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/dnotify.h>
25 #include <linux/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/vs_base.h>
32 #include <asm/namei.h>
33 #include <asm/uaccess.h>
35 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
37 /* [Feb-1997 T. Schoebel-Theuer]
38 * Fundamental changes in the pathname lookup mechanisms (namei)
39 * were necessary because of omirr. The reason is that omirr needs
40 * to know the _real_ pathname, not the user-supplied one, in case
41 * of symlinks (and also when transname replacements occur).
43 * The new code replaces the old recursive symlink resolution with
44 * an iterative one (in case of non-nested symlink chains). It does
45 * this with calls to <fs>_follow_link().
46 * As a side effect, dir_namei(), _namei() and follow_link() are now
47 * replaced with a single function lookup_dentry() that can handle all
48 * the special cases of the former code.
50 * With the new dcache, the pathname is stored at each inode, at least as
51 * long as the refcount of the inode is positive. As a side effect, the
52 * size of the dcache depends on the inode cache and thus is dynamic.
54 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
55 * resolution to correspond with current state of the code.
57 * Note that the symlink resolution is not *completely* iterative.
58 * There is still a significant amount of tail- and mid- recursion in
59 * the algorithm. Also, note that <fs>_readlink() is not used in
60 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
61 * may return different results than <fs>_follow_link(). Many virtual
62 * filesystems (including /proc) exhibit this behavior.
65 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
66 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
67 * and the name already exists in form of a symlink, try to create the new
68 * name indicated by the symlink. The old code always complained that the
69 * name already exists, due to not following the symlink even if its target
70 * is nonexistent. The new semantics affects also mknod() and link() when
71 * the name is a symlink pointing to a non-existant name.
73 * I don't know which semantics is the right one, since I have no access
74 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
75 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
76 * "old" one. Personally, I think the new semantics is much more logical.
77 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
78 * file does succeed in both HP-UX and SunOs, but not in Solaris
79 * and in the old Linux semantics.
82 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
83 * semantics. See the comments in "open_namei" and "do_link" below.
85 * [10-Sep-98 Alan Modra] Another symlink change.
88 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
89 * inside the path - always follow.
90 * in the last component in creation/removal/renaming - never follow.
91 * if LOOKUP_FOLLOW passed - follow.
92 * if the pathname has trailing slashes - follow.
93 * otherwise - don't follow.
94 * (applied in that order).
96 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
97 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
98 * During the 2.4 we need to fix the userland stuff depending on it -
99 * hopefully we will be able to get rid of that wart in 2.5. So far only
100 * XEmacs seems to be relying on it...
103 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
104 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
105 * any extra contention...
108 /* In order to reduce some races, while at the same time doing additional
109 * checking and hopefully speeding things up, we copy filenames to the
110 * kernel data space before using them..
112 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
113 * PATH_MAX includes the nul terminator --RR.
115 static inline int do_getname(const char __user *filename, char *page)
118 unsigned long len = PATH_MAX;
120 if ((unsigned long) filename >= TASK_SIZE) {
121 if (!segment_eq(get_fs(), KERNEL_DS))
123 } else if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
124 len = TASK_SIZE - (unsigned long) filename;
126 retval = strncpy_from_user((char *)page, filename, len);
130 return -ENAMETOOLONG;
136 char * getname(const char __user * filename)
140 result = ERR_PTR(-ENOMEM);
143 int retval = do_getname(filename, tmp);
148 result = ERR_PTR(retval);
151 if (unlikely(current->audit_context) && !IS_ERR(result) && result)
152 audit_getname(result);
159 * is used to check for read/write/execute permissions on a file.
160 * We use "fsuid" for this, letting us set arbitrary permissions
161 * for filesystem access without changing the "normal" uids which
162 * are used for other things..
164 int vfs_permission(struct inode * inode, int mask)
166 umode_t mode = inode->i_mode;
168 /* Prevent vservers from escaping chroot() barriers */
169 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN))
172 if (mask & MAY_WRITE) {
174 * Nobody gets write access to a read-only fs.
176 if (IS_RDONLY(inode) &&
177 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
181 * Nobody gets write access to an immutable file.
183 if (IS_IMMUTABLE(inode))
187 if (current->fsuid == inode->i_uid)
189 else if (in_group_p(inode->i_gid))
193 * If the DACs are ok we don't need any capability check.
195 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
199 * Read/write DACs are always overridable.
200 * Executable DACs are overridable if at least one exec bit is set.
202 if (!(mask & MAY_EXEC) ||
203 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
204 if (capable(CAP_DAC_OVERRIDE))
208 * Searching includes executable on directories, else just read.
210 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
211 if (capable(CAP_DAC_READ_SEARCH))
217 int permission(struct inode * inode,int mask, struct nameidata *nd)
221 umode_t mode = inode->i_mode;
223 /* Ordinary permission routines do not understand MAY_APPEND. */
224 submask = mask & ~MAY_APPEND;
226 if (nd && (mask & MAY_WRITE) && MNT_IS_RDONLY(nd->mnt) &&
227 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
230 if (inode->i_op && inode->i_op->permission)
231 retval = inode->i_op->permission(inode, submask, nd);
233 retval = vfs_permission(inode, submask);
237 return security_inode_permission(inode, mask, nd);
241 * get_write_access() gets write permission for a file.
242 * put_write_access() releases this write permission.
243 * This is used for regular files.
244 * We cannot support write (and maybe mmap read-write shared) accesses and
245 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
246 * can have the following values:
247 * 0: no writers, no VM_DENYWRITE mappings
248 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
249 * > 0: (i_writecount) users are writing to the file.
251 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
252 * except for the cases where we don't hold i_writecount yet. Then we need to
253 * use {get,deny}_write_access() - these functions check the sign and refuse
254 * to do the change if sign is wrong. Exclusion between them is provided by
255 * the inode->i_lock spinlock.
