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(!audit_dummy_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 vx_barrier(struct inode *inode)
233 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN|VX_WATCH)) {
234 vxwprintk(1, "xid=%d did hit the barrier.",
241 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
243 if (vx_barrier(inode))
245 if (inode->i_xid == 0)
247 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
250 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
251 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
256 int permission(struct inode *inode, int mask, struct nameidata *nd)
258 umode_t mode = inode->i_mode;
261 if (mask & MAY_WRITE) {
264 * Nobody gets write access to a read-only fs.
266 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
267 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
271 * Nobody gets write access to an immutable file.
273 if (IS_IMMUTABLE(inode))
279 * MAY_EXEC on regular files requires special handling: We override
280 * filesystem execute permissions if the mode bits aren't set.
282 if ((mask & MAY_EXEC) && S_ISREG(mode) && !(mode & S_IXUGO))
285 /* Ordinary permission routines do not understand MAY_APPEND. */
286 submask = mask & ~MAY_APPEND;
287 if ((retval = xid_permission(inode, mask, nd)))
289 if (inode->i_op && inode->i_op->permission)
290 retval = inode->i_op->permission(inode, submask, nd);
292 retval = generic_permission(inode, submask, NULL);
296 return security_inode_permission(inode, mask, nd);
300 * vfs_permission - check for access rights to a given path
301 * @nd: lookup result that describes the path
302 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
304 * Used to check for read/write/execute permissions on a path.
305 * We use "fsuid" for this, letting us set arbitrary permissions
306 * for filesystem access without changing the "normal" uids which
307 * are used for other things.
309 int vfs_permission(struct nameidata *nd, int mask)
311 return permission(nd->dentry->d_inode, mask, nd);
315 * file_permission - check for additional access rights to a given file
316 * @file: file to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
319 * Used to check for read/write/execute permissions on an already opened
323 * Do not use this function in new code. All access checks should
324 * be done using vfs_permission().
326 int file_permission(struct file *file, int mask)
328 return permission(file->f_dentry->d_inode, mask, NULL);
332 * get_write_access() gets write permission for a file.
333 * put_write_access() releases this write permission.
334 * This is used for regular files.
335 * We cannot support write (and maybe mmap read-write shared) accesses and
336 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
337 * can have the following values:
338 * 0: no writers, no VM_DENYWRITE mappings
339 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
340 * > 0: (i_writecount) users are writing to the file.
342 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
343 * except for the cases where we don't hold i_writecount yet. Then we need to
344 * use {get,deny}_write_access() - these functions check the sign and refuse
345 * to do the change if sign is wrong. Exclusion between them is provided by
346 * the inode->i_lock spinlock.
349 int get_write_access(struct inode * inode)
351 spin_lock(&inode->i_lock);
352 if (atomic_read(&inode->i_writecount) < 0) {
353 spin_unlock(&inode->i_lock);
356 atomic_inc(&inode->i_writecount);
357 spin_unlock(&inode->i_lock);
362 int deny_write_access(struct file * file)
364 struct inode *inode = file->f_dentry->d_inode;
366 spin_lock(&inode->i_lock);
367 if (atomic_read(&inode->i_writecount) > 0) {
368 spin_unlock(&inode->i_lock);
371 atomic_dec(&inode->i_writecount);
372 spin_unlock(&inode->i_lock);
377 void path_release(struct nameidata *nd)
384 * umount() mustn't call path_release()/mntput() as that would clear
387 void path_release_on_umount(struct nameidata *nd)
390 mntput_no_expire(nd->mnt);
394 * release_open_intent - free up open intent resources
395 * @nd: pointer to nameidata
397 void release_open_intent(struct nameidata *nd)
399 if (nd->intent.open.file->f_dentry == NULL)
400 put_filp(nd->intent.open.file);
402 fput(nd->intent.open.file);
405 static inline struct dentry *do_revalidate(struct dentry *dentry, struct nameidata *nd)
407 int status = dentry->d_op->d_revalidate(dentry, nd);
408 if (unlikely(status <= 0)) {
410 * The dentry failed validation.
411 * If d_revalidate returned 0 attempt to invalidate
412 * the dentry otherwise d_revalidate is asking us
413 * to return a fail status.
416 if (!d_invalidate(dentry)) {
422 dentry = ERR_PTR(status);
429 * Internal lookup() using the new generic dcache.
432 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
434 struct dentry * dentry = __d_lookup(parent, name);
436 /* lockess __d_lookup may fail due to concurrent d_move()
437 * in some unrelated directory, so try with d_lookup
440 dentry = d_lookup(parent, name);
442 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
443 dentry = do_revalidate(dentry, nd);
449 * Short-cut version of permission(), for calling by
450 * path_walk(), when dcache lock is held. Combines parts
451 * of permission() and generic_permission(), and tests ONLY for
452 * MAY_EXEC permission.
454 * If appropriate, check DAC only. If not appropriate, or
455 * short-cut DAC fails, then call permission() to do more
456 * complete permission check.
458 static int exec_permission_lite(struct inode *inode,
459 struct nameidata *nd)
461 umode_t mode = inode->i_mode;
463 if (vx_barrier(inode))
465 if (inode->i_op && inode->i_op->permission)
468 if (current->fsuid == inode->i_uid)
470 else if (in_group_p(inode->i_gid))
476 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
479 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
482 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
487 return security_inode_permission(inode, MAY_EXEC, nd);
491 * This is called when everything else fails, and we actually have
492 * to go to the low-level filesystem to find out what we should do..
494 * We get the directory semaphore, and after getting that we also
495 * make sure that nobody added the entry to the dcache in the meantime..
498 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
500 struct dentry * result;
501 struct inode *dir = parent->d_inode;
503 mutex_lock(&dir->i_mutex);
505 * First re-do the cached lookup just in case it was created
506 * while we waited for the directory semaphore..
508 * FIXME! This could use version numbering or similar to
509 * avoid unnecessary cache lookups.
511 * The "dcache_lock" is purely to protect the RCU list walker
512 * from concurrent renames at this point (we mustn't get false
513 * negatives from the RCU list walk here, unlike the optimistic
516 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
518 result = d_lookup(parent, name);
520 struct dentry * dentry = d_alloc(parent, name);
521 result = ERR_PTR(-ENOMEM);
523 result = dir->i_op->lookup(dir, dentry, nd);
529 mutex_unlock(&dir->i_mutex);
534 * Uhhuh! Nasty case: the cache was re-populated while
535 * we waited on the semaphore. Need to revalidate.
