP_CONTINUE;
849 err = exec_permission_lite(inode);
854 c = *(const unsigned char *)name;
856 hash = init_name_hash();
859 hash = partial_name_hash(c, hash);
860 c = *(const unsigned char *)name;
861 } while (c && (c != '/'));
862 this.len = name - (const char *) this.name;
863 this.hash = end_name_hash(hash);
865 /* remove trailing slashes? */
868 while (*++name == '/');
870 goto last_with_slashes;
873 * "." and ".." are special - ".." especially so because it has
874 * to be able to know about the current root directory and
875 * parent relationships.
877 if (this.name[0] == '.') switch (this.len) {
881 if (this.name[1] != '.')
884 inode = nd->path.dentry->d_inode;
890 * See if the low-level filesystem might want
891 * to use its own hash..
893 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
894 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
899 /* This does the actual lookups.. */
900 err = do_lookup(nd, &this, &next);
905 inode = next.dentry->d_inode;
909 if (inode->i_op->follow_link) {
910 err = do_follow_link(&next, nd);
914 inode = nd->path.dentry->d_inode;
918 path_to_nameidata(&next, nd);
920 if (!inode->i_op->lookup)
923 /* here ends the main loop */
926 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
928 /* Clear LOOKUP_CONTINUE iff it was previously unset */
929 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
930 if (lookup_flags & LOOKUP_PARENT)
932 if (this.name[0] == '.') switch (this.len) {
936 if (this.name[1] != '.')
939 inode = nd->path.dentry->d_inode;
944 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
945 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
950 err = do_lookup(nd, &this, &next);
953 inode = next.dentry->d_inode;
954 if ((lookup_flags & LOOKUP_FOLLOW)
955 && inode && inode->i_op->follow_link) {
956 err = do_follow_link(&next, nd);
959 inode = nd->path.dentry->d_inode;
961 path_to_nameidata(&next, nd);
965 if (lookup_flags & LOOKUP_DIRECTORY) {
967 if (!inode->i_op->lookup)
973 nd->last_type = LAST_NORM;
974 if (this.name[0] != '.')
977 nd->last_type = LAST_DOT;
978 else if (this.len == 2 && this.name[1] == '.')
979 nd->last_type = LAST_DOTDOT;
984 * We bypassed the ordinary revalidation routines.
985 * We may need to check the cached dentry for staleness.
987 if (nd->path.dentry && nd->path.dentry->d_sb &&
988 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
990 /* Note: we do not d_invalidate() */
991 if (!nd->path.dentry->d_op->d_revalidate(
992 nd->path.dentry, nd))
998 path_put_conditional(&next, nd);
1001 path_put(&nd->path);
1006 static int path_walk(const char *name, struct nameidata *nd)
1008 current->total_link_count = 0;
1009 return link_path_walk(name, nd);
1012 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1018 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1021 nd->root.mnt = NULL;
1025 nd->path = nd->root;
1026 path_get(&nd->root);
1027 } else if (dfd == AT_FDCWD) {
1028 struct fs_struct *fs = current->fs;
1029 read_lock(&fs->lock);
1032 read_unlock(&fs->lock);
1034 struct dentry *dentry;
1036 file = fget_light(dfd, &fput_needed);
1041 dentry = file->f_path.dentry;
1044 if (!S_ISDIR(dentry->d_inode->i_mode))
1047 retval = file_permission(file, MAY_EXEC);
1051 nd->path = file->f_path;
1052 path_get(&file->f_path);
1054 fput_light(file, fput_needed);
1059 fput_light(file, fput_needed);
1064 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1065 static int do_path_lookup(int dfd, const char *name,
1066 unsigned int flags, struct nameidata *nd)
1068 int retval = path_init(dfd, name, flags, nd);
1070 retval = path_walk(name, nd);
1071 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1072 nd->path.dentry->d_inode))
1073 audit_inode(name, nd->path.dentry);
1075 path_put(&nd->root);
1076 nd->root.mnt = NULL;
1081 int path_lookup(const char *name, unsigned int flags,
1082 struct nameidata *nd)
1084 return do_path_lookup(AT_FDCWD, name, flags, nd);
1087 int kern_path(const char *name, unsigned int flags, struct path *path)
1089 struct nameidata nd;
1090 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1097 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1098 * @dentry: pointer to dentry of the base directory
1099 * @mnt: pointer to vfs mount of the base directory
1100 * @name: pointer to file name
1101 * @flags: lookup flags
1102 * @nd: pointer to nameidata
1104 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1105 const char *name, unsigned int flags,
1106 struct nameidata *nd)
1110 /* same as do_path_lookup */
1111 nd->last_type = LAST_ROOT;
1115 nd->path.dentry = dentry;
1117 path_get(&nd->path);
1118 nd->root = nd->path;
1119 path_get(&nd->root);
1121 retval = path_walk(name, nd);
1122 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1123 nd->path.dentry->d_inode))
1124 audit_inode(name, nd->path.dentry);
1126 path_put(&nd->root);
1127 nd->root.mnt = NULL;
1133 * path_lookup_open - lookup a file path with open intent
1134 * @dfd: the directory to use as base, or AT_FDCWD
1135 * @name: pointer to file name
1136 * @lookup_flags: lookup intent flags
1137 * @nd: pointer to nameidata
1138 * @open_flags: open intent flags
1140 static int path_lookup_open(int dfd, const char *name,
1141 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1143 struct file *filp = get_empty_filp();
1148 nd->intent.open.file = filp;
1149 nd->intent.open.flags = open_flags;
1150 nd->intent.open.create_mode = 0;
1151 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1152 if (IS_ERR(nd->intent.open.file)) {
1154 err = PTR_ERR(nd->intent.open.file);
1155 path_put(&nd->path);
1157 } else if (err != 0)
1158 release_open_intent(nd);
1162 static struct dentry *__lookup_hash(struct qstr *name,
1163 struct dentry *base, struct nameidata *nd)
1165 struct dentry *dentry;
1166 struct inode *inode;
1169 inode = base->d_inode;
1172 * See if the low-level filesystem might want
1173 * to use its own hash..