258 int get_write_access(struct inode * inode)
260 spin_lock(&inode->i_lock);
261 if (atomic_read(&inode->i_writecount) < 0) {
262 spin_unlock(&inode->i_lock);
265 atomic_inc(&inode->i_writecount);
266 spin_unlock(&inode->i_lock);
271 int deny_write_access(struct file * file)
273 struct inode *inode = file->f_dentry->d_inode;
275 spin_lock(&inode->i_lock);
276 if (atomic_read(&inode->i_writecount) > 0) {
277 spin_unlock(&inode->i_lock);
280 atomic_dec(&inode->i_writecount);
281 spin_unlock(&inode->i_lock);
286 void path_release(struct nameidata *nd)
293 * umount() mustn't call path_release()/mntput() as that would clear
296 void path_release_on_umount(struct nameidata *nd)
303 * Internal lookup() using the new generic dcache.
306 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
308 struct dentry * dentry = __d_lookup(parent, name);
310 /* lockess __d_lookup may fail due to concurrent d_move()
311 * in some unrelated directory, so try with d_lookup
314 dentry = d_lookup(parent, name);
316 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
317 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
326 * Short-cut version of permission(), for calling by
327 * path_walk(), when dcache lock is held. Combines parts
328 * of permission() and vfs_permission(), and tests ONLY for
329 * MAY_EXEC permission.
331 * If appropriate, check DAC only. If not appropriate, or
332 * short-cut DAC fails, then call permission() to do more
333 * complete permission check.
335 static inline int exec_permission_lite(struct inode *inode,
336 struct nameidata *nd)
338 umode_t mode = inode->i_mode;
340 if (inode->i_op && inode->i_op->permission)
343 if (current->fsuid == inode->i_uid)
345 else if (in_group_p(inode->i_gid))
351 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
354 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
357 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
362 return security_inode_permission(inode, MAY_EXEC, nd);
366 * This is called when everything else fails, and we actually have
367 * to go to the low-level filesystem to find out what we should do..
369 * We get the directory semaphore, and after getting that we also
370 * make sure that nobody added the entry to the dcache in the meantime..
373 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
375 struct dentry * result;
376 struct inode *dir = parent->d_inode;
380 * First re-do the cached lookup just in case it was created
381 * while we waited for the directory semaphore..
383 * FIXME! This could use version numbering or similar to
384 * avoid unnecessary cache lookups.
386 * The "dcache_lock" is purely to protect the RCU list walker
387 * from concurrent renames at this point (we mustn't get false
388 * negatives from the RCU list walk here, unlike the optimistic
391 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
393 result = d_lookup(parent, name);
395 struct dentry * dentry = d_alloc(parent, name);
396 result = ERR_PTR(-ENOMEM);
398 result = dir->i_op->lookup(dir, dentry, nd);
409 * Uhhuh! Nasty case: the cache was re-populated while
410 * we waited on the semaphore. Need to revalidate.
413 if (result->d_op && result->d_op->d_revalidate) {
414 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
416 result = ERR_PTR(-ENOENT);
422 static int __emul_lookup_dentry(const char *, struct nameidata *);
426 walk_init_root(const char *name, struct nameidata *nd)
428 read_lock(¤t->fs->lock);
429 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
430 nd->mnt = mntget(current->fs->altrootmnt);
431 nd->dentry = dget(current->fs->altroot);
432 read_unlock(¤t->fs->lock);
433 if (__emul_lookup_dentry(name,nd))
435 read_lock(¤t->fs->lock);
437 nd->mnt = mntget(current->fs->rootmnt);
438 nd->dentry = dget(current->fs->root);
439 read_unlock(¤t->fs->lock);
443 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
452 if (!walk_init_root(link, nd))
453 /* weird __emul_prefix() stuff did it */
456 res = link_path_walk(link, nd);
458 if (nd->depth || res || nd->last_type!=LAST_NORM)
461 * If it is an iterative symlinks resolution in open_namei() we
462 * have to copy the last component. And all that crap because of
463 * bloody create() on broken symlinks. Furrfu...
466 if (unlikely(!name)) {
470 strcpy(name, nd->last.name);
471 nd->last.name = name;
475 return PTR_ERR(link);
479 * This limits recursive symlink follows to 8, while
480 * limiting consecutive symlinks to 40.
482 * Without that kind of total limit, nasty chains of consecutive
483 * symlinks can cause almost arbitrarily long lookups.
485 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
488 if (current->link_count >= MAX_NESTED_LINKS)
490 if (current->total_link_count >= 40)
492 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
494 err = security_inode_follow_link(dentry, nd);
497 current->link_count++;
498 current->total_link_count++;
500 touch_atime(nd->mnt, dentry);
501 nd_set_link(nd, NULL);
502 err = dentry->d_inode->i_op->follow_link(dentry, nd);
504 char *s = nd_get_link(nd);
506 err = __vfs_follow_link(nd, s);
507 if (dentry->d_inode->i_op->put_link)
508 dentry->d_inode->i_op->put_link(dentry, nd);
510 current->link_count--;
518 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
520 struct vfsmount *parent;
521 struct dentry *mountpoint;
522 spin_lock(&vfsmount_lock);
523 parent=(*mnt)->mnt_parent;
524 if (parent == *mnt) {
525 spin_unlock(&vfsmount_lock);
529 mountpoint=dget((*mnt)->mnt_mountpoint);
530 spin_unlock(&vfsmount_lock);
532 *dentry = mountpoint;
538 /* no need for dcache_lock, as serialization is taken care in
541 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
544 while (d_mountpoint(*dentry)) {
545 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
551 *dentry = dget(mounted->mnt_root);
557 /* no need for dcache_lock, as serialization is taken care in
560 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
562 struct vfsmount *mounted;
564 mounted = lookup_mnt(*mnt, *dentry);
569 *dentry = dget(mounted->mnt_root);
575 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
577 return __follow_down(mnt,dentry);
580 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
583 struct vfsmount *parent;
584 struct dentry *old = *dentry;
586 read_lock(¤t->fs->lock);
587 if (*dentry == current->fs->root &&
588 *mnt == current->fs->rootmnt) {
589 read_unlock(¤t->fs->lock);
592 read_unlock(¤t->fs->lock);
593 spin_lock(&dcache_lock);
594 if (*dentry != (*mnt)->mnt_root) {
595 *dentry = dget((*dentry)->d_parent);
596 spin_unlock(&dcache_lock);
600 spin_unlock(&dcache_lock);
601 spin_lock(&vfsmount_lock);
602 parent = (*mnt)->mnt_parent;
603 if (parent == *mnt) {
604 spin_unlock(&vfsmount_lock);
608 *dentry = dget((*mnt)->mnt_mountpoint);
609 spin_unlock(&vfsmount_lock);
614 follow_mount(mnt, dentry);
618 struct vfsmount *mnt;
619 struct dentry *dentry;
623 * It's more convoluted than I'd like it to be, but... it's still fairly
624 * small and for now I'd prefer to have fast path as straight as possible.