537 mutex_unlock(&dir->i_mutex);
538 if (result->d_op && result->d_op->d_revalidate) {
539 result = do_revalidate(result, nd);
541 result = ERR_PTR(-ENOENT);
546 static int __emul_lookup_dentry(const char *, struct nameidata *);
549 static __always_inline int
550 walk_init_root(const char *name, struct nameidata *nd)
552 read_lock(¤t->fs->lock);
553 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
554 nd->mnt = mntget(current->fs->altrootmnt);
555 nd->dentry = dget(current->fs->altroot);
556 read_unlock(¤t->fs->lock);
557 if (__emul_lookup_dentry(name,nd))
559 read_lock(¤t->fs->lock);
561 nd->mnt = mntget(current->fs->rootmnt);
562 nd->dentry = dget(current->fs->root);
563 read_unlock(¤t->fs->lock);
567 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
576 if (!walk_init_root(link, nd))
577 /* weird __emul_prefix() stuff did it */
580 res = link_path_walk(link, nd);
582 if (nd->depth || res || nd->last_type!=LAST_NORM)
585 * If it is an iterative symlinks resolution in open_namei() we
586 * have to copy the last component. And all that crap because of
587 * bloody create() on broken symlinks. Furrfu...
590 if (unlikely(!name)) {
594 strcpy(name, nd->last.name);
595 nd->last.name = name;
599 return PTR_ERR(link);
603 struct vfsmount *mnt;
604 struct dentry *dentry;
607 static inline void dput_path(struct path *path, struct nameidata *nd)
610 if (path->mnt != nd->mnt)
614 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
617 if (nd->mnt != path->mnt)
620 nd->dentry = path->dentry;
623 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
627 struct dentry *dentry = path->dentry;
629 touch_atime(path->mnt, dentry);
630 nd_set_link(nd, NULL);
632 if (path->mnt != nd->mnt) {
633 path_to_nameidata(path, nd);
637 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
638 error = PTR_ERR(cookie);
639 if (!IS_ERR(cookie)) {
640 char *s = nd_get_link(nd);
643 error = __vfs_follow_link(nd, s);
644 if (dentry->d_inode->i_op->put_link)
645 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
654 * This limits recursive symlink follows to 8, while
655 * limiting consecutive symlinks to 40.
657 * Without that kind of total limit, nasty chains of consecutive
658 * symlinks can cause almost arbitrarily long lookups.
660 static inline int do_follow_link(struct path *path, struct nameidata *nd)
663 if (current->link_count >= MAX_NESTED_LINKS)
665 if (current->total_link_count >= 40)
667 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
669 err = security_inode_follow_link(path->dentry, nd);
672 current->link_count++;
673 current->total_link_count++;
675 err = __do_follow_link(path, nd);
676 current->link_count--;
685 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
687 struct vfsmount *parent;
688 struct dentry *mountpoint;
689 spin_lock(&vfsmount_lock);
690 parent=(*mnt)->mnt_parent;
691 if (parent == *mnt) {
692 spin_unlock(&vfsmount_lock);
696 mountpoint=dget((*mnt)->mnt_mountpoint);
697 spin_unlock(&vfsmount_lock);
699 *dentry = mountpoint;
705 /* no need for dcache_lock, as serialization is taken care in
708 static int __follow_mount(struct path *path)
711 while (d_mountpoint(path->dentry)) {
712 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
719 path->dentry = dget(mounted->mnt_root);
725 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
727 while (d_mountpoint(*dentry)) {
728 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
734 *dentry = dget(mounted->mnt_root);
738 /* no need for dcache_lock, as serialization is taken care in
741 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
743 struct vfsmount *mounted;
745 mounted = lookup_mnt(*mnt, *dentry);
750 *dentry = dget(mounted->mnt_root);
756 static __always_inline void follow_dotdot(struct nameidata *nd)
759 struct vfsmount *parent;
760 struct dentry *old = nd->dentry;
762 read_lock(¤t->fs->lock);
763 if (nd->dentry == current->fs->root &&
764 nd->mnt == current->fs->rootmnt) {
765 read_unlock(¤t->fs->lock);
766 /* for sane '/' avoid follow_mount() */
769 read_unlock(¤t->fs->lock);
770 spin_lock(&dcache_lock);
771 if (nd->dentry != nd->mnt->mnt_root) {
772 nd->dentry = dget(nd->dentry->d_parent);
773 spin_unlock(&dcache_lock);
777 spin_unlock(&dcache_lock);
778 spin_lock(&vfsmount_lock);
779 parent = nd->mnt->mnt_parent;
780 if (parent == nd->mnt) {
781 spin_unlock(&vfsmount_lock);
785 nd->dentry = dget(nd->mnt->mnt_mountpoint);
786 spin_unlock(&vfsmount_lock);
791 follow_mount(&nd->mnt, &nd->dentry);
795 * It's more convoluted than I'd like it to be, but... it's still fairly
796 * small and for now I'd prefer to have fast path as straight as possible.
797 * It _is_ time-critical.
799 static int do_lookup(struct nameidata *nd, struct qstr *name,
800 struct path *path, int atomic)
802 struct vfsmount *mnt = nd->mnt;
803 struct dentry *dentry = __d_lookup(nd->dentry, name);
808 if (dentry->d_op && dentry->d_op->d_revalidate)
809 goto need_revalidate;
811 inode = dentry->d_inode;
814 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
816 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
817 struct proc_dir_entry *de = PDE(inode);
819 if (de && !vx_hide_check(0, de->vx_flags))
824 path->dentry = dentry;
825 __follow_mount(path);
828 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
829 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
830 vxd_path(dentry, mnt));
836 return -EWOULDBLOCKIO;
837 dentry = real_lookup(nd->dentry, name, nd);
844 return -EWOULDBLOCKIO;
845 dentry = do_revalidate(dentry, nd);
853 return PTR_ERR(dentry);
858 * This is the basic name resolution function, turning a pathname into
859 * the final dentry. We expect 'base' to be positive and a directory.
861 * Returns 0 and nd will have valid dentry and mnt on success.
862 * Returns error and drops reference to input namei data on failure.
