[PATCH] arch/alpha: Use ARRAY_SIZE macro
[pohmelfs.git] / fs / namei.c
blobe01070d7bf58a39e27a50ade9fcfd598574f43aa
1 /*
2 * linux/fs/namei.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
8 * Some corrections by tytso.
9 */
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
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>
20 #include <linux/fs.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/namei.h>
35 #include <asm/namei.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
120 int retval;
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
140 char * getname(const char __user * filename)
142 char *tmp, *result;
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
155 audit_getname(result);
156 return result;
159 #ifdef CONFIG_AUDITSYSCALL
160 void putname(const char *name)
162 if (unlikely(current->audit_context))
163 audit_putname(name);
164 else
165 __putname(name);
167 EXPORT_SYMBOL(putname);
168 #endif
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
182 int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
185 umode_t mode = inode->i_mode;
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
203 * If the DACs are ok we don't need any capability check.
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
208 check_capabilities:
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
219 * Searching includes executable on directories, else just read.
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
225 return -EACCES;
228 int permission(struct inode *inode, int mask, struct nameidata *nd)
230 int retval, submask;
232 if (mask & MAY_WRITE) {
233 umode_t mode = inode->i_mode;
236 * Nobody gets write access to a read-only fs.
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
243 * Nobody gets write access to an immutable file.
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
250 /* Ordinary permission routines do not understand MAY_APPEND. */
251 submask = mask & ~MAY_APPEND;
252 if (inode->i_op && inode->i_op->permission)
253 retval = inode->i_op->permission(inode, submask, nd);
254 else
255 retval = generic_permission(inode, submask, NULL);
256 if (retval)
257 return retval;
259 return security_inode_permission(inode, mask, nd);
263 * vfs_permission - check for access rights to a given path
264 * @nd: lookup result that describes the path
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
267 * Used to check for read/write/execute permissions on a path.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
270 * are used for other things.
272 int vfs_permission(struct nameidata *nd, int mask)
274 return permission(nd->dentry->d_inode, mask, nd);
278 * file_permission - check for additional access rights to a given file
279 * @file: file to check access rights for
280 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
282 * Used to check for read/write/execute permissions on an already opened
283 * file.
285 * Note:
286 * Do not use this function in new code. All access checks should
287 * be done using vfs_permission().
289 int file_permission(struct file *file, int mask)
291 return permission(file->f_dentry->d_inode, mask, NULL);
295 * get_write_access() gets write permission for a file.
296 * put_write_access() releases this write permission.
297 * This is used for regular files.
298 * We cannot support write (and maybe mmap read-write shared) accesses and
299 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
300 * can have the following values:
301 * 0: no writers, no VM_DENYWRITE mappings
302 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
303 * > 0: (i_writecount) users are writing to the file.
305 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
306 * except for the cases where we don't hold i_writecount yet. Then we need to
307 * use {get,deny}_write_access() - these functions check the sign and refuse
308 * to do the change if sign is wrong. Exclusion between them is provided by
309 * the inode->i_lock spinlock.
312 int get_write_access(struct inode * inode)
314 spin_lock(&inode->i_lock);
315 if (atomic_read(&inode->i_writecount) < 0) {
316 spin_unlock(&inode->i_lock);
317 return -ETXTBSY;
319 atomic_inc(&inode->i_writecount);
320 spin_unlock(&inode->i_lock);
322 return 0;
325 int deny_write_access(struct file * file)
327 struct inode *inode = file->f_dentry->d_inode;
329 spin_lock(&inode->i_lock);
330 if (atomic_read(&inode->i_writecount) > 0) {
331 spin_unlock(&inode->i_lock);
332 return -ETXTBSY;
334 atomic_dec(&inode->i_writecount);
335 spin_unlock(&inode->i_lock);
337 return 0;
340 void path_release(struct nameidata *nd)
342 dput(nd->dentry);
343 mntput(nd->mnt);
347 * umount() mustn't call path_release()/mntput() as that would clear
348 * mnt_expiry_mark
350 void path_release_on_umount(struct nameidata *nd)
352 dput(nd->dentry);
353 mntput_no_expire(nd->mnt);
357 * release_open_intent - free up open intent resources
358 * @nd: pointer to nameidata
360 void release_open_intent(struct nameidata *nd)
362 if (nd->intent.open.file->f_dentry == NULL)
363 put_filp(nd->intent.open.file);
364 else
365 fput(nd->intent.open.file);
369 * Internal lookup() using the new generic dcache.
370 * SMP-safe
372 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
374 struct dentry * dentry = __d_lookup(parent, name);
376 /* lockess __d_lookup may fail due to concurrent d_move()
377 * in some unrelated directory, so try with d_lookup
379 if (!dentry)
380 dentry = d_lookup(parent, name);
382 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
383 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
384 dput(dentry);
385 dentry = NULL;
388 return dentry;
392 * Short-cut version of permission(), for calling by
393 * path_walk(), when dcache lock is held. Combines parts
394 * of permission() and generic_permission(), and tests ONLY for
395 * MAY_EXEC permission.