625 * It _is_ time-critical.
627 static int do_lookup(struct nameidata *nd, struct qstr *name,
628 struct path *path, int atomic)
630 struct vfsmount *mnt = nd->mnt;
631 struct dentry *dentry = __d_lookup(nd->dentry, name);
635 if (dentry->d_op && dentry->d_op->d_revalidate)
636 goto need_revalidate;
639 path->dentry = dentry;
644 return -EWOULDBLOCKIO;
645 dentry = real_lookup(nd->dentry, name, nd);
652 return -EWOULDBLOCKIO;
653 if (dentry->d_op->d_revalidate(dentry, nd))
655 if (d_invalidate(dentry))
661 return PTR_ERR(dentry);
667 * This is the basic name resolution function, turning a pathname
668 * into the final dentry.
670 * We expect 'base' to be positive and a directory.
672 int fastcall link_path_walk(const char * name, struct nameidata *nd)
677 unsigned int lookup_flags = nd->flags;
679 atomic = (lookup_flags & LOOKUP_ATOMIC);
686 inode = nd->dentry->d_inode;
688 lookup_flags = LOOKUP_FOLLOW;
690 /* At this point we know we have a real path component. */
696 err = exec_permission_lite(inode, nd);
697 if (err == -EAGAIN) {
698 err = permission(inode, MAY_EXEC, nd);
704 c = *(const unsigned char *)name;
706 hash = init_name_hash();
709 hash = partial_name_hash(c, hash);
710 c = *(const unsigned char *)name;
711 } while (c && (c != '/'));
712 this.len = name - (const char *) this.name;
713 this.hash = end_name_hash(hash);
715 /* remove trailing slashes? */
718 while (*++name == '/');
720 goto last_with_slashes;
723 * "." and ".." are special - ".." especially so because it has
724 * to be able to know about the current root directory and
725 * parent relationships.
727 if (this.name[0] == '.') switch (this.len) {
731 if (this.name[1] != '.')
733 follow_dotdot(&nd->mnt, &nd->dentry);
734 inode = nd->dentry->d_inode;
740 * See if the low-level filesystem might want
741 * to use its own hash..
743 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
744 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
748 nd->flags |= LOOKUP_CONTINUE;
749 /* This does the actual lookups.. */
750 err = do_lookup(nd, &this, &next, atomic);
753 /* Check mountpoints.. */
754 follow_mount(&next.mnt, &next.dentry);
757 inode = next.dentry->d_inode;
764 if (inode->i_op->follow_link) {
766 err = do_follow_link(next.dentry, nd);
772 inode = nd->dentry->d_inode;
781 nd->dentry = next.dentry;
784 if (!inode->i_op->lookup)
787 /* here ends the main loop */
790 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
792 nd->flags &= ~LOOKUP_CONTINUE;
793 if (lookup_flags & LOOKUP_PARENT)
795 if (this.name[0] == '.') switch (this.len) {
799 if (this.name[1] != '.')
801 follow_dotdot(&nd->mnt, &nd->dentry);
802 inode = nd->dentry->d_inode;
807 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
808 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
812 err = do_lookup(nd, &this, &next, atomic);
815 follow_mount(&next.mnt, &next.dentry);
816 inode = next.dentry->d_inode;
817 if ((lookup_flags & LOOKUP_FOLLOW)
818 && inode && inode->i_op && inode->i_op->follow_link) {
820 err = do_follow_link(next.dentry, nd);
825 inode = nd->dentry->d_inode;
829 nd->dentry = next.dentry;
834 if (lookup_flags & LOOKUP_DIRECTORY) {
836 if (!inode->i_op || !inode->i_op->lookup)
842 nd->last_type = LAST_NORM;
843 if (this.name[0] != '.')
846 nd->last_type = LAST_DOT;
847 else if (this.len == 2 && this.name[1] == '.')
848 nd->last_type = LAST_DOTDOT;
853 * We bypassed the ordinary revalidation routines.
854 * We may need to check the cached dentry for staleness.
856 if (nd->dentry && nd->dentry->d_sb &&
857 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
859 /* Note: we do not d_invalidate() */
860 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
874 int fastcall path_walk(const char * name, struct nameidata *nd)
876 current->total_link_count = 0;
877 return link_path_walk(name, nd);
881 /* returns 1 if everything is done */
882 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
884 if (path_walk(name, nd))
885 return 0; /* something went wrong... */
887 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
888 struct dentry *old_dentry = nd->dentry;
889 struct vfsmount *old_mnt = nd->mnt;
890 struct qstr last = nd->last;
891 int last_type = nd->last_type;
893 * NAME was not found in alternate root or it's a directory. Try to find
894 * it in the normal root:
896 nd->last_type = LAST_ROOT;
897 read_lock(¤t->fs->lock);
898 nd->mnt = mntget(current->fs->rootmnt);
899 nd->dentry = dget(current->fs->root);
900 read_unlock(¤t->fs->lock);
901 if (path_walk(name, nd) == 0) {
902 if (nd->dentry->d_inode) {
909 nd->dentry = old_dentry;
912 nd->last_type = last_type;
917 void set_fs_altroot(void)
919 char *emul = __emul_prefix();
921 struct vfsmount *mnt = NULL, *oldmnt;
922 struct dentry *dentry = NULL, *olddentry;
927 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
933 write_lock(¤t->fs->lock);
934 oldmnt = current->fs->altrootmnt;
935 olddentry = current->fs->altroot;
936 current->fs->altrootmnt = mnt;
937 current->fs->altroot = dentry;
938 write_unlock(¤t->fs->lock);
945 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
949 nd->last_type = LAST_ROOT; /* if there are only slashes... */
953 read_lock(¤t->fs->lock);
955 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
956 nd->mnt = mntget(current->fs->altrootmnt);
957 nd->dentry = dget(current->fs->altroot);
958 read_unlock(¤t->fs->lock);
959 if (__emul_lookup_dentry(name,nd))
961 read_lock(¤t->fs->lock);
963 nd->mnt = mntget(current->fs->rootmnt);
964 nd->dentry = dget(current->fs->root);
966 nd->mnt = mntget(current->fs->pwdmnt);
967 nd->dentry = dget(current->fs->pwd);
969 read_unlock(¤t->fs->lock);
970 current->total_link_count = 0;
971 retval = link_path_walk(name, nd);
972 if (unlikely(current->audit_context
973 && nd && nd->dentry && nd->dentry->d_inode))
975 nd->dentry->d_inode->i_ino,
976 nd->dentry->d_inode->i_rdev);
981 * Restricted form of lookup. Doesn't follow links, single-component only,
982 * needs parent already locked. Doesn't follow mounts.
985 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
987 struct dentry * dentry;
991 inode = base->d_inode;
992 err = permission(inode, MAY_EXEC, nd);
993 dentry = ERR_PTR(err);
998 * See if the low-level filesystem might want
999 * to use its own hash..
1001 if (base->d_op && base->d_op->d_hash) {
1002 err = base->d_op->d_hash(base, name);
1003 dentry = ERR_PTR(err);
1008 dentry = cached_lookup(base, name, nd);
1010 struct dentry *new = d_alloc(base, name);
1011 dentry = ERR_PTR(-ENOMEM);
1014 dentry = inode->i_op->lookup(inode, new, nd);
1024 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1026 return __lookup_hash(name, base, NULL);
1030 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1041 hash = init_name_hash();
1043 c = *(const unsigned char *)name++;
1044 if (c == '/' || c == '\0')
1046 hash = partial_name_hash(c, hash);
1048 this.hash = end_name_hash(hash);
1050 return lookup_hash(&this, base);
1052 return ERR_PTR(-EACCES);
1058 * is used by most simple commands to get the inode of a specified name.
1059 * Open, link etc use their own routines, but this is enough for things
1062 * namei exists in two versions: namei/lnamei. The only difference is
1063 * that namei follows links, while lnamei does not.
1066 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1068 char *tmp = getname(name);
1069 int err = PTR_ERR(tmp);
1072 err = path_lookup(tmp, flags, nd);
1079 * It's inline, so penalty for filesystems that don't use sticky bit is
1082 static inline int check_sticky(struct inode *dir, struct inode *inode)
1084 if (!(dir->i_mode & S_ISVTX))
1086 if (inode->i_uid == current->fsuid)
1088 if (dir->i_uid == current->fsuid)
1090 return !capable(CAP_FOWNER);
1094 * Check whether we can remove a link victim from directory dir, check
1095 * whether the type of victim is right.
1096 * 1. We can't do it if dir is read-only (done in permission())
1097 * 2. We should have write and exec permissions on dir
1098 * 3. We can't remove anything from append-only dir
1099 * 4. We can't do anything with immutable dir (done in permission())
1100 * 5. If the sticky bit on dir is set we should either
1101 * a. be owner of dir, or
1102 * b. be owner of victim, or
1103 * c. have CAP_FOWNER capability
1104 * 6. If the victim is append-only or immutable we can't do antyhing with
1105 * links pointing to it.
1106 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1107 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1108 * 9. We can't remove a root or mountpoint.
1109 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1110 * nfs_async_unlink().
1112 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1116 if (!victim->d_inode)
1119 BUG_ON(victim->d_parent->d_inode != dir);
1121 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1126 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1127 IS_IXORUNLINK(victim->d_inode))
1130 if (!S_ISDIR(victim->d_inode->i_mode))
1132 if (IS_ROOT(victim))
1134 } else if (S_ISDIR(victim->d_inode->i_mode))
1136 if (IS_DEADDIR(dir))
1138 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1143 /* Check whether we can create an object with dentry child in directory
1145 * 1. We can't do it if child already exists (open has special treatment for
1146 * this case, but since we are inlined it's OK)
1147 * 2. We can't do it if dir is read-only (done in permission())
1148 * 3. We should have write and exec permissions on dir
1149 * 4. We can't do it if dir is immutable (done in permission())
1151 static inline int may_create(struct inode *dir, struct dentry *child,
1152 struct nameidata *nd)
1156 if (IS_DEADDIR(dir))
1158 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1161 static inline int mnt_may_create(struct vfsmount *mnt, struct inode *dir, struct dentry *child) {
1164 if (IS_DEADDIR(dir))
1166 if (mnt->mnt_flags & MNT_RDONLY)
1171 static inline int mnt_may_unlink(struct vfsmount *mnt, struct inode *dir, struct dentry *child) {
1172 if (!child->d_inode)
1174 if (mnt->mnt_flags & MNT_RDONLY)
1180 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1183 * O_DIRECTORY translates into forcing a directory lookup.
1185 static inline int lookup_flags(unsigned int f)
1187 unsigned long retval = LOOKUP_FOLLOW;
1190 retval &= ~LOOKUP_FOLLOW;
1192 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1193 retval &= ~LOOKUP_FOLLOW;
1195 if (f & O_DIRECTORY)
1196 retval |= LOOKUP_DIRECTORY;
1197 if (f & O_ATOMICLOOKUP)
1198 retval |= LOOKUP_ATOMIC;
1204 * p1 and p2 should be directories on the same fs.
1206 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1211 down(&p1->d_inode->i_sem);
1215 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1217 for (p = p1; p->d_parent != p; p = p->d_parent) {
1218 if (p->d_parent == p2) {
1219 down(&p2->d_inode->i_sem);
1220 down(&p1->d_inode->i_sem);
1225 for (p = p2; p->d_parent != p; p = p->d_parent) {
1226 if (p->d_parent == p1) {
1227 down(&p1->d_inode->i_sem);
1228 down(&p2->d_inode->i_sem);
1233 down(&p1->d_inode->i_sem);
1234 down(&p2->d_inode->i_sem);
1238 void unlock_rename(struct dentry *p1, struct dentry *p2)
1240 up(&p1->d_inode->i_sem);
1242 up(&p2->d_inode->i_sem);
1243 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1247 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1248 struct nameidata *nd)
1250 int error = may_create(dir, dentry, nd);
1255 if (!dir->i_op || !dir->i_op->create)
1256 return -EACCES; /* shouldn't it be ENOSYS? */
1259 error = security_inode_create(dir, dentry, mode);
1263 error = dir->i_op->create(dir, dentry, mode, nd);
1265 inode_dir_notify(dir, DN_CREATE);
1266 security_inode_post_create(dir, dentry, mode);
1271 int may_open(struct nameidata *nd, int acc_mode, int flag)
1273 struct dentry *dentry = nd->dentry;
1274 struct inode *inode = dentry->d_inode;
1280 if (S_ISLNK(inode->i_mode))
1283 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1286 error = permission(inode, acc_mode, nd);
1291 * FIFO's, sockets and device files are special: they don't
1292 * actually live on the filesystem itself, and as such you
1293 * can write to them even if the filesystem is read-only.
1295 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1297 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1298 if (nd->mnt->mnt_flags & MNT_NODEV)
1302 } else if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt)))
1303 && (flag & FMODE_WRITE))
1306 * An append-only file must be opened in append mode for writing.
1308 if (IS_APPEND(inode)) {
1309 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1315 /* O_NOATIME can only be set by the owner or superuser */
1316 if (flag & O_NOATIME)
1317 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1321 * Ensure there are no outstanding leases on the file.
1323 error = break_lease(inode, flag);
1327 if (flag & O_TRUNC) {
1328 error = get_write_access(inode);
1333 * Refuse to truncate files with mandatory locks held on them.
1335 error = locks_verify_locked(inode);
1339 error = do_truncate(dentry, 0);
1341 put_write_access(inode);
1345 if (flag & FMODE_WRITE)
1354 * namei for open - this is in fact almost the whole open-routine.
1356 * Note that the low bits of "flag" aren't the same as in the open
1357 * system call - they are 00 - no permissions needed
1358 * 01 - read permission needed
1359 * 10 - write permission needed
1360 * 11 - read/write permissions needed
1361 * which is a lot more logical, and also allows the "no perm" needed
1362 * for symlinks (where the permissions are checked later).
1365 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1367 int acc_mode, error = 0;
1368 struct dentry *dentry;
1372 acc_mode = ACC_MODE(flag);
1374 /* Allow the LSM permission hook to distinguish append
1375 access from general write access. */
1376 if (flag & O_APPEND)
1377 acc_mode |= MAY_APPEND;
1379 /* Fill in the open() intent data */
1380 nd->intent.open.flags = flag;
1381 nd->intent.open.create_mode = mode;
1384 * The simplest case - just a plain lookup.
1386 if (!(flag & O_CREAT)) {
1387 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1394 * Create - we need to know the parent.
1396 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1401 * We have the parent and last component. First of all, check
1402 * that we are not asked to creat(2) an obvious directory - that
1406 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1410 nd->flags &= ~LOOKUP_PARENT;
1411 down(&dir->d_inode->i_sem);
1412 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1415 error = PTR_ERR(dentry);
1416 if (IS_ERR(dentry)) {
1417 up(&dir->d_inode->i_sem);
1421 /* Negative dentry, just create the file */
1422 if (!dentry->d_inode) {
1423 if (!IS_POSIXACL(dir->d_inode))
1424 mode &= ~current->fs->umask;
1425 error = vfs_create(dir->d_inode, dentry, mode, nd);
1426 up(&dir->d_inode->i_sem);
1428 nd->dentry = dentry;
1431 /* Don't check for write permission, don't truncate */
1438 * It already exists.
1440 up(&dir->d_inode->i_sem);
1446 if (d_mountpoint(dentry)) {
1448 if (flag & O_NOFOLLOW)
1450 while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
1453 if (!dentry->d_inode)
1455 if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1459 nd->dentry = dentry;
1461 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1464 error = may_open(nd, acc_mode, flag);
1477 if (flag & O_NOFOLLOW)
1480 * This is subtle. Instead of calling do_follow_link() we do the
1481 * thing by hands. The reason is that this way we have zero link_count
1482 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1483 * After that we have the parent and last component, i.e.
1484 * we are in the same situation as after the first path_walk().
1485 * Well, almost - if the last component is normal we get its copy
1486 * stored in nd->last.name and we will have to putname() it when we
1487 * are done. Procfs-like symlinks just set LAST_BIND.
1489 nd->flags |= LOOKUP_PARENT;
1490 error = security_inode_follow_link(dentry, nd);
1493 touch_atime(nd->mnt, dentry);
1494 nd_set_link(nd, NULL);
1495 error = dentry->d_inode->i_op->follow_link(dentry, nd);
1497 char *s = nd_get_link(nd);
1499 error = __vfs_follow_link(nd, s);
1500 if (dentry->d_inode->i_op->put_link)
1501 dentry->d_inode->i_op->put_link(dentry, nd);
1506 nd->flags &= ~LOOKUP_PARENT;
1507 if (nd->last_type == LAST_BIND) {
1508 dentry = nd->dentry;
1512 if (nd->last_type != LAST_NORM)
1514 if (nd->last.name[nd->last.len]) {
1515 putname(nd->last.name);
1520 putname(nd->last.name);
1524 down(&dir->d_inode->i_sem);
1525 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1526 putname(nd->last.name);
1531 * lookup_create - lookup a dentry, creating it if it doesn't exist
1532 * @nd: nameidata info
1533 * @is_dir: directory flag
1535 * Simple function to lookup and return a dentry and create it
1536 * if it doesn't exist. Is SMP-safe.
1538 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1540 struct dentry *dentry;
1543 down(&nd->dentry->d_inode->i_sem);
1545 if (nd->last_type != LAST_NORM)
1547 nd->flags &= ~LOOKUP_PARENT;
1548 dentry = lookup_hash(&nd->last, nd->dentry);
1551 error = mnt_may_create(nd->mnt, nd->dentry->d_inode, dentry);
1555 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1562 return ERR_PTR(error);
1565 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1567 int error = may_create(dir, dentry, NULL);
1572 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1575 if (!dir->i_op || !dir->i_op->mknod)
1578 error = security_inode_mknod(dir, dentry, mode, dev);
1583 error = dir->i_op->mknod(dir, dentry, mode, dev);
1585 inode_dir_notify(dir, DN_CREATE);
1586 security_inode_post_mknod(dir, dentry, mode, dev);
1591 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1595 struct dentry * dentry;
1596 struct nameidata nd;
1600 tmp = getname(filename);
1602 return PTR_ERR(tmp);
1604 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1607 dentry = lookup_create(&nd, 0);
1608 error = PTR_ERR(dentry);
1610 if (!IS_POSIXACL(nd.dentry->d_inode))
1611 mode &= ~current->fs->umask;
1612 if (!IS_ERR(dentry)) {
1613 switch (mode & S_IFMT) {
1614 case 0: case S_IFREG:
1615 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1617 case S_IFCHR: case S_IFBLK:
1618 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1619 new_decode_dev(dev));
1621 case S_IFIFO: case S_IFSOCK:
1622 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1632 up(&nd.dentry->d_inode->i_sem);
1640 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1642 int error = may_create(dir, dentry, NULL);
1647 if (!dir->i_op || !dir->i_op->mkdir)
1650 mode &= (S_IRWXUGO|S_ISVTX);
1651 error = security_inode_mkdir(dir, dentry, mode);
1656 error = dir->i_op->mkdir(dir, dentry, mode);
1658 inode_dir_notify(dir, DN_CREATE);
1659 security_inode_post_mkdir(dir,dentry, mode);
1664 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1669 tmp = getname(pathname);
1670 error = PTR_ERR(tmp);
1672 struct dentry *dentry;
1673 struct nameidata nd;
1675 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1678 dentry = lookup_create(&nd, 1);
1679 error = PTR_ERR(dentry);
1680 if (!IS_ERR(dentry)) {
1681 if (!IS_POSIXACL(nd.dentry->d_inode))
1682 mode &= ~current->fs->umask;
1683 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1686 up(&nd.dentry->d_inode->i_sem);
1696 * We try to drop the dentry early: we should have
1697 * a usage count of 2 if we're the only user of this
1698 * dentry, and if that is true (possibly after pruning
1699 * the dcache), then we drop the dentry now.
1701 * A low-level filesystem can, if it choses, legally
1704 * if (!d_unhashed(dentry))
1707 * if it cannot handle the case of removing a directory
1708 * that is still in use by something else..
1710 void dentry_unhash(struct dentry *dentry)
1713 spin_lock(&dcache_lock);
1714 switch (atomic_read(&dentry->d_count)) {
1716 spin_unlock(&dcache_lock);
1717 shrink_dcache_parent(dentry);
1718 spin_lock(&dcache_lock);
1719 if (atomic_read(&dentry->d_count) != 2)
1724 spin_unlock(&dcache_lock);
1727 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1729 int error = may_delete(dir, dentry, 1);
1734 if (!dir->i_op || !dir->i_op->rmdir)
1739 down(&dentry->d_inode->i_sem);
1740 dentry_unhash(dentry);
1741 if (d_mountpoint(dentry))
1744 error = security_inode_rmdir(dir, dentry);
1746 error = dir->i_op->rmdir(dir, dentry);
1748 dentry->d_inode->i_flags |= S_DEAD;
1751 up(&dentry->d_inode->i_sem);
1753 inode_dir_notify(dir, DN_DELETE);
1761 asmlinkage long sys_rmdir(const char __user * pathname)
1765 struct dentry *dentry;
1766 struct nameidata nd;
1768 name = getname(pathname);
1770 return PTR_ERR(name);
1772 error = path_lookup(name, LOOKUP_PARENT, &nd);
1776 switch(nd.last_type) {
1787 down(&nd.dentry->d_inode->i_sem);
1788 dentry = lookup_hash(&nd.last, nd.dentry);
1789 error = PTR_ERR(dentry);
1790 if (!IS_ERR(dentry)) {
1791 error = mnt_may_unlink(nd.mnt, nd.dentry->d_inode, dentry);
1794 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1798 up(&nd.dentry->d_inode->i_sem);
1806 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1808 int error = may_delete(dir, dentry, 0);
1813 if (!dir->i_op || !dir->i_op->unlink)
1818 down(&dentry->d_inode->i_sem);
1819 if (d_mountpoint(dentry))
1822 error = security_inode_unlink(dir, dentry);
1824 error = dir->i_op->unlink(dir, dentry);
1826 up(&dentry->d_inode->i_sem);
1828 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1829 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1831 inode_dir_notify(dir, DN_DELETE);
1837 * Make sure that the actual truncation of the file will occur outside its
1838 * directory's i_sem. Truncate can take a long time if there is a lot of
1839 * writeout happening, and we don't want to prevent access to the directory
1840 * while waiting on the I/O.
1842 asmlinkage long sys_unlink(const char __user * pathname)
1846 struct dentry *dentry;
1847 struct nameidata nd;
1848 struct inode *inode = NULL;
1850 name = getname(pathname);
1852 return PTR_ERR(name);
1854 error = path_lookup(name, LOOKUP_PARENT, &nd);
1858 if (nd.last_type != LAST_NORM)
1860 down(&nd.dentry->d_inode->i_sem);
1861 dentry = lookup_hash(&nd.last, nd.dentry);
1862 error = PTR_ERR(dentry);
1863 if (!IS_ERR(dentry)) {
1864 /* Why not before? Because we want correct error value */
1865 if (nd.last.name[nd.last.len])
1867 error = mnt_may_unlink(nd.mnt, nd.dentry->d_inode, dentry);
1870 inode = dentry->d_inode;
1872 atomic_inc(&inode->i_count);
1873 error = vfs_unlink(nd.dentry->d_inode, dentry);
1877 up(&nd.dentry->d_inode->i_sem);
1879 iput(inode); /* truncate the inode here */
1887 error = !dentry->d_inode ? -ENOENT :
1888 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1892 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1894 int error = may_create(dir, dentry, NULL);
1899 if (!dir->i_op || !dir->i_op->symlink)
1902 error = security_inode_symlink(dir, dentry, oldname);
1907 error = dir->i_op->symlink(dir, dentry, oldname);
1909 inode_dir_notify(dir, DN_CREATE);
1910 security_inode_post_symlink(dir, dentry, oldname);
1915 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1921 from = getname(oldname);
1923 return PTR_ERR(from);
1924 to = getname(newname);
1925 error = PTR_ERR(to);
1927 struct dentry *dentry;
1928 struct nameidata nd;
1930 error = path_lookup(to, LOOKUP_PARENT, &nd);
1933 dentry = lookup_create(&nd, 0);
1934 error = PTR_ERR(dentry);
1935 if (!IS_ERR(dentry)) {
1936 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1939 up(&nd.dentry->d_inode->i_sem);
1948 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1950 struct inode *inode = old_dentry->d_inode;
1956 error = may_create(dir, new_dentry, NULL);
1960 if (dir->i_sb != inode->i_sb)
1964 * A link to an append-only or immutable file cannot be created.
1966 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
1968 if (!dir->i_op || !dir->i_op->link)
1970 if (S_ISDIR(old_dentry->d_inode->i_mode))
1973 error = security_inode_link(old_dentry, dir, new_dentry);
1977 down(&old_dentry->d_inode->i_sem);
1979 error = dir->i_op->link(old_dentry, dir, new_dentry);
1980 up(&old_dentry->d_inode->i_sem);
1982 inode_dir_notify(dir, DN_CREATE);
1983 security_inode_post_link(old_dentry, dir, new_dentry);
1989 * Hardlinks are often used in delicate situations. We avoid
1990 * security-related surprises by not following symlinks on the
1993 * We don't follow them on the oldname either to be compatible
1994 * with linux 2.0, and to avoid hard-linking to directories
1995 * and other special files. --ADM
1997 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
1999 struct dentry *new_dentry;
2000 struct nameidata nd, old_nd;
2004 to = getname(newname);
2008 error = __user_walk(oldname, 0, &old_nd);
2011 error = path_lookup(to, LOOKUP_PARENT, &nd);
2015 * We allow hard-links to be created to a bind-mount as long
2016 * as the bind-mount is not read-only. Checking for cross-dev
2017 * links is subsumed by the superblock check in vfs_link().
2020 if (MNT_IS_RDONLY(old_nd.mnt))
2022 new_dentry = lookup_create(&nd, 0);
2023 error = PTR_ERR(new_dentry);
2024 if (!IS_ERR(new_dentry)) {
2025 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2028 up(&nd.dentry->d_inode->i_sem);
2032 path_release(&old_nd);
2040 * The worst of all namespace operations - renaming directory. "Perverted"
2041 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2043 * a) we can get into loop creation. Check is done in is_subdir().
2044 * b) race potential - two innocent renames can create a loop together.
2045 * That's where 4.4 screws up. Current fix: serialization on
2046 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2048 * c) we have to lock _three_ objects - parents and victim (if it exists).
2049 * And that - after we got ->i_sem on parents (until then we don't know
2050 * whether the target exists). Solution: try to be smart with locking
2051 * order for inodes. We rely on the fact that tree topology may change
2052 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2053 * move will be locked. Thus we can rank directories by the tree
2054 * (ancestors first) and rank all non-directories after them.
2055 * That works since everybody except rename does "lock parent, lookup,
2056 * lock child" and rename is under ->s_vfs_rename_sem.
2057 * HOWEVER, it relies on the assumption that any object with ->lookup()
2058 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2059 * we'd better make sure that there's no link(2) for them.
2060 * d) some filesystems don't support opened-but-unlinked directories,
2061 * either because of layout or because they are not ready to deal with
2062 * all cases correctly. The latter will be fixed (taking this sort of
2063 * stuff into VFS), but the former is not going away. Solution: the same
2064 * trick as in rmdir().
2065 * e) conversion from fhandle to dentry may come in the wrong moment - when
2066 * we are removing the target. Solution: we will have to grab ->i_sem
2067 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2068 * ->i_sem on parents, which works but leads to some truely excessive
2071 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2072 struct inode *new_dir, struct dentry *new_dentry)
2075 struct inode *target;
2078 * If we are going to change the parent - check write permissions,
2079 * we'll need to flip '..'.
