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 ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
451 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
454 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
459 return security_inode_permission(inode, MAY_EXEC);
463 * This is called when everything else fails, and we actually have
464 * to go to the low-level filesystem to find out what we should do..
466 * We get the directory semaphore, and after getting that we also
467 * make sure that nobody added the entry to the dcache in the meantime..
470 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
472 struct dentry * result;
473 struct inode *dir = parent->d_inode;
475 mutex_lock(&dir->i_mutex);
477 * First re-do the cached lookup just in case it was created
478 * while we waited for the directory semaphore..
480 * FIXME! This could use version numbering or similar to
481 * avoid unnecessary cache lookups.
483 * The "dcache_lock" is purely to protect the RCU list walker
484 * from concurrent renames at this point (we mustn't get false
485 * negatives from the RCU list walk here, unlike the optimistic
488 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
490 result = d_lookup(parent, name);
492 struct dentry *dentry;
494 /* Don't create child dentry for a dead directory. */
495 result = ERR_PTR(-ENOENT);
499 dentry = d_alloc(parent, name);
500 result = ERR_PTR(-ENOMEM);
502 result = dir->i_op->lookup(dir, dentry, nd);
509 mutex_unlock(&dir->i_mutex);
514 * Uhhuh! Nasty case: the cache was re-populated while
515 * we waited on the semaphore. Need to revalidate.
517 mutex_unlock(&dir->i_mutex);
518 if (result->d_op && result->d_op->d_revalidate) {
519 result = do_revalidate(result, nd);
521 result = ERR_PTR(-ENOENT);
527 * Wrapper to retry pathname resolution whenever the underlying
528 * file system returns an ESTALE.
530 * Retry the whole path once, forcing real lookup requests
531 * instead of relying on the dcache.
533 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
535 struct path save = nd->path;
538 /* make sure the stuff we saved doesn't go away */
541 result = __link_path_walk(name, nd);
542 if (result == -ESTALE) {
543 /* nd->path had been dropped */
546 nd->flags |= LOOKUP_REVAL;
547 result = __link_path_walk(name, nd);
555 static __always_inline void set_root(struct nameidata *nd)
558 struct fs_struct *fs = current->fs;
559 read_lock(&fs->lock);
562 read_unlock(&fs->lock);
566 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
580 res = link_path_walk(link, nd);
581 if (nd->depth || res || nd->last_type!=LAST_NORM)
584 * If it is an iterative symlinks resolution in open_namei() we
585 * have to copy the last component. And all that crap because of
586 * bloody create() on broken symlinks. Furrfu...
589 if (unlikely(!name)) {
593 strcpy(name, nd->last.name);
594 nd->last.name = name;
598 return PTR_ERR(link);
601 static void path_put_conditional(struct path *path, struct nameidata *nd)
604 if (path->mnt != nd->path.mnt)
608 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
613 nd->path.mnt = path->mnt;
614 nd->path.dentry = path->dentry;
617 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
621 struct dentry *dentry = path->dentry;
623 touch_atime(path->mnt, dentry);
624 nd_set_link(nd, NULL);
626 if (path->mnt != nd->path.mnt) {
627 path_to_nameidata(path, nd);
631 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
632 error = PTR_ERR(cookie);
633 if (!IS_ERR(cookie)) {
634 char *s = nd_get_link(nd);
637 error = __vfs_follow_link(nd, s);
638 if (dentry->d_inode->i_op->put_link)
639 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
647 * This limits recursive symlink follows to 8, while
648 * limiting consecutive symlinks to 40.
650 * Without that kind of total limit, nasty chains of consecutive
651 * symlinks can cause almost arbitrarily long lookups.
653 static inline int do_follow_link(struct path *path, struct nameidata *nd)
656 if (current->link_count >= MAX_NESTED_LINKS)
658 if (current->total_link_count >= 40)
660 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
662 err = security_inode_follow_link(path->dentry, nd);
665 current->link_count++;
666 current->total_link_count++;
668 err = __do_follow_link(path, nd);
669 current->link_count--;
673 path_put_conditional(path, nd);
678 int follow_up(struct path *path)
680 struct vfsmount *parent;
681 struct dentry *mountpoint;
682 spin_lock(&vfsmount_lock);
683 parent = path->mnt->mnt_parent;
684 if (parent == path->mnt) {
685 spin_unlock(&vfsmount_lock);
689 mountpoint = dget(path->mnt->mnt_mountpoint);
690 spin_unlock(&vfsmount_lock);
692 path->dentry = mountpoint;
698 /* no need for dcache_lock, as serialization is taken care in
701 static int __follow_mount(struct path *path)
704 while (d_mountpoint(path->dentry)) {
705 struct vfsmount *mounted = lookup_mnt(path);
712 path->dentry = dget(mounted->mnt_root);
718 static void follow_mount(struct path *path)
720 while (d_mountpoint(path->dentry)) {
721 struct vfsmount *mounted = lookup_mnt(path);
727 path->dentry = dget(mounted->mnt_root);
731 /* no need for dcache_lock, as serialization is taken care in
734 int follow_down(struct path *path)
736 struct vfsmount *mounted;
738 mounted = lookup_mnt(path);
743 path->dentry = dget(mounted->mnt_root);
749 static __always_inline void follow_dotdot(struct nameidata *nd)
754 struct vfsmount *parent;
755 struct dentry *old = nd->path.dentry;
757 if (nd->path.dentry == nd->root.dentry &&
758 nd->path.mnt == nd->root.mnt) {
761 spin_lock(&dcache_lock);
762 if (nd->path.dentry != nd->path.mnt->mnt_root) {
763 nd->path.dentry = dget(nd->path.dentry->d_parent);
764 spin_unlock(&dcache_lock);
768 spin_unlock(&dcache_lock);
769 spin_lock(&vfsmount_lock);
770 parent = nd->path.mnt->mnt_parent;
771 if (parent == nd->path.mnt) {
772 spin_unlock(&vfsmount_lock);
776 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
777 spin_unlock(&vfsmount_lock);
779 mntput(nd->path.mnt);
780 nd->path.mnt = parent;
782 follow_mount(&nd->path);
786 * It's more convoluted than I'd like it to be, but... it's still fairly
787 * small and for now I'd prefer to have fast path as straight as possible.