397 * If appropriate, check DAC only. If not appropriate, or
398 * short-cut DAC fails, then call permission() to do more
399 * complete permission check.
401 static int exec_permission_lite(struct inode *inode,
402 struct nameidata *nd)
404 umode_t mode = inode->i_mode;
406 if (inode->i_op && inode->i_op->permission)
407 return -EAGAIN;
409 if (current->fsuid == inode->i_uid)
410 mode >>= 6;
411 else if (in_group_p(inode->i_gid))
412 mode >>= 3;
414 if (mode & MAY_EXEC)
415 goto ok;
417 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
421 goto ok;
423 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
424 goto ok;
426 return -EACCES;
428 return security_inode_permission(inode, MAY_EXEC, nd);
432 * This is called when everything else fails, and we actually have
433 * to go to the low-level filesystem to find out what we should do..
435 * We get the directory semaphore, and after getting that we also
436 * make sure that nobody added the entry to the dcache in the meantime..
437 * SMP-safe
439 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
441 struct dentry * result;
442 struct inode *dir = parent->d_inode;
444 mutex_lock(&dir->i_mutex);
446 * First re-do the cached lookup just in case it was created
447 * while we waited for the directory semaphore..
449 * FIXME! This could use version numbering or similar to
450 * avoid unnecessary cache lookups.
452 * The "dcache_lock" is purely to protect the RCU list walker
453 * from concurrent renames at this point (we mustn't get false
454 * negatives from the RCU list walk here, unlike the optimistic
455 * fast walk).
457 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
459 result = d_lookup(parent, name);
460 if (!result) {
461 struct dentry * dentry = d_alloc(parent, name);
462 result = ERR_PTR(-ENOMEM);
463 if (dentry) {
464 result = dir->i_op->lookup(dir, dentry, nd);
465 if (result)
466 dput(dentry);
467 else
468 result = dentry;
470 mutex_unlock(&dir->i_mutex);
471 return result;
475 * Uhhuh! Nasty case: the cache was re-populated while
476 * we waited on the semaphore. Need to revalidate.
478 mutex_unlock(&dir->i_mutex);
479 if (result->d_op && result->d_op->d_revalidate) {
480 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
481 dput(result);
482 result = ERR_PTR(-ENOENT);
485 return result;
488 static int __emul_lookup_dentry(const char *, struct nameidata *);
490 /* SMP-safe */
491 static __always_inline int
492 walk_init_root(const char *name, struct nameidata *nd)
494 read_lock(&current->fs->lock);
495 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
496 nd->mnt = mntget(current->fs->altrootmnt);
497 nd->dentry = dget(current->fs->altroot);
498 read_unlock(&current->fs->lock);
499 if (__emul_lookup_dentry(name,nd))
500 return 0;
501 read_lock(&current->fs->lock);
503 nd->mnt = mntget(current->fs->rootmnt);
504 nd->dentry = dget(current->fs->root);
505 read_unlock(&current->fs->lock);
506 return 1;
509 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
511 int res = 0;
512 char *name;
513 if (IS_ERR(link))
514 goto fail;
516 if (*link == '/') {
517 path_release(nd);
518 if (!walk_init_root(link, nd))
519 /* weird __emul_prefix() stuff did it */
520 goto out;
522 res = link_path_walk(link, nd);
523 out:
524 if (nd->depth || res || nd->last_type!=LAST_NORM)
525 return res;
527 * If it is an iterative symlinks resolution in open_namei() we
528 * have to copy the last component. And all that crap because of
529 * bloody create() on broken symlinks. Furrfu...
531 name = __getname();
532 if (unlikely(!name)) {
533 path_release(nd);
534 return -ENOMEM;
536 strcpy(name, nd->last.name);
537 nd->last.name = name;
538 return 0;
539 fail:
540 path_release(nd);
541 return PTR_ERR(link);
544 struct path {
545 struct vfsmount *mnt;
546 struct dentry *dentry;
549 static inline void dput_path(struct path *path, struct nameidata *nd)
551 dput(path->dentry);
552 if (path->mnt != nd->mnt)
553 mntput(path->mnt);
556 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
558 dput(nd->dentry);
559 if (nd->mnt != path->mnt)
560 mntput(nd->mnt);
561 nd->mnt = path->mnt;
562 nd->dentry = path->dentry;
565 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
567 int error;
568 void *cookie;
569 struct dentry *dentry = path->dentry;
571 touch_atime(path->mnt, dentry);
572 nd_set_link(nd, NULL);
574 if (path->mnt != nd->mnt) {
575 path_to_nameidata(path, nd);
576 dget(dentry);
578 mntget(path->mnt);
579 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
580 error = PTR_ERR(cookie);
581 if (!IS_ERR(cookie)) {
582 char *s = nd_get_link(nd);
583 error = 0;
584 if (s)
585 error = __vfs_follow_link(nd, s);
586 if (dentry->d_inode->i_op->put_link)
587 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
589 dput(dentry);
590 mntput(path->mnt);
592 return error;
596 * This limits recursive symlink follows to 8, while
597 * limiting consecutive symlinks to 40.