2081 if (new_dir != old_dir) {
2082 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2087 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2091 target = new_dentry->d_inode;
2093 down(&target->i_sem);
2094 dentry_unhash(new_dentry);
2096 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2099 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2102 target->i_flags |= S_DEAD;
2104 if (d_unhashed(new_dentry))
2105 d_rehash(new_dentry);
2109 d_move(old_dentry,new_dentry);
2110 security_inode_post_rename(old_dir, old_dentry,
2111 new_dir, new_dentry);
2116 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2117 struct inode *new_dir, struct dentry *new_dentry)
2119 struct inode *target;
2122 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2127 target = new_dentry->d_inode;
2129 down(&target->i_sem);
2130 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2133 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2135 /* The following d_move() should become unconditional */
2136 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2137 d_move(old_dentry, new_dentry);
2138 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2146 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2147 struct inode *new_dir, struct dentry *new_dentry)
2150 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2152 if (old_dentry->d_inode == new_dentry->d_inode)
2155 error = may_delete(old_dir, old_dentry, is_dir);
2159 if (!new_dentry->d_inode)
2160 error = may_create(new_dir, new_dentry, NULL);
2162 error = may_delete(new_dir, new_dentry, is_dir);
2166 if (!old_dir->i_op || !old_dir->i_op->rename)
2169 DQUOT_INIT(old_dir);
2170 DQUOT_INIT(new_dir);
2173 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2175 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2177 if (old_dir == new_dir)
2178 inode_dir_notify(old_dir, DN_RENAME);
2180 inode_dir_notify(old_dir, DN_DELETE);
2181 inode_dir_notify(new_dir, DN_CREATE);
2187 static inline int do_rename(const char * oldname, const char * newname)
2190 struct dentry * old_dir, * new_dir;
2191 struct dentry * old_dentry, *new_dentry;
2192 struct dentry * trap;
2193 struct nameidata oldnd, newnd;
2195 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2199 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2204 if (oldnd.mnt != newnd.mnt)
2207 old_dir = oldnd.dentry;
2209 if (oldnd.last_type != LAST_NORM)
2212 new_dir = newnd.dentry;
2213 if (newnd.last_type != LAST_NORM)
2216 trap = lock_rename(new_dir, old_dir);
2218 old_dentry = lookup_hash(&oldnd.last, old_dir);
2219 error = PTR_ERR(old_dentry);
2220 if (IS_ERR(old_dentry))
2222 /* source must exist */
2224 if (!old_dentry->d_inode)
2226 /* unless the source is a directory trailing slashes give -ENOTDIR */
2227 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2229 if (oldnd.last.name[oldnd.last.len])
2231 if (newnd.last.name[newnd.last.len])
2234 /* source should not be ancestor of target */
2236 if (old_dentry == trap)
2239 if (MNT_IS_RDONLY(newnd.mnt))
2241 new_dentry = lookup_hash(&newnd.last, new_dir);
2242 error = PTR_ERR(new_dentry);
2243 if (IS_ERR(new_dentry))
2245 /* target should not be an ancestor of source */
2247 if (new_dentry == trap)
2250 error = vfs_rename(old_dir->d_inode, old_dentry,
2251 new_dir->d_inode, new_dentry);
2257 unlock_rename(new_dir, old_dir);
2259 path_release(&newnd);
2261 path_release(&oldnd);
2266 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2272 from = getname(oldname);
2274 return PTR_ERR(from);
2275 to = getname(newname);
2276 error = PTR_ERR(to);
2278 error = do_rename(from,to);
2285 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2289 len = PTR_ERR(link);
2294 if (len > (unsigned) buflen)
2296 if (copy_to_user(buffer, link, len))
2303 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2304 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2305 * using) it for any given inode is up to filesystem.
2307 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2309 struct nameidata nd;
2312 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2314 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2315 if (dentry->d_inode->i_op->put_link)
2316 dentry->d_inode->i_op->put_link(dentry, &nd);
2321 int vfs_follow_link(struct nameidata *nd, const char *link)
2323 return __vfs_follow_link(nd, link);
2326 /* get the link contents into pagecache */
2327 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2330 struct address_space *mapping = dentry->d_inode->i_mapping;
2331 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2335 wait_on_page_locked(page);
2336 if (!PageUptodate(page))
2342 page_cache_release(page);
2343 return ERR_PTR(-EIO);
2349 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2351 struct page *page = NULL;
2352 char *s = page_getlink(dentry, &page);
2353 int res = vfs_readlink(dentry,buffer,buflen,s);
2356 page_cache_release(page);
2361 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2364 nd_set_link(nd, page_getlink(dentry, &page));
2368 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2370 if (!IS_ERR(nd_get_link(nd))) {
2372 page = find_get_page(dentry->d_inode->i_mapping, 0);
2376 page_cache_release(page);
2377 page_cache_release(page);
2381 int page_symlink(struct inode *inode, const char *symname, int len)
2383 struct address_space *mapping = inode->i_mapping;
2384 struct page *page = grab_cache_page(mapping, 0);
2390 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2393 kaddr = kmap_atomic(page, KM_USER0);
2394 memcpy(kaddr, symname, len-1);
2395 kunmap_atomic(kaddr, KM_USER0);
2396 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2398 * Notice that we are _not_ going to block here - end of page is
2399 * unmapped, so this will only try to map the rest of page, see
2400 * that it is unmapped (typically even will not look into inode -
2401 * ->i_size will be enough for everything) and zero it out.
2402 * OTOH it's obviously correct and should make the page up-to-date.
2404 if (!PageUptodate(page)) {
2405 err = mapping->a_ops->readpage(NULL, page);
2406 wait_on_page_locked(page);
2410 page_cache_release(page);
2413 mark_inode_dirty(inode);
2417 page_cache_release(page);
2422 struct inode_operations page_symlink_inode_operations = {
2423 .readlink = generic_readlink,
2424 .follow_link = page_follow_link_light,
2425 .put_link = page_put_link,
2428 EXPORT_SYMBOL(__user_walk);
2429 EXPORT_SYMBOL(follow_down);
2430 EXPORT_SYMBOL(follow_up);
2431 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2432 EXPORT_SYMBOL(getname);
2433 EXPORT_SYMBOL(lock_rename);
2434 EXPORT_SYMBOL(lookup_hash);
2435 EXPORT_SYMBOL(lookup_one_len);
2436 EXPORT_SYMBOL(page_follow_link_light);
2437 EXPORT_SYMBOL(page_put_link);
2438 EXPORT_SYMBOL(page_readlink);
2439 EXPORT_SYMBOL(page_symlink);
2440 EXPORT_SYMBOL(page_symlink_inode_operations);
2441 EXPORT_SYMBOL(path_lookup);
2442 EXPORT_SYMBOL(path_release);
2443 EXPORT_SYMBOL(path_walk);
2444 EXPORT_SYMBOL(permission);
2445 EXPORT_SYMBOL(unlock_rename);
2446 EXPORT_SYMBOL(vfs_create);
2447 EXPORT_SYMBOL(vfs_follow_link);
2448 EXPORT_SYMBOL(vfs_link);
2449 EXPORT_SYMBOL(vfs_mkdir);
2450 EXPORT_SYMBOL(vfs_mknod);
2451 EXPORT_SYMBOL(vfs_permission);
2452 EXPORT_SYMBOL(vfs_readlink);
2453 EXPORT_SYMBOL(vfs_rename);
2454 EXPORT_SYMBOL(vfs_rmdir);
2455 EXPORT_SYMBOL(vfs_symlink);
2456 EXPORT_SYMBOL(vfs_unlink);
2457 EXPORT_SYMBOL(dentry_unhash);
2458 EXPORT_SYMBOL(generic_readlink);