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/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.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/namei.h>
35 #include <linux/proc_fs.h>
36 #include <linux/vserver/inode.h>
37 #include <linux/vserver/debug.h>
38 #include <asm/namei.h>
39 #include <asm/uaccess.h>
41 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existant name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
121 static int do_getname(const char __user *filename, char *page)
124 unsigned long len = PATH_MAX;
126 if (!segment_eq(get_fs(), KERNEL_DS)) {
127 if ((unsigned long) filename >= TASK_SIZE)
129 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
130 len = TASK_SIZE - (unsigned long) filename;
133 retval = strncpy_from_user(page, filename, len);
137 return -ENAMETOOLONG;
143 char * getname(const char __user * filename)
147 result = ERR_PTR(-ENOMEM);
150 int retval = do_getname(filename, tmp);
155 result = ERR_PTR(retval);
158 audit_getname(result);
162 #ifdef CONFIG_AUDITSYSCALL
163 void putname(const char *name)
165 if (unlikely(current->audit_context))
170 EXPORT_SYMBOL(putname);
175 * generic_permission - check for access rights on a Posix-like filesystem
176 * @inode: inode to check access rights for
177 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
178 * @check_acl: optional callback to check for Posix ACLs
180 * Used to check for read/write/execute permissions on a file.
181 * We use "fsuid" for this, letting us set arbitrary permissions
182 * for filesystem access without changing the "normal" uids which
183 * are used for other things..
185 int generic_permission(struct inode *inode, int mask,
186 int (*check_acl)(struct inode *inode, int mask))
188 umode_t mode = inode->i_mode;
190 if (current->fsuid == inode->i_uid)
193 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
194 int error = check_acl(inode, mask);
195 if (error == -EACCES)
196 goto check_capabilities;
197 else if (error != -EAGAIN)
201 if (in_group_p(inode->i_gid))
206 * If the DACs are ok we don't need any capability check.
208 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
213 * Read/write DACs are always overridable.
214 * Executable DACs are overridable if at least one exec bit is set.
216 if (!(mask & MAY_EXEC) ||
217 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
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))
231 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
233 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN)) {
234 vxwprintk(1, "xid=%d did hit the barrier.",
238 if (inode->i_xid == 0)
240 #ifdef CONFIG_VSERVER_FILESHARING
241 /* MEF: PlanetLab FS module assumes that any file that can be
242 * named (e.g., via a cross mount) is not hidden from another
243 * context or the admin context.
245 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC))
248 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
251 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
252 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
257 int permission(struct inode *inode, int mask, struct nameidata *nd)
261 if (mask & MAY_WRITE) {
262 umode_t mode = inode->i_mode;
265 * Nobody gets write access to a read-only fs.
267 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode))
279 /* Ordinary permission routines do not understand MAY_APPEND. */
280 submask = mask & ~MAY_APPEND;
281 if ((retval = xid_permission(inode, mask, nd)))
283 if (inode->i_op && inode->i_op->permission)
284 retval = inode->i_op->permission(inode, submask, nd);
286 retval = generic_permission(inode, submask, NULL);
290 return security_inode_permission(inode, mask, nd);
294 * vfs_permission - check for access rights to a given path
295 * @nd: lookup result that describes the path
296 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 * Used to check for read/write/execute permissions on a path.
299 * We use "fsuid" for this, letting us set arbitrary permissions
300 * for filesystem access without changing the "normal" uids which
301 * are used for other things.
303 int vfs_permission(struct nameidata *nd, int mask)
305 return permission(nd->dentry->d_inode, mask, nd);
309 * file_permission - check for additional access rights to a given file
310 * @file: file to check access rights for
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
313 * Used to check for read/write/execute permissions on an already opened
317 * Do not use this function in new code. All access checks should
318 * be done using vfs_permission().
320 int file_permission(struct file *file, int mask)
322 return permission(file->f_dentry->d_inode, mask, NULL);
326 * get_write_access() gets write permission for a file.
327 * put_write_access() releases this write permission.
328 * This is used for regular files.
329 * We cannot support write (and maybe mmap read-write shared) accesses and
330 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
331 * can have the following values:
332 * 0: no writers, no VM_DENYWRITE mappings
333 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
334 * > 0: (i_writecount) users are writing to the file.
336 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
337 * except for the cases where we don't hold i_writecount yet. Then we need to
338 * use {get,deny}_write_access() - these functions check the sign and refuse
339 * to do the change if sign is wrong. Exclusion between them is provided by
340 * the inode->i_lock spinlock.
343 int get_write_access(struct inode * inode)
345 spin_lock(&inode->i_lock);
346 if (atomic_read(&inode->i_writecount) < 0) {
347 spin_unlock(&inode->i_lock);
350 atomic_inc(&inode->i_writecount);
351 spin_unlock(&inode->i_lock);
356 int deny_write_access(struct file * file)
358 struct inode *inode = file->f_dentry->d_inode;
360 spin_lock(&inode->i_lock);
361 if (atomic_read(&inode->i_writecount) > 0) {
362 spin_unlock(&inode->i_lock);
365 atomic_dec(&inode->i_writecount);
366 spin_unlock(&inode->i_lock);
371 void path_release(struct nameidata *nd)
378 * umount() mustn't call path_release()/mntput() as that would clear
381 void path_release_on_umount(struct nameidata *nd)
384 mntput_no_expire(nd->mnt);
388 * release_open_intent - free up open intent resources
389 * @nd: pointer to nameidata
391 void release_open_intent(struct nameidata *nd)
393 if (nd->intent.open.file->f_dentry == NULL)
394 put_filp(nd->intent.open.file);
396 fput(nd->intent.open.file);
400 * Internal lookup() using the new generic dcache.
403 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
405 struct dentry * dentry = __d_lookup(parent, name);
407 /* lockess __d_lookup may fail due to concurrent d_move()
408 * in some unrelated directory, so try with d_lookup
411 dentry = d_lookup(parent, name);
413 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
414 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
423 * Short-cut version of permission(), for calling by
424 * path_walk(), when dcache lock is held. Combines parts
425 * of permission() and generic_permission(), and tests ONLY for
426 * MAY_EXEC permission.
428 * If appropriate, check DAC only. If not appropriate, or
429 * short-cut DAC fails, then call permission() to do more
430 * complete permission check.
432 static int exec_permission_lite(struct inode *inode,
433 struct nameidata *nd)
435 umode_t mode = inode->i_mode;
437 if (inode->i_op && inode->i_op->permission)
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, nd);
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 = d_alloc(parent, name);
493 result = ERR_PTR(-ENOMEM);
495 result = dir->i_op->lookup(dir, dentry, nd);
501 mutex_unlock(&dir->i_mutex);
506 * Uhhuh! Nasty case: the cache was re-populated while
507 * we waited on the semaphore. Need to revalidate.