599 * Without that kind of total limit, nasty chains of consecutive
600 * symlinks can cause almost arbitrarily long lookups.
602 static inline int do_follow_link(struct path *path, struct nameidata *nd)
604 int err = -ELOOP;
605 if (current->link_count >= MAX_NESTED_LINKS)
606 goto loop;
607 if (current->total_link_count >= 40)
608 goto loop;
609 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
610 cond_resched();
611 err = security_inode_follow_link(path->dentry, nd);
612 if (err)
613 goto loop;
614 current->link_count++;
615 current->total_link_count++;
616 nd->depth++;
617 err = __do_follow_link(path, nd);
618 current->link_count--;
619 nd->depth--;
620 return err;
621 loop:
622 dput_path(path, nd);
623 path_release(nd);
624 return err;
627 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
629 struct vfsmount *parent;
630 struct dentry *mountpoint;
631 spin_lock(&vfsmount_lock);
632 parent=(*mnt)->mnt_parent;
633 if (parent == *mnt) {
634 spin_unlock(&vfsmount_lock);
635 return 0;
637 mntget(parent);
638 mountpoint=dget((*mnt)->mnt_mountpoint);
639 spin_unlock(&vfsmount_lock);
640 dput(*dentry);
641 *dentry = mountpoint;
642 mntput(*mnt);
643 *mnt = parent;
644 return 1;
647 /* no need for dcache_lock, as serialization is taken care in
648 * namespace.c
650 static int __follow_mount(struct path *path)
652 int res = 0;
653 while (d_mountpoint(path->dentry)) {
654 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
655 if (!mounted)
656 break;
657 dput(path->dentry);
658 if (res)
659 mntput(path->mnt);
660 path->mnt = mounted;
661 path->dentry = dget(mounted->mnt_root);
662 res = 1;
664 return res;
667 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
669 while (d_mountpoint(*dentry)) {
670 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
671 if (!mounted)
672 break;
673 dput(*dentry);
674 mntput(*mnt);
675 *mnt = mounted;
676 *dentry = dget(mounted->mnt_root);
680 /* no need for dcache_lock, as serialization is taken care in
681 * namespace.c
683 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
685 struct vfsmount *mounted;
687 mounted = lookup_mnt(*mnt, *dentry);
688 if (mounted) {
689 dput(*dentry);
690 mntput(*mnt);
691 *mnt = mounted;
692 *dentry = dget(mounted->mnt_root);
693 return 1;
695 return 0;
698 static __always_inline void follow_dotdot(struct nameidata *nd)
700 while(1) {
701 struct vfsmount *parent;
702 struct dentry *old = nd->dentry;
704 read_lock(&current->fs->lock);
705 if (nd->dentry == current->fs->root &&
706 nd->mnt == current->fs->rootmnt) {
707 read_unlock(&current->fs->lock);
708 break;
710 read_unlock(&current->fs->lock);
711 spin_lock(&dcache_lock);
712 if (nd->dentry != nd->mnt->mnt_root) {
713 nd->dentry = dget(nd->dentry->d_parent);
714 spin_unlock(&dcache_lock);
715 dput(old);
716 break;
718 spin_unlock(&dcache_lock);
719 spin_lock(&vfsmount_lock);
720 parent = nd->mnt->mnt_parent;
721 if (parent == nd->mnt) {
722 spin_unlock(&vfsmount_lock);
723 break;
725 mntget(parent);
726 nd->dentry = dget(nd->mnt->mnt_mountpoint);
727 spin_unlock(&vfsmount_lock);
728 dput(old);
729 mntput(nd->mnt);
730 nd->mnt = parent;
732 follow_mount(&nd->mnt, &nd->dentry);
736 * It's more convoluted than I'd like it to be, but... it's still fairly
737 * small and for now I'd prefer to have fast path as straight as possible.
738 * It _is_ time-critical.
740 static int do_lookup(struct nameidata *nd, struct qstr *name,
741 struct path *path)
743 struct vfsmount *mnt = nd->mnt;
744 struct dentry *dentry = __d_lookup(nd->dentry, name);
746 if (!dentry)
747 goto need_lookup;
748 if (dentry->d_op && dentry->d_op->d_revalidate)
749 goto need_revalidate;
750 done:
751 path->mnt = mnt;
752 path->dentry = dentry;
753 __follow_mount(path);
754 return 0;
756 need_lookup:
757 dentry = real_lookup(nd->dentry, name, nd);
758 if (IS_ERR(dentry))
759 goto fail;
760 goto done;
762 need_revalidate:
763 if (dentry->d_op->d_revalidate(dentry, nd))
764 goto done;
765 if (d_invalidate(dentry))
766 goto done;
767 dput(dentry);
768 goto need_lookup;
770 fail:
771 return PTR_ERR(dentry);
775 * Name resolution.
776 * This is the basic name resolution function, turning a pathname into
777 * the final dentry. We expect 'base' to be positive and a directory.