788 * It _is_ time-critical.
790 static int do_lookup(struct nameidata *nd, struct qstr *name,
793 struct vfsmount *mnt = nd->path.mnt;
794 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
798 if (dentry->d_op && dentry->d_op->d_revalidate)
799 goto need_revalidate;
802 path->dentry = dentry;
803 __follow_mount(path);
807 dentry = real_lookup(nd->path.dentry, name, nd);
813 dentry = do_revalidate(dentry, nd);
821 return PTR_ERR(dentry);
826 * This is the basic name resolution function, turning a pathname into
827 * the final dentry. We expect 'base' to be positive and a directory.
829 * Returns 0 and nd will have valid dentry and mnt on success.
830 * Returns error and drops reference to input namei data on failure.
832 static int __link_path_walk(const char *name, struct nameidata *nd)
837 unsigned int lookup_flags = nd->flags;
844 inode = nd->path.dentry->d_inode;
846 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
848 /* At this point we know we have a real path component. */
854 nd->flags |= LOOKUP_CONTINUE;
855 err = exec_permission_lite(inode);
860 c = *(const unsigned char *)name;
862 hash = init_name_hash();
865 hash = partial_name_hash(c, hash);
866 c = *(const unsigned char *)name;
867 } while (c && (c != '/'));
868 this.len = name - (const char *) this.name;
869 this.hash = end_name_hash(hash);
871 /* remove trailing slashes? */
874 while (*++name == '/');
876 goto last_with_slashes;
879 * "." and ".." are special - ".." especially so because it has
880 * to be able to know about the current root directory and
881 * parent relationships.
883 if (this.name[0] == '.') switch (this.len) {
887 if (this.name[1] != '.')
890 inode = nd->path.dentry->d_inode;
896 * See if the low-level filesystem might want
897 * to use its own hash..
899 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
900 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
905 /* This does the actual lookups.. */
906 err = do_lookup(nd, &this, &next);
911 inode = next.dentry->d_inode;
915 if (inode->i_op->follow_link) {
916 err = do_follow_link(&next, nd);
920 inode = nd->path.dentry->d_inode;
924 path_to_nameidata(&next, nd);
926 if (!inode->i_op->lookup)
929 /* here ends the main loop */
932 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
934 /* Clear LOOKUP_CONTINUE iff it was previously unset */
935 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
936 if (lookup_flags & LOOKUP_PARENT)
938 if (this.name[0] == '.') switch (this.len) {
942 if (this.name[1] != '.')
945 inode = nd->path.dentry->d_inode;
950 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
951 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
956 err = do_lookup(nd, &this, &next);
959 inode = next.dentry->d_inode;
960 if ((lookup_flags & LOOKUP_FOLLOW)
961 && inode && inode->i_op->follow_link) {
962 err = do_follow_link(&next, nd);
965 inode = nd->path.dentry->d_inode;
967 path_to_nameidata(&next, nd);
971 if (lookup_flags & LOOKUP_DIRECTORY) {
973 if (!inode->i_op->lookup)
979 nd->last_type = LAST_NORM;
980 if (this.name[0] != '.')
983 nd->last_type = LAST_DOT;
984 else if (this.len == 2 && this.name[1] == '.')
985 nd->last_type = LAST_DOTDOT;
990 * We bypassed the ordinary revalidation routines.
991 * We may need to check the cached dentry for staleness.
993 if (nd->path.dentry && nd->path.dentry->d_sb &&
994 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
996 /* Note: we do not d_invalidate() */
997 if (!nd->path.dentry->d_op->d_revalidate(
998 nd->path.dentry, nd))
1004 path_put_conditional(&next, nd);
1007 path_put(&nd->path);
1012 static int path_walk(const char *name, struct nameidata *nd)
1014 current->total_link_count = 0;
1015 return link_path_walk(name, nd);
1018 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1024 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1027 nd->root.mnt = NULL;
1031 nd->path = nd->root;
1032 path_get(&nd->root);
1033 } else if (dfd == AT_FDCWD) {
1034 struct fs_struct *fs = current->fs;
1035 read_lock(&fs->lock);
1038 read_unlock(&fs->lock);
1040 struct dentry *dentry;
1042 file = fget_light(dfd, &fput_needed);
1047 dentry = file->f_path.dentry;
1050 if (!S_ISDIR(dentry->d_inode->i_mode))
1053 retval = file_permission(file, MAY_EXEC);
1057 nd->path = file->f_path;
1058 path_get(&file->f_path);
1060 fput_light(file, fput_needed);
1065 fput_light(file, fput_needed);
1070 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1071 static int do_path_lookup(int dfd, const char *name,
1072 unsigned int flags, struct nameidata *nd)
1074 int retval = path_init(dfd, name, flags, nd);
1076 retval = path_walk(name, nd);
1077 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1078 nd->path.dentry->d_inode))
1079 audit_inode(name, nd->path.dentry);
1081 path_put(&nd->root);
1082 nd->root.mnt = NULL;
1087 int path_lookup(const char *name, unsigned int flags,
1088 struct nameidata *nd)
1090 return do_path_lookup(AT_FDCWD, name, flags, nd);
1093 int kern_path(const char *name, unsigned int flags, struct path *path)
1095 struct nameidata nd;
1096 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1103 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1104 * @dentry: pointer to dentry of the base directory
1105 * @mnt: pointer to vfs mount of the base directory
1106 * @name: pointer to file name
1107 * @flags: lookup flags
1108 * @nd: pointer to nameidata
1110 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1111 const char *name, unsigned int flags,
1112 struct nameidata *nd)
1116 /* same as do_path_lookup */
1117 nd->last_type = LAST_ROOT;
1121 nd->path.dentry = dentry;
1123 path_get(&nd->path);
1124 nd->root = nd->path;
1125 path_get(&nd->root);
1127 retval = path_walk(name, nd);
1128 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1129 nd->path.dentry->d_inode))
1130 audit_inode(name, nd->path.dentry);
1132 path_put(&nd->root);
1133 nd->root.mnt = NULL;
1139 * path_lookup_open - lookup a file path with open intent
1140 * @dfd: the directory to use as base, or AT_FDCWD
1141 * @name: pointer to file name
1142 * @lookup_flags: lookup intent flags
1143 * @nd: pointer to nameidata
1144 * @open_flags: open intent flags
1146 static int path_lookup_open(int dfd, const char *name,
1147 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1149 struct file *filp = get_empty_filp();
1154 nd->intent.open.file = filp;
1155 nd->intent.open.flags = open_flags;
1156 nd->intent.open.create_mode = 0;
1157 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1158 if (IS_ERR(nd->intent.open.file)) {
1160 err = PTR_ERR(nd->intent.open.file);
1161 path_put(&nd->path);
1163 } else if (err != 0)
1164 release_open_intent(nd);
1168 static struct dentry *__lookup_hash(struct qstr *name,
1169 struct dentry *base, struct nameidata *nd)
1171 struct dentry *dentry;
1172 struct inode *inode;
1175 inode = base->d_inode;
1178 * See if the low-level filesystem might want
1179 * to use its own hash..