864 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
869 unsigned int lookup_flags = nd->flags;
871 atomic = (lookup_flags & LOOKUP_ATOMIC);
878 inode = nd->dentry->d_inode;
880 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
882 /* At this point we know we have a real path component. */
888 nd->flags |= LOOKUP_CONTINUE;
889 err = exec_permission_lite(inode, nd);
891 err = vfs_permission(nd, MAY_EXEC);
896 c = *(const unsigned char *)name;
898 hash = init_name_hash();
901 hash = partial_name_hash(c, hash);
902 c = *(const unsigned char *)name;
903 } while (c && (c != '/'));
904 this.len = name - (const char *) this.name;
905 this.hash = end_name_hash(hash);
907 /* remove trailing slashes? */
910 while (*++name == '/');
912 goto last_with_slashes;
915 * "." and ".." are special - ".." especially so because it has
916 * to be able to know about the current root directory and
917 * parent relationships.
919 if (this.name[0] == '.') switch (this.len) {
923 if (this.name[1] != '.')
926 inode = nd->dentry->d_inode;
932 * See if the low-level filesystem might want
933 * to use its own hash..
935 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
936 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
940 /* This does the actual lookups.. */
941 err = do_lookup(nd, &this, &next, atomic);
946 inode = next.dentry->d_inode;
953 if (inode->i_op->follow_link) {
954 err = do_follow_link(&next, nd);
958 inode = nd->dentry->d_inode;
965 path_to_nameidata(&next, nd);
967 if (!inode->i_op->lookup)
970 /* here ends the main loop */
973 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
975 /* Clear LOOKUP_CONTINUE iff it was previously unset */
976 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
977 if (lookup_flags & LOOKUP_PARENT)
979 if (this.name[0] == '.') switch (this.len) {
983 if (this.name[1] != '.')
986 inode = nd->dentry->d_inode;
991 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
992 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
996 err = do_lookup(nd, &this, &next, atomic);
999 inode = next.dentry->d_inode;
1000 if ((lookup_flags & LOOKUP_FOLLOW)
1001 && inode && inode->i_op && inode->i_op->follow_link) {
1002 err = do_follow_link(&next, nd);
1005 inode = nd->dentry->d_inode;
1007 path_to_nameidata(&next, nd);
1011 if (lookup_flags & LOOKUP_DIRECTORY) {
1013 if (!inode->i_op || !inode->i_op->lookup)
1019 nd->last_type = LAST_NORM;
1020 if (this.name[0] != '.')
1023 nd->last_type = LAST_DOT;
1024 else if (this.len == 2 && this.name[1] == '.')
1025 nd->last_type = LAST_DOTDOT;
1030 * We bypassed the ordinary revalidation routines.
1031 * We may need to check the cached dentry for staleness.
1033 if (nd->dentry && nd->dentry->d_sb &&
1034 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1036 /* Note: we do not d_invalidate() */
1037 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1043 dput_path(&next, nd);
1052 * Wrapper to retry pathname resolution whenever the underlying
1053 * file system returns an ESTALE.
1055 * Retry the whole path once, forcing real lookup requests
1056 * instead of relying on the dcache.
1058 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1060 struct nameidata save = *nd;
1063 /* make sure the stuff we saved doesn't go away */
1067 result = __link_path_walk(name, nd);
1068 if (result == -ESTALE) {
1072 nd->flags |= LOOKUP_REVAL;
1073 result = __link_path_walk(name, nd);
1082 int fastcall path_walk(const char * name, struct nameidata *nd)
1084 current->total_link_count = 0;
1085 return link_path_walk(name, nd);
1089 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1090 * everything is done. Returns 0 and drops input nd, if lookup failed;
1092 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1094 if (path_walk(name, nd))
1095 return 0; /* something went wrong... */
1097 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1098 struct dentry *old_dentry = nd->dentry;
1099 struct vfsmount *old_mnt = nd->mnt;
1100 struct qstr last = nd->last;
1101 int last_type = nd->last_type;
1103 * NAME was not found in alternate root or it's a directory. Try to find
1104 * it in the normal root:
1106 nd->last_type = LAST_ROOT;
1107 read_lock(¤t->fs->lock);
1108 nd->mnt = mntget(current->fs->rootmnt);
1109 nd->dentry = dget(current->fs->root);
1110 read_unlock(¤t->fs->lock);
1111 if (path_walk(name, nd) == 0) {
1112 if (nd->dentry->d_inode) {
1119 nd->dentry = old_dentry;
1122 nd->last_type = last_type;
1127 void set_fs_altroot(void)
1129 char *emul = __emul_prefix();
1130 struct nameidata nd;
1131 struct vfsmount *mnt = NULL, *oldmnt;
1132 struct dentry *dentry = NULL, *olddentry;
1137 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1143 write_lock(¤t->fs->lock);
1144 oldmnt = current->fs->altrootmnt;
1145 olddentry = current->fs->altroot;
1146 current->fs->altrootmnt = mnt;
1147 current->fs->altroot = dentry;
1148 write_unlock(¤t->fs->lock);
1155 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1156 static int fastcall do_path_lookup(int dfd, const char *name,
1157 unsigned int flags, struct nameidata *nd)
1163 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1168 read_lock(¤t->fs->lock);
1169 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1170 nd->mnt = mntget(current->fs->altrootmnt);
1171 nd->dentry = dget(current->fs->altroot);
1172 read_unlock(¤t->fs->lock);
1173 if (__emul_lookup_dentry(name,nd))
1174 goto out; /* found in altroot */
1175 read_lock(¤t->fs->lock);
1177 nd->mnt = mntget(current->fs->rootmnt);
1178 nd->dentry = dget(current->fs->root);
1179 read_unlock(¤t->fs->lock);
1180 } else if (dfd == AT_FDCWD) {
1181 read_lock(¤t->fs->lock);
1182 nd->mnt = mntget(current->fs->pwdmnt);
1183 nd->dentry = dget(current->fs->pwd);
1184 read_unlock(¤t->fs->lock);
1186 struct dentry *dentry;
1188 file = fget_light(dfd, &fput_needed);
1193 dentry = file->f_dentry;
1196 if (!S_ISDIR(dentry->d_inode->i_mode))
1199 retval = file_permission(file, MAY_EXEC);
1203 nd->mnt = mntget(file->f_vfsmnt);
1204 nd->dentry = dget(dentry);
1206 fput_light(file, fput_needed);
1208 current->total_link_count = 0;
1209 retval = link_path_walk(name, nd);
1211 if (likely(retval == 0)) {
1212 if (unlikely(!audit_dummy_context() && nd && nd->dentry &&
1213 nd->dentry->d_inode))
1214 audit_inode(name, nd->dentry->d_inode);
1220 fput_light(file, fput_needed);