779 * Returns 0 and nd will have valid dentry and mnt on success.
780 * Returns error and drops reference to input namei data on failure.
782 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
784 struct path next;
785 struct inode *inode;
786 int err;
787 unsigned int lookup_flags = nd->flags;
789 while (*name=='/')
790 name++;
791 if (!*name)
792 goto return_reval;
794 inode = nd->dentry->d_inode;
795 if (nd->depth)
796 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
798 /* At this point we know we have a real path component. */
799 for(;;) {
800 unsigned long hash;
801 struct qstr this;
802 unsigned int c;
804 nd->flags |= LOOKUP_CONTINUE;
805 err = exec_permission_lite(inode, nd);
806 if (err == -EAGAIN)
807 err = vfs_permission(nd, MAY_EXEC);
808 if (err)
809 break;
811 this.name = name;
812 c = *(const unsigned char *)name;
814 hash = init_name_hash();
815 do {
816 name++;
817 hash = partial_name_hash(c, hash);
818 c = *(const unsigned char *)name;
819 } while (c && (c != '/'));
820 this.len = name - (const char *) this.name;
821 this.hash = end_name_hash(hash);
823 /* remove trailing slashes? */
824 if (!c)
825 goto last_component;
826 while (*++name == '/');
827 if (!*name)
828 goto last_with_slashes;
831 * "." and ".." are special - ".." especially so because it has
832 * to be able to know about the current root directory and
833 * parent relationships.
835 if (this.name[0] == '.') switch (this.len) {
836 default:
837 break;
838 case 2:
839 if (this.name[1] != '.')
840 break;
841 follow_dotdot(nd);
842 inode = nd->dentry->d_inode;
843 /* fallthrough */
844 case 1:
845 continue;
848 * See if the low-level filesystem might want
849 * to use its own hash..
851 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
852 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
853 if (err < 0)
854 break;
856 /* This does the actual lookups.. */
857 err = do_lookup(nd, &this, &next);
858 if (err)
859 break;
861 err = -ENOENT;
862 inode = next.dentry->d_inode;
863 if (!inode)
864 goto out_dput;
865 err = -ENOTDIR;
866 if (!inode->i_op)
867 goto out_dput;
869 if (inode->i_op->follow_link) {
870 err = do_follow_link(&next, nd);
871 if (err)
872 goto return_err;
873 err = -ENOENT;
874 inode = nd->dentry->d_inode;
875 if (!inode)
876 break;
877 err = -ENOTDIR;
878 if (!inode->i_op)
879 break;
880 } else
881 path_to_nameidata(&next, nd);
882 err = -ENOTDIR;
883 if (!inode->i_op->lookup)
884 break;
885 continue;
886 /* here ends the main loop */
888 last_with_slashes:
889 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
890 last_component:
891 /* Clear LOOKUP_CONTINUE iff it was previously unset */
892 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
893 if (lookup_flags & LOOKUP_PARENT)
894 goto lookup_parent;
895 if (this.name[0] == '.') switch (this.len) {
896 default:
897 break;
898 case 2:
899 if (this.name[1] != '.')
900 break;
901 follow_dotdot(nd);
902 inode = nd->dentry->d_inode;
903 /* fallthrough */
904 case 1:
905 goto return_reval;
907 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
908 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
909 if (err < 0)
910 break;
912 err = do_lookup(nd, &this, &next);
913 if (err)
914 break;
915 inode = next.dentry->d_inode;
916 if ((lookup_flags & LOOKUP_FOLLOW)
917 && inode && inode->i_op && inode->i_op->follow_link) {
918 err = do_follow_link(&next, nd);
919 if (err)
920 goto return_err;
921 inode = nd->dentry->d_inode;
922 } else
923 path_to_nameidata(&next, nd);
924 err = -ENOENT;
925 if (!inode)
926 break;
927 if (lookup_flags & LOOKUP_DIRECTORY) {
928 err = -ENOTDIR;
929 if (!inode->i_op || !inode->i_op->lookup)
930 break;
932 goto return_base;
933 lookup_parent:
934 nd->last = this;
935 nd->last_type = LAST_NORM;
936 if (this.name[0] != '.')
937 goto return_base;
938 if (this.len == 1)
939 nd->last_type = LAST_DOT;
940 else if (this.len == 2 && this.name[1] == '.')
941 nd->last_type = LAST_DOTDOT;
942 else
943 goto return_base;
944 return_reval:
946 * We bypassed the ordinary revalidation routines.
947 * We may need to check the cached dentry for staleness.
949 if (nd->dentry && nd->dentry->d_sb &&
950 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
951 err = -ESTALE;
952 /* Note: we do not d_invalidate() */
953 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
954 break;
956 return_base:
957 return 0;
958 out_dput:
959 dput_path(&next, nd);
960 break;
962 path_release(nd);
963 return_err:
964 return err;
968 * Wrapper to retry pathname resolution whenever the underlying
969 * file system returns an ESTALE.
971 * Retry the whole path once, forcing real lookup requests
972 * instead of relying on the dcache.