509 mutex_unlock(&dir->i_mutex);
510 if (result->d_op && result->d_op->d_revalidate) {
511 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
513 result = ERR_PTR(-ENOENT);
519 static int __emul_lookup_dentry(const char *, struct nameidata *);
522 static __always_inline int
523 walk_init_root(const char *name, struct nameidata *nd)
525 read_lock(¤t->fs->lock);
526 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
527 nd->mnt = mntget(current->fs->altrootmnt);
528 nd->dentry = dget(current->fs->altroot);
529 read_unlock(¤t->fs->lock);
530 if (__emul_lookup_dentry(name,nd))
532 read_lock(¤t->fs->lock);
534 nd->mnt = mntget(current->fs->rootmnt);
535 nd->dentry = dget(current->fs->root);
536 read_unlock(¤t->fs->lock);
540 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
549 if (!walk_init_root(link, nd))
550 /* weird __emul_prefix() stuff did it */
553 res = link_path_walk(link, nd);
555 if (nd->depth || res || nd->last_type!=LAST_NORM)
558 * If it is an iterative symlinks resolution in open_namei() we
559 * have to copy the last component. And all that crap because of
560 * bloody create() on broken symlinks. Furrfu...
563 if (unlikely(!name)) {
567 strcpy(name, nd->last.name);
568 nd->last.name = name;
572 return PTR_ERR(link);
576 struct vfsmount *mnt;
577 struct dentry *dentry;
580 static inline void dput_path(struct path *path, struct nameidata *nd)
583 if (path->mnt != nd->mnt)
587 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
590 if (nd->mnt != path->mnt)
593 nd->dentry = path->dentry;
596 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
600 struct dentry *dentry = path->dentry;
602 touch_atime(path->mnt, dentry);
603 nd_set_link(nd, NULL);
605 if (path->mnt != nd->mnt) {
606 path_to_nameidata(path, nd);
610 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
611 error = PTR_ERR(cookie);
612 if (!IS_ERR(cookie)) {
613 char *s = nd_get_link(nd);
616 error = __vfs_follow_link(nd, s);
617 if (dentry->d_inode->i_op->put_link)
618 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
627 * This limits recursive symlink follows to 8, while
628 * limiting consecutive symlinks to 40.
630 * Without that kind of total limit, nasty chains of consecutive
631 * symlinks can cause almost arbitrarily long lookups.
633 static inline int do_follow_link(struct path *path, struct nameidata *nd)
636 if (current->link_count >= MAX_NESTED_LINKS)
638 if (current->total_link_count >= 40)
640 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
642 err = security_inode_follow_link(path->dentry, nd);
645 current->link_count++;
646 current->total_link_count++;
648 err = __do_follow_link(path, nd);
649 current->link_count--;
658 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
660 struct vfsmount *parent;
661 struct dentry *mountpoint;
662 spin_lock(&vfsmount_lock);
663 parent=(*mnt)->mnt_parent;
664 if (parent == *mnt) {
665 spin_unlock(&vfsmount_lock);
669 mountpoint=dget((*mnt)->mnt_mountpoint);
670 spin_unlock(&vfsmount_lock);
672 *dentry = mountpoint;
678 /* no need for dcache_lock, as serialization is taken care in
681 static int __follow_mount(struct path *path)
684 while (d_mountpoint(path->dentry)) {
685 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
692 path->dentry = dget(mounted->mnt_root);
698 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
700 while (d_mountpoint(*dentry)) {
701 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
707 *dentry = dget(mounted->mnt_root);
711 /* no need for dcache_lock, as serialization is taken care in
714 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
716 struct vfsmount *mounted;
718 mounted = lookup_mnt(*mnt, *dentry);
723 *dentry = dget(mounted->mnt_root);
729 static __always_inline void follow_dotdot(struct nameidata *nd)
732 struct vfsmount *parent;
733 struct dentry *old = nd->dentry;
735 read_lock(¤t->fs->lock);
736 if (nd->dentry == current->fs->root &&
737 nd->mnt == current->fs->rootmnt) {
738 read_unlock(¤t->fs->lock);
739 /* for sane '/' avoid follow_mount() */
742 read_unlock(¤t->fs->lock);
743 spin_lock(&dcache_lock);
744 if (nd->dentry != nd->mnt->mnt_root) {
745 nd->dentry = dget(nd->dentry->d_parent);
746 spin_unlock(&dcache_lock);
750 spin_unlock(&dcache_lock);
751 spin_lock(&vfsmount_lock);
752 parent = nd->mnt->mnt_parent;
753 if (parent == nd->mnt) {
754 spin_unlock(&vfsmount_lock);
758 nd->dentry = dget(nd->mnt->mnt_mountpoint);
759 spin_unlock(&vfsmount_lock);
764 follow_mount(&nd->mnt, &nd->dentry);
768 * It's more convoluted than I'd like it to be, but... it's still fairly
769 * small and for now I'd prefer to have fast path as straight as possible.
770 * It _is_ time-critical.
772 static int do_lookup(struct nameidata *nd, struct qstr *name,
773 struct path *path, int atomic)
775 struct vfsmount *mnt = nd->mnt;
776 struct dentry *dentry = __d_lookup(nd->dentry, name);
781 if (dentry->d_op && dentry->d_op->d_revalidate)
782 goto need_revalidate;
783 inode = dentry->d_inode;
786 #ifdef CONFIG_VSERVER_FILESHARING
787 /* MEF: PlanetLab FS module assumes that any file that can be
788 * named (e.g., via a cross mount) is not hidden from another
789 * context or the admin context.
791 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC|VX_ADMIN)) {
794 else /* do the following check */
796 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
798 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
799 struct proc_dir_entry *de = PDE(inode);
801 if (de && !vx_hide_check(0, de->vx_flags))
806 path->dentry = dentry;
807 __follow_mount(path);
810 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
811 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
812 vxd_path(dentry, mnt));
818 return -EWOULDBLOCKIO;
819 dentry = real_lookup(nd->dentry, name, nd);
826 return -EWOULDBLOCKIO;
827 if (dentry->d_op->d_revalidate(dentry, nd))
829 if (d_invalidate(dentry))
835 return PTR_ERR(dentry);
840 * This is the basic name resolution function, turning a pathname into
841 * the final dentry. We expect 'base' to be positive and a directory.
843 * Returns 0 and nd will have valid dentry and mnt on success.
844 * Returns error and drops reference to input namei data on failure.
846 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
851 unsigned int lookup_flags = nd->flags;
853 atomic = (lookup_flags & LOOKUP_ATOMIC);
860 inode = nd->dentry->d_inode;
862 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
864 /* At this point we know we have a real path component. */
870 nd->flags |= LOOKUP_CONTINUE;
871 err = exec_permission_lite(inode, nd);
873 err = vfs_permission(nd, MAY_EXEC);
878 c = *(const unsigned char *)name;
880 hash = init_name_hash();
883 hash = partial_name_hash(c, hash);
884 c = *(const unsigned char *)name;
885 } while (c && (c != '/'));
886 this.len = name - (const char *) this.name;
887 this.hash = end_name_hash(hash);
889 /* remove trailing slashes? */
892 while (*++name == '/');
894 goto last_with_slashes;
897 * "." and ".." are special - ".." especially so because it has
898 * to be able to know about the current root directory and
899 * parent relationships.
901 if (this.name[0] == '.') switch (this.len) {
905 if (this.name[1] != '.')
908 inode = nd->dentry->d_inode;
914 * See if the low-level filesystem might want
915 * to use its own hash..
917 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
918 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
922 /* This does the actual lookups.. */
923 err = do_lookup(nd, &this, &next, atomic);
928 inode = next.dentry->d_inode;
935 if (inode->i_op->follow_link) {
936 err = do_follow_link(&next, nd);
940 inode = nd->dentry->d_inode;
947 path_to_nameidata(&next, nd);
949 if (!inode->i_op->lookup)
952 /* here ends the main loop */
955 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
957 /* Clear LOOKUP_CONTINUE iff it was previously unset */
958 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
959 if (lookup_flags & LOOKUP_PARENT)
961 if (this.name[0] == '.') switch (this.len) {
965 if (this.name[1] != '.')
968 inode = nd->dentry->d_inode;
973 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
974 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
978 err = do_lookup(nd, &this, &next, atomic);
981 inode = next.dentry->d_inode;
982 if ((lookup_flags & LOOKUP_FOLLOW)
983 && inode && inode->i_op && inode->i_op->follow_link) {
984 err = do_follow_link(&next, nd);
987 inode = nd->dentry->d_inode;
989 path_to_nameidata(&next, nd);
993 if (lookup_flags & LOOKUP_DIRECTORY) {
995 if (!inode->i_op || !inode->i_op->lookup)
1001 nd->last_type = LAST_NORM;
1002 if (this.name[0] != '.')
1005 nd->last_type = LAST_DOT;
1006 else if (this.len == 2 && this.name[1] == '.')
1007 nd->last_type = LAST_DOTDOT;
1012 * We bypassed the ordinary revalidation routines.
1013 * We may need to check the cached dentry for staleness.
1015 if (nd->dentry && nd->dentry->d_sb &&
1016 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1018 /* Note: we do not d_invalidate() */
1019 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1025 dput_path(&next, nd);
1034 * Wrapper to retry pathname resolution whenever the underlying
1035 * file system returns an ESTALE.
1037 * Retry the whole path once, forcing real lookup requests
1038 * instead of relying on the dcache.
1040 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1042 struct nameidata save = *nd;
1045 /* make sure the stuff we saved doesn't go away */
1049 result = __link_path_walk(name, nd);
1050 if (result == -ESTALE) {
1054 nd->flags |= LOOKUP_REVAL;
1055 result = __link_path_walk(name, nd);
1064 int fastcall path_walk(const char * name, struct nameidata *nd)
1066 current->total_link_count = 0;
1067 return link_path_walk(name, nd);
1071 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1072 * everything is done. Returns 0 and drops input nd, if lookup failed;
1074 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1076 if (path_walk(name, nd))
1077 return 0; /* something went wrong... */
1079 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1080 struct dentry *old_dentry = nd->dentry;
1081 struct vfsmount *old_mnt = nd->mnt;
1082 struct qstr last = nd->last;
1083 int last_type = nd->last_type;
1085 * NAME was not found in alternate root or it's a directory. Try to find
1086 * it in the normal root:
1088 nd->last_type = LAST_ROOT;
1089 read_lock(¤t->fs->lock);
1090 nd->mnt = mntget(current->fs->rootmnt);
1091 nd->dentry = dget(current->fs->root);
1092 read_unlock(¤t->fs->lock);
1093 if (path_walk(name, nd) == 0) {
1094 if (nd->dentry->d_inode) {
1101 nd->dentry = old_dentry;
1104 nd->last_type = last_type;
1109 void set_fs_altroot(void)
1111 char *emul = __emul_prefix();
1112 struct nameidata nd;
1113 struct vfsmount *mnt = NULL, *oldmnt;
1114 struct dentry *dentry = NULL, *olddentry;
1119 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1125 write_lock(¤t->fs->lock);
1126 oldmnt = current->fs->altrootmnt;
1127 olddentry = current->fs->altroot;
1128 current->fs->altrootmnt = mnt;
1129 current->fs->altroot = dentry;
1130 write_unlock(¤t->fs->lock);
1137 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1138 static int fastcall do_path_lookup(int dfd, const char *name,
1139 unsigned int flags, struct nameidata *nd)
1145 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1150 read_lock(¤t->fs->lock);
1151 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1152 nd->mnt = mntget(current->fs->altrootmnt);
1153 nd->dentry = dget(current->fs->altroot);
1154 read_unlock(¤t->fs->lock);
1155 if (__emul_lookup_dentry(name,nd))
1156 goto out; /* found in altroot */
1157 read_lock(¤t->fs->lock);
1159 nd->mnt = mntget(current->fs->rootmnt);
1160 nd->dentry = dget(current->fs->root);
1161 read_unlock(¤t->fs->lock);
1162 } else if (dfd == AT_FDCWD) {
1163 read_lock(¤t->fs->lock);
1164 nd->mnt = mntget(current->fs->pwdmnt);
1165 nd->dentry = dget(current->fs->pwd);
1166 read_unlock(¤t->fs->lock);
1168 struct dentry *dentry;
1170 file = fget_light(dfd, &fput_needed);
1175 dentry = file->f_dentry;
1178 if (!S_ISDIR(dentry->d_inode->i_mode))
1181 retval = file_permission(file, MAY_EXEC);
1185 nd->mnt = mntget(file->f_vfsmnt);
1186 nd->dentry = dget(dentry);
1188 fput_light(file, fput_needed);
1190 current->total_link_count = 0;
1191 retval = link_path_walk(name, nd);
1193 if (likely(retval == 0)) {
1194 if (unlikely(current->audit_context && nd && nd->dentry &&
1195 nd->dentry->d_inode))
1196 audit_inode(name, nd->dentry->d_inode, flags);
1202 fput_light(file, fput_needed);
1206 int fastcall path_lookup(const char *name, unsigned int flags,
1207 struct nameidata *nd)
1209 return do_path_lookup(AT_FDCWD, name, flags, nd);
1212 static int __path_lookup_intent_open(int dfd, const char *name,
1213 unsigned int lookup_flags, struct nameidata *nd,
1214 int open_flags, int create_mode)
1216 struct file *filp = get_empty_filp();
1221 nd->intent.open.file = filp;
1222 nd->intent.open.flags = open_flags;
1223 nd->intent.open.create_mode = create_mode;
1224 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1225 if (IS_ERR(nd->intent.open.file)) {
1227 err = PTR_ERR(nd->intent.open.file);
1230 } else if (err != 0)
1231 release_open_intent(nd);
1236 * path_lookup_open - lookup a file path with open intent
1237 * @dfd: the directory to use as base, or AT_FDCWD
1238 * @name: pointer to file name
1239 * @lookup_flags: lookup intent flags
1240 * @nd: pointer to nameidata
1241 * @open_flags: open intent flags
1243 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1244 struct nameidata *nd, int open_flags)
1246 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1251 * path_lookup_create - lookup a file path with open + create intent
1252 * @dfd: the directory to use as base, or AT_FDCWD
1253 * @name: pointer to file name
1254 * @lookup_flags: lookup intent flags
1255 * @nd: pointer to nameidata
1256 * @open_flags: open intent flags
1257 * @create_mode: create intent flags
1259 static int path_lookup_create(int dfd, const char *name,
1260 unsigned int lookup_flags, struct nameidata *nd,
1261 int open_flags, int create_mode)
1263 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1264 nd, open_flags, create_mode);
1267 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1268 struct nameidata *nd, int open_flags)
1270 char *tmp = getname(name);
1271 int err = PTR_ERR(tmp);
1274 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1281 * Restricted form of lookup. Doesn't follow links, single-component only,
1282 * needs parent already locked. Doesn't follow mounts.
1285 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1287 struct dentry * dentry;
1288 struct inode *inode;
1291 inode = base->d_inode;
1292 err = permission(inode, MAY_EXEC, nd);
1293 dentry = ERR_PTR(err);
1298 * See if the low-level filesystem might want
1299 * to use its own hash..
1301 if (base->d_op && base->d_op->d_hash) {
1302 err = base->d_op->d_hash(base, name);
1303 dentry = ERR_PTR(err);
1308 dentry = cached_lookup(base, name, nd);
1310 struct dentry *new = d_alloc(base, name);
1311 dentry = ERR_PTR(-ENOMEM);
1314 dentry = inode->i_op->lookup(inode, new, nd);
1324 static struct dentry *lookup_hash(struct nameidata *nd)
1326 return __lookup_hash(&nd->last, nd->dentry, nd);
1330 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1341 hash = init_name_hash();
1343 c = *(const unsigned char *)name++;
1344 if (c == '/' || c == '\0')
1346 hash = partial_name_hash(c, hash);
1348 this.hash = end_name_hash(hash);
1350 return __lookup_hash(&this, base, NULL);
1352 return ERR_PTR(-EACCES);
1358 * is used by most simple commands to get the inode of a specified name.
1359 * Open, link etc use their own routines, but this is enough for things
1362 * namei exists in two versions: namei/lnamei. The only difference is
1363 * that namei follows links, while lnamei does not.
1366 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1367 struct nameidata *nd)
1369 char *tmp = getname(name);
1370 int err = PTR_ERR(tmp);
1373 err = do_path_lookup(dfd, tmp, flags, nd);
1379 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1381 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1385 * It's inline, so penalty for filesystems that don't use sticky bit is
1388 static inline int check_sticky(struct inode *dir, struct inode *inode)
1390 if (!(dir->i_mode & S_ISVTX))
1392 if (inode->i_uid == current->fsuid)
1394 if (dir->i_uid == current->fsuid)
1396 return !capable(CAP_FOWNER);
1400 * Check whether we can remove a link victim from directory dir, check
1401 * whether the type of victim is right.
1402 * 1. We can't do it if dir is read-only (done in permission())
1403 * 2. We should have write and exec permissions on dir
1404 * 3. We can't remove anything from append-only dir
1405 * 4. We can't do anything with immutable dir (done in permission())
1406 * 5. If the sticky bit on dir is set we should either
1407 * a. be owner of dir, or
1408 * b. be owner of victim, or
1409 * c. have CAP_FOWNER capability
1410 * 6. If the victim is append-only or immutable we can't do antyhing with
1411 * links pointing to it.
1412 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1413 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1414 * 9. We can't remove a root or mountpoint.
1415 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1416 * nfs_async_unlink().
1418 static int may_delete(struct inode *dir, struct dentry *victim,
1419 int isdir, struct nameidata *nd)
1423 if (!victim->d_inode)
1426 BUG_ON(victim->d_parent->d_inode != dir);
1427 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
1429 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1434 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1435 IS_IXORUNLINK(victim->d_inode))
1438 if (!S_ISDIR(victim->d_inode->i_mode))
1440 if (IS_ROOT(victim))
1442 } else if (S_ISDIR(victim->d_inode->i_mode))
1444 if (IS_DEADDIR(dir))
1446 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1451 /* Check whether we can create an object with dentry child in directory
1453 * 1. We can't do it if child already exists (open has special treatment for
1454 * this case, but since we are inlined it's OK)
1455 * 2. We can't do it if dir is read-only (done in permission())
1456 * 3. We should have write and exec permissions on dir
1457 * 4. We can't do it if dir is immutable (done in permission())
1459 static inline int may_create(struct inode *dir, struct dentry *child,
1460 struct nameidata *nd)
1464 if (IS_DEADDIR(dir))
1466 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1470 * O_DIRECTORY translates into forcing a directory lookup.
1472 static inline int lookup_flags(unsigned int f)
1474 unsigned long retval = LOOKUP_FOLLOW;
1477 retval &= ~LOOKUP_FOLLOW;
1479 if (f & O_DIRECTORY)
1480 retval |= LOOKUP_DIRECTORY;
1481 if (f & O_ATOMICLOOKUP)
1482 retval |= LOOKUP_ATOMIC;
1488 * p1 and p2 should be directories on the same fs.
1490 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1495 mutex_lock(&p1->d_inode->i_mutex);
1499 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1501 for (p = p1; p->d_parent != p; p = p->d_parent) {
1502 if (p->d_parent == p2) {
1503 mutex_lock(&p2->d_inode->i_mutex);
1504 mutex_lock(&p1->d_inode->i_mutex);
1509 for (p = p2; p->d_parent != p; p = p->d_parent) {
1510 if (p->d_parent == p1) {
1511 mutex_lock(&p1->d_inode->i_mutex);
1512 mutex_lock(&p2->d_inode->i_mutex);
1517 mutex_lock(&p1->d_inode->i_mutex);
1518 mutex_lock(&p2->d_inode->i_mutex);
1522 void unlock_rename(struct dentry *p1, struct dentry *p2)
1524 mutex_unlock(&p1->d_inode->i_mutex);
1526 mutex_unlock(&p2->d_inode->i_mutex);
1527 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1531 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1532 struct nameidata *nd)
1534 int error = may_create(dir, dentry, nd);
1539 if (!dir->i_op || !dir->i_op->create)
1540 return -EACCES; /* shouldn't it be ENOSYS? */
1543 error = security_inode_create(dir, dentry, mode);
1547 error = dir->i_op->create(dir, dentry, mode, nd);
1549 fsnotify_create(dir, dentry);
1553 int may_open(struct nameidata *nd, int acc_mode, int flag)
1555 struct dentry *dentry = nd->dentry;
1556 struct inode *inode = dentry->d_inode;
1562 if (S_ISLNK(inode->i_mode))
1565 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1568 error = vfs_permission(nd, acc_mode);
1573 * FIFO's, sockets and device files are special: they don't
1574 * actually live on the filesystem itself, and as such you
1575 * can write to them even if the filesystem is read-only.
1577 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1579 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1580 if (nd->mnt->mnt_flags & MNT_NODEV)
1584 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1585 && (flag & FMODE_WRITE))
1588 * An append-only file must be opened in append mode for writing.
1590 if (IS_APPEND(inode)) {
1591 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1597 /* O_NOATIME can only be set by the owner or superuser */
1598 if (flag & O_NOATIME)
1599 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1603 * Ensure there are no outstanding leases on the file.
1605 error = break_lease(inode, flag);
1609 if (flag & O_TRUNC) {
1610 error = get_write_access(inode);
1615 * Refuse to truncate files with mandatory locks held on them.
1617 error = locks_verify_locked(inode);
1621 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1623 put_write_access(inode);
1627 if (flag & FMODE_WRITE)
1636 * namei for open - this is in fact almost the whole open-routine.
1638 * Note that the low bits of "flag" aren't the same as in the open
1639 * system call - they are 00 - no permissions needed
1640 * 01 - read permission needed
1641 * 10 - write permission needed
1642 * 11 - read/write permissions needed
1643 * which is a lot more logical, and also allows the "no perm" needed
1644 * for symlinks (where the permissions are checked later).
1647 int open_namei(int dfd, const char *pathname, int flag,
1648 int mode, struct nameidata *nd)
1650 int acc_mode, error;
1655 acc_mode = ACC_MODE(flag);
1657 /* O_TRUNC implies we need access checks for write permissions */
1659 acc_mode |= MAY_WRITE;
1661 /* Allow the LSM permission hook to distinguish append
1662 access from general write access. */
1663 if (flag & O_APPEND)
1664 acc_mode |= MAY_APPEND;
1667 * The simplest case - just a plain lookup.
1669 if (!(flag & O_CREAT)) {
1670 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1678 * Create - we need to know the parent.
1680 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1685 * We have the parent and last component. First of all, check
1686 * that we are not asked to creat(2) an obvious directory - that
1690 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1694 nd->flags &= ~LOOKUP_PARENT;
1695 mutex_lock(&dir->d_inode->i_mutex);
1696 path.dentry = lookup_hash(nd);
1700 error = PTR_ERR(path.dentry);
1701 if (IS_ERR(path.dentry)) {
1702 mutex_unlock(&dir->d_inode->i_mutex);
1706 if (IS_ERR(nd->intent.open.file)) {
1707 mutex_unlock(&dir->d_inode->i_mutex);
1708 error = PTR_ERR(nd->intent.open.file);
1712 /* Negative dentry, just create the file */
1713 if (!path.dentry->d_inode) {
1714 if (!IS_POSIXACL(dir->d_inode))
1715 mode &= ~current->fs->umask;
1716 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1717 mutex_unlock(&dir->d_inode->i_mutex);
1719 nd->dentry = path.dentry;
1722 /* Don't check for write permission, don't truncate */
1729 * It already exists.
1731 mutex_unlock(&dir->d_inode->i_mutex);
1737 if (__follow_mount(&path)) {
1739 if (flag & O_NOFOLLOW)
1743 if (!path.dentry->d_inode)
1745 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1748 path_to_nameidata(&path, nd);
1750 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1753 error = may_open(nd, acc_mode, flag);
1759 dput_path(&path, nd);
1761 if (!IS_ERR(nd->intent.open.file))
1762 release_open_intent(nd);
1768 if (flag & O_NOFOLLOW)
1771 * This is subtle. Instead of calling do_follow_link() we do the
1772 * thing by hands. The reason is that this way we have zero link_count
1773 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1774 * After that we have the parent and last component, i.e.
1775 * we are in the same situation as after the first path_walk().
1776 * Well, almost - if the last component is normal we get its copy
1777 * stored in nd->last.name and we will have to putname() it when we
1778 * are done. Procfs-like symlinks just set LAST_BIND.
1780 nd->flags |= LOOKUP_PARENT;
1781 error = security_inode_follow_link(path.dentry, nd);
1784 error = __do_follow_link(&path, nd);
1787 nd->flags &= ~LOOKUP_PARENT;
1788 if (nd->last_type == LAST_BIND)
1791 if (nd->last_type != LAST_NORM)
1793 if (nd->last.name[nd->last.len]) {
1794 __putname(nd->last.name);
1799 __putname(nd->last.name);
1803 mutex_lock(&dir->d_inode->i_mutex);
1804 path.dentry = lookup_hash(nd);
1806 __putname(nd->last.name);
1811 * lookup_create - lookup a dentry, creating it if it doesn't exist
1812 * @nd: nameidata info
1813 * @is_dir: directory flag
1815 * Simple function to lookup and return a dentry and create it
1816 * if it doesn't exist. Is SMP-safe.
1818 * Returns with nd->dentry->d_inode->i_mutex locked.
1820 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1822 struct dentry *dentry = ERR_PTR(-EEXIST);
1824 mutex_lock(&nd->dentry->d_inode->i_mutex);
1826 * Yucky last component or no last component at all?
1827 * (foo/., foo/.., /////)
1829 if (nd->last_type != LAST_NORM)
1831 nd->flags &= ~LOOKUP_PARENT;
1834 * Do the final lookup.
1836 dentry = lookup_hash(nd);
1841 * Special case - lookup gave negative, but... we had foo/bar/
1842 * From the vfs_mknod() POV we just have a negative dentry -
1843 * all is fine. Let's be bastards - you had / on the end, you've
1844 * been asking for (non-existent) directory. -ENOENT for you.
1846 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1851 dentry = ERR_PTR(-ENOENT);
1855 EXPORT_SYMBOL_GPL(lookup_create);
1857 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1858 int mode, dev_t dev, struct nameidata *nd)
1860 int error = may_create(dir, dentry, nd);
1865 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1868 if (!dir->i_op || !dir->i_op->mknod)
1871 error = security_inode_mknod(dir, dentry, mode, dev);
1876 error = dir->i_op->mknod(dir, dentry, mode, dev);
1878 fsnotify_create(dir, dentry);
1882 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1887 struct dentry * dentry;
1888 struct nameidata nd;
1892 tmp = getname(filename);
1894 return PTR_ERR(tmp);
1896 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1899 dentry = lookup_create(&nd, 0);
1900 error = PTR_ERR(dentry);
1902 if (!IS_POSIXACL(nd.dentry->d_inode))
1903 mode &= ~current->fs->umask;
1904 if (!IS_ERR(dentry)) {
1905 switch (mode & S_IFMT) {
1906 case 0: case S_IFREG:
1907 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1909 case S_IFCHR: case S_IFBLK:
1910 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1911 new_decode_dev(dev), &nd);
1913 case S_IFIFO: case S_IFSOCK:
1914 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1925 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1933 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1935 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1938 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1939 int mode, struct nameidata *nd)
1941 int error = may_create(dir, dentry, nd);
1946 if (!dir->i_op || !dir->i_op->mkdir)
1949 mode &= (S_IRWXUGO|S_ISVTX);
1950 error = security_inode_mkdir(dir, dentry, mode);
1955 error = dir->i_op->mkdir(dir, dentry, mode);
1957 fsnotify_mkdir(dir, dentry);
1961 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1966 tmp = getname(pathname);
1967 error = PTR_ERR(tmp);
1969 struct dentry *dentry;
1970 struct nameidata nd;
1972 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1975 dentry = lookup_create(&nd, 1);
1976 error = PTR_ERR(dentry);
1977 if (!IS_ERR(dentry)) {
1978 if (!IS_POSIXACL(nd.dentry->d_inode))
1979 mode &= ~current->fs->umask;
1980 error = vfs_mkdir(nd.dentry->d_inode, dentry,
1984 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1993 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1995 return sys_mkdirat(AT_FDCWD, pathname, mode);
1999 * We try to drop the dentry early: we should have
2000 * a usage count of 2 if we're the only user of this
2001 * dentry, and if that is true (possibly after pruning
2002 * the dcache), then we drop the dentry now.
2004 * A low-level filesystem can, if it choses, legally
2007 * if (!d_unhashed(dentry))
2010 * if it cannot handle the case of removing a directory
2011 * that is still in use by something else..
2013 void dentry_unhash(struct dentry *dentry)
2016 if (atomic_read(&dentry->d_count))
2017 shrink_dcache_parent(dentry);
2018 spin_lock(&dcache_lock);
2019 spin_lock(&dentry->d_lock);
2020 if (atomic_read(&dentry->d_count) == 2)
2022 spin_unlock(&dentry->d_lock);
2023 spin_unlock(&dcache_lock);
2026 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2027 struct nameidata *nd)
2029 int error = may_delete(dir, dentry, 1, nd);
2034 if (!dir->i_op || !dir->i_op->rmdir)
2039 mutex_lock(&dentry->d_inode->i_mutex);
2040 dentry_unhash(dentry);
2041 if (d_mountpoint(dentry))
2044 error = security_inode_rmdir(dir, dentry);
2046 error = dir->i_op->rmdir(dir, dentry);
2048 dentry->d_inode->i_flags |= S_DEAD;
2051 mutex_unlock(&dentry->d_inode->i_mutex);
2060 static long do_rmdir(int dfd, const char __user *pathname)
2064 struct dentry *dentry;
2065 struct nameidata nd;
2067 name = getname(pathname);
2069 return PTR_ERR(name);
2071 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2075 switch(nd.last_type) {
2086 mutex_lock(&nd.dentry->d_inode->i_mutex);
2087 dentry = lookup_hash(&nd);
2088 error = PTR_ERR(dentry);
2089 if (!IS_ERR(dentry)) {
2090 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2093 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2101 asmlinkage long sys_rmdir(const char __user *pathname)
2103 return do_rmdir(AT_FDCWD, pathname);
2106 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2107 struct nameidata *nd)
2109 int error = may_delete(dir, dentry, 0, nd);
2114 if (!dir->i_op || !dir->i_op->unlink)
2119 mutex_lock(&dentry->d_inode->i_mutex);
2120 if (d_mountpoint(dentry))
2123 error = security_inode_unlink(dir, dentry);
2125 error = dir->i_op->unlink(dir, dentry);
2127 mutex_unlock(&dentry->d_inode->i_mutex);
2129 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2130 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2138 * Make sure that the actual truncation of the file will occur outside its
2139 * directory's i_mutex. Truncate can take a long time if there is a lot of
2140 * writeout happening, and we don't want to prevent access to the directory
2141 * while waiting on the I/O.
2143 static long do_unlinkat(int dfd, const char __user *pathname)
2147 struct dentry *dentry;
2148 struct nameidata nd;
2149 struct inode *inode = NULL;
2151 name = getname(pathname);
2153 return PTR_ERR(name);
2155 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2159 if (nd.last_type != LAST_NORM)
2161 mutex_lock(&nd.dentry->d_inode->i_mutex);
2162 dentry = lookup_hash(&nd);
2163 error = PTR_ERR(dentry);
2164 if (!IS_ERR(dentry)) {
2165 /* Why not before? Because we want correct error value */
2166 if (nd.last.name[nd.last.len])
2168 inode = dentry->d_inode;
2170 atomic_inc(&inode->i_count);
2171 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2175 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2177 iput(inode); /* truncate the inode here */
2185 error = !dentry->d_inode ? -ENOENT :
2186 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2190 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2192 if ((flag & ~AT_REMOVEDIR) != 0)
2195 if (flag & AT_REMOVEDIR)
2196 return do_rmdir(dfd, pathname);
2198 return do_unlinkat(dfd, pathname);
2201 asmlinkage long sys_unlink(const char __user *pathname)
2203 return do_unlinkat(AT_FDCWD, pathname);
2206 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2207 const char *oldname, int mode, struct nameidata *nd)
2209 int error = may_create(dir, dentry, nd);
2214 if (!dir->i_op || !dir->i_op->symlink)
2217 error = security_inode_symlink(dir, dentry, oldname);
2222 error = dir->i_op->symlink(dir, dentry, oldname);
2224 fsnotify_create(dir, dentry);
2228 asmlinkage long sys_symlinkat(const char __user *oldname,
2229 int newdfd, const char __user *newname)
2235 from = getname(oldname);
2237 return PTR_ERR(from);
2238 to = getname(newname);
2239 error = PTR_ERR(to);
2241 struct dentry *dentry;
2242 struct nameidata nd;
2244 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2247 dentry = lookup_create(&nd, 0);
2248 error = PTR_ERR(dentry);
2249 if (!IS_ERR(dentry)) {
2250 error = vfs_symlink(nd.dentry->d_inode, dentry,
2251 from, S_IALLUGO, &nd);
2254 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2263 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2265 return sys_symlinkat(oldname, AT_FDCWD, newname);
2268 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2269 struct dentry *new_dentry, struct nameidata *nd)
2271 struct inode *inode = old_dentry->d_inode;
2277 error = may_create(dir, new_dentry, nd);
2281 if (dir->i_sb != inode->i_sb)
2285 * A link to an append-only or immutable file cannot be created.
2287 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2289 if (!dir->i_op || !dir->i_op->link)
2291 if (S_ISDIR(old_dentry->d_inode->i_mode))
2294 error = security_inode_link(old_dentry, dir, new_dentry);
2298 mutex_lock(&old_dentry->d_inode->i_mutex);
2300 error = dir->i_op->link(old_dentry, dir, new_dentry);
2301 mutex_unlock(&old_dentry->d_inode->i_mutex);
2303 fsnotify_create(dir, new_dentry);
2308 * Hardlinks are often used in delicate situations. We avoid
2309 * security-related surprises by not following symlinks on the
2312 * We don't follow them on the oldname either to be compatible
2313 * with linux 2.0, and to avoid hard-linking to directories
2314 * and other special files. --ADM
2316 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2317 int newdfd, const char __user *newname,
2320 struct dentry *new_dentry;
2321 struct nameidata nd, old_nd;
2328 to = getname(newname);
2332 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
2335 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2339 * We allow hard-links to be created to a bind-mount as long
2340 * as the bind-mount is not read-only. Checking for cross-dev
2341 * links is subsumed by the superblock check in vfs_link().
2344 if (MNT_IS_RDONLY(old_nd.mnt))
2346 new_dentry = lookup_create(&nd, 0);
2347 error = PTR_ERR(new_dentry);
2348 if (!IS_ERR(new_dentry)) {
2349 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2353 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2357 path_release(&old_nd);
2364 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2366 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2370 * The worst of all namespace operations - renaming directory. "Perverted"
2371 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2373 * a) we can get into loop creation. Check is done in is_subdir().
2374 * b) race potential - two innocent renames can create a loop together.
2375 * That's where 4.4 screws up. Current fix: serialization on
2376 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2378 * c) we have to lock _three_ objects - parents and victim (if it exists).
2379 * And that - after we got ->i_mutex on parents (until then we don't know
2380 * whether the target exists). Solution: try to be smart with locking
2381 * order for inodes. We rely on the fact that tree topology may change
2382 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2383 * move will be locked. Thus we can rank directories by the tree
2384 * (ancestors first) and rank all non-directories after them.
2385 * That works since everybody except rename does "lock parent, lookup,
2386 * lock child" and rename is under ->s_vfs_rename_mutex.
2387 * HOWEVER, it relies on the assumption that any object with ->lookup()
2388 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2389 * we'd better make sure that there's no link(2) for them.
2390 * d) some filesystems don't support opened-but-unlinked directories,
2391 * either because of layout or because they are not ready to deal with
2392 * all cases correctly. The latter will be fixed (taking this sort of
2393 * stuff into VFS), but the former is not going away. Solution: the same
2394 * trick as in rmdir().
2395 * e) conversion from fhandle to dentry may come in the wrong moment - when
2396 * we are removing the target. Solution: we will have to grab ->i_mutex
2397 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2398 * ->i_mutex on parents, which works but leads to some truely excessive
2401 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2402 struct inode *new_dir, struct dentry *new_dentry)
2405 struct inode *target;
2408 * If we are going to change the parent - check write permissions,
2409 * we'll need to flip '..'.
2411 if (new_dir != old_dir) {
2412 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2417 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2421 target = new_dentry->d_inode;
2423 mutex_lock(&target->i_mutex);
2424 dentry_unhash(new_dentry);
2426 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2429 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2432 target->i_flags |= S_DEAD;
2433 mutex_unlock(&target->i_mutex);
2434 if (d_unhashed(new_dentry))
2435 d_rehash(new_dentry);
2439 d_move(old_dentry,new_dentry);
2443 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2444 struct inode *new_dir, struct dentry *new_dentry)
2446 struct inode *target;
2449 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2454 target = new_dentry->d_inode;
2456 mutex_lock(&target->i_mutex);
2457 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2460 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2462 /* The following d_move() should become unconditional */
2463 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2464 d_move(old_dentry, new_dentry);
2467 mutex_unlock(&target->i_mutex);
2472 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2473 struct inode *new_dir, struct dentry *new_dentry)
2476 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2477 const char *old_name;
2479 if (old_dentry->d_inode == new_dentry->d_inode)
2482 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2486 if (!new_dentry->d_inode)
2487 error = may_create(new_dir, new_dentry, NULL);
2489 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2493 if (!old_dir->i_op || !old_dir->i_op->rename)
2496 DQUOT_INIT(old_dir);
2497 DQUOT_INIT(new_dir);
2499 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2502 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2504 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2506 const char *new_name = old_dentry->d_name.name;
2507 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2508 new_dentry->d_inode, old_dentry->d_inode);
2510 fsnotify_oldname_free(old_name);
2515 static int do_rename(int olddfd, const char *oldname,
2516 int newdfd, const char *newname)
2519 struct dentry * old_dir, * new_dir;
2520 struct dentry * old_dentry, *new_dentry;
2521 struct dentry * trap;
2522 struct nameidata oldnd, newnd;
2524 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2528 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2533 if (oldnd.mnt != newnd.mnt)
2536 old_dir = oldnd.dentry;
2538 if (oldnd.last_type != LAST_NORM)
2541 new_dir = newnd.dentry;
2542 if (newnd.last_type != LAST_NORM)
2545 trap = lock_rename(new_dir, old_dir);
2547 old_dentry = lookup_hash(&oldnd);
2548 error = PTR_ERR(old_dentry);
2549 if (IS_ERR(old_dentry))
2551 /* source must exist */
2553 if (!old_dentry->d_inode)
2555 /* unless the source is a directory trailing slashes give -ENOTDIR */
2556 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2558 if (oldnd.last.name[oldnd.last.len])
2560 if (newnd.last.name[newnd.last.len])
2563 /* source should not be ancestor of target */
2565 if (old_dentry == trap)
2568 if (MNT_IS_RDONLY(newnd.mnt))
2570 new_dentry = lookup_hash(&newnd);
2571 error = PTR_ERR(new_dentry);
2572 if (IS_ERR(new_dentry))
2574 /* target should not be an ancestor of source */
2576 if (new_dentry == trap)
2579 error = vfs_rename(old_dir->d_inode, old_dentry,
2580 new_dir->d_inode, new_dentry);
2586 unlock_rename(new_dir, old_dir);
2588 path_release(&newnd);
2590 path_release(&oldnd);
2595 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2596 int newdfd, const char __user *newname)
2602 from = getname(oldname);
2604 return PTR_ERR(from);
2605 to = getname(newname);
2606 error = PTR_ERR(to);
2608 error = do_rename(olddfd, from, newdfd, to);
2615 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2617 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2620 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2624 len = PTR_ERR(link);
2629 if (len > (unsigned) buflen)
2631 if (copy_to_user(buffer, link, len))
2638 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2639 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2640 * using) it for any given inode is up to filesystem.
2642 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2644 struct nameidata nd;
2648 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2649 if (!IS_ERR(cookie)) {
2650 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2651 if (dentry->d_inode->i_op->put_link)
2652 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2653 cookie = ERR_PTR(res);
2655 return PTR_ERR(cookie);
2658 int vfs_follow_link(struct nameidata *nd, const char *link)
2660 return __vfs_follow_link(nd, link);
2664 /* get the link contents into pagecache */
2665 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2668 struct address_space *mapping = dentry->d_inode->i_mapping;
2669 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2673 wait_on_page_locked(page);
2674 if (!PageUptodate(page))
2680 page_cache_release(page);
2681 return ERR_PTR(-EIO);
2687 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2689 struct page *page = NULL;
2690 char *s = page_getlink(dentry, &page);
2691 int res = vfs_readlink(dentry,buffer,buflen,s);
2694 page_cache_release(page);
2699 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2701 struct page *page = NULL;
2702 nd_set_link(nd, page_getlink(dentry, &page));
2706 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2708 struct page *page = cookie;
2712 page_cache_release(page);
2716 int __page_symlink(struct inode *inode, const char *symname, int len,
2719 struct address_space *mapping = inode->i_mapping;
2725 page = find_or_create_page(mapping, 0, gfp_mask);
2728 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2729 if (err == AOP_TRUNCATED_PAGE) {
2730 page_cache_release(page);
2735 kaddr = kmap_atomic(page, KM_USER0);
2736 memcpy(kaddr, symname, len-1);
2737 kunmap_atomic(kaddr, KM_USER0);
2738 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2739 if (err == AOP_TRUNCATED_PAGE) {
2740 page_cache_release(page);
2746 * Notice that we are _not_ going to block here - end of page is
2747 * unmapped, so this will only try to map the rest of page, see
2748 * that it is unmapped (typically even will not look into inode -
2749 * ->i_size will be enough for everything) and zero it out.
2750 * OTOH it's obviously correct and should make the page up-to-date.
2752 if (!PageUptodate(page)) {
2753 err = mapping->a_ops->readpage(NULL, page);
2754 if (err != AOP_TRUNCATED_PAGE)
2755 wait_on_page_locked(page);
2759 page_cache_release(page);
2762 mark_inode_dirty(inode);
2766 page_cache_release(page);
2771 int page_symlink(struct inode *inode, const char *symname, int len)
2773 return __page_symlink(inode, symname, len,
2774 mapping_gfp_mask(inode->i_mapping));
2777 struct inode_operations page_symlink_inode_operations = {
2778 .readlink = generic_readlink,
2779 .follow_link = page_follow_link_light,
2780 .put_link = page_put_link,
2783 EXPORT_SYMBOL(__user_walk);
2784 EXPORT_SYMBOL(__user_walk_fd);
2785 EXPORT_SYMBOL(follow_down);
2786 EXPORT_SYMBOL(follow_up);
2787 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2788 EXPORT_SYMBOL(getname);
2789 EXPORT_SYMBOL(lock_rename);
2790 EXPORT_SYMBOL(lookup_one_len);
2791 EXPORT_SYMBOL(page_follow_link_light);
2792 EXPORT_SYMBOL(page_put_link);
2793 EXPORT_SYMBOL(page_readlink);
2794 EXPORT_SYMBOL(__page_symlink);
2795 EXPORT_SYMBOL(page_symlink);
2796 EXPORT_SYMBOL(page_symlink_inode_operations);
2797 EXPORT_SYMBOL(path_lookup);
2798 EXPORT_SYMBOL(path_release);
2799 EXPORT_SYMBOL(path_walk);
2800 EXPORT_SYMBOL(permission);
2801 EXPORT_SYMBOL(vfs_permission);
2802 EXPORT_SYMBOL(file_permission);
2803 EXPORT_SYMBOL(unlock_rename);
2804 EXPORT_SYMBOL(vfs_create);
2805 EXPORT_SYMBOL(vfs_follow_link);
2806 EXPORT_SYMBOL(vfs_link);
2807 EXPORT_SYMBOL(vfs_mkdir);
2808 EXPORT_SYMBOL(vfs_mknod);
2809 EXPORT_SYMBOL(generic_permission);
2810 EXPORT_SYMBOL(vfs_readlink);
2811 EXPORT_SYMBOL(vfs_rename);
2812 EXPORT_SYMBOL(vfs_rmdir);
2813 EXPORT_SYMBOL(vfs_symlink);
2814 EXPORT_SYMBOL(vfs_unlink);
2815 EXPORT_SYMBOL(dentry_unhash);
2816 EXPORT_SYMBOL(generic_readlink);