1175 if (base->d_op && base->d_op->d_hash) {
1176 err = base->d_op->d_hash(base, name);
1177 dentry = ERR_PTR(err);
1182 dentry = cached_lookup(base, name, nd);
1186 /* Don't create child dentry for a dead directory. */
1187 dentry = ERR_PTR(-ENOENT);
1188 if (IS_DEADDIR(inode))
1191 new = d_alloc(base, name);
1192 dentry = ERR_PTR(-ENOMEM);
1195 dentry = inode->i_op->lookup(inode, new, nd);
1206 * Restricted form of lookup. Doesn't follow links, single-component only,
1207 * needs parent already locked. Doesn't follow mounts.
1210 static struct dentry *lookup_hash(struct nameidata *nd)
1214 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1216 return ERR_PTR(err);
1217 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1220 static int __lookup_one_len(const char *name, struct qstr *this,
1221 struct dentry *base, int len)
1231 hash = init_name_hash();
1233 c = *(const unsigned char *)name++;
1234 if (c == '/' || c == '\0')
1236 hash = partial_name_hash(c, hash);
1238 this->hash = end_name_hash(hash);
1243 * lookup_one_len - filesystem helper to lookup single pathname component
1244 * @name: pathname component to lookup
1245 * @base: base directory to lookup from
1246 * @len: maximum length @len should be interpreted to
1248 * Note that this routine is purely a helper for filesystem usage and should
1249 * not be called by generic code. Also note that by using this function the
1250 * nameidata argument is passed to the filesystem methods and a filesystem
1251 * using this helper needs to be prepared for that.
1253 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1258 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1260 err = __lookup_one_len(name, &this, base, len);
1262 return ERR_PTR(err);
1264 err = inode_permission(base->d_inode, MAY_EXEC);
1266 return ERR_PTR(err);
1267 return __lookup_hash(&this, base, NULL);
1271 * lookup_one_noperm - bad hack for sysfs
1272 * @name: pathname component to lookup
1273 * @base: base directory to lookup from
1275 * This is a variant of lookup_one_len that doesn't perform any permission
1276 * checks. It's a horrible hack to work around the braindead sysfs
1277 * architecture and should not be used anywhere else.
1279 * DON'T USE THIS FUNCTION EVER, thanks.
1281 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1286 err = __lookup_one_len(name, &this, base, strlen(name));
1288 return ERR_PTR(err);
1289 return __lookup_hash(&this, base, NULL);
1292 int user_path_at(int dfd, const char __user *name, unsigned flags,
1295 struct nameidata nd;
1296 char *tmp = getname(name);
1297 int err = PTR_ERR(tmp);
1300 BUG_ON(flags & LOOKUP_PARENT);
1302 err = do_path_lookup(dfd, tmp, flags, &nd);
1310 static int user_path_parent(int dfd, const char __user *path,
1311 struct nameidata *nd, char **name)
1313 char *s = getname(path);
1319 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1329 * It's inline, so penalty for filesystems that don't use sticky bit is
1332 static inline int check_sticky(struct inode *dir, struct inode *inode)
1334 uid_t fsuid = current_fsuid();
1336 if (!(dir->i_mode & S_ISVTX))
1338 if (inode->i_uid == fsuid)
1340 if (dir->i_uid == fsuid)
1342 return !capable(CAP_FOWNER);
1346 * Check whether we can remove a link victim from directory dir, check
1347 * whether the type of victim is right.
1348 * 1. We can't do it if dir is read-only (done in permission())
1349 * 2. We should have write and exec permissions on dir
1350 * 3. We can't remove anything from append-only dir
1351 * 4. We can't do anything with immutable dir (done in permission())
1352 * 5. If the sticky bit on dir is set we should either
1353 * a. be owner of dir, or
1354 * b. be owner of victim, or
1355 * c. have CAP_FOWNER capability
1356 * 6. If the victim is append-only or immutable we can't do antyhing with
1357 * links pointing to it.
1358 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1359 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1360 * 9. We can't remove a root or mountpoint.