974 int fastcall link_path_walk(const char *name, struct nameidata *nd)
976 struct nameidata save = *nd;
977 int result;
979 /* make sure the stuff we saved doesn't go away */
980 dget(save.dentry);
981 mntget(save.mnt);
983 result = __link_path_walk(name, nd);
984 if (result == -ESTALE) {
985 *nd = save;
986 dget(nd->dentry);
987 mntget(nd->mnt);
988 nd->flags |= LOOKUP_REVAL;
989 result = __link_path_walk(name, nd);
992 dput(save.dentry);
993 mntput(save.mnt);
995 return result;
998 int fastcall path_walk(const char * name, struct nameidata *nd)
1000 current->total_link_count = 0;
1001 return link_path_walk(name, nd);
1005 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1006 * everything is done. Returns 0 and drops input nd, if lookup failed;
1008 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1010 if (path_walk(name, nd))
1011 return 0; /* something went wrong... */
1013 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1014 struct dentry *old_dentry = nd->dentry;
1015 struct vfsmount *old_mnt = nd->mnt;
1016 struct qstr last = nd->last;
1017 int last_type = nd->last_type;
1019 * NAME was not found in alternate root or it's a directory. Try to find
1020 * it in the normal root:
1022 nd->last_type = LAST_ROOT;
1023 read_lock(&current->fs->lock);
1024 nd->mnt = mntget(current->fs->rootmnt);
1025 nd->dentry = dget(current->fs->root);
1026 read_unlock(&current->fs->lock);
1027 if (path_walk(name, nd) == 0) {
1028 if (nd->dentry->d_inode) {
1029 dput(old_dentry);
1030 mntput(old_mnt);
1031 return 1;
1033 path_release(nd);
1035 nd->dentry = old_dentry;
1036 nd->mnt = old_mnt;
1037 nd->last = last;
1038 nd->last_type = last_type;
1040 return 1;
1043 void set_fs_altroot(void)
1045 char *emul = __emul_prefix();
1046 struct nameidata nd;
1047 struct vfsmount *mnt = NULL, *oldmnt;
1048 struct dentry *dentry = NULL, *olddentry;
1049 int err;
1051 if (!emul)
1052 goto set_it;
1053 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1054 if (!err) {
1055 mnt = nd.mnt;
1056 dentry = nd.dentry;
1058 set_it:
1059 write_lock(&current->fs->lock);
1060 oldmnt = current->fs->altrootmnt;
1061 olddentry = current->fs->altroot;
1062 current->fs->altrootmnt = mnt;
1063 current->fs->altroot = dentry;
1064 write_unlock(&current->fs->lock);
1065 if (olddentry) {
1066 dput(olddentry);
1067 mntput(oldmnt);
1071 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1072 static int fastcall do_path_lookup(int dfd, const char *name,
1073 unsigned int flags, struct nameidata *nd)
1075 int retval = 0;
1076 int fput_needed;
1077 struct file *file;
1079 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1080 nd->flags = flags;
1081 nd->depth = 0;
1083 if (*name=='/') {
1084 read_lock(&current->fs->lock);
1085 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1086 nd->mnt = mntget(current->fs->altrootmnt);
1087 nd->dentry = dget(current->fs->altroot);
1088 read_unlock(&current->fs->lock);
1089 if (__emul_lookup_dentry(name,nd))
1090 goto out; /* found in altroot */
1091 read_lock(&current->fs->lock);
1093 nd->mnt = mntget(current->fs->rootmnt);
1094 nd->dentry = dget(current->fs->root);
1095 read_unlock(&current->fs->lock);
1096 } else if (dfd == AT_FDCWD) {
1097 read_lock(&current->fs->lock);
1098 nd->mnt = mntget(current->fs->pwdmnt);
1099 nd->dentry = dget(current->fs->pwd);
1100 read_unlock(&current->fs->lock);
1101 } else {
1102 struct dentry *dentry;
1104 file = fget_light(dfd, &fput_needed);
1105 retval = -EBADF;
1106 if (!file)
1107 goto out_fail;
1109 dentry = file->f_dentry;
1111 retval = -ENOTDIR;
1112 if (!S_ISDIR(dentry->d_inode->i_mode))
1113 goto fput_fail;
1115 retval = file_permission(file, MAY_EXEC);
1116 if (retval)
1117 goto fput_fail;
1119 nd->mnt = mntget(file->f_vfsmnt);
1120 nd->dentry = dget(dentry);
1122 fput_light(file, fput_needed);
1124 current->total_link_count = 0;
1125 retval = link_path_walk(name, nd);
1126 out:
1127 if (likely(retval == 0)) {
1128 if (unlikely(current->audit_context && nd && nd->dentry &&
1129 nd->dentry->d_inode))
1130 audit_inode(name, nd->dentry->d_inode);
1132 out_fail:
1133 return retval;
1135 fput_fail:
1136 fput_light(file, fput_needed);
1137 goto out_fail;
1140 int fastcall path_lookup(const char *name, unsigned int flags,
1141 struct nameidata *nd)
1143 return do_path_lookup(AT_FDCWD, name, flags, nd);
1146 static int __path_lookup_intent_open(int dfd, const char *name,
1147 unsigned int lookup_flags, struct nameidata *nd,
1148 int open_flags, int create_mode)
1150 struct file *filp = get_empty_filp();