1181 if (base->d_op && base->d_op->d_hash) {
1182 err = base->d_op->d_hash(base, name);
1183 dentry = ERR_PTR(err);
1188 dentry = cached_lookup(base, name, nd);
1192 /* Don't create child dentry for a dead directory. */
1193 dentry = ERR_PTR(-ENOENT);
1194 if (IS_DEADDIR(inode))
1197 new = d_alloc(base, name);
1198 dentry = ERR_PTR(-ENOMEM);
1201 dentry = inode->i_op->lookup(inode, new, nd);
1212 * Restricted form of lookup. Doesn't follow links, single-component only,
1213 * needs parent already locked. Doesn't follow mounts.
1216 static struct dentry *lookup_hash(struct nameidata *nd)
1220 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1222 return ERR_PTR(err);
1223 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1226 static int __lookup_one_len(const char *name, struct qstr *this,
1227 struct dentry *base, int len)
1237 hash = init_name_hash();
1239 c = *(const unsigned char *)name++;
1240 if (c == '/' || c == '\0')
1242 hash = partial_name_hash(c, hash);
1244 this->hash = end_name_hash(hash);
1249 * lookup_one_len - filesystem helper to lookup single pathname component
1250 * @name: pathname component to lookup
1251 * @base: base directory to lookup from
1252 * @len: maximum length @len should be interpreted to
1254 * Note that this routine is purely a helper for filesystem usage and should
1255 * not be called by generic code. Also note that by using this function the
1256 * nameidata argument is passed to the filesystem methods and a filesystem
1257 * using this helper needs to be prepared for that.
1259 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1264 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1266 err = __lookup_one_len(name, &this, base, len);
1268 return ERR_PTR(err);
1270 err = inode_permission(base->d_inode, MAY_EXEC);
1272 return ERR_PTR(err);
1273 return __lookup_hash(&this, base, NULL);
1277 * lookup_one_noperm - bad hack for sysfs
1278 * @name: pathname component to lookup
1279 * @base: base directory to lookup from
1281 * This is a variant of lookup_one_len that doesn't perform any permission
1282 * checks. It's a horrible hack to work around the braindead sysfs
1283 * architecture and should not be used anywhere else.
1285 * DON'T USE THIS FUNCTION EVER, thanks.
1287 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1292 err = __lookup_one_len(name, &this, base, strlen(name));
1294 return ERR_PTR(err);
1295 return __lookup_hash(&this, base, NULL);
1298 int user_path_at(int dfd, const char __user *name, unsigned flags,
1301 struct nameidata nd;
1302 char *tmp = getname(name);
1303 int err = PTR_ERR(tmp);
1306 BUG_ON(flags & LOOKUP_PARENT);
1308 err = do_path_lookup(dfd, tmp, flags, &nd);
1316 static int user_path_parent(int dfd, const char __user *path,
1317 struct nameidata *nd, char **name)
1319 char *s = getname(path);
1325 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1335 * It's inline, so penalty for filesystems that don't use sticky bit is
1338 static inline int check_sticky(struct inode *dir, struct inode *inode)
1340 uid_t fsuid = current_fsuid();
1342 if (!(dir->i_mode & S_ISVTX))
1344 if (inode->i_uid == fsuid)
1346 if (dir->i_uid == fsuid)
1348 return !capable(CAP_FOWNER);
1352 * Check whether we can remove a link victim from directory dir, check
1353 * whether the type of victim is right.
1354 * 1. We can't do it if dir is read-only (done in permission())
1355 * 2. We should have write and exec permissions on dir
1356 * 3. We can't remove anything from append-only dir
1357 * 4. We can't do anything with immutable dir (done in permission())
1358 * 5. If the sticky bit on dir is set we should either
1359 * a. be owner of dir, or
1360 * b. be owner of victim, or
1361 * c. have CAP_FOWNER capability
1362 * 6. If the victim is append-only or immutable we can't do antyhing with
1363 * links pointing to it.
1364 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1365 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1366 * 9. We can't remove a root or mountpoint.
1367 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1368 * nfs_async_unlink().
1370 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1374 if (!victim->d_inode)
1377 BUG_ON(victim->d_parent->d_inode != dir);
1378 audit_inode_child(victim->d_name.name, victim, dir);
1380 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1385 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1386 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1389 if (!S_ISDIR(victim->d_inode->i_mode))
1391 if (IS_ROOT(victim))
1393 } else if (S_ISDIR(victim->d_inode->i_mode))
1395 if (IS_DEADDIR(dir))
1397 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1402 /* Check whether we can create an object with dentry child in directory
1404 * 1. We can't do it if child already exists (open has special treatment for
1405 * this case, but since we are inlined it's OK)
1406 * 2. We can't do it if dir is read-only (done in permission())
1407 * 3. We should have write and exec permissions on dir
1408 * 4. We can't do it if dir is immutable (done in permission())
1410 static inline int may_create(struct inode *dir, struct dentry *child)
1414 if (IS_DEADDIR(dir))
1416 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1420 * O_DIRECTORY translates into forcing a directory lookup.
1422 static inline int lookup_flags(unsigned int f)
1424 unsigned long retval = LOOKUP_FOLLOW;
1427 retval &= ~LOOKUP_FOLLOW;
1429 if (f & O_DIRECTORY)
1430 retval |= LOOKUP_DIRECTORY;
1436 * p1 and p2 should be directories on the same fs.
1438 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1443 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1447 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1449 p = d_ancestor(p2, p1);
1451 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1452 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1456 p = d_ancestor(p1, p2);
1458 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1459 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1463 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1464 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1468 void unlock_rename(struct dentry *p1, struct dentry *p2)
1470 mutex_unlock(&p1->d_inode->i_mutex);
1472 mutex_unlock(&p2->d_inode->i_mutex);
1473 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1477 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1478 struct nameidata *nd)
1480 int error = may_create(dir, dentry);
1485 if (!dir->i_op->create)
1486 return -EACCES; /* shouldn't it be ENOSYS? */
1489 error = security_inode_create(dir, dentry, mode);
1493 error = dir->i_op->create(dir, dentry, mode, nd);
1495 fsnotify_create(dir, dentry);
1499 int may_open(struct path *path, int acc_mode, int flag)
1501 struct dentry *dentry = path->dentry;
1502 struct inode *inode = dentry->d_inode;
1508 switch (inode->i_mode & S_IFMT) {
1512 if (acc_mode & MAY_WRITE)
1517 if (path->mnt->mnt_flags & MNT_NODEV)
1526 error = inode_permission(inode, acc_mode);
1530 error = ima_path_check(path,
1531 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC),
1536 * An append-only file must be opened in append mode for writing.
1538 if (IS_APPEND(inode)) {
1540 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1546 /* O_NOATIME can only be set by the owner or superuser */
1547 if (flag & O_NOATIME)
1548 if (!is_owner_or_cap(inode)) {
1554 * Ensure there are no outstanding leases on the file.
1556 error = break_lease(inode, flag);
1560 if (flag & O_TRUNC) {
1561 error = get_write_access(inode);
1566 * Refuse to truncate files with mandatory locks held on them.
1568 error = locks_verify_locked(inode);
1570 error = security_path_truncate(path, 0,
1571 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1575 error = do_truncate(dentry, 0,
1576 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1579 put_write_access(inode);
1583 if (flag & FMODE_WRITE)
1588 ima_counts_put(path, acc_mode ?
1589 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1590 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1595 * Be careful about ever adding any more callers of this
1596 * function. Its flags must be in the namei format, not
1597 * what get passed to sys_open().
1599 static int __open_namei_create(struct nameidata *nd, struct path *path,
1603 struct dentry *dir = nd->path.dentry;
1605 if (!IS_POSIXACL(dir->d_inode))
1606 mode &= ~current_umask();
1607 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1610 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1612 mutex_unlock(&dir->d_inode->i_mutex);
1613 dput(nd->path.dentry);
1614 nd->path.dentry = path->dentry;
1617 /* Don't check for write permission, don't truncate */
1618 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1622 * Note that while the flag value (low two bits) for sys_open means:
1627 * it is changed into
1628 * 00 - no permissions needed
1629 * 01 - read-permission
1630 * 10 - write-permission
1632 * for the internal routines (ie open_namei()/follow_link() etc)
1633 * This is more logical, and also allows the 00 "no perm needed"
1634 * to be used for symlinks (where the permissions are checked
1638 static inline int open_to_namei_flags(int flag)
1640 if ((flag+1) & O_ACCMODE)
1645 static int open_will_write_to_fs(int flag, struct inode *inode)
1648 * We'll never write to the fs underlying
1651 if (special_file(inode->i_mode))
1653 return (flag & O_TRUNC);
1657 * Note that the low bits of the passed in "open_flag"
1658 * are not the same as in the local variable "flag". See
1659 * open_to_namei_flags() for more details.
1661 struct file *do_filp_open(int dfd, const char *pathname,
1662 int open_flag, int mode, int acc_mode)
1665 struct nameidata nd;
1671 int flag = open_to_namei_flags(open_flag);
1674 acc_mode = MAY_OPEN | ACC_MODE(flag);
1676 /* O_TRUNC implies we need access checks for write permissions */
1678 acc_mode |= MAY_WRITE;
1680 /* Allow the LSM permission hook to distinguish append
1681 access from general write access. */
1682 if (flag & O_APPEND)
1683 acc_mode |= MAY_APPEND;
1686 * The simplest case - just a plain lookup.
1688 if (!(flag & O_CREAT)) {
1689 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1692 return ERR_PTR(error);
1697 * Create - we need to know the parent.
1699 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1701 return ERR_PTR(error);
1702 error = path_walk(pathname, &nd);
1706 return ERR_PTR(error);
1708 if (unlikely(!audit_dummy_context()))
1709 audit_inode(pathname, nd.path.dentry);
1712 * We have the parent and last component. First of all, check
1713 * that we are not asked to creat(2) an obvious directory - that
1717 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1721 filp = get_empty_filp();
1724 nd.intent.open.file = filp;
1725 nd.intent.open.flags = flag;
1726 nd.intent.open.create_mode = mode;
1727 dir = nd.path.dentry;
1728 nd.flags &= ~LOOKUP_PARENT;
1729 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1731 nd.flags |= LOOKUP_EXCL;
1732 mutex_lock(&dir->d_inode->i_mutex);
1733 path.dentry = lookup_hash(&nd);
1734 path.mnt = nd.path.mnt;
1737 error = PTR_ERR(path.dentry);
1738 if (IS_ERR(path.dentry)) {
1739 mutex_unlock(&dir->d_inode->i_mutex);
1743 if (IS_ERR(nd.intent.open.file)) {
1744 error = PTR_ERR(nd.intent.open.file);
1745 goto exit_mutex_unlock;
1748 /* Negative dentry, just create the file */
1749 if (!path.dentry->d_inode) {
1751 * This write is needed to ensure that a
1752 * ro->rw transition does not occur between
1753 * the time when the file is created and when
1754 * a permanent write count is taken through
1755 * the 'struct file' in nameidata_to_filp().
1757 error = mnt_want_write(nd.path.mnt);
1759 goto exit_mutex_unlock;
1760 error = __open_namei_create(&nd, &path, flag, mode);
1762 mnt_drop_write(nd.path.mnt);
1765 filp = nameidata_to_filp(&nd, open_flag);
1767 ima_counts_put(&nd.path,
1768 acc_mode & (MAY_READ | MAY_WRITE |
1770 mnt_drop_write(nd.path.mnt);
1777 * It already exists.
1779 mutex_unlock(&dir->d_inode->i_mutex);
1780 audit_inode(pathname, path.dentry);
1786 if (__follow_mount(&path)) {
1788 if (flag & O_NOFOLLOW)
1793 if (!path.dentry->d_inode)
1795 if (path.dentry->d_inode->i_op->follow_link)
1798 path_to_nameidata(&path, &nd);
1800 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1805 * 1. may_open() truncates a file
1806 * 2. a rw->ro mount transition occurs
1807 * 3. nameidata_to_filp() fails due to
1809 * That would be inconsistent, and should
1810 * be avoided. Taking this mnt write here
1811 * ensures that (2) can not occur.
1813 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1815 error = mnt_want_write(nd.path.mnt);
1819 error = may_open(&nd.path, acc_mode, flag);
1822 mnt_drop_write(nd.path.mnt);
1825 filp = nameidata_to_filp(&nd, open_flag);
1827 ima_counts_put(&nd.path,
1828 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1830 * It is now safe to drop the mnt write
1831 * because the filp has had a write taken
1835 mnt_drop_write(nd.path.mnt);
1841 mutex_unlock(&dir->d_inode->i_mutex);
1843 path_put_conditional(&path, &nd);
1845 if (!IS_ERR(nd.intent.open.file))
1846 release_open_intent(&nd);
1851 return ERR_PTR(error);
1855 if (flag & O_NOFOLLOW)
1858 * This is subtle. Instead of calling do_follow_link() we do the
1859 * thing by hands. The reason is that this way we have zero link_count
1860 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1861 * After that we have the parent and last component, i.e.
1862 * we are in the same situation as after the first path_walk().
1863 * Well, almost - if the last component is normal we get its copy
1864 * stored in nd->last.name and we will have to putname() it when we
1865 * are done. Procfs-like symlinks just set LAST_BIND.
1867 nd.flags |= LOOKUP_PARENT;
1868 error = security_inode_follow_link(path.dentry, &nd);
1871 error = __do_follow_link(&path, &nd);
1873 /* Does someone understand code flow here? Or it is only
1874 * me so stupid? Anathema to whoever designed this non-sense
1875 * with "intent.open".
1877 release_open_intent(&nd);
1880 return ERR_PTR(error);
1882 nd.flags &= ~LOOKUP_PARENT;
1883 if (nd.last_type == LAST_BIND)
1886 if (nd.last_type != LAST_NORM)
1888 if (nd.last.name[nd.last.len]) {
1889 __putname(nd.last.name);
1894 __putname(nd.last.name);
1897 dir = nd.path.dentry;
1898 mutex_lock(&dir->d_inode->i_mutex);
1899 path.dentry = lookup_hash(&nd);
1900 path.mnt = nd.path.mnt;
1901 __putname(nd.last.name);
1906 * filp_open - open file and return file pointer
1908 * @filename: path to open
1909 * @flags: open flags as per the open(2) second argument
1910 * @mode: mode for the new file if O_CREAT is set, else ignored
1912 * This is the helper to open a file from kernelspace if you really
1913 * have to. But in generally you should not do this, so please move
1914 * along, nothing to see here..
1916 struct file *filp_open(const char *filename, int flags, int mode)
1918 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1920 EXPORT_SYMBOL(filp_open);
1923 * lookup_create - lookup a dentry, creating it if it doesn't exist
1924 * @nd: nameidata info
1925 * @is_dir: directory flag
1927 * Simple function to lookup and return a dentry and create it
1928 * if it doesn't exist. Is SMP-safe.
1930 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1932 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1934 struct dentry *dentry = ERR_PTR(-EEXIST);
1936 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1938 * Yucky last component or no last component at all?
1939 * (foo/., foo/.., /////)
1941 if (nd->last_type != LAST_NORM)
1943 nd->flags &= ~LOOKUP_PARENT;
1944 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1945 nd->intent.open.flags = O_EXCL;
1948 * Do the final lookup.
1950 dentry = lookup_hash(nd);
1954 if (dentry->d_inode)
1957 * Special case - lookup gave negative, but... we had foo/bar/
1958 * From the vfs_mknod() POV we just have a negative dentry -
1959 * all is fine. Let's be bastards - you had / on the end, you've
1960 * been asking for (non-existent) directory. -ENOENT for you.
1962 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1964 dentry = ERR_PTR(-ENOENT);
1969 dentry = ERR_PTR(-EEXIST);
1973 EXPORT_SYMBOL_GPL(lookup_create);
1975 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1977 int error = may_create(dir, dentry);
1982 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1985 if (!dir->i_op->mknod)
1988 error = devcgroup_inode_mknod(mode, dev);
1992 error = security_inode_mknod(dir, dentry, mode, dev);
1997 error = dir->i_op->mknod(dir, dentry, mode, dev);
1999 fsnotify_create(dir, dentry);
2003 static int may_mknod(mode_t mode)
2005 switch (mode & S_IFMT) {
2011 case 0: /* zero mode translates to S_IFREG */
2020 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2025 struct dentry *dentry;
2026 struct nameidata nd;
2031 error = user_path_parent(dfd, filename, &nd, &tmp);
2035 dentry = lookup_create(&nd, 0);
2036 if (IS_ERR(dentry)) {
2037 error = PTR_ERR(dentry);
2040 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2041 mode &= ~current_umask();
2042 error = may_mknod(mode);
2045 error = mnt_want_write(nd.path.mnt);
2048 error = security_path_mknod(&nd.path, dentry, mode, dev);
2050 goto out_drop_write;
2051 switch (mode & S_IFMT) {
2052 case 0: case S_IFREG:
2053 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2055 case S_IFCHR: case S_IFBLK:
2056 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2057 new_decode_dev(dev));
2059 case S_IFIFO: case S_IFSOCK:
2060 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2064 mnt_drop_write(nd.path.mnt);
2068 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2075 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2077 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2080 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2082 int error = may_create(dir, dentry);
2087 if (!dir->i_op->mkdir)
2090 mode &= (S_IRWXUGO|S_ISVTX);
2091 error = security_inode_mkdir(dir, dentry, mode);
2096 error = dir->i_op->mkdir(dir, dentry, mode);
2098 fsnotify_mkdir(dir, dentry);
2102 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2106 struct dentry *dentry;
2107 struct nameidata nd;
2109 error = user_path_parent(dfd, pathname, &nd, &tmp);
2113 dentry = lookup_create(&nd, 1);
2114 error = PTR_ERR(dentry);
2118 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2119 mode &= ~current_umask();
2120 error = mnt_want_write(nd.path.mnt);
2123 error = security_path_mkdir(&nd.path, dentry, mode);
2125 goto out_drop_write;
2126 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2128 mnt_drop_write(nd.path.mnt);
2132 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2139 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2141 return sys_mkdirat(AT_FDCWD, pathname, mode);
2145 * We try to drop the dentry early: we should have
2146 * a usage count of 2 if we're the only user of this
2147 * dentry, and if that is true (possibly after pruning
2148 * the dcache), then we drop the dentry now.
2150 * A low-level filesystem can, if it choses, legally
2153 * if (!d_unhashed(dentry))
2156 * if it cannot handle the case of removing a directory
2157 * that is still in use by something else..
2159 void dentry_unhash(struct dentry *dentry)
2162 shrink_dcache_parent(dentry);
2163 spin_lock(&dcache_lock);
2164 spin_lock(&dentry->d_lock);
2165 if (atomic_read(&dentry->d_count) == 2)
2167 spin_unlock(&dentry->d_lock);
2168 spin_unlock(&dcache_lock);
2171 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2173 int error = may_delete(dir, dentry, 1);
2178 if (!dir->i_op->rmdir)
2183 mutex_lock(&dentry->d_inode->i_mutex);
2184 dentry_unhash(dentry);
2185 if (d_mountpoint(dentry))
2188 error = security_inode_rmdir(dir, dentry);
2190 error = dir->i_op->rmdir(dir, dentry);
2192 dentry->d_inode->i_flags |= S_DEAD;
2195 mutex_unlock(&dentry->d_inode->i_mutex);
2204 static long do_rmdir(int dfd, const char __user *pathname)
2208 struct dentry *dentry;
2209 struct nameidata nd;
2211 error = user_path_parent(dfd, pathname, &nd, &name);
2215 switch(nd.last_type) {
2227 nd.flags &= ~LOOKUP_PARENT;
2229 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2230 dentry = lookup_hash(&nd);
2231 error = PTR_ERR(dentry);
2234 error = mnt_want_write(nd.path.mnt);
2237 error = security_path_rmdir(&nd.path, dentry);
2240 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2242 mnt_drop_write(nd.path.mnt);
2246 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2253 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2255 return do_rmdir(AT_FDCWD, pathname);
2258 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2260 int error = may_delete(dir, dentry, 0);
2265 if (!dir->i_op->unlink)
2270 mutex_lock(&dentry->d_inode->i_mutex);
2271 if (d_mountpoint(dentry))
2274 error = security_inode_unlink(dir, dentry);
2276 error = dir->i_op->unlink(dir, dentry);
2278 mutex_unlock(&dentry->d_inode->i_mutex);
2280 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2281 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2282 fsnotify_link_count(dentry->d_inode);
2290 * Make sure that the actual truncation of the file will occur outside its
2291 * directory's i_mutex. Truncate can take a long time if there is a lot of
2292 * writeout happening, and we don't want to prevent access to the directory
2293 * while waiting on the I/O.
2295 static long do_unlinkat(int dfd, const char __user *pathname)
2299 struct dentry *dentry;
2300 struct nameidata nd;
2301 struct inode *inode = NULL;
2303 error = user_path_parent(dfd, pathname, &nd, &name);
2308 if (nd.last_type != LAST_NORM)
2311 nd.flags &= ~LOOKUP_PARENT;
2313 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2314 dentry = lookup_hash(&nd);
2315 error = PTR_ERR(dentry);
2316 if (!IS_ERR(dentry)) {
2317 /* Why not before? Because we want correct error value */
2318 if (nd.last.name[nd.last.len])
2320 inode = dentry->d_inode;
2322 atomic_inc(&inode->i_count);
2323 error = mnt_want_write(nd.path.mnt);
2326 error = security_path_unlink(&nd.path, dentry);
2329 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2331 mnt_drop_write(nd.path.mnt);
2335 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2337 iput(inode); /* truncate the inode here */
2344 error = !dentry->d_inode ? -ENOENT :
2345 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2349 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2351 if ((flag & ~AT_REMOVEDIR) != 0)
2354 if (flag & AT_REMOVEDIR)
2355 return do_rmdir(dfd, pathname);
2357 return do_unlinkat(dfd, pathname);
2360 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2362 return do_unlinkat(AT_FDCWD, pathname);
2365 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2367 int error = may_create(dir, dentry);
2372 if (!dir->i_op->symlink)
2375 error = security_inode_symlink(dir, dentry, oldname);
2380 error = dir->i_op->symlink(dir, dentry, oldname);
2382 fsnotify_create(dir, dentry);
2386 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2387 int, newdfd, const char __user *, newname)
2392 struct dentry *dentry;
2393 struct nameidata nd;
2395 from = getname(oldname);
2397 return PTR_ERR(from);
2399 error = user_path_parent(newdfd, newname, &nd, &to);
2403 dentry = lookup_create(&nd, 0);
2404 error = PTR_ERR(dentry);
2408 error = mnt_want_write(nd.path.mnt);
2411 error = security_path_symlink(&nd.path, dentry, from);
2413 goto out_drop_write;
2414 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2416 mnt_drop_write(nd.path.mnt);
2420 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2428 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2430 return sys_symlinkat(oldname, AT_FDCWD, newname);
2433 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2435 struct inode *inode = old_dentry->d_inode;
2441 error = may_create(dir, new_dentry);
2445 if (dir->i_sb != inode->i_sb)
2449 * A link to an append-only or immutable file cannot be created.
2451 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2453 if (!dir->i_op->link)
2455 if (S_ISDIR(inode->i_mode))
2458 error = security_inode_link(old_dentry, dir, new_dentry);
2462 mutex_lock(&inode->i_mutex);
2464 error = dir->i_op->link(old_dentry, dir, new_dentry);
2465 mutex_unlock(&inode->i_mutex);
2467 fsnotify_link(dir, inode, new_dentry);
2472 * Hardlinks are often used in delicate situations. We avoid
2473 * security-related surprises by not following symlinks on the
2476 * We don't follow them on the oldname either to be compatible
2477 * with linux 2.0, and to avoid hard-linking to directories
2478 * and other special files. --ADM
2480 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2481 int, newdfd, const char __user *, newname, int, flags)
2483 struct dentry *new_dentry;
2484 struct nameidata nd;
2485 struct path old_path;
2489 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2492 error = user_path_at(olddfd, oldname,
2493 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2498 error = user_path_parent(newdfd, newname, &nd, &to);
2502 if (old_path.mnt != nd.path.mnt)
2504 new_dentry = lookup_create(&nd, 0);
2505 error = PTR_ERR(new_dentry);
2506 if (IS_ERR(new_dentry))
2508 error = mnt_want_write(nd.path.mnt);
2511 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2513 goto out_drop_write;
2514 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2516 mnt_drop_write(nd.path.mnt);
2520 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2525 path_put(&old_path);
2530 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2532 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2536 * The worst of all namespace operations - renaming directory. "Perverted"
2537 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2539 * a) we can get into loop creation. Check is done in is_subdir().
2540 * b) race potential - two innocent renames can create a loop together.
2541 * That's where 4.4 screws up. Current fix: serialization on
2542 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2544 * c) we have to lock _three_ objects - parents and victim (if it exists).
2545 * And that - after we got ->i_mutex on parents (until then we don't know
2546 * whether the target exists). Solution: try to be smart with locking
2547 * order for inodes. We rely on the fact that tree topology may change
2548 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2549 * move will be locked. Thus we can rank directories by the tree
2550 * (ancestors first) and rank all non-directories after them.
2551 * That works since everybody except rename does "lock parent, lookup,
2552 * lock child" and rename is under ->s_vfs_rename_mutex.
2553 * HOWEVER, it relies on the assumption that any object with ->lookup()
2554 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2555 * we'd better make sure that there's no link(2) for them.
2556 * d) some filesystems don't support opened-but-unlinked directories,
2557 * either because of layout or because they are not ready to deal with
2558 * all cases correctly. The latter will be fixed (taking this sort of
2559 * stuff into VFS), but the former is not going away. Solution: the same
2560 * trick as in rmdir().
2561 * e) conversion from fhandle to dentry may come in the wrong moment - when
2562 * we are removing the target. Solution: we will have to grab ->i_mutex
2563 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2564 * ->i_mutex on parents, which works but leads to some truely excessive
2567 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2568 struct inode *new_dir, struct dentry *new_dentry)
2571 struct inode *target;
2574 * If we are going to change the parent - check write permissions,
2575 * we'll need to flip '..'.
2577 if (new_dir != old_dir) {
2578 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2583 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2587 target = new_dentry->d_inode;
2589 mutex_lock(&target->i_mutex);
2590 dentry_unhash(new_dentry);
2592 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2595 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2598 target->i_flags |= S_DEAD;
2599 mutex_unlock(&target->i_mutex);
2600 if (d_unhashed(new_dentry))
2601 d_rehash(new_dentry);
2605 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2606 d_move(old_dentry,new_dentry);
2610 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2611 struct inode *new_dir, struct dentry *new_dentry)
2613 struct inode *target;
2616 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2621 target = new_dentry->d_inode;
2623 mutex_lock(&target->i_mutex);
2624 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2627 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2629 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2630 d_move(old_dentry, new_dentry);
2633 mutex_unlock(&target->i_mutex);
2638 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2639 struct inode *new_dir, struct dentry *new_dentry)
2642 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2643 const char *old_name;
2645 if (old_dentry->d_inode == new_dentry->d_inode)
2648 error = may_delete(old_dir, old_dentry, is_dir);
2652 if (!new_dentry->d_inode)
2653 error = may_create(new_dir, new_dentry);
2655 error = may_delete(new_dir, new_dentry, is_dir);
2659 if (!old_dir->i_op->rename)
2662 vfs_dq_init(old_dir);
2663 vfs_dq_init(new_dir);
2665 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2668 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2670 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2672 const char *new_name = old_dentry->d_name.name;
2673 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2674 new_dentry->d_inode, old_dentry);
2676 fsnotify_oldname_free(old_name);
2681 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2682 int, newdfd, const char __user *, newname)
2684 struct dentry *old_dir, *new_dir;
2685 struct dentry *old_dentry, *new_dentry;
2686 struct dentry *trap;
2687 struct nameidata oldnd, newnd;
2692 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2696 error = user_path_parent(newdfd, newname, &newnd, &to);
2701 if (oldnd.path.mnt != newnd.path.mnt)
2704 old_dir = oldnd.path.dentry;
2706 if (oldnd.last_type != LAST_NORM)
2709 new_dir = newnd.path.dentry;
2710 if (newnd.last_type != LAST_NORM)
2713 oldnd.flags &= ~LOOKUP_PARENT;
2714 newnd.flags &= ~LOOKUP_PARENT;
2715 newnd.flags |= LOOKUP_RENAME_TARGET;
2717 trap = lock_rename(new_dir, old_dir);
2719 old_dentry = lookup_hash(&oldnd);
2720 error = PTR_ERR(old_dentry);
2721 if (IS_ERR(old_dentry))
2723 /* source must exist */
2725 if (!old_dentry->d_inode)
2727 /* unless the source is a directory trailing slashes give -ENOTDIR */
2728 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2730 if (oldnd.last.name[oldnd.last.len])
2732 if (newnd.last.name[newnd.last.len])
2735 /* source should not be ancestor of target */
2737 if (old_dentry == trap)
2739 new_dentry = lookup_hash(&newnd);
2740 error = PTR_ERR(new_dentry);
2741 if (IS_ERR(new_dentry))
2743 /* target should not be an ancestor of source */
2745 if (new_dentry == trap)
2748 error = mnt_want_write(oldnd.path.mnt);
2751 error = security_path_rename(&oldnd.path, old_dentry,
2752 &newnd.path, new_dentry);
2755 error = vfs_rename(old_dir->d_inode, old_dentry,
2756 new_dir->d_inode, new_dentry);
2758 mnt_drop_write(oldnd.path.mnt);
2764 unlock_rename(new_dir, old_dir);
2766 path_put(&newnd.path);
2769 path_put(&oldnd.path);
2775 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2777 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2780 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2784 len = PTR_ERR(link);
2789 if (len > (unsigned) buflen)
2791 if (copy_to_user(buffer, link, len))
2798 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2799 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2800 * using) it for any given inode is up to filesystem.
2802 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2804 struct nameidata nd;
2809 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2811 return PTR_ERR(cookie);
2813 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2814 if (dentry->d_inode->i_op->put_link)
2815 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2819 int vfs_follow_link(struct nameidata *nd, const char *link)
2821 return __vfs_follow_link(nd, link);
2824 /* get the link contents into pagecache */
2825 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2829 struct address_space *mapping = dentry->d_inode->i_mapping;
2830 page = read_mapping_page(mapping, 0, NULL);
2835 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2839 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2841 struct page *page = NULL;
2842 char *s = page_getlink(dentry, &page);
2843 int res = vfs_readlink(dentry,buffer,buflen,s);
2846 page_cache_release(page);
2851 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2853 struct page *page = NULL;
2854 nd_set_link(nd, page_getlink(dentry, &page));
2858 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2860 struct page *page = cookie;
2864 page_cache_release(page);
2869 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2871 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2873 struct address_space *mapping = inode->i_mapping;
2878 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2880 flags |= AOP_FLAG_NOFS;
2883 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2884 flags, &page, &fsdata);
2888 kaddr = kmap_atomic(page, KM_USER0);
2889 memcpy(kaddr, symname, len-1);
2890 kunmap_atomic(kaddr, KM_USER0);
2892 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2899 mark_inode_dirty(inode);
2905 int page_symlink(struct inode *inode, const char *symname, int len)
2907 return __page_symlink(inode, symname, len,
2908 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2911 const struct inode_operations page_symlink_inode_operations = {
2912 .readlink = generic_readlink,
2913 .follow_link = page_follow_link_light,
2914 .put_link = page_put_link,
2917 EXPORT_SYMBOL(user_path_at);
2918 EXPORT_SYMBOL(follow_down);
2919 EXPORT_SYMBOL(follow_up);
2920 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2921 EXPORT_SYMBOL(getname);
2922 EXPORT_SYMBOL(lock_rename);
2923 EXPORT_SYMBOL(lookup_one_len);
2924 EXPORT_SYMBOL(page_follow_link_light);
2925 EXPORT_SYMBOL(page_put_link);
2926 EXPORT_SYMBOL(page_readlink);
2927 EXPORT_SYMBOL(__page_symlink);
2928 EXPORT_SYMBOL(page_symlink);
2929 EXPORT_SYMBOL(page_symlink_inode_operations);
2930 EXPORT_SYMBOL(path_lookup);
2931 EXPORT_SYMBOL(kern_path);
2932 EXPORT_SYMBOL(vfs_path_lookup);
2933 EXPORT_SYMBOL(inode_permission);
2934 EXPORT_SYMBOL(file_permission);
2935 EXPORT_SYMBOL(unlock_rename);
2936 EXPORT_SYMBOL(vfs_create);
2937 EXPORT_SYMBOL(vfs_follow_link);
2938 EXPORT_SYMBOL(vfs_link);
2939 EXPORT_SYMBOL(vfs_mkdir);
2940 EXPORT_SYMBOL(vfs_mknod);
2941 EXPORT_SYMBOL(generic_permission);
2942 EXPORT_SYMBOL(vfs_readlink);
2943 EXPORT_SYMBOL(vfs_rename);
2944 EXPORT_SYMBOL(vfs_rmdir);
2945 EXPORT_SYMBOL(vfs_symlink);
2946 EXPORT_SYMBOL(vfs_unlink);
2947 EXPORT_SYMBOL(dentry_unhash);
2948 EXPORT_SYMBOL(generic_readlink);