1224 int fastcall path_lookup(const char *name, unsigned int flags,
1225 struct nameidata *nd)
1227 return do_path_lookup(AT_FDCWD, name, flags, nd);
1230 static int __path_lookup_intent_open(int dfd, const char *name,
1231 unsigned int lookup_flags, struct nameidata *nd,
1232 int open_flags, int create_mode)
1234 struct file *filp = get_empty_filp();
1239 nd->intent.open.file = filp;
1240 nd->intent.open.flags = open_flags;
1241 nd->intent.open.create_mode = create_mode;
1242 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1243 if (IS_ERR(nd->intent.open.file)) {
1245 err = PTR_ERR(nd->intent.open.file);
1248 } else if (err != 0)
1249 release_open_intent(nd);
1254 * path_lookup_open - lookup a file path with open intent
1255 * @dfd: the directory to use as base, or AT_FDCWD
1256 * @name: pointer to file name
1257 * @lookup_flags: lookup intent flags
1258 * @nd: pointer to nameidata
1259 * @open_flags: open intent flags
1261 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1262 struct nameidata *nd, int open_flags)
1264 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1269 * path_lookup_create - lookup a file path with open + create intent
1270 * @dfd: the directory to use as base, or AT_FDCWD
1271 * @name: pointer to file name
1272 * @lookup_flags: lookup intent flags
1273 * @nd: pointer to nameidata
1274 * @open_flags: open intent flags
1275 * @create_mode: create intent flags
1277 static int path_lookup_create(int dfd, const char *name,
1278 unsigned int lookup_flags, struct nameidata *nd,
1279 int open_flags, int create_mode)
1281 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1282 nd, open_flags, create_mode);
1285 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1286 struct nameidata *nd, int open_flags)
1288 char *tmp = getname(name);
1289 int err = PTR_ERR(tmp);
1292 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1299 * Restricted form of lookup. Doesn't follow links, single-component only,
1300 * needs parent already locked. Doesn't follow mounts.
1303 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1305 struct dentry * dentry;
1306 struct inode *inode;
1309 inode = base->d_inode;
1310 err = permission(inode, MAY_EXEC, nd);
1311 dentry = ERR_PTR(err);
1316 * See if the low-level filesystem might want
1317 * to use its own hash..
1319 if (base->d_op && base->d_op->d_hash) {
1320 err = base->d_op->d_hash(base, name);
1321 dentry = ERR_PTR(err);
1326 dentry = cached_lookup(base, name, nd);
1328 struct dentry *new = d_alloc(base, name);
1329 dentry = ERR_PTR(-ENOMEM);
1332 dentry = inode->i_op->lookup(inode, new, nd);
1342 static struct dentry *lookup_hash(struct nameidata *nd)
1344 return __lookup_hash(&nd->last, nd->dentry, nd);
1348 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1359 hash = init_name_hash();
1361 c = *(const unsigned char *)name++;
1362 if (c == '/' || c == '\0')
1364 hash = partial_name_hash(c, hash);
1366 this.hash = end_name_hash(hash);
1368 return __lookup_hash(&this, base, NULL);
1370 return ERR_PTR(-EACCES);
1376 * is used by most simple commands to get the inode of a specified name.
1377 * Open, link etc use their own routines, but this is enough for things
1380 * namei exists in two versions: namei/lnamei. The only difference is
1381 * that namei follows links, while lnamei does not.
1384 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1385 struct nameidata *nd)
1387 char *tmp = getname(name);
1388 int err = PTR_ERR(tmp);
1391 err = do_path_lookup(dfd, tmp, flags, nd);
1397 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1399 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1403 * It's inline, so penalty for filesystems that don't use sticky bit is
1406 static inline int check_sticky(struct inode *dir, struct inode *inode)
1408 if (!(dir->i_mode & S_ISVTX))
1410 if (inode->i_uid == current->fsuid)
1412 if (dir->i_uid == current->fsuid)
1414 return !capable(CAP_FOWNER);
1418 * Check whether we can remove a link victim from directory dir, check
1419 * whether the type of victim is right.
1420 * 1. We can't do it if dir is read-only (done in permission())
1421 * 2. We should have write and exec permissions on dir
1422 * 3. We can't remove anything from append-only dir
1423 * 4. We can't do anything with immutable dir (done in permission())
1424 * 5. If the sticky bit on dir is set we should either
1425 * a. be owner of dir, or
1426 * b. be owner of victim, or
1427 * c. have CAP_FOWNER capability
1428 * 6. If the victim is append-only or immutable we can't do antyhing with
1429 * links pointing to it.
1430 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1431 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1432 * 9. We can't remove a root or mountpoint.
1433 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1434 * nfs_async_unlink().
1436 static int may_delete(struct inode *dir, struct dentry *victim,
1437 int isdir, struct nameidata *nd)
1441 if (!victim->d_inode)
1444 BUG_ON(victim->d_parent->d_inode != dir);
1445 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
1447 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1452 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1453 IS_IXORUNLINK(victim->d_inode))
1456 if (!S_ISDIR(victim->d_inode->i_mode))
1458 if (IS_ROOT(victim))
1460 } else if (S_ISDIR(victim->d_inode->i_mode))
1462 if (IS_DEADDIR(dir))
1464 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1469 /* Check whether we can create an object with dentry child in directory
1471 * 1. We can't do it if child already exists (open has special treatment for
1472 * this case, but since we are inlined it's OK)
1473 * 2. We can't do it if dir is read-only (done in permission())
1474 * 3. We should have write and exec permissions on dir
1475 * 4. We can't do it if dir is immutable (done in permission())
1477 static inline int may_create(struct inode *dir, struct dentry *child,
1478 struct nameidata *nd)
1482 if (IS_DEADDIR(dir))
1484 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1488 * O_DIRECTORY translates into forcing a directory lookup.
1490 static inline int lookup_flags(unsigned int f)
1492 unsigned long retval = LOOKUP_FOLLOW;
1495 retval &= ~LOOKUP_FOLLOW;
1497 if (f & O_DIRECTORY)
1498 retval |= LOOKUP_DIRECTORY;
1499 if (f & O_ATOMICLOOKUP)
1500 retval |= LOOKUP_ATOMIC;
1506 * p1 and p2 should be directories on the same fs.
1508 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1513 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1517 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1519 for (p = p1; p->d_parent != p; p = p->d_parent) {
1520 if (p->d_parent == p2) {
1521 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1522 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1527 for (p = p2; p->d_parent != p; p = p->d_parent) {
1528 if (p->d_parent == p1) {
1529 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1530 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1535 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1536 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1540 void unlock_rename(struct dentry *p1, struct dentry *p2)
1542 mutex_unlock(&p1->d_inode->i_mutex);
1544 mutex_unlock(&p2->d_inode->i_mutex);
1545 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1549 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1550 struct nameidata *nd)
1552 int error = may_create(dir, dentry, nd);
1557 if (!dir->i_op || !dir->i_op->create)
1558 return -EACCES; /* shouldn't it be ENOSYS? */
1561 error = security_inode_create(dir, dentry, mode);
1565 error = dir->i_op->create(dir, dentry, mode, nd);
1567 fsnotify_create(dir, dentry);
1571 int may_open(struct nameidata *nd, int acc_mode, int flag)
1573 struct dentry *dentry = nd->dentry;
1574 struct inode *inode = dentry->d_inode;
1580 if (S_ISLNK(inode->i_mode))
1583 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1586 error = vfs_permission(nd, acc_mode);
1591 * FIFO's, sockets and device files are special: they don't
1592 * actually live on the filesystem itself, and as such you
1593 * can write to them even if the filesystem is read-only.
1595 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1597 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1598 if (nd->mnt->mnt_flags & MNT_NODEV)
1602 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1603 && (flag & FMODE_WRITE))
1606 * An append-only file must be opened in append mode for writing.
1608 if (IS_APPEND(inode)) {
1609 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1615 /* O_NOATIME can only be set by the owner or superuser */
1616 if (flag & O_NOATIME)
1617 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1621 * Ensure there are no outstanding leases on the file.
1623 error = break_lease(inode, flag);
1627 if (flag & O_TRUNC) {
1628 error = get_write_access(inode);
1633 * Refuse to truncate files with mandatory locks held on them.
1635 error = locks_verify_locked(inode);
1639 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1641 put_write_access(inode);
1645 if (flag & FMODE_WRITE)
1654 * namei for open - this is in fact almost the whole open-routine.
1656 * Note that the low bits of "flag" aren't the same as in the open
1657 * system call - they are 00 - no permissions needed
1658 * 01 - read permission needed
1659 * 10 - write permission needed
1660 * 11 - read/write permissions needed
1661 * which is a lot more logical, and also allows the "no perm" needed
1662 * for symlinks (where the permissions are checked later).
1665 int open_namei(int dfd, const char *pathname, int flag,
1666 int mode, struct nameidata *nd)
1668 int acc_mode, error;
1673 acc_mode = ACC_MODE(flag);
1675 /* O_TRUNC implies we need access checks for write permissions */
1677 acc_mode |= MAY_WRITE;
1679 /* Allow the LSM permission hook to distinguish append
1680 access from general write access. */
1681 if (flag & O_APPEND)
1682 acc_mode |= MAY_APPEND;
1685 * The simplest case - just a plain lookup.
1687 if (!(flag & O_CREAT)) {
1688 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1696 * Create - we need to know the parent.
1698 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1703 * We have the parent and last component. First of all, check
1704 * that we are not asked to creat(2) an obvious directory - that
1708 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1712 nd->flags &= ~LOOKUP_PARENT;
1713 mutex_lock(&dir->d_inode->i_mutex);
1714 path.dentry = lookup_hash(nd);
1718 error = PTR_ERR(path.dentry);
1719 if (IS_ERR(path.dentry)) {
1720 mutex_unlock(&dir->d_inode->i_mutex);
1724 if (IS_ERR(nd->intent.open.file)) {
1725 mutex_unlock(&dir->d_inode->i_mutex);
1726 error = PTR_ERR(nd->intent.open.file);
1730 /* Negative dentry, just create the file */
1731 if (!path.dentry->d_inode) {
1732 if (!IS_POSIXACL(dir->d_inode))
1733 mode &= ~current->fs->umask;
1734 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1735 mutex_unlock(&dir->d_inode->i_mutex);
1737 nd->dentry = path.dentry;
1740 /* Don't check for write permission, don't truncate */
1747 * It already exists.
1749 mutex_unlock(&dir->d_inode->i_mutex);
1750 audit_inode_update(path.dentry->d_inode);
1756 if (__follow_mount(&path)) {
1758 if (flag & O_NOFOLLOW)
1763 if (!path.dentry->d_inode)
1765 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1768 path_to_nameidata(&path, nd);
1770 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1773 error = may_open(nd, acc_mode, flag);
1779 dput_path(&path, nd);
1781 if (!IS_ERR(nd->intent.open.file))
1782 release_open_intent(nd);
1788 if (flag & O_NOFOLLOW)
1791 * This is subtle. Instead of calling do_follow_link() we do the
1792 * thing by hands. The reason is that this way we have zero link_count
1793 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1794 * After that we have the parent and last component, i.e.
1795 * we are in the same situation as after the first path_walk().
1796 * Well, almost - if the last component is normal we get its copy
1797 * stored in nd->last.name and we will have to putname() it when we
1798 * are done. Procfs-like symlinks just set LAST_BIND.
1800 nd->flags |= LOOKUP_PARENT;
1801 error = security_inode_follow_link(path.dentry, nd);
1804 error = __do_follow_link(&path, nd);
1806 /* Does someone understand code flow here? Or it is only
1807 * me so stupid? Anathema to whoever designed this non-sense
1808 * with "intent.open".
1810 release_open_intent(nd);
1813 nd->flags &= ~LOOKUP_PARENT;
1814 if (nd->last_type == LAST_BIND)
1817 if (nd->last_type != LAST_NORM)
1819 if (nd->last.name[nd->last.len]) {
1820 __putname(nd->last.name);
1825 __putname(nd->last.name);
1829 mutex_lock(&dir->d_inode->i_mutex);
1830 path.dentry = lookup_hash(nd);
1832 __putname(nd->last.name);
1837 * lookup_create - lookup a dentry, creating it if it doesn't exist
1838 * @nd: nameidata info
1839 * @is_dir: directory flag
1841 * Simple function to lookup and return a dentry and create it
1842 * if it doesn't exist. Is SMP-safe.
1844 * Returns with nd->dentry->d_inode->i_mutex locked.
1846 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1848 struct dentry *dentry = ERR_PTR(-EEXIST);
1850 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1852 * Yucky last component or no last component at all?
1853 * (foo/., foo/.., /////)
1855 if (nd->last_type != LAST_NORM)
1857 nd->flags &= ~LOOKUP_PARENT;
1858 nd->flags |= LOOKUP_CREATE;
1859 nd->intent.open.flags = O_EXCL;
1862 * Do the final lookup.
1864 dentry = lookup_hash(nd);
1869 * Special case - lookup gave negative, but... we had foo/bar/
1870 * From the vfs_mknod() POV we just have a negative dentry -
1871 * all is fine. Let's be bastards - you had / on the end, you've
1872 * been asking for (non-existent) directory. -ENOENT for you.
1874 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1879 dentry = ERR_PTR(-ENOENT);
1883 EXPORT_SYMBOL_GPL(lookup_create);
1885 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1886 int mode, dev_t dev, struct nameidata *nd)
1888 int error = may_create(dir, dentry, nd);
1893 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1896 if (!dir->i_op || !dir->i_op->mknod)
1899 error = security_inode_mknod(dir, dentry, mode, dev);
1904 error = dir->i_op->mknod(dir, dentry, mode, dev);
1906 fsnotify_create(dir, dentry);
1910 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1915 struct dentry * dentry;
1916 struct nameidata nd;
1920 tmp = getname(filename);
1922 return PTR_ERR(tmp);
1924 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1927 dentry = lookup_create(&nd, 0);
1928 error = PTR_ERR(dentry);
1930 if (!IS_POSIXACL(nd.dentry->d_inode))
1931 mode &= ~current->fs->umask;
1932 if (!IS_ERR(dentry)) {
1933 switch (mode & S_IFMT) {
1934 case 0: case S_IFREG:
1935 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1937 case S_IFCHR: case S_IFBLK:
1938 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1939 new_decode_dev(dev), &nd);
1941 case S_IFIFO: case S_IFSOCK:
1942 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1953 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1961 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1963 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1966 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1967 int mode, struct nameidata *nd)
1969 int error = may_create(dir, dentry, nd);
1974 if (!dir->i_op || !dir->i_op->mkdir)
1977 mode &= (S_IRWXUGO|S_ISVTX);
1978 error = security_inode_mkdir(dir, dentry, mode);
1983 error = dir->i_op->mkdir(dir, dentry, mode);
1985 fsnotify_mkdir(dir, dentry);
1989 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1994 tmp = getname(pathname);
1995 error = PTR_ERR(tmp);
1997 struct dentry *dentry;
1998 struct nameidata nd;
2000 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2003 dentry = lookup_create(&nd, 1);
2004 error = PTR_ERR(dentry);
2005 if (!IS_ERR(dentry)) {
2006 if (!IS_POSIXACL(nd.dentry->d_inode))
2007 mode &= ~current->fs->umask;
2008 error = vfs_mkdir(nd.dentry->d_inode, dentry,
2012 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2021 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2023 return sys_mkdirat(AT_FDCWD, pathname, mode);
2027 * We try to drop the dentry early: we should have
2028 * a usage count of 2 if we're the only user of this
2029 * dentry, and if that is true (possibly after pruning
2030 * the dcache), then we drop the dentry now.
2032 * A low-level filesystem can, if it choses, legally
2035 * if (!d_unhashed(dentry))
2038 * if it cannot handle the case of removing a directory
2039 * that is still in use by something else..
2041 void dentry_unhash(struct dentry *dentry)
2044 if (atomic_read(&dentry->d_count))
2045 shrink_dcache_parent(dentry);
2046 spin_lock(&dcache_lock);
2047 spin_lock(&dentry->d_lock);
2048 if (atomic_read(&dentry->d_count) == 2)
2050 spin_unlock(&dentry->d_lock);
2051 spin_unlock(&dcache_lock);
2054 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2055 struct nameidata *nd)
2057 int error = may_delete(dir, dentry, 1, nd);
2062 if (!dir->i_op || !dir->i_op->rmdir)
2067 mutex_lock(&dentry->d_inode->i_mutex);
2068 dentry_unhash(dentry);
2069 if (d_mountpoint(dentry))
2072 error = security_inode_rmdir(dir, dentry);
2074 error = dir->i_op->rmdir(dir, dentry);
2076 dentry->d_inode->i_flags |= S_DEAD;
2079 mutex_unlock(&dentry->d_inode->i_mutex);
2088 static long do_rmdir(int dfd, const char __user *pathname)
2092 struct dentry *dentry;
2093 struct nameidata nd;
2095 name = getname(pathname);
2097 return PTR_ERR(name);
2099 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2103 switch(nd.last_type) {
2114 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2115 dentry = lookup_hash(&nd);
2116 error = PTR_ERR(dentry);
2117 if (!IS_ERR(dentry)) {
2118 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2121 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2129 asmlinkage long sys_rmdir(const char __user *pathname)
2131 return do_rmdir(AT_FDCWD, pathname);
2134 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2135 struct nameidata *nd)
2137 int error = may_delete(dir, dentry, 0, nd);
2142 if (!dir->i_op || !dir->i_op->unlink)
2147 mutex_lock(&dentry->d_inode->i_mutex);
2148 if (d_mountpoint(dentry))
2151 error = security_inode_unlink(dir, dentry);
2153 error = dir->i_op->unlink(dir, dentry);
2155 mutex_unlock(&dentry->d_inode->i_mutex);
2157 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2158 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2166 * Make sure that the actual truncation of the file will occur outside its
2167 * directory's i_mutex. Truncate can take a long time if there is a lot of
2168 * writeout happening, and we don't want to prevent access to the directory
2169 * while waiting on the I/O.