1151 int err;
1153 if (filp == NULL)
1154 return -ENFILE;
1155 nd->intent.open.file = filp;
1156 nd->intent.open.flags = open_flags;
1157 nd->intent.open.create_mode = create_mode;
1158 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1159 if (IS_ERR(nd->intent.open.file)) {
1160 if (err == 0) {
1161 err = PTR_ERR(nd->intent.open.file);
1162 path_release(nd);
1164 } else if (err != 0)
1165 release_open_intent(nd);
1166 return err;
1170 * path_lookup_open - lookup a file path with open intent
1171 * @dfd: the directory to use as base, or AT_FDCWD
1172 * @name: pointer to file name
1173 * @lookup_flags: lookup intent flags
1174 * @nd: pointer to nameidata
1175 * @open_flags: open intent flags
1177 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1178 struct nameidata *nd, int open_flags)
1180 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1181 open_flags, 0);
1185 * path_lookup_create - lookup a file path with open + create intent
1186 * @dfd: the directory to use as base, or AT_FDCWD
1187 * @name: pointer to file name
1188 * @lookup_flags: lookup intent flags
1189 * @nd: pointer to nameidata
1190 * @open_flags: open intent flags
1191 * @create_mode: create intent flags
1193 static int path_lookup_create(int dfd, const char *name,
1194 unsigned int lookup_flags, struct nameidata *nd,
1195 int open_flags, int create_mode)
1197 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1198 nd, open_flags, create_mode);
1201 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1202 struct nameidata *nd, int open_flags)
1204 char *tmp = getname(name);
1205 int err = PTR_ERR(tmp);
1207 if (!IS_ERR(tmp)) {
1208 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1209 putname(tmp);
1211 return err;
1215 * Restricted form of lookup. Doesn't follow links, single-component only,
1216 * needs parent already locked. Doesn't follow mounts.
1217 * SMP-safe.
1219 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1221 struct dentry * dentry;
1222 struct inode *inode;
1223 int err;
1225 inode = base->d_inode;
1226 err = permission(inode, MAY_EXEC, nd);
1227 dentry = ERR_PTR(err);
1228 if (err)
1229 goto out;
1232 * See if the low-level filesystem might want
1233 * to use its own hash..
1235 if (base->d_op && base->d_op->d_hash) {
1236 err = base->d_op->d_hash(base, name);
1237 dentry = ERR_PTR(err);
1238 if (err < 0)
1239 goto out;
1242 dentry = cached_lookup(base, name, nd);
1243 if (!dentry) {
1244 struct dentry *new = d_alloc(base, name);
1245 dentry = ERR_PTR(-ENOMEM);
1246 if (!new)
1247 goto out;
1248 dentry = inode->i_op->lookup(inode, new, nd);
1249 if (!dentry)
1250 dentry = new;
1251 else
1252 dput(new);
1254 out:
1255 return dentry;
1258 static struct dentry *lookup_hash(struct nameidata *nd)
1260 return __lookup_hash(&nd->last, nd->dentry, nd);
1263 /* SMP-safe */
1264 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1266 unsigned long hash;
1267 struct qstr this;
1268 unsigned int c;
1270 this.name = name;
1271 this.len = len;
1272 if (!len)
1273 goto access;
1275 hash = init_name_hash();
1276 while (len--) {
1277 c = *(const unsigned char *)name++;
1278 if (c == '/' || c == '\0')
1279 goto access;
1280 hash = partial_name_hash(c, hash);
1282 this.hash = end_name_hash(hash);
1284 return __lookup_hash(&this, base, NULL);
1285 access:
1286 return ERR_PTR(-EACCES);
1290 * namei()
1292 * is used by most simple commands to get the inode of a specified name.
1293 * Open, link etc use their own routines, but this is enough for things
1294 * like 'chmod' etc.
1296 * namei exists in two versions: namei/lnamei. The only difference is
1297 * that namei follows links, while lnamei does not.
1298 * SMP-safe
1300 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1301 struct nameidata *nd)
1303 char *tmp = getname(name);
1304 int err = PTR_ERR(tmp);
1306 if (!IS_ERR(tmp)) {
1307 err = do_path_lookup(dfd, tmp, flags, nd);
1308 putname(tmp);
1310 return err;
1313 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1315 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1319 * It's inline, so penalty for filesystems that don't use sticky bit is
1320 * minimal.
1322 static inline int check_sticky(struct inode *dir, struct inode *inode)
1324 if (!(dir->i_mode & S_ISVTX))
1325 return 0;
1326 if (inode->i_uid == current->fsuid)
1327 return 0;
1328 if (dir->i_uid == current->fsuid)
1329 return 0;
1330 return !capable(CAP_FOWNER);
1334 * Check whether we can remove a link victim from directory dir, check
1335 * whether the type of victim is right.
