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 <asm/namei.h>
31 #include <asm/uaccess.h>
33 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
35 /* [Feb-1997 T. Schoebel-Theuer]
36 * Fundamental changes in the pathname lookup mechanisms (namei)
37 * were necessary because of omirr. The reason is that omirr needs
38 * to know the _real_ pathname, not the user-supplied one, in case
39 * of symlinks (and also when transname replacements occur).
41 * The new code replaces the old recursive symlink resolution with
42 * an iterative one (in case of non-nested symlink chains). It does
43 * this with calls to <fs>_follow_link().
44 * As a side effect, dir_namei(), _namei() and follow_link() are now
45 * replaced with a single function lookup_dentry() that can handle all
46 * the special cases of the former code.
48 * With the new dcache, the pathname is stored at each inode, at least as
49 * long as the refcount of the inode is positive. As a side effect, the
50 * size of the dcache depends on the inode cache and thus is dynamic.
52 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
53 * resolution to correspond with current state of the code.
55 * Note that the symlink resolution is not *completely* iterative.
56 * There is still a significant amount of tail- and mid- recursion in
57 * the algorithm. Also, note that <fs>_readlink() is not used in
58 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
59 * may return different results than <fs>_follow_link(). Many virtual
60 * filesystems (including /proc) exhibit this behavior.
63 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
64 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
65 * and the name already exists in form of a symlink, try to create the new
66 * name indicated by the symlink. The old code always complained that the
67 * name already exists, due to not following the symlink even if its target
68 * is nonexistent. The new semantics affects also mknod() and link() when
69 * the name is a symlink pointing to a non-existant name.
71 * I don't know which semantics is the right one, since I have no access
72 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
73 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
74 * "old" one. Personally, I think the new semantics is much more logical.
75 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
76 * file does succeed in both HP-UX and SunOs, but not in Solaris
77 * and in the old Linux semantics.
80 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
81 * semantics. See the comments in "open_namei" and "do_link" below.
83 * [10-Sep-98 Alan Modra] Another symlink change.
86 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
87 * inside the path - always follow.
88 * in the last component in creation/removal/renaming - never follow.
89 * if LOOKUP_FOLLOW passed - follow.
90 * if the pathname has trailing slashes - follow.
91 * otherwise - don't follow.
92 * (applied in that order).
94 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
95 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
96 * During the 2.4 we need to fix the userland stuff depending on it -
97 * hopefully we will be able to get rid of that wart in 2.5. So far only
98 * XEmacs seems to be relying on it...
101 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
102 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
103 * any extra contention...
106 /* In order to reduce some races, while at the same time doing additional
107 * checking and hopefully speeding things up, we copy filenames to the
108 * kernel data space before using them..
110 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
111 * PATH_MAX includes the nul terminator --RR.
113 static inline int do_getname(const char __user *filename, char *page)
116 unsigned long len = PATH_MAX;
118 if ((unsigned long) filename >= TASK_SIZE) {
119 if (!segment_eq(get_fs(), KERNEL_DS))
121 } else if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
122 len = TASK_SIZE - (unsigned long) filename;
124 retval = strncpy_from_user((char *)page, filename, len);
128 return -ENAMETOOLONG;
134 char * getname(const char __user * filename)
138 result = ERR_PTR(-ENOMEM);
141 int retval = do_getname(filename, tmp);
146 result = ERR_PTR(retval);
149 if (unlikely(current->audit_context) && !IS_ERR(result) && result)
150 audit_getname(result);
157 * is used to check for read/write/execute permissions on a file.
158 * We use "fsuid" for this, letting us set arbitrary permissions
159 * for filesystem access without changing the "normal" uids which
160 * are used for other things..
162 int vfs_permission(struct inode * inode, int mask)
164 umode_t mode = inode->i_mode;
166 if (mask & MAY_WRITE) {
168 * Nobody gets write access to a read-only fs.
170 if (IS_RDONLY(inode) &&
171 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
175 * Nobody gets write access to an immutable file.
177 if (IS_IMMUTABLE(inode))
181 if (current->fsuid == inode->i_uid)
183 else if (in_group_p(inode->i_gid))
187 * If the DACs are ok we don't need any capability check.
189 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
193 * Read/write DACs are always overridable.
194 * Executable DACs are overridable if at least one exec bit is set.
196 if (!(mask & MAY_EXEC) ||
197 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
198 if (capable(CAP_DAC_OVERRIDE))
202 * Searching includes executable on directories, else just read.
204 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
205 if (capable(CAP_DAC_READ_SEARCH))
211 int permission(struct inode * inode,int mask, struct nameidata *nd)
216 /* Ordinary permission routines do not understand MAY_APPEND. */
217 submask = mask & ~MAY_APPEND;
219 if (inode->i_op && inode->i_op->permission)
220 retval = inode->i_op->permission(inode, submask, nd);
222 retval = vfs_permission(inode, submask);
226 return security_inode_permission(inode, mask, nd);
230 * get_write_access() gets write permission for a file.
231 * put_write_access() releases this write permission.
232 * This is used for regular files.
233 * We cannot support write (and maybe mmap read-write shared) accesses and
234 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
235 * can have the following values:
236 * 0: no writers, no VM_DENYWRITE mappings
237 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
238 * > 0: (i_writecount) users are writing to the file.
240 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
241 * except for the cases where we don't hold i_writecount yet. Then we need to
242 * use {get,deny}_write_access() - these functions check the sign and refuse
243 * to do the change if sign is wrong. Exclusion between them is provided by
244 * the inode->i_lock spinlock.
247 int get_write_access(struct inode * inode)
249 spin_lock(&inode->i_lock);
250 if (atomic_read(&inode->i_writecount) < 0) {
251 spin_unlock(&inode->i_lock);
254 atomic_inc(&inode->i_writecount);
255 spin_unlock(&inode->i_lock);
260 int deny_write_access(struct file * file)
262 struct inode *inode = file->f_dentry->d_inode;
264 spin_lock(&inode->i_lock);
265 if (atomic_read(&inode->i_writecount) > 0) {
266 spin_unlock(&inode->i_lock);
269 atomic_dec(&inode->i_writecount);
270 spin_unlock(&inode->i_lock);
275 void path_release(struct nameidata *nd)
282 * Internal lookup() using the new generic dcache.
285 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
287 struct dentry * dentry = __d_lookup(parent, name);
289 /* lockess __d_lookup may fail due to concurrent d_move()
290 * in some unrelated directory, so try with d_lookup
293 dentry = d_lookup(parent, name);
295 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
296 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
305 * Short-cut version of permission(), for calling by
306 * path_walk(), when dcache lock is held. Combines parts
307 * of permission() and vfs_permission(), and tests ONLY for
308 * MAY_EXEC permission.
310 * If appropriate, check DAC only. If not appropriate, or
311 * short-cut DAC fails, then call permission() to do more
312 * complete permission check.
314 static inline int exec_permission_lite(struct inode *inode,
315 struct nameidata *nd)
317 umode_t mode = inode->i_mode;
319 if ((inode->i_op && inode->i_op->permission))
322 if (current->fsuid == inode->i_uid)
324 else if (in_group_p(inode->i_gid))
330 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
333 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
338 return security_inode_permission(inode, MAY_EXEC, nd);
342 * This is called when everything else fails, and we actually have
343 * to go to the low-level filesystem to find out what we should do..
345 * We get the directory semaphore, and after getting that we also
346 * make sure that nobody added the entry to the dcache in the meantime..
349 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
351 struct dentry * result;
352 struct inode *dir = parent->d_inode;
356 * First re-do the cached lookup just in case it was created
357 * while we waited for the directory semaphore..
359 * FIXME! This could use version numbering or similar to
360 * avoid unnecessary cache lookups.
362 * The "dcache_lock" is purely to protect the RCU list walker
363 * from concurrent renames at this point (we mustn't get false
364 * negatives from the RCU list walk here, unlike the optimistic
367 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
369 result = d_lookup(parent, name);
371 struct dentry * dentry = d_alloc(parent, name);
372 result = ERR_PTR(-ENOMEM);
374 result = dir->i_op->lookup(dir, dentry, nd);
385 * Uhhuh! Nasty case: the cache was re-populated while
386 * we waited on the semaphore. Need to revalidate.
389 if (result->d_op && result->d_op->d_revalidate) {
390 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
392 result = ERR_PTR(-ENOENT);
399 * This limits recursive symlink follows to 8, while
400 * limiting consecutive symlinks to 40.
402 * Without that kind of total limit, nasty chains of consecutive
403 * symlinks can cause almost arbitrarily long lookups.
405 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
408 if (current->link_count >= 5)
410 if (current->total_link_count >= 40)
413 err = security_inode_follow_link(dentry, nd);
416 current->link_count++;
417 current->total_link_count++;
418 touch_atime(nd->mnt, dentry);
419 err = dentry->d_inode->i_op->follow_link(dentry, nd);
420 current->link_count--;
427 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
429 struct vfsmount *parent;
430 struct dentry *mountpoint;
431 spin_lock(&vfsmount_lock);
432 parent=(*mnt)->mnt_parent;
433 if (parent == *mnt) {
434 spin_unlock(&vfsmount_lock);
438 mountpoint=dget((*mnt)->mnt_mountpoint);
439 spin_unlock(&vfsmount_lock);
441 *dentry = mountpoint;
447 /* no need for dcache_lock, as serialization is taken care in
450 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
453 while (d_mountpoint(*dentry)) {
454 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
460 *dentry = dget(mounted->mnt_root);
466 /* no need for dcache_lock, as serialization is taken care in
469 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
471 struct vfsmount *mounted;
473 mounted = lookup_mnt(*mnt, *dentry);
478 *dentry = dget(mounted->mnt_root);
484 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
486 return __follow_down(mnt,dentry);
489 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
492 struct vfsmount *parent;
493 struct dentry *old = *dentry;
495 read_lock(¤t->fs->lock);
496 if (*dentry == current->fs->root &&
497 *mnt == current->fs->rootmnt) {
498 read_unlock(¤t->fs->lock);
501 read_unlock(¤t->fs->lock);
502 spin_lock(&dcache_lock);
503 if (*dentry != (*mnt)->mnt_root) {
504 *dentry = dget((*dentry)->d_parent);
505 spin_unlock(&dcache_lock);
509 spin_unlock(&dcache_lock);
510 spin_lock(&vfsmount_lock);
511 parent = (*mnt)->mnt_parent;
512 if (parent == *mnt) {
513 spin_unlock(&vfsmount_lock);
517 *dentry = dget((*mnt)->mnt_mountpoint);
518 spin_unlock(&vfsmount_lock);
523 follow_mount(mnt, dentry);
527 struct vfsmount *mnt;
528 struct dentry *dentry;
532 * It's more convoluted than I'd like it to be, but... it's still fairly
533 * small and for now I'd prefer to have fast path as straight as possible.
534 * It _is_ time-critical.
536 static int do_lookup(struct nameidata *nd, struct qstr *name,
539 struct vfsmount *mnt = nd->mnt;
540 struct dentry *dentry = __d_lookup(nd->dentry, name);
544 if (dentry->d_op && dentry->d_op->d_revalidate)
545 goto need_revalidate;
548 path->dentry = dentry;
552 dentry = real_lookup(nd->dentry, name, nd);
558 if (dentry->d_op->d_revalidate(dentry, nd))
560 if (d_invalidate(dentry))
566 return PTR_ERR(dentry);
572 * This is the basic name resolution function, turning a pathname
573 * into the final dentry.
575 * We expect 'base' to be positive and a directory.
577 int fastcall link_path_walk(const char * name, struct nameidata *nd)
582 unsigned int lookup_flags = nd->flags;
589 inode = nd->dentry->d_inode;
590 if (current->link_count)
591 lookup_flags = LOOKUP_FOLLOW;
593 /* At this point we know we have a real path component. */
599 err = exec_permission_lite(inode, nd);
600 if (err == -EAGAIN) {
601 err = permission(inode, MAY_EXEC, nd);
607 c = *(const unsigned char *)name;
609 hash = init_name_hash();
612 hash = partial_name_hash(c, hash);
613 c = *(const unsigned char *)name;
614 } while (c && (c != '/'));
615 this.len = name - (const char *) this.name;
616 this.hash = end_name_hash(hash);
618 /* remove trailing slashes? */
621 while (*++name == '/');
623 goto last_with_slashes;
626 * "." and ".." are special - ".." especially so because it has
627 * to be able to know about the current root directory and
628 * parent relationships.
630 if (this.name[0] == '.') switch (this.len) {
634 if (this.name[1] != '.')
636 follow_dotdot(&nd->mnt, &nd->dentry);
637 inode = nd->dentry->d_inode;
643 * See if the low-level filesystem might want
644 * to use its own hash..
646 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
647 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
651 nd->flags |= LOOKUP_CONTINUE;
652 /* This does the actual lookups.. */
653 err = do_lookup(nd, &this, &next);
656 /* Check mountpoints.. */
657 follow_mount(&next.mnt, &next.dentry);
660 inode = next.dentry->d_inode;
667 if (inode->i_op->follow_link) {
669 err = do_follow_link(next.dentry, nd);
675 inode = nd->dentry->d_inode;
684 nd->dentry = next.dentry;
687 if (!inode->i_op->lookup)
690 /* here ends the main loop */
693 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
695 nd->flags &= ~LOOKUP_CONTINUE;
696 if (lookup_flags & LOOKUP_PARENT)
698 if (this.name[0] == '.') switch (this.len) {
702 if (this.name[1] != '.')
704 follow_dotdot(&nd->mnt, &nd->dentry);
705 inode = nd->dentry->d_inode;
710 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
711 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
715 err = do_lookup(nd, &this, &next);
718 follow_mount(&next.mnt, &next.dentry);
719 inode = next.dentry->d_inode;
720 if ((lookup_flags & LOOKUP_FOLLOW)
721 && inode && inode->i_op && inode->i_op->follow_link) {
723 err = do_follow_link(next.dentry, nd);
728 inode = nd->dentry->d_inode;
732 nd->dentry = next.dentry;
737 if (lookup_flags & LOOKUP_DIRECTORY) {
739 if (!inode->i_op || !inode->i_op->lookup)
745 nd->last_type = LAST_NORM;
746 if (this.name[0] != '.')
749 nd->last_type = LAST_DOT;
750 else if (this.len == 2 && this.name[1] == '.')
751 nd->last_type = LAST_DOTDOT;
756 * We bypassed the ordinary revalidation routines.
757 * We may need to check the cached dentry for staleness.
759 if (nd->dentry && nd->dentry->d_sb &&
760 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
762 /* Note: we do not d_invalidate() */
763 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
777 int fastcall path_walk(const char * name, struct nameidata *nd)
779 current->total_link_count = 0;
780 return link_path_walk(name, nd);
784 /* returns 1 if everything is done */
785 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
787 if (path_walk(name, nd))
788 return 0; /* something went wrong... */
790 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
791 struct nameidata nd_root;
793 * NAME was not found in alternate root or it's a directory. Try to find
794 * it in the normal root:
796 nd_root.last_type = LAST_ROOT;
797 nd_root.flags = nd->flags;
798 memcpy(&nd_root.intent, &nd->intent, sizeof(nd_root.intent));
799 read_lock(¤t->fs->lock);
800 nd_root.mnt = mntget(current->fs->rootmnt);
801 nd_root.dentry = dget(current->fs->root);
802 read_unlock(¤t->fs->lock);
803 if (path_walk(name, &nd_root))
805 if (nd_root.dentry->d_inode) {
807 nd->dentry = nd_root.dentry;
808 nd->mnt = nd_root.mnt;
809 nd->last = nd_root.last;
812 path_release(&nd_root);
817 void set_fs_altroot(void)
819 char *emul = __emul_prefix();
821 struct vfsmount *mnt = NULL, *oldmnt;
822 struct dentry *dentry = NULL, *olddentry;
827 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
833 write_lock(¤t->fs->lock);
834 oldmnt = current->fs->altrootmnt;
835 olddentry = current->fs->altroot;
836 current->fs->altrootmnt = mnt;
837 current->fs->altroot = dentry;
838 write_unlock(¤t->fs->lock);
847 walk_init_root(const char *name, struct nameidata *nd)
849 read_lock(¤t->fs->lock);
850 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
851 nd->mnt = mntget(current->fs->altrootmnt);
852 nd->dentry = dget(current->fs->altroot);
853 read_unlock(¤t->fs->lock);
854 if (__emul_lookup_dentry(name,nd))
856 read_lock(¤t->fs->lock);
858 nd->mnt = mntget(current->fs->rootmnt);
859 nd->dentry = dget(current->fs->root);
860 read_unlock(¤t->fs->lock);
864 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
868 nd->last_type = LAST_ROOT; /* if there are only slashes... */
871 read_lock(¤t->fs->lock);
873 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
874 nd->mnt = mntget(current->fs->altrootmnt);
875 nd->dentry = dget(current->fs->altroot);
876 read_unlock(¤t->fs->lock);
877 if (__emul_lookup_dentry(name,nd))
879 read_lock(¤t->fs->lock);
881 nd->mnt = mntget(current->fs->rootmnt);
882 nd->dentry = dget(current->fs->root);
885 nd->mnt = mntget(current->fs->pwdmnt);
886 nd->dentry = dget(current->fs->pwd);
888 read_unlock(¤t->fs->lock);
889 current->total_link_count = 0;
890 retval = link_path_walk(name, nd);
891 if (unlikely(current->audit_context
892 && nd && nd->dentry && nd->dentry->d_inode))
894 nd->dentry->d_inode->i_ino,
895 nd->dentry->d_inode->i_rdev);
900 * Restricted form of lookup. Doesn't follow links, single-component only,
901 * needs parent already locked. Doesn't follow mounts.
904 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
906 struct dentry * dentry;
910 inode = base->d_inode;
911 err = permission(inode, MAY_EXEC, nd);
912 dentry = ERR_PTR(err);
917 * See if the low-level filesystem might want
918 * to use its own hash..
920 if (base->d_op && base->d_op->d_hash) {
921 err = base->d_op->d_hash(base, name);
922 dentry = ERR_PTR(err);
927 dentry = cached_lookup(base, name, nd);
929 struct dentry *new = d_alloc(base, name);
930 dentry = ERR_PTR(-ENOMEM);
933 dentry = inode->i_op->lookup(inode, new, nd);
943 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
945 return __lookup_hash(name, base, NULL);
949 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
960 hash = init_name_hash();
962 c = *(const unsigned char *)name++;
963 if (c == '/' || c == '\0')
965 hash = partial_name_hash(c, hash);
967 this.hash = end_name_hash(hash);
969 return lookup_hash(&this, base);
971 return ERR_PTR(-EACCES);
977 * is used by most simple commands to get the inode of a specified name.
978 * Open, link etc use their own routines, but this is enough for things
981 * namei exists in two versions: namei/lnamei. The only difference is
982 * that namei follows links, while lnamei does not.
985 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
987 char *tmp = getname(name);
988 int err = PTR_ERR(tmp);
991 err = path_lookup(tmp, flags, nd);
998 * It's inline, so penalty for filesystems that don't use sticky bit is
1001 static inline int check_sticky(struct inode *dir, struct inode *inode)
1003 if (!(dir->i_mode & S_ISVTX))
1005 if (inode->i_uid == current->fsuid)
1007 if (dir->i_uid == current->fsuid)
1009 return !capable(CAP_FOWNER);
1013 * Check whether we can remove a link victim from directory dir, check
1014 * whether the type of victim is right.
1015 * 1. We can't do it if dir is read-only (done in permission())
1016 * 2. We should have write and exec permissions on dir
1017 * 3. We can't remove anything from append-only dir
1018 * 4. We can't do anything with immutable dir (done in permission())
1019 * 5. If the sticky bit on dir is set we should either
1020 * a. be owner of dir, or
1021 * b. be owner of victim, or
1022 * c. have CAP_FOWNER capability
1023 * 6. If the victim is append-only or immutable we can't do antyhing with
1024 * links pointing to it.
1025 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1026 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1027 * 9. We can't remove a root or mountpoint.
1028 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1029 * nfs_async_unlink().
1031 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1034 if (!victim->d_inode || victim->d_parent->d_inode != dir)
1036 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1041 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1042 IS_IMMUTABLE(victim->d_inode))
1045 if (!S_ISDIR(victim->d_inode->i_mode))
1047 if (IS_ROOT(victim))
1049 } else if (S_ISDIR(victim->d_inode->i_mode))
1051 if (IS_DEADDIR(dir))
1053 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1058 /* Check whether we can create an object with dentry child in directory
1060 * 1. We can't do it if child already exists (open has special treatment for
1061 * this case, but since we are inlined it's OK)
1062 * 2. We can't do it if dir is read-only (done in permission())
1063 * 3. We should have write and exec permissions on dir
1064 * 4. We can't do it if dir is immutable (done in permission())
1066 static inline int may_create(struct inode *dir, struct dentry *child,
1067 struct nameidata *nd)
1071 if (IS_DEADDIR(dir))
1073 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1077 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1080 * O_DIRECTORY translates into forcing a directory lookup.
1082 static inline int lookup_flags(unsigned int f)
1084 unsigned long retval = LOOKUP_FOLLOW;
1087 retval &= ~LOOKUP_FOLLOW;
1089 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1090 retval &= ~LOOKUP_FOLLOW;
1092 if (f & O_DIRECTORY)
1093 retval |= LOOKUP_DIRECTORY;
1099 * p1 and p2 should be directories on the same fs.
1101 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1106 down(&p1->d_inode->i_sem);
1110 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1112 for (p = p1; p->d_parent != p; p = p->d_parent) {
1113 if (p->d_parent == p2) {
1114 down(&p2->d_inode->i_sem);
1115 down(&p1->d_inode->i_sem);
1120 for (p = p2; p->d_parent != p; p = p->d_parent) {
1121 if (p->d_parent == p1) {
1122 down(&p1->d_inode->i_sem);
1123 down(&p2->d_inode->i_sem);
1128 down(&p1->d_inode->i_sem);
1129 down(&p2->d_inode->i_sem);
1133 void unlock_rename(struct dentry *p1, struct dentry *p2)
1135 up(&p1->d_inode->i_sem);
1137 up(&p2->d_inode->i_sem);
1138 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1142 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1143 struct nameidata *nd)
1145 int error = may_create(dir, dentry, nd);
1150 if (!dir->i_op || !dir->i_op->create)
1151 return -EACCES; /* shouldn't it be ENOSYS? */
1154 error = security_inode_create(dir, dentry, mode);
1158 error = dir->i_op->create(dir, dentry, mode, nd);
1160 inode_dir_notify(dir, DN_CREATE);
1161 security_inode_post_create(dir, dentry, mode);
1166 int may_open(struct nameidata *nd, int acc_mode, int flag)
1168 struct dentry *dentry = nd->dentry;
1169 struct inode *inode = dentry->d_inode;
1175 if (S_ISLNK(inode->i_mode))
1178 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1181 error = permission(inode, acc_mode, nd);
1186 * FIFO's, sockets and device files are special: they don't
1187 * actually live on the filesystem itself, and as such you
1188 * can write to them even if the filesystem is read-only.
1190 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1192 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1193 if (nd->mnt->mnt_flags & MNT_NODEV)
1197 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1200 * An append-only file must be opened in append mode for writing.
1202 if (IS_APPEND(inode)) {
1203 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1210 * Ensure there are no outstanding leases on the file.
1212 error = break_lease(inode, flag);
1216 if (flag & O_TRUNC) {
1217 error = get_write_access(inode);
1222 * Refuse to truncate files with mandatory locks held on them.
1224 error = locks_verify_locked(inode);
1228 error = do_truncate(dentry, 0);
1230 put_write_access(inode);
1234 if (flag & FMODE_WRITE)
1243 * namei for open - this is in fact almost the whole open-routine.
1245 * Note that the low bits of "flag" aren't the same as in the open
1246 * system call - they are 00 - no permissions needed
1247 * 01 - read permission needed
1248 * 10 - write permission needed
1249 * 11 - read/write permissions needed
1250 * which is a lot more logical, and also allows the "no perm" needed
1251 * for symlinks (where the permissions are checked later).
1254 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1256 int acc_mode, error = 0;
1257 struct dentry *dentry;
1261 acc_mode = ACC_MODE(flag);
1263 /* Allow the LSM permission hook to distinguish append
1264 access from general write access. */
1265 if (flag & O_APPEND)
1266 acc_mode |= MAY_APPEND;
1268 /* Fill in the open() intent data */
1269 nd->intent.open.flags = flag;
1270 nd->intent.open.create_mode = mode;
1273 * The simplest case - just a plain lookup.
1275 if (!(flag & O_CREAT)) {
1276 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1283 * Create - we need to know the parent.
1285 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1290 * We have the parent and last component. First of all, check
1291 * that we are not asked to creat(2) an obvious directory - that
1295 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1299 nd->flags &= ~LOOKUP_PARENT;
1300 down(&dir->d_inode->i_sem);
1301 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1304 error = PTR_ERR(dentry);
1305 if (IS_ERR(dentry)) {
1306 up(&dir->d_inode->i_sem);
1310 /* Negative dentry, just create the file */
1311 if (!dentry->d_inode) {
1312 if (!IS_POSIXACL(dir->d_inode))
1313 mode &= ~current->fs->umask;
1314 error = vfs_create(dir->d_inode, dentry, mode, nd);
1315 up(&dir->d_inode->i_sem);
1317 nd->dentry = dentry;
1320 /* Don't check for write permission, don't truncate */
1327 * It already exists.
1329 up(&dir->d_inode->i_sem);
1335 if (d_mountpoint(dentry)) {
1337 if (flag & O_NOFOLLOW)
1339 while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
1342 if (!dentry->d_inode)
1344 if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1348 nd->dentry = dentry;
1350 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1353 error = may_open(nd, acc_mode, flag);
1366 if (flag & O_NOFOLLOW)
1369 * This is subtle. Instead of calling do_follow_link() we do the
1370 * thing by hands. The reason is that this way we have zero link_count
1371 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1372 * After that we have the parent and last component, i.e.
1373 * we are in the same situation as after the first path_walk().
1374 * Well, almost - if the last component is normal we get its copy
1375 * stored in nd->last.name and we will have to putname() it when we
1376 * are done. Procfs-like symlinks just set LAST_BIND.
1378 nd->flags |= LOOKUP_PARENT;
1379 error = security_inode_follow_link(dentry, nd);
1382 touch_atime(nd->mnt, dentry);
1383 error = dentry->d_inode->i_op->follow_link(dentry, nd);
1387 nd->flags &= ~LOOKUP_PARENT;
1388 if (nd->last_type == LAST_BIND) {
1389 dentry = nd->dentry;
1393 if (nd->last_type != LAST_NORM)
1395 if (nd->last.name[nd->last.len]) {
1396 putname(nd->last.name);
1401 putname(nd->last.name);
1405 down(&dir->d_inode->i_sem);
1406 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1407 putname(nd->last.name);
1412 * lookup_create - lookup a dentry, creating it if it doesn't exist
1413 * @nd: nameidata info
1414 * @is_dir: directory flag
1416 * Simple function to lookup and return a dentry and create it
1417 * if it doesn't exist. Is SMP-safe.
1419 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1421 struct dentry *dentry;
1423 down(&nd->dentry->d_inode->i_sem);
1424 dentry = ERR_PTR(-EEXIST);
1425 if (nd->last_type != LAST_NORM)
1427 nd->flags &= ~LOOKUP_PARENT;
1428 dentry = lookup_hash(&nd->last, nd->dentry);
1431 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1436 dentry = ERR_PTR(-ENOENT);
1441 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1443 int error = may_create(dir, dentry, NULL);
1448 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1451 if (!dir->i_op || !dir->i_op->mknod)
1454 error = security_inode_mknod(dir, dentry, mode, dev);
1459 error = dir->i_op->mknod(dir, dentry, mode, dev);
1461 inode_dir_notify(dir, DN_CREATE);
1462 security_inode_post_mknod(dir, dentry, mode, dev);
1467 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1471 struct dentry * dentry;
1472 struct nameidata nd;
1476 tmp = getname(filename);
1478 return PTR_ERR(tmp);
1480 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1483 dentry = lookup_create(&nd, 0);
1484 error = PTR_ERR(dentry);
1486 if (!IS_POSIXACL(nd.dentry->d_inode))
1487 mode &= ~current->fs->umask;
1488 if (!IS_ERR(dentry)) {
1489 switch (mode & S_IFMT) {
1490 case 0: case S_IFREG:
1491 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1493 case S_IFCHR: case S_IFBLK:
1494 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1495 new_decode_dev(dev));
1497 case S_IFIFO: case S_IFSOCK:
1498 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1508 up(&nd.dentry->d_inode->i_sem);
1516 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1518 int error = may_create(dir, dentry, NULL);
1523 if (!dir->i_op || !dir->i_op->mkdir)
1526 mode &= (S_IRWXUGO|S_ISVTX);
1527 error = security_inode_mkdir(dir, dentry, mode);
1532 error = dir->i_op->mkdir(dir, dentry, mode);
1534 inode_dir_notify(dir, DN_CREATE);
1535 security_inode_post_mkdir(dir,dentry, mode);
1540 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1545 tmp = getname(pathname);
1546 error = PTR_ERR(tmp);
1548 struct dentry *dentry;
1549 struct nameidata nd;
1551 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1554 dentry = lookup_create(&nd, 1);
1555 error = PTR_ERR(dentry);
1556 if (!IS_ERR(dentry)) {
1557 if (!IS_POSIXACL(nd.dentry->d_inode))
1558 mode &= ~current->fs->umask;
1559 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1562 up(&nd.dentry->d_inode->i_sem);
1572 * We try to drop the dentry early: we should have
1573 * a usage count of 2 if we're the only user of this
1574 * dentry, and if that is true (possibly after pruning
1575 * the dcache), then we drop the dentry now.
1577 * A low-level filesystem can, if it choses, legally
1580 * if (!d_unhashed(dentry))
1583 * if it cannot handle the case of removing a directory
1584 * that is still in use by something else..
1586 static void d_unhash(struct dentry *dentry)
1589 spin_lock(&dcache_lock);
1590 switch (atomic_read(&dentry->d_count)) {
1592 spin_unlock(&dcache_lock);
1593 shrink_dcache_parent(dentry);
1594 spin_lock(&dcache_lock);
1595 if (atomic_read(&dentry->d_count) != 2)
1600 spin_unlock(&dcache_lock);
1603 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1605 int error = may_delete(dir, dentry, 1);
1610 if (!dir->i_op || !dir->i_op->rmdir)
1615 down(&dentry->d_inode->i_sem);
1617 if (d_mountpoint(dentry))
1620 error = security_inode_rmdir(dir, dentry);
1622 error = dir->i_op->rmdir(dir, dentry);
1624 dentry->d_inode->i_flags |= S_DEAD;
1627 up(&dentry->d_inode->i_sem);
1629 inode_dir_notify(dir, DN_DELETE);
1637 asmlinkage long sys_rmdir(const char __user * pathname)
1641 struct dentry *dentry;
1642 struct nameidata nd;
1644 name = getname(pathname);
1646 return PTR_ERR(name);
1648 error = path_lookup(name, LOOKUP_PARENT, &nd);
1652 switch(nd.last_type) {
1663 down(&nd.dentry->d_inode->i_sem);
1664 dentry = lookup_hash(&nd.last, nd.dentry);
1665 error = PTR_ERR(dentry);
1666 if (!IS_ERR(dentry)) {
1667 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1670 up(&nd.dentry->d_inode->i_sem);
1678 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1680 int error = may_delete(dir, dentry, 0);
1685 if (!dir->i_op || !dir->i_op->unlink)
1690 down(&dentry->d_inode->i_sem);
1691 if (d_mountpoint(dentry))
1694 error = security_inode_unlink(dir, dentry);
1696 error = dir->i_op->unlink(dir, dentry);
1698 up(&dentry->d_inode->i_sem);
1700 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1701 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1703 inode_dir_notify(dir, DN_DELETE);
1709 * Make sure that the actual truncation of the file will occur outside its
1710 * directory's i_sem. Truncate can take a long time if there is a lot of
1711 * writeout happening, and we don't want to prevent access to the directory
1712 * while waiting on the I/O.
1714 asmlinkage long sys_unlink(const char __user * pathname)
1718 struct dentry *dentry;
1719 struct nameidata nd;
1720 struct inode *inode = NULL;
1722 name = getname(pathname);
1724 return PTR_ERR(name);
1726 error = path_lookup(name, LOOKUP_PARENT, &nd);
1730 if (nd.last_type != LAST_NORM)
1732 down(&nd.dentry->d_inode->i_sem);
1733 dentry = lookup_hash(&nd.last, nd.dentry);
1734 error = PTR_ERR(dentry);
1735 if (!IS_ERR(dentry)) {
1736 /* Why not before? Because we want correct error value */
1737 if (nd.last.name[nd.last.len])
1739 inode = dentry->d_inode;
1741 atomic_inc(&inode->i_count);
1742 error = vfs_unlink(nd.dentry->d_inode, dentry);
1746 up(&nd.dentry->d_inode->i_sem);
1753 iput(inode); /* truncate the inode here */
1757 error = !dentry->d_inode ? -ENOENT :
1758 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1762 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1764 int error = may_create(dir, dentry, NULL);
1769 if (!dir->i_op || !dir->i_op->symlink)
1772 error = security_inode_symlink(dir, dentry, oldname);
1777 error = dir->i_op->symlink(dir, dentry, oldname);
1779 inode_dir_notify(dir, DN_CREATE);
1780 security_inode_post_symlink(dir, dentry, oldname);
1785 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1791 from = getname(oldname);
1793 return PTR_ERR(from);
1794 to = getname(newname);
1795 error = PTR_ERR(to);
1797 struct dentry *dentry;
1798 struct nameidata nd;
1800 error = path_lookup(to, LOOKUP_PARENT, &nd);
1803 dentry = lookup_create(&nd, 0);
1804 error = PTR_ERR(dentry);
1805 if (!IS_ERR(dentry)) {
1806 error = vfs_symlink(nd.dentry->d_inode, dentry, from);
1809 up(&nd.dentry->d_inode->i_sem);
1818 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1820 struct inode *inode = old_dentry->d_inode;
1826 error = may_create(dir, new_dentry, NULL);
1830 if (dir->i_sb != inode->i_sb)
1834 * A link to an append-only or immutable file cannot be created.
1836 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1838 if (!dir->i_op || !dir->i_op->link)
1840 if (S_ISDIR(old_dentry->d_inode->i_mode))
1843 error = security_inode_link(old_dentry, dir, new_dentry);
1847 down(&old_dentry->d_inode->i_sem);
1849 error = dir->i_op->link(old_dentry, dir, new_dentry);
1850 up(&old_dentry->d_inode->i_sem);
1852 inode_dir_notify(dir, DN_CREATE);
1853 security_inode_post_link(old_dentry, dir, new_dentry);
1859 * Hardlinks are often used in delicate situations. We avoid
1860 * security-related surprises by not following symlinks on the
1863 * We don't follow them on the oldname either to be compatible
1864 * with linux 2.0, and to avoid hard-linking to directories
1865 * and other special files. --ADM
1867 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
1869 struct dentry *new_dentry;
1870 struct nameidata nd, old_nd;
1874 to = getname(newname);
1878 error = __user_walk(oldname, 0, &old_nd);
1881 error = path_lookup(to, LOOKUP_PARENT, &nd);
1885 if (old_nd.mnt != nd.mnt)
1887 new_dentry = lookup_create(&nd, 0);
1888 error = PTR_ERR(new_dentry);
1889 if (!IS_ERR(new_dentry)) {
1890 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
1893 up(&nd.dentry->d_inode->i_sem);
1897 path_release(&old_nd);
1905 * The worst of all namespace operations - renaming directory. "Perverted"
1906 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
1908 * a) we can get into loop creation. Check is done in is_subdir().
1909 * b) race potential - two innocent renames can create a loop together.
1910 * That's where 4.4 screws up. Current fix: serialization on
1911 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
1913 * c) we have to lock _three_ objects - parents and victim (if it exists).
1914 * And that - after we got ->i_sem on parents (until then we don't know
1915 * whether the target exists). Solution: try to be smart with locking
1916 * order for inodes. We rely on the fact that tree topology may change
1917 * only under ->s_vfs_rename_sem _and_ that parent of the object we
1918 * move will be locked. Thus we can rank directories by the tree
1919 * (ancestors first) and rank all non-directories after them.
1920 * That works since everybody except rename does "lock parent, lookup,
1921 * lock child" and rename is under ->s_vfs_rename_sem.
1922 * HOWEVER, it relies on the assumption that any object with ->lookup()
1923 * has no more than 1 dentry. If "hybrid" objects will ever appear,
1924 * we'd better make sure that there's no link(2) for them.
1925 * d) some filesystems don't support opened-but-unlinked directories,
1926 * either because of layout or because they are not ready to deal with
1927 * all cases correctly. The latter will be fixed (taking this sort of
1928 * stuff into VFS), but the former is not going away. Solution: the same
1929 * trick as in rmdir().
1930 * e) conversion from fhandle to dentry may come in the wrong moment - when
1931 * we are removing the target. Solution: we will have to grab ->i_sem
1932 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1933 * ->i_sem on parents, which works but leads to some truely excessive
1936 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
1937 struct inode *new_dir, struct dentry *new_dentry)
1940 struct inode *target;
1943 * If we are going to change the parent - check write permissions,
1944 * we'll need to flip '..'.
1946 if (new_dir != old_dir) {
1947 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
1952 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
1956 target = new_dentry->d_inode;
1958 down(&target->i_sem);
1959 d_unhash(new_dentry);
1961 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
1964 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1967 target->i_flags |= S_DEAD;
1969 if (d_unhashed(new_dentry))
1970 d_rehash(new_dentry);
1974 d_move(old_dentry,new_dentry);
1975 security_inode_post_rename(old_dir, old_dentry,
1976 new_dir, new_dentry);
1981 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
1982 struct inode *new_dir, struct dentry *new_dentry)
1984 struct inode *target;
1987 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
1992 target = new_dentry->d_inode;
1994 down(&target->i_sem);
1995 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
1998 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2000 /* The following d_move() should become unconditional */
2001 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2002 d_move(old_dentry, new_dentry);
2003 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2011 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2012 struct inode *new_dir, struct dentry *new_dentry)
2015 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2017 if (old_dentry->d_inode == new_dentry->d_inode)
2020 error = may_delete(old_dir, old_dentry, is_dir);
2024 if (!new_dentry->d_inode)
2025 error = may_create(new_dir, new_dentry, NULL);
2027 error = may_delete(new_dir, new_dentry, is_dir);
2031 if (!old_dir->i_op || !old_dir->i_op->rename)
2034 DQUOT_INIT(old_dir);
2035 DQUOT_INIT(new_dir);
2038 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2040 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2042 if (old_dir == new_dir)
2043 inode_dir_notify(old_dir, DN_RENAME);
2045 inode_dir_notify(old_dir, DN_DELETE);
2046 inode_dir_notify(new_dir, DN_CREATE);
2052 static inline int do_rename(const char * oldname, const char * newname)
2055 struct dentry * old_dir, * new_dir;
2056 struct dentry * old_dentry, *new_dentry;
2057 struct dentry * trap;
2058 struct nameidata oldnd, newnd;
2060 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2064 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2069 if (oldnd.mnt != newnd.mnt)
2072 old_dir = oldnd.dentry;
2074 if (oldnd.last_type != LAST_NORM)
2077 new_dir = newnd.dentry;
2078 if (newnd.last_type != LAST_NORM)
2081 trap = lock_rename(new_dir, old_dir);
2083 old_dentry = lookup_hash(&oldnd.last, old_dir);
2084 error = PTR_ERR(old_dentry);
2085 if (IS_ERR(old_dentry))
2087 /* source must exist */
2089 if (!old_dentry->d_inode)
2091 /* unless the source is a directory trailing slashes give -ENOTDIR */
2092 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2094 if (oldnd.last.name[oldnd.last.len])
2096 if (newnd.last.name[newnd.last.len])
2099 /* source should not be ancestor of target */
2101 if (old_dentry == trap)
2103 new_dentry = lookup_hash(&newnd.last, new_dir);
2104 error = PTR_ERR(new_dentry);
2105 if (IS_ERR(new_dentry))
2107 /* target should not be an ancestor of source */
2109 if (new_dentry == trap)
2112 error = vfs_rename(old_dir->d_inode, old_dentry,
2113 new_dir->d_inode, new_dentry);
2119 unlock_rename(new_dir, old_dir);
2121 path_release(&newnd);
2123 path_release(&oldnd);
2128 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2134 from = getname(oldname);
2136 return PTR_ERR(from);
2137 to = getname(newname);
2138 error = PTR_ERR(to);
2140 error = do_rename(from,to);
2147 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2151 len = PTR_ERR(link);
2156 if (len > (unsigned) buflen)
2158 if (copy_to_user(buffer, link, len))
2165 __vfs_follow_link(struct nameidata *nd, const char *link)
2174 if (!walk_init_root(link, nd))
2175 /* weird __emul_prefix() stuff did it */
2178 res = link_path_walk(link, nd);
2180 if (current->link_count || res || nd->last_type!=LAST_NORM)
2183 * If it is an iterative symlinks resolution in open_namei() we
2184 * have to copy the last component. And all that crap because of
2185 * bloody create() on broken symlinks. Furrfu...
2188 if (unlikely(!name)) {
2192 strcpy(name, nd->last.name);
2193 nd->last.name = name;
2197 return PTR_ERR(link);
2200 int vfs_follow_link(struct nameidata *nd, const char *link)
2202 return __vfs_follow_link(nd, link);
2205 /* get the link contents into pagecache */
2206 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2209 struct address_space *mapping = dentry->d_inode->i_mapping;
2210 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2214 wait_on_page_locked(page);
2215 if (!PageUptodate(page))
2221 page_cache_release(page);
2222 return ERR_PTR(-EIO);
2228 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2230 struct page *page = NULL;
2231 char *s = page_getlink(dentry, &page);
2232 int res = vfs_readlink(dentry,buffer,buflen,s);
2235 page_cache_release(page);
2240 int page_follow_link(struct dentry *dentry, struct nameidata *nd)
2242 struct page *page = NULL;
2243 char *s = page_getlink(dentry, &page);
2244 int res = __vfs_follow_link(nd, s);
2247 page_cache_release(page);
2252 int page_symlink(struct inode *inode, const char *symname, int len)
2254 struct address_space *mapping = inode->i_mapping;
2255 struct page *page = grab_cache_page(mapping, 0);
2261 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2264 kaddr = kmap_atomic(page, KM_USER0);
2265 memcpy(kaddr, symname, len-1);
2266 kunmap_atomic(kaddr, KM_USER0);
2267 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2269 * Notice that we are _not_ going to block here - end of page is
2270 * unmapped, so this will only try to map the rest of page, see
2271 * that it is unmapped (typically even will not look into inode -
2272 * ->i_size will be enough for everything) and zero it out.
2273 * OTOH it's obviously correct and should make the page up-to-date.
2275 if (!PageUptodate(page)) {
2276 err = mapping->a_ops->readpage(NULL, page);
2277 wait_on_page_locked(page);
2281 page_cache_release(page);
2284 mark_inode_dirty(inode);
2288 page_cache_release(page);
2293 struct inode_operations page_symlink_inode_operations = {
2294 .readlink = page_readlink,
2295 .follow_link = page_follow_link,
2298 EXPORT_SYMBOL(__user_walk);
2299 EXPORT_SYMBOL(follow_down);
2300 EXPORT_SYMBOL(follow_up);
2301 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2302 EXPORT_SYMBOL(getname);
2303 EXPORT_SYMBOL(lock_rename);
2304 EXPORT_SYMBOL(lookup_create);
2305 EXPORT_SYMBOL(lookup_hash);
2306 EXPORT_SYMBOL(lookup_one_len);
2307 EXPORT_SYMBOL(page_follow_link);
2308 EXPORT_SYMBOL(page_readlink);
2309 EXPORT_SYMBOL(page_symlink);
2310 EXPORT_SYMBOL(page_symlink_inode_operations);
2311 EXPORT_SYMBOL(path_lookup);
2312 EXPORT_SYMBOL(path_release);
2313 EXPORT_SYMBOL(path_walk);
2314 EXPORT_SYMBOL(permission);
2315 EXPORT_SYMBOL(unlock_rename);
2316 EXPORT_SYMBOL(vfs_create);
2317 EXPORT_SYMBOL(vfs_follow_link);
2318 EXPORT_SYMBOL(vfs_link);
2319 EXPORT_SYMBOL(vfs_mkdir);
2320 EXPORT_SYMBOL(vfs_mknod);
2321 EXPORT_SYMBOL(vfs_permission);
2322 EXPORT_SYMBOL(vfs_readlink);
2323 EXPORT_SYMBOL(vfs_rename);
2324 EXPORT_SYMBOL(vfs_rmdir);
2325 EXPORT_SYMBOL(vfs_symlink);
2326 EXPORT_SYMBOL(vfs_unlink);