1361 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1362 * nfs_async_unlink().
1364 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1368 if (!victim->d_inode)
1371 BUG_ON(victim->d_parent->d_inode != dir);
1372 audit_inode_child(victim->d_name.name, victim, dir);
1374 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1379 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1380 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1383 if (!S_ISDIR(victim->d_inode->i_mode))
1385 if (IS_ROOT(victim))
1387 } else if (S_ISDIR(victim->d_inode->i_mode))
1389 if (IS_DEADDIR(dir))
1391 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1396 /* Check whether we can create an object with dentry child in directory
1398 * 1. We can't do it if child already exists (open has special treatment for
1399 * this case, but since we are inlined it's OK)
1400 * 2. We can't do it if dir is read-only (done in permission())
1401 * 3. We should have write and exec permissions on dir
1402 * 4. We can't do it if dir is immutable (done in permission())
1404 static inline int may_create(struct inode *dir, struct dentry *child)
1408 if (IS_DEADDIR(dir))
1410 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1414 * O_DIRECTORY translates into forcing a directory lookup.
1416 static inline int lookup_flags(unsigned int f)
1418 unsigned long retval = LOOKUP_FOLLOW;
1421 retval &= ~LOOKUP_FOLLOW;
1423 if (f & O_DIRECTORY)
1424 retval |= LOOKUP_DIRECTORY;
1430 * p1 and p2 should be directories on the same fs.
1432 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1437 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1441 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1443 p = d_ancestor(p2, p1);
1445 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1446 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1450 p = d_ancestor(p1, p2);
1452 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1453 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1457 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1458 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1462 void unlock_rename(struct dentry *p1, struct dentry *p2)
1464 mutex_unlock(&p1->d_inode->i_mutex);
1466 mutex_unlock(&p2->d_inode->i_mutex);
1467 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1471 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1472 struct nameidata *nd)
1474 int error = may_create(dir, dentry);
1479 if (!dir->i_op->create)
1480 return -EACCES; /* shouldn't it be ENOSYS? */
1483 error = security_inode_create(dir, dentry, mode);
1487 error = dir->i_op->create(dir, dentry, mode, nd);
1489 fsnotify_create(dir, dentry);
1493 int may_open(struct path *path, int acc_mode, int flag)
1495 struct dentry *dentry = path->dentry;
1496 struct inode *inode = dentry->d_inode;
1502 switch (inode->i_mode & S_IFMT) {
1506 if (acc_mode & MAY_WRITE)
1511 if (path->mnt->mnt_flags & MNT_NODEV)
1520 error = inode_permission(inode, acc_mode);
1524 error = ima_path_check(path,
1525 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC),
1530 * An append-only file must be opened in append mode for writing.
1532 if (IS_APPEND(inode)) {
1534 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1540 /* O_NOATIME can only be set by the owner or superuser */
1541 if (flag & O_NOATIME)
1542 if (!is_owner_or_cap(inode)) {
1548 * Ensure there are no outstanding leases on the file.
1550 error = break_lease(inode, flag);
1554 if (flag & O_TRUNC) {
1555 error = get_write_access(inode);
1560 * Refuse to truncate files with mandatory locks held on them.
1562 error = locks_verify_locked(inode);
1564 error = security_path_truncate(path, 0,
1565 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1569 error = do_truncate(dentry, 0,
1570 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1573 put_write_access(inode);
1577 if (flag & FMODE_WRITE)
1582 ima_counts_put(path, acc_mode ?
1583 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1584 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1589 * Be careful about ever adding any more callers of this
1590 * function. Its flags must be in the namei format, not
1591 * what get passed to sys_open().
1593 static int __open_namei_create(struct nameidata *nd, struct path *path,
1597 struct dentry *dir = nd->path.dentry;
1599 if (!IS_POSIXACL(dir->d_inode))
1600 mode &= ~current_umask();
1601 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1604 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1606 mutex_unlock(&dir->d_inode->i_mutex);
1607 dput(nd->path.dentry);
1608 nd->path.dentry = path->dentry;
1611 /* Don't check for write permission, don't truncate */
1612 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1616 * Note that while the flag value (low two bits) for sys_open means:
1621 * it is changed into
1622 * 00 - no permissions needed
1623 * 01 - read-permission
1624 * 10 - write-permission
1626 * for the internal routines (ie open_namei()/follow_link() etc)
1627 * This is more logical, and also allows the 00 "no perm needed"
1628 * to be used for symlinks (where the permissions are checked
1632 static inline int open_to_namei_flags(int flag)
1634 if ((flag+1) & O_ACCMODE)
1639 static int open_will_write_to_fs(int flag, struct inode *inode)
1642 * We'll never write to the fs underlying
1645 if (special_file(inode->i_mode))
1647 return (flag & O_TRUNC);
1651 * Note that the low bits of the passed in "open_flag"
1652 * are not the same as in the local variable "flag". See
1653 * open_to_namei_flags() for more details.