1336 * 1. We can't do it if dir is read-only (done in permission())
1337 * 2. We should have write and exec permissions on dir
1338 * 3. We can't remove anything from append-only dir
1339 * 4. We can't do anything with immutable dir (done in permission())
1340 * 5. If the sticky bit on dir is set we should either
1341 * a. be owner of dir, or
1342 * b. be owner of victim, or
1343 * c. have CAP_FOWNER capability
1344 * 6. If the victim is append-only or immutable we can't do antyhing with
1345 * links pointing to it.
1346 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1347 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1348 * 9. We can't remove a root or mountpoint.
1349 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1350 * nfs_async_unlink().
1352 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1354 int error;
1356 if (!victim->d_inode)
1357 return -ENOENT;
1359 BUG_ON(victim->d_parent->d_inode != dir);
1360 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
1362 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1363 if (error)
1364 return error;
1365 if (IS_APPEND(dir))
1366 return -EPERM;
1367 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1368 IS_IMMUTABLE(victim->d_inode))
1369 return -EPERM;
1370 if (isdir) {
1371 if (!S_ISDIR(victim->d_inode->i_mode))
1372 return -ENOTDIR;
1373 if (IS_ROOT(victim))
1374 return -EBUSY;
1375 } else if (S_ISDIR(victim->d_inode->i_mode))
1376 return -EISDIR;
1377 if (IS_DEADDIR(dir))
1378 return -ENOENT;
1379 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1380 return -EBUSY;
1381 return 0;
1384 /* Check whether we can create an object with dentry child in directory
1385 * dir.
1386 * 1. We can't do it if child already exists (open has special treatment for
1387 * this case, but since we are inlined it's OK)
1388 * 2. We can't do it if dir is read-only (done in permission())
1389 * 3. We should have write and exec permissions on dir
1390 * 4. We can't do it if dir is immutable (done in permission())
1392 static inline int may_create(struct inode *dir, struct dentry *child,
1393 struct nameidata *nd)
1395 if (child->d_inode)
1396 return -EEXIST;
1397 if (IS_DEADDIR(dir))
1398 return -ENOENT;
1399 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1403 * O_DIRECTORY translates into forcing a directory lookup.
1405 static inline int lookup_flags(unsigned int f)
1407 unsigned long retval = LOOKUP_FOLLOW;
1409 if (f & O_NOFOLLOW)
1410 retval &= ~LOOKUP_FOLLOW;
1412 if (f & O_DIRECTORY)
1413 retval |= LOOKUP_DIRECTORY;
1415 return retval;
1419 * p1 and p2 should be directories on the same fs.
1421 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1423 struct dentry *p;
1425 if (p1 == p2) {
1426 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1427 return NULL;
1430 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1432 for (p = p1; p->d_parent != p; p = p->d_parent) {
1433 if (p->d_parent == p2) {
1434 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1435 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1436 return p;
1440 for (p = p2; p->d_parent != p; p = p->d_parent) {
1441 if (p->d_parent == p1) {
1442 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1443 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1444 return p;
1448 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1449 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1450 return NULL;
1453 void unlock_rename(struct dentry *p1, struct dentry *p2)
1455 mutex_unlock(&p1->d_inode->i_mutex);
1456 if (p1 != p2) {
1457 mutex_unlock(&p2->d_inode->i_mutex);
1458 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1462 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1463 struct nameidata *nd)
1465 int error = may_create(dir, dentry, nd);
1467 if (error)
1468 return error;
1470 if (!dir->i_op || !dir->i_op->create)
1471 return -EACCES; /* shouldn't it be ENOSYS? */
1472 mode &= S_IALLUGO;
1473 mode |= S_IFREG;
1474 error = security_inode_create(dir, dentry, mode);
1475 if (error)
1476 return error;
1477 DQUOT_INIT(dir);
1478 error = dir->i_op->create(dir, dentry, mode, nd);
1479 if (!error)
1480 fsnotify_create(dir, dentry);
1481 return error;
1484 int may_open(struct nameidata *nd, int acc_mode, int flag)
1486 struct dentry *dentry = nd->dentry;
1487 struct inode *inode = dentry->d_inode;
1488 int error;
1490 if (!inode)
1491 return -ENOENT;
1493 if (S_ISLNK(inode->i_mode))
1494 return -ELOOP;
1496 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1497 return -EISDIR;
1499 error = vfs_permission(nd, acc_mode);
1500 if (error)
1501 return error;
1504 * FIFO's, sockets and device files are special: they don't
1505 * actually live on the filesystem itself, and as such you
1506 * can write to them even if the filesystem is read-only.
1508 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1509 flag &= ~O_TRUNC;
1510 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1511 if (nd->mnt->mnt_flags & MNT_NODEV)
1512 return -EACCES;
1514 flag &= ~O_TRUNC;
1515 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1516 return -EROFS;
1518 * An append-only file must be opened in append mode for writing.