2171 static long do_unlinkat(int dfd, const char __user *pathname)
2175 struct dentry *dentry;
2176 struct nameidata nd;
2177 struct inode *inode = NULL;
2179 name = getname(pathname);
2181 return PTR_ERR(name);
2183 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2187 if (nd.last_type != LAST_NORM)
2189 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2190 dentry = lookup_hash(&nd);
2191 error = PTR_ERR(dentry);
2192 if (!IS_ERR(dentry)) {
2193 /* Why not before? Because we want correct error value */
2194 if (nd.last.name[nd.last.len])
2196 inode = dentry->d_inode;
2198 atomic_inc(&inode->i_count);
2199 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2203 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2205 iput(inode); /* truncate the inode here */
2213 error = !dentry->d_inode ? -ENOENT :
2214 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2218 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2220 if ((flag & ~AT_REMOVEDIR) != 0)
2223 if (flag & AT_REMOVEDIR)
2224 return do_rmdir(dfd, pathname);
2226 return do_unlinkat(dfd, pathname);
2229 asmlinkage long sys_unlink(const char __user *pathname)
2231 return do_unlinkat(AT_FDCWD, pathname);
2234 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2235 const char *oldname, int mode, struct nameidata *nd)
2237 int error = may_create(dir, dentry, nd);
2242 if (!dir->i_op || !dir->i_op->symlink)
2245 error = security_inode_symlink(dir, dentry, oldname);
2250 error = dir->i_op->symlink(dir, dentry, oldname);
2252 fsnotify_create(dir, dentry);
2256 asmlinkage long sys_symlinkat(const char __user *oldname,
2257 int newdfd, const char __user *newname)
2263 from = getname(oldname);
2265 return PTR_ERR(from);
2266 to = getname(newname);
2267 error = PTR_ERR(to);
2269 struct dentry *dentry;
2270 struct nameidata nd;
2272 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2275 dentry = lookup_create(&nd, 0);
2276 error = PTR_ERR(dentry);
2277 if (!IS_ERR(dentry)) {
2278 error = vfs_symlink(nd.dentry->d_inode, dentry,
2279 from, S_IALLUGO, &nd);
2282 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2291 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2293 return sys_symlinkat(oldname, AT_FDCWD, newname);
2296 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2297 struct dentry *new_dentry, struct nameidata *nd)
2299 struct inode *inode = old_dentry->d_inode;
2305 error = may_create(dir, new_dentry, nd);
2309 if (dir->i_sb != inode->i_sb)
2313 * A link to an append-only or immutable file cannot be created.
2315 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2317 if (!dir->i_op || !dir->i_op->link)
2319 if (S_ISDIR(old_dentry->d_inode->i_mode))
2322 error = security_inode_link(old_dentry, dir, new_dentry);
2326 mutex_lock(&old_dentry->d_inode->i_mutex);
2328 error = dir->i_op->link(old_dentry, dir, new_dentry);
2329 mutex_unlock(&old_dentry->d_inode->i_mutex);
2331 fsnotify_create(dir, new_dentry);
2336 * Hardlinks are often used in delicate situations. We avoid
2337 * security-related surprises by not following symlinks on the
2340 * We don't follow them on the oldname either to be compatible
2341 * with linux 2.0, and to avoid hard-linking to directories
2342 * and other special files. --ADM
2344 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2345 int newdfd, const char __user *newname,
2348 struct dentry *new_dentry;
2349 struct nameidata nd, old_nd;
2353 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2356 to = getname(newname);
2360 error = __user_walk_fd(olddfd, oldname,
2361 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2365 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2369 if (old_nd.mnt != nd.mnt)
2371 new_dentry = lookup_create(&nd, 0);
2372 error = PTR_ERR(new_dentry);
2373 if (!IS_ERR(new_dentry)) {
2374 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2378 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2382 path_release(&old_nd);
2389 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2391 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2395 * The worst of all namespace operations - renaming directory. "Perverted"
2396 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2398 * a) we can get into loop creation. Check is done in is_subdir().
2399 * b) race potential - two innocent renames can create a loop together.
2400 * That's where 4.4 screws up. Current fix: serialization on
2401 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2403 * c) we have to lock _three_ objects - parents and victim (if it exists).
2404 * And that - after we got ->i_mutex on parents (until then we don't know
2405 * whether the target exists). Solution: try to be smart with locking
2406 * order for inodes. We rely on the fact that tree topology may change
2407 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2408 * move will be locked. Thus we can rank directories by the tree
2409 * (ancestors first) and rank all non-directories after them.
2410 * That works since everybody except rename does "lock parent, lookup,
2411 * lock child" and rename is under ->s_vfs_rename_mutex.
2412 * HOWEVER, it relies on the assumption that any object with ->lookup()
2413 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2414 * we'd better make sure that there's no link(2) for them.
2415 * d) some filesystems don't support opened-but-unlinked directories,
2416 * either because of layout or because they are not ready to deal with
2417 * all cases correctly. The latter will be fixed (taking this sort of
2418 * stuff into VFS), but the former is not going away. Solution: the same
2419 * trick as in rmdir().
2420 * e) conversion from fhandle to dentry may come in the wrong moment - when
2421 * we are removing the target. Solution: we will have to grab ->i_mutex
2422 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2423 * ->i_mutex on parents, which works but leads to some truely excessive
2426 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2427 struct inode *new_dir, struct dentry *new_dentry)
2430 struct inode *target;
2433 * If we are going to change the parent - check write permissions,
2434 * we'll need to flip '..'.
2436 if (new_dir != old_dir) {
2437 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2442 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2446 target = new_dentry->d_inode;
2448 mutex_lock(&target->i_mutex);
2449 dentry_unhash(new_dentry);
2451 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2454 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2457 target->i_flags |= S_DEAD;
2458 mutex_unlock(&target->i_mutex);
2459 if (d_unhashed(new_dentry))
2460 d_rehash(new_dentry);
2464 d_move(old_dentry,new_dentry);
2468 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2469 struct inode *new_dir, struct dentry *new_dentry)
2471 struct inode *target;
2474 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2479 target = new_dentry->d_inode;
2481 mutex_lock(&target->i_mutex);
2482 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2485 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2487 /* The following d_move() should become unconditional */
2488 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2489 d_move(old_dentry, new_dentry);
2492 mutex_unlock(&target->i_mutex);
2497 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2498 struct inode *new_dir, struct dentry *new_dentry)
2501 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2502 const char *old_name;
2504 if (old_dentry->d_inode == new_dentry->d_inode)
2507 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2511 if (!new_dentry->d_inode)
2512 error = may_create(new_dir, new_dentry, NULL);
2514 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2518 if (!old_dir->i_op || !old_dir->i_op->rename)
2521 DQUOT_INIT(old_dir);
2522 DQUOT_INIT(new_dir);
2524 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2527 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2529 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2531 const char *new_name = old_dentry->d_name.name;
2532 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2533 new_dentry->d_inode, old_dentry->d_inode);
2535 fsnotify_oldname_free(old_name);
2540 static int do_rename(int olddfd, const char *oldname,
2541 int newdfd, const char *newname)
2544 struct dentry * old_dir, * new_dir;
2545 struct dentry * old_dentry, *new_dentry;
2546 struct dentry * trap;
2547 struct nameidata oldnd, newnd;
2549 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2553 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2558 if (oldnd.mnt != newnd.mnt)
2561 old_dir = oldnd.dentry;
2563 if (oldnd.last_type != LAST_NORM)
2566 new_dir = newnd.dentry;
2567 if (newnd.last_type != LAST_NORM)
2570 trap = lock_rename(new_dir, old_dir);
2572 old_dentry = lookup_hash(&oldnd);
2573 error = PTR_ERR(old_dentry);
2574 if (IS_ERR(old_dentry))
2576 /* source must exist */
2578 if (!old_dentry->d_inode)
2580 /* unless the source is a directory trailing slashes give -ENOTDIR */
2581 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2583 if (oldnd.last.name[oldnd.last.len])
2585 if (newnd.last.name[newnd.last.len])
2588 /* source should not be ancestor of target */
2590 if (old_dentry == trap)
2593 if (MNT_IS_RDONLY(newnd.mnt))
2595 new_dentry = lookup_hash(&newnd);
2596 error = PTR_ERR(new_dentry);
2597 if (IS_ERR(new_dentry))
2599 /* target should not be an ancestor of source */
2601 if (new_dentry == trap)
2604 error = vfs_rename(old_dir->d_inode, old_dentry,
2605 new_dir->d_inode, new_dentry);
2611 unlock_rename(new_dir, old_dir);
2613 path_release(&newnd);
2615 path_release(&oldnd);
2620 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2621 int newdfd, const char __user *newname)
2627 from = getname(oldname);
2629 return PTR_ERR(from);
2630 to = getname(newname);
2631 error = PTR_ERR(to);
2633 error = do_rename(olddfd, from, newdfd, to);
2640 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2642 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2645 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2649 len = PTR_ERR(link);
2654 if (len > (unsigned) buflen)
2656 if (copy_to_user(buffer, link, len))
2663 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2664 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2665 * using) it for any given inode is up to filesystem.
2667 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2669 struct nameidata nd;
2673 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2674 if (!IS_ERR(cookie)) {
2675 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2676 if (dentry->d_inode->i_op->put_link)
2677 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2678 cookie = ERR_PTR(res);
2680 return PTR_ERR(cookie);
2683 int vfs_follow_link(struct nameidata *nd, const char *link)
2685 return __vfs_follow_link(nd, link);
2688 /* get the link contents into pagecache */
2689 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2692 struct address_space *mapping = dentry->d_inode->i_mapping;
2693 page = read_mapping_page(mapping, 0, NULL);
2696 wait_on_page_locked(page);
2697 if (!PageUptodate(page))
2703 page_cache_release(page);
2704 return ERR_PTR(-EIO);
2710 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2712 struct page *page = NULL;
2713 char *s = page_getlink(dentry, &page);
2714 int res = vfs_readlink(dentry,buffer,buflen,s);
2717 page_cache_release(page);
2722 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2724 struct page *page = NULL;
2725 nd_set_link(nd, page_getlink(dentry, &page));
2729 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2731 struct page *page = cookie;
2735 page_cache_release(page);
2739 int __page_symlink(struct inode *inode, const char *symname, int len,
2742 struct address_space *mapping = inode->i_mapping;
2748 page = find_or_create_page(mapping, 0, gfp_mask);
2751 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2752 if (err == AOP_TRUNCATED_PAGE) {
2753 page_cache_release(page);
2758 kaddr = kmap_atomic(page, KM_USER0);
2759 memcpy(kaddr, symname, len-1);
2760 kunmap_atomic(kaddr, KM_USER0);
2761 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2762 if (err == AOP_TRUNCATED_PAGE) {
2763 page_cache_release(page);
2769 * Notice that we are _not_ going to block here - end of page is
2770 * unmapped, so this will only try to map the rest of page, see
2771 * that it is unmapped (typically even will not look into inode -
2772 * ->i_size will be enough for everything) and zero it out.
2773 * OTOH it's obviously correct and should make the page up-to-date.
2775 if (!PageUptodate(page)) {
2776 err = mapping->a_ops->readpage(NULL, page);
2777 if (err != AOP_TRUNCATED_PAGE)
2778 wait_on_page_locked(page);
2782 page_cache_release(page);
2785 mark_inode_dirty(inode);
2789 page_cache_release(page);
2794 int page_symlink(struct inode *inode, const char *symname, int len)
2796 return __page_symlink(inode, symname, len,
2797 mapping_gfp_mask(inode->i_mapping));
2800 struct inode_operations page_symlink_inode_operations = {
2801 .readlink = generic_readlink,
2802 .follow_link = page_follow_link_light,
2803 .put_link = page_put_link,
2806 EXPORT_SYMBOL(__user_walk);
2807 EXPORT_SYMBOL(__user_walk_fd);
2808 EXPORT_SYMBOL(follow_down);
2809 EXPORT_SYMBOL(follow_up);
2810 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2811 EXPORT_SYMBOL(getname);
2812 EXPORT_SYMBOL(lock_rename);
2813 EXPORT_SYMBOL(lookup_one_len);
2814 EXPORT_SYMBOL(page_follow_link_light);
2815 EXPORT_SYMBOL(page_put_link);
2816 EXPORT_SYMBOL(page_readlink);
2817 EXPORT_SYMBOL(__page_symlink);
2818 EXPORT_SYMBOL(page_symlink);
2819 EXPORT_SYMBOL(page_symlink_inode_operations);
2820 EXPORT_SYMBOL(path_lookup);
2821 EXPORT_SYMBOL(path_release);
2822 EXPORT_SYMBOL(path_walk);
2823 EXPORT_SYMBOL(permission);
2824 EXPORT_SYMBOL(vfs_permission);
2825 EXPORT_SYMBOL(file_permission);
2826 EXPORT_SYMBOL(unlock_rename);
2827 EXPORT_SYMBOL(vfs_create);
2828 EXPORT_SYMBOL(vfs_follow_link);
2829 EXPORT_SYMBOL(vfs_link);
2830 EXPORT_SYMBOL(vfs_mkdir);
2831 EXPORT_SYMBOL(vfs_mknod);
2832 EXPORT_SYMBOL(generic_permission);
2833 EXPORT_SYMBOL(vfs_readlink);
2834 EXPORT_SYMBOL(vfs_rename);
2835 EXPORT_SYMBOL(vfs_rmdir);
2836 EXPORT_SYMBOL(vfs_symlink);
2837 EXPORT_SYMBOL(vfs_unlink);
2838 EXPORT_SYMBOL(dentry_unhash);
2839 EXPORT_SYMBOL(generic_readlink);