1655 struct file *do_filp_open(int dfd, const char *pathname,
1656 int open_flag, int mode, int acc_mode)
1659 struct nameidata nd;
1665 int flag = open_to_namei_flags(open_flag);
1668 acc_mode = MAY_OPEN | ACC_MODE(flag);
1670 /* O_TRUNC implies we need access checks for write permissions */
1672 acc_mode |= MAY_WRITE;
1674 /* Allow the LSM permission hook to distinguish append
1675 access from general write access. */
1676 if (flag & O_APPEND)
1677 acc_mode |= MAY_APPEND;
1680 * The simplest case - just a plain lookup.
1682 if (!(flag & O_CREAT)) {
1683 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1686 return ERR_PTR(error);
1691 * Create - we need to know the parent.
1693 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1695 return ERR_PTR(error);
1696 error = path_walk(pathname, &nd);
1700 return ERR_PTR(error);
1702 if (unlikely(!audit_dummy_context()))
1703 audit_inode(pathname, nd.path.dentry);
1706 * We have the parent and last component. First of all, check
1707 * that we are not asked to creat(2) an obvious directory - that
1711 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1715 filp = get_empty_filp();
1718 nd.intent.open.file = filp;
1719 nd.intent.open.flags = flag;
1720 nd.intent.open.create_mode = mode;
1721 dir = nd.path.dentry;
1722 nd.flags &= ~LOOKUP_PARENT;
1723 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1725 nd.flags |= LOOKUP_EXCL;
1726 mutex_lock(&dir->d_inode->i_mutex);
1727 path.dentry = lookup_hash(&nd);
1728 path.mnt = nd.path.mnt;
1731 error = PTR_ERR(path.dentry);
1732 if (IS_ERR(path.dentry)) {
1733 mutex_unlock(&dir->d_inode->i_mutex);
1737 if (IS_ERR(nd.intent.open.file)) {
1738 error = PTR_ERR(nd.intent.open.file);
1739 goto exit_mutex_unlock;
1742 /* Negative dentry, just create the file */
1743 if (!path.dentry->d_inode) {
1745 * This write is needed to ensure that a
1746 * ro->rw transition does not occur between
1747 * the time when the file is created and when
1748 * a permanent write count is taken through
1749 * the 'struct file' in nameidata_to_filp().
1751 error = mnt_want_write(nd.path.mnt);
1753 goto exit_mutex_unlock;
1754 error = __open_namei_create(&nd, &path, flag, mode);
1756 mnt_drop_write(nd.path.mnt);
1759 filp = nameidata_to_filp(&nd, open_flag);
1761 ima_counts_put(&nd.path,
1762 acc_mode & (MAY_READ | MAY_WRITE |
1764 mnt_drop_write(nd.path.mnt);
1771 * It already exists.
1773 mutex_unlock(&dir->d_inode->i_mutex);
1774 audit_inode(pathname, path.dentry);
1780 if (__follow_mount(&path)) {
1782 if (flag & O_NOFOLLOW)
1787 if (!path.dentry->d_inode)
1789 if (path.dentry->d_inode->i_op->follow_link)
1792 path_to_nameidata(&path, &nd);
1794 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1799 * 1. may_open() truncates a file
1800 * 2. a rw->ro mount transition occurs
1801 * 3. nameidata_to_filp() fails due to
1803 * That would be inconsistent, and should
1804 * be avoided. Taking this mnt write here
1805 * ensures that (2) can not occur.
1807 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1809 error = mnt_want_write(nd.path.mnt);
1813 error = may_open(&nd.path, acc_mode, flag);
1816 mnt_drop_write(nd.path.mnt);
1819 filp = nameidata_to_filp(&nd, open_flag);
1821 ima_counts_put(&nd.path,
1822 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1824 * It is now safe to drop the mnt write
1825 * because the filp has had a write taken
1829 mnt_drop_write(nd.path.mnt);
1835 mutex_unlock(&dir->d_inode->i_mutex);
1837 path_put_conditional(&path, &nd);
1839 if (!IS_ERR(nd.intent.open.file))
1840 release_open_intent(&nd);
1845 return ERR_PTR(error);
1849 if (flag & O_NOFOLLOW)
1852 * This is subtle. Instead of calling do_follow_link() we do the
1853 * thing by hands. The reason is that this way we have zero link_count
1854 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1855 * After that we have the parent and last component, i.e.
1856 * we are in the same situation as after the first path_walk().
1857 * Well, almost - if the last component is normal we get its copy
1858 * stored in nd->last.name and we will have to putname() it when we
1859 * are done. Procfs-like symlinks just set LAST_BIND.
1861 nd.flags |= LOOKUP_PARENT;
1862 error = security_inode_follow_link(path.dentry, &nd);
1865 error = __do_follow_link(&path, &nd);
1867 /* Does someone understand code flow here? Or it is only
1868 * me so stupid? Anathema to whoever designed this non-sense
1869 * with "intent.open".
1871 release_open_intent(&nd);
1874 return ERR_PTR(error);
1876 nd.flags &= ~LOOKUP_PARENT;
1877 if (nd.last_type == LAST_BIND)
1880 if (nd.last_type != LAST_NORM)
1882 if (nd.last.name[nd.last.len]) {
1883 __putname(nd.last.name);
1888 __putname(nd.last.name);
1891 dir = nd.path.dentry;
1892 mutex_lock(&dir->d_inode->i_mutex);
1893 path.dentry = lookup_hash(&nd);
1894 path.mnt = nd.path.mnt;
1895 __putname(nd.last.name);
1900 * filp_open - open file and return file pointer
1902 * @filename: path to open
1903 * @flags: open flags as per the open(2) second argument
1904 * @mode: mode for the new file if O_CREAT is set, else ignored
1906 * This is the helper to open a file from kernelspace if you really
1907 * have to. But in generally you should not do this, so please move
1908 * along, nothing to see here..
1910 struct file *filp_open(const char *filename, int flags, int mode)
1912 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1914 EXPORT_SYMBOL(filp_open);
1917 * lookup_create - lookup a dentry, creating it if it doesn't exist
1918 * @nd: nameidata info
1919 * @is_dir: directory flag
1921 * Simple function to lookup and return a dentry and create it
1922 * if it doesn't exist. Is SMP-safe.
1924 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1926 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1928 struct dentry *dentry = ERR_PTR(-EEXIST);
1930 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1932 * Yucky last component or no last component at all?
1933 * (foo/., foo/.., /////)
1935 if (nd->last_type != LAST_NORM)
1937 nd->flags &= ~LOOKUP_PARENT;
1938 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1939 nd->intent.open.flags = O_EXCL;
1942 * Do the final lookup.
1944 dentry = lookup_hash(nd);
1948 if (dentry->d_inode)
1951 * Special case - lookup gave negative, but... we had foo/bar/
1952 * From the vfs_mknod() POV we just have a negative dentry -
1953 * all is fine. Let's be bastards - you had / on the end, you've
1954 * been asking for (non-existent) directory. -ENOENT for you.
1956 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1958 dentry = ERR_PTR(-ENOENT);
1963 dentry = ERR_PTR(-EEXIST);
1967 EXPORT_SYMBOL_GPL(lookup_create);
1969 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1971 int error = may_create(dir, dentry);
1976 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1979 if (!dir->i_op->mknod)
1982 error = devcgroup_inode_mknod(mode, dev);
1986 error = security_inode_mknod(dir, dentry, mode, dev);
1991 error = dir->i_op->mknod(dir, dentry, mode, dev);
1993 fsnotify_create(dir, dentry);
1997 static int may_mknod(mode_t mode)
1999 switch (mode & S_IFMT) {
2005 case 0: /* zero mode translates to S_IFREG */
2014 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2019 struct dentry *dentry;
2020 struct nameidata nd;
2025 error = user_path_parent(dfd, filename, &nd, &tmp);
2029 dentry = lookup_create(&nd, 0);
2030 if (IS_ERR(dentry)) {
2031 error = PTR_ERR(dentry);
2034 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2035 mode &= ~current_umask();
2036 error = may_mknod(mode);
2039 error = mnt_want_write(nd.path.mnt);
2042 error = security_path_mknod(&nd.path, dentry, mode, dev);
2044 goto out_drop_write;
2045 switch (mode & S_IFMT) {
2046 case 0: case S_IFREG:
2047 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2049 case S_IFCHR: case S_IFBLK:
2050 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2051 new_decode_dev(dev));
2053 case S_IFIFO: case S_IFSOCK:
2054 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2058 mnt_drop_write(nd.path.mnt);
2062 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2069 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2071 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2074 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2076 int error = may_create(dir, dentry);
2081 if (!dir->i_op->mkdir)
2084 mode &= (S_IRWXUGO|S_ISVTX);
2085 error = security_inode_mkdir(dir, dentry, mode);
2090 error = dir->i_op->mkdir(dir, dentry, mode);
2092 fsnotify_mkdir(dir, dentry);
2096 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2100 struct dentry *dentry;
2101 struct nameidata nd;
2103 error = user_path_parent(dfd, pathname, &nd, &tmp);
2107 dentry = lookup_create(&nd, 1);
2108 error = PTR_ERR(dentry);
2112 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2113 mode &= ~current_umask();
2114 error = mnt_want_write(nd.path.mnt);
2117 error = security_path_mkdir(&nd.path, dentry, mode);
2119 goto out_drop_write;
2120 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2122 mnt_drop_write(nd.path.mnt);
2126 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2133 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2135 return sys_mkdirat(AT_FDCWD, pathname, mode);
2139 * We try to drop the dentry early: we should have
2140 * a usage count of 2 if we're the only user of this
2141 * dentry, and if that is true (possibly after pruning
2142 * the dcache), then we drop the dentry now.
2144 * A low-level filesystem can, if it choses, legally
2147 * if (!d_unhashed(dentry))
2150 * if it cannot handle the case of removing a directory
2151 * that is still in use by something else..
2153 void dentry_unhash(struct dentry *dentry)
2156 shrink_dcache_parent(dentry);
2157 spin_lock(&dcache_lock);
2158 spin_lock(&dentry->d_lock);
2159 if (atomic_read(&dentry->d_count) == 2)
2161 spin_unlock(&dentry->d_lock);
2162 spin_unlock(&dcache_lock);
2165 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2167 int error = may_delete(dir, dentry, 1);
2172 if (!dir->i_op->rmdir)
2177 mutex_lock(&dentry->d_inode->i_mutex);
2178 dentry_unhash(dentry);
2179 if (d_mountpoint(dentry))
2182 error = security_inode_rmdir(dir, dentry);
2184 error = dir->i_op->rmdir(dir, dentry);
2186 dentry->d_inode->i_flags |= S_DEAD;
2189 mutex_unlock(&dentry->d_inode->i_mutex);
2198 static long do_rmdir(int dfd, const char __user *pathname)
2202 struct dentry *dentry;
2203 struct nameidata nd;
2205 error = user_path_parent(dfd, pathname, &nd, &name);
2209 switch(nd.last_type) {
2221 nd.flags &= ~LOOKUP_PARENT;
2223 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2224 dentry = lookup_hash(&nd);
2225 error = PTR_ERR(dentry);
2228 error = mnt_want_write(nd.path.mnt);
2231 error = security_path_rmdir(&nd.path, dentry);
2234 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2236 mnt_drop_write(nd.path.mnt);
2240 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2247 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2249 return do_rmdir(AT_FDCWD, pathname);
2252 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2254 int error = may_delete(dir, dentry, 0);
2259 if (!dir->i_op->unlink)
2264 mutex_lock(&dentry->d_inode->i_mutex);
2265 if (d_mountpoint(dentry))
2268 error = security_inode_unlink(dir, dentry);
2270 error = dir->i_op->unlink(dir, dentry);
2272 mutex_unlock(&dentry->d_inode->i_mutex);
2274 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2275 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2276 fsnotify_link_count(dentry->d_inode);
2284 * Make sure that the actual truncation of the file will occur outside its
2285 * directory's i_mutex. Truncate can take a long time if there is a lot of
2286 * writeout happening, and we don't want to prevent access to the directory
2287 * while waiting on the I/O.
2289 static long do_unlinkat(int dfd, const char __user *pathname)
2293 struct dentry *dentry;
2294 struct nameidata nd;
2295 struct inode *inode = NULL;
2297 error = user_path_parent(dfd, pathname, &nd, &name);
2302 if (nd.last_type != LAST_NORM)
2305 nd.flags &= ~LOOKUP_PARENT;
2307 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2308 dentry = lookup_hash(&nd);
2309 error = PTR_ERR(dentry);
2310 if (!IS_ERR(dentry)) {
2311 /* Why not before? Because we want correct error value */
2312 if (nd.last.name[nd.last.len])
2314 inode = dentry->d_inode;
2316 atomic_inc(&inode->i_count);
2317 error = mnt_want_write(nd.path.mnt);
2320 error = security_path_unlink(&nd.path, dentry);
2323 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2325 mnt_drop_write(nd.path.mnt);
2329 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2331 iput(inode); /* truncate the inode here */
2338 error = !dentry->d_inode ? -ENOENT :
2339 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2343 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2345 if ((flag & ~AT_REMOVEDIR) != 0)
2348 if (flag & AT_REMOVEDIR)
2349 return do_rmdir(dfd, pathname);
2351 return do_unlinkat(dfd, pathname);
2354 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2356 return do_unlinkat(AT_FDCWD, pathname);
2359 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2361 int error = may_create(dir, dentry);
2366 if (!dir->i_op->symlink)
2369 error = security_inode_symlink(dir, dentry, oldname);
2374 error = dir->i_op->symlink(dir, dentry, oldname);
2376 fsnotify_create(dir, dentry);
2380 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2381 int, newdfd, const char __user *, newname)
2386 struct dentry *dentry;
2387 struct nameidata nd;
2389 from = getname(oldname);
2391 return PTR_ERR(from);
2393 error = user_path_parent(newdfd, newname, &nd, &to);
2397 dentry = lookup_create(&nd, 0);
2398 error = PTR_ERR(dentry);
2402 error = mnt_want_write(nd.path.mnt);
2405 error = security_path_symlink(&nd.path, dentry, from);
2407 goto out_drop_write;
2408 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2410 mnt_drop_write(nd.path.mnt);
2414 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2422 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2424 return sys_symlinkat(oldname, AT_FDCWD, newname);
2427 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2429 struct inode *inode = old_dentry->d_inode;
2435 error = may_create(dir, new_dentry);
2439 if (dir->i_sb != inode->i_sb)
2443 * A link to an append-only or immutable file cannot be created.
2445 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2447 if (!dir->i_op->link)
2449 if (S_ISDIR(inode->i_mode))
2452 error = security_inode_link(old_dentry, dir, new_dentry);
2456 mutex_lock(&inode->i_mutex);
2458 error = dir->i_op->link(old_dentry, dir, new_dentry);
2459 mutex_unlock(&inode->i_mutex);
2461 fsnotify_link(dir, inode, new_dentry);
2466 * Hardlinks are often used in delicate situations. We avoid
2467 * security-related surprises by not following symlinks on the
2470 * We don't follow them on the oldname either to be compatible
2471 * with linux 2.0, and to avoid hard-linking to directories
2472 * and other special files. --ADM
2474 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2475 int, newdfd, const char __user *, newname, int, flags)
2477 struct dentry *new_dentry;
2478 struct nameidata nd;
2479 struct path old_path;
2483 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2486 error = user_path_at(olddfd, oldname,
2487 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2492 error = user_path_parent(newdfd, newname, &nd, &to);
2496 if (old_path.mnt != nd.path.mnt)
2498 new_dentry = lookup_create(&nd, 0);
2499 error = PTR_ERR(new_dentry);
2500 if (IS_ERR(new_dentry))
2502 error = mnt_want_write(nd.path.mnt);
2505 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2507 goto out_drop_write;
2508 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2510 mnt_drop_write(nd.path.mnt);
2514 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2519 path_put(&old_path);
2524 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2526 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2530 * The worst of all namespace operations - renaming directory. "Perverted"
2531 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2533 * a) we can get into loop creation. Check is done in is_subdir().
2534 * b) race potential - two innocent renames can create a loop together.
2535 * That's where 4.4 screws up. Current fix: serialization on
2536 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2538 * c) we have to lock _three_ objects - parents and victim (if it exists).
2539 * And that - after we got ->i_mutex on parents (until then we don't know
2540 * whether the target exists). Solution: try to be smart with locking
2541 * order for inodes. We rely on the fact that tree topology may change
2542 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2543 * move will be locked. Thus we can rank directories by the tree
2544 * (ancestors first) and rank all non-directories after them.
2545 * That works since everybody except rename does "lock parent, lookup,
2546 * lock child" and rename is under ->s_vfs_rename_mutex.
2547 * HOWEVER, it relies on the assumption that any object with ->lookup()
2548 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2549 * we'd better make sure that there's no link(2) for them.
2550 * d) some filesystems don't support opened-but-unlinked directories,
2551 * either because of layout or because they are not ready to deal with
2552 * all cases correctly. The latter will be fixed (taking this sort of
2553 * stuff into VFS), but the former is not going away. Solution: the same
2554 * trick as in rmdir().
2555 * e) conversion from fhandle to dentry may come in the wrong moment - when
2556 * we are removing the target. Solution: we will have to grab ->i_mutex
2557 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2558 * ->i_mutex on parents, which works but leads to some truely excessive
2561 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2562 struct inode *new_dir, struct dentry *new_dentry)
2565 struct inode *target;
2568 * If we are going to change the parent - check write permissions,
2569 * we'll need to flip '..'.
2571 if (new_dir != old_dir) {
2572 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2577 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2581 target = new_dentry->d_inode;
2583 mutex_lock(&target->i_mutex);
2584 dentry_unhash(new_dentry);
2586 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2589 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2592 target->i_flags |= S_DEAD;
2593 mutex_unlock(&target->i_mutex);
2594 if (d_unhashed(new_dentry))
2595 d_rehash(new_dentry);
2599 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2600 d_move(old_dentry,new_dentry);
2604 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2605 struct inode *new_dir, struct dentry *new_dentry)
2607 struct inode *target;
2610 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2615 target = new_dentry->d_inode;
2617 mutex_lock(&target->i_mutex);
2618 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2621 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2623 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2624 d_move(old_dentry, new_dentry);
2627 mutex_unlock(&target->i_mutex);
2632 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2633 struct inode *new_dir, struct dentry *new_dentry)
2636 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2637 const char *old_name;
2639 if (old_dentry->d_inode == new_dentry->d_inode)
2642 error = may_delete(old_dir, old_dentry, is_dir);
2646 if (!new_dentry->d_inode)
2647 error = may_create(new_dir, new_dentry);
2649 error = may_delete(new_dir, new_dentry, is_dir);
2653 if (!old_dir->i_op->rename)
2656 vfs_dq_init(old_dir);
2657 vfs_dq_init(new_dir);
2659 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2662 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2664 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2666 const char *new_name = old_dentry->d_name.name;
2667 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2668 new_dentry->d_inode, old_dentry);
2670 fsnotify_oldname_free(old_name);
2675 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2676 int, newdfd, const char __user *, newname)
2678 struct dentry *old_dir, *new_dir;
2679 struct dentry *old_dentry, *new_dentry;
2680 struct dentry *trap;
2681 struct nameidata oldnd, newnd;
2686 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2690 error = user_path_parent(newdfd, newname, &newnd, &to);
2695 if (oldnd.path.mnt != newnd.path.mnt)
2698 old_dir = oldnd.path.dentry;
2700 if (oldnd.last_type != LAST_NORM)
2703 new_dir = newnd.path.dentry;
2704 if (newnd.last_type != LAST_NORM)
2707 oldnd.flags &= ~LOOKUP_PARENT;
2708 newnd.flags &= ~LOOKUP_PARENT;
2709 newnd.flags |= LOOKUP_RENAME_TARGET;
2711 trap = lock_rename(new_dir, old_dir);
2713 old_dentry = lookup_hash(&oldnd);
2714 error = PTR_ERR(old_dentry);
2715 if (IS_ERR(old_dentry))
2717 /* source must exist */
2719 if (!old_dentry->d_inode)
2721 /* unless the source is a directory trailing slashes give -ENOTDIR */
2722 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2724 if (oldnd.last.name[oldnd.last.len])
2726 if (newnd.last.name[newnd.last.len])
2729 /* source should not be ancestor of target */
2731 if (old_dentry == trap)
2733 new_dentry = lookup_hash(&newnd);
2734 error = PTR_ERR(new_dentry);
2735 if (IS_ERR(new_dentry))
2737 /* target should not be an ancestor of source */
2739 if (new_dentry == trap)
2742 error = mnt_want_write(oldnd.path.mnt);
2745 error = security_path_rename(&oldnd.path, old_dentry,
2746 &newnd.path, new_dentry);
2749 error = vfs_rename(old_dir->d_inode, old_dentry,
2750 new_dir->d_inode, new_dentry);
2752 mnt_drop_write(oldnd.path.mnt);
2758 unlock_rename(new_dir, old_dir);
2760 path_put(&newnd.path);
2763 path_put(&oldnd.path);
2769 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2771 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2774 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2778 len = PTR_ERR(link);
2783 if (len > (unsigned) buflen)
2785 if (copy_to_user(buffer, link, len))
2792 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2793 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2794 * using) it for any given inode is up to filesystem.
2796 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2798 struct nameidata nd;
2803 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2805 return PTR_ERR(cookie);
2807 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2808 if (dentry->d_inode->i_op->put_link)
2809 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2813 int vfs_follow_link(struct nameidata *nd, const char *link)
2815 return __vfs_follow_link(nd, link);
2818 /* get the link contents into pagecache */
2819 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2823 struct address_space *mapping = dentry->d_inode->i_mapping;
2824 page = read_mapping_page(mapping, 0, NULL);
2829 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2833 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2835 struct page *page = NULL;
2836 char *s = page_getlink(dentry, &page);
2837 int res = vfs_readlink(dentry,buffer,buflen,s);
2840 page_cache_release(page);
2845 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2847 struct page *page = NULL;
2848 nd_set_link(nd, page_getlink(dentry, &page));
2852 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2854 struct page *page = cookie;
2858 page_cache_release(page);
2863 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2865 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2867 struct address_space *mapping = inode->i_mapping;
2872 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2874 flags |= AOP_FLAG_NOFS;
2877 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2878 flags, &page, &fsdata);
2882 kaddr = kmap_atomic(page, KM_USER0);
2883 memcpy(kaddr, symname, len-1);
2884 kunmap_atomic(kaddr, KM_USER0);
2886 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2893 mark_inode_dirty(inode);
2899 int page_symlink(struct inode *inode, const char *symname, int len)
2901 return __page_symlink(inode, symname, len,
2902 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2905 const struct inode_operations page_symlink_inode_operations = {
2906 .readlink = generic_readlink,
2907 .follow_link = page_follow_link_light,
2908 .put_link = page_put_link,
2911 EXPORT_SYMBOL(user_path_at);
2912 EXPORT_SYMBOL(follow_down);
2913 EXPORT_SYMBOL(follow_up);
2914 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2915 EXPORT_SYMBOL(getname);
2916 EXPORT_SYMBOL(lock_rename);
2917 EXPORT_SYMBOL(lookup_one_len);
2918 EXPORT_SYMBOL(page_follow_link_light);
2919 EXPORT_SYMBOL(page_put_link);
2920 EXPORT_SYMBOL(page_readlink);
2921 EXPORT_SYMBOL(__page_symlink);
2922 EXPORT_SYMBOL(page_symlink);
2923 EXPORT_SYMBOL(page_symlink_inode_operations);
2924 EXPORT_SYMBOL(path_lookup);
2925 EXPORT_SYMBOL(kern_path);
2926 EXPORT_SYMBOL(vfs_path_lookup);
2927 EXPORT_SYMBOL(inode_permission);
2928 EXPORT_SYMBOL(file_permission);
2929 EXPORT_SYMBOL(unlock_rename);
2930 EXPORT_SYMBOL(vfs_create);
2931 EXPORT_SYMBOL(vfs_follow_link);
2932 EXPORT_SYMBOL(vfs_link);
2933 EXPORT_SYMBOL(vfs_mkdir);
2934 EXPORT_SYMBOL(vfs_mknod);
2935 EXPORT_SYMBOL(generic_permission);
2936 EXPORT_SYMBOL(vfs_readlink);
2937 EXPORT_SYMBOL(vfs_rename);
2938 EXPORT_SYMBOL(vfs_rmdir);
2939 EXPORT_SYMBOL(vfs_symlink);
2940 EXPORT_SYMBOL(vfs_unlink);
2941 EXPORT_SYMBOL(dentry_unhash);
2942 EXPORT_SYMBOL(generic_readlink);