1520 if (IS_APPEND(inode)) {
1521 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1522 return -EPERM;
1523 if (flag & O_TRUNC)
1524 return -EPERM;
1527 /* O_NOATIME can only be set by the owner or superuser */
1528 if (flag & O_NOATIME)
1529 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1530 return -EPERM;
1533 * Ensure there are no outstanding leases on the file.
1535 error = break_lease(inode, flag);
1536 if (error)
1537 return error;
1539 if (flag & O_TRUNC) {
1540 error = get_write_access(inode);
1541 if (error)
1542 return error;
1545 * Refuse to truncate files with mandatory locks held on them.
1547 error = locks_verify_locked(inode);
1548 if (!error) {
1549 DQUOT_INIT(inode);
1551 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1553 put_write_access(inode);
1554 if (error)
1555 return error;
1556 } else
1557 if (flag & FMODE_WRITE)
1558 DQUOT_INIT(inode);
1560 return 0;
1564 * open_namei()
1566 * namei for open - this is in fact almost the whole open-routine.
1568 * Note that the low bits of "flag" aren't the same as in the open
1569 * system call - they are 00 - no permissions needed
1570 * 01 - read permission needed
1571 * 10 - write permission needed
1572 * 11 - read/write permissions needed
1573 * which is a lot more logical, and also allows the "no perm" needed
1574 * for symlinks (where the permissions are checked later).
1575 * SMP-safe
1577 int open_namei(int dfd, const char *pathname, int flag,
1578 int mode, struct nameidata *nd)
1580 int acc_mode, error;
1581 struct path path;
1582 struct dentry *dir;
1583 int count = 0;
1585 acc_mode = ACC_MODE(flag);
1587 /* O_TRUNC implies we need access checks for write permissions */
1588 if (flag & O_TRUNC)
1589 acc_mode |= MAY_WRITE;
1591 /* Allow the LSM permission hook to distinguish append
1592 access from general write access. */
1593 if (flag & O_APPEND)
1594 acc_mode |= MAY_APPEND;
1597 * The simplest case - just a plain lookup.
1599 if (!(flag & O_CREAT)) {
1600 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1601 nd, flag);
1602 if (error)
1603 return error;
1604 goto ok;
1608 * Create - we need to know the parent.
1610 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1611 if (error)
1612 return error;
1615 * We have the parent and last component. First of all, check
1616 * that we are not asked to creat(2) an obvious directory - that
1617 * will not do.
1619 error = -EISDIR;
1620 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1621 goto exit;
1623 dir = nd->dentry;
1624 nd->flags &= ~LOOKUP_PARENT;
1625 mutex_lock(&dir->d_inode->i_mutex);
1626 path.dentry = lookup_hash(nd);
1627 path.mnt = nd->mnt;
1629 do_last:
1630 error = PTR_ERR(path.dentry);
1631 if (IS_ERR(path.dentry)) {
1632 mutex_unlock(&dir->d_inode->i_mutex);
1633 goto exit;
1636 if (IS_ERR(nd->intent.open.file)) {
1637 mutex_unlock(&dir->d_inode->i_mutex);
1638 error = PTR_ERR(nd->intent.open.file);
1639 goto exit_dput;
1642 /* Negative dentry, just create the file */
1643 if (!path.dentry->d_inode) {
1644 if (!IS_POSIXACL(dir->d_inode))
1645 mode &= ~current->fs->umask;
1646 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1647 mutex_unlock(&dir->d_inode->i_mutex);
1648 dput(nd->dentry);
1649 nd->dentry = path.dentry;
1650 if (error)
1651 goto exit;
1652 /* Don't check for write permission, don't truncate */
1653 acc_mode = 0;
1654 flag &= ~O_TRUNC;
1655 goto ok;
1659 * It already exists.
1661 mutex_unlock(&dir->d_inode->i_mutex);
1663 error = -EEXIST;
1664 if (flag & O_EXCL)
1665 goto exit_dput;
1667 if (__follow_mount(&path)) {
1668 error = -ELOOP;
1669 if (flag & O_NOFOLLOW)
1670 goto exit_dput;
1672 error = -ENOENT;
1673 if (!path.dentry->d_inode)
1674 goto exit_dput;
1675 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1676 goto do_link;
1678 path_to_nameidata(&path, nd);
1679 error = -EISDIR;
1680 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1681 goto exit;
1683 error = may_open(nd, acc_mode, flag);
1684 if (error)
1685 goto exit;
1686 return 0;
1688 exit_dput:
1689 dput_path(&path, nd);
1690 exit:
1691 if (!IS_ERR(nd->intent.open.file))
1692 release_open_intent(nd);
1693 path_release(nd);
1694 return error;
1696 do_link:
1697 error = -ELOOP;
1698 if (flag & O_NOFOLLOW)
1699 goto exit_dput;
1701 * This is subtle. Instead of calling do_follow_link() we do the
1702 * thing by hands. The reason is that this way we have zero link_count
1703 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1704 * After that we have the parent and last component, i.e.
1705 * we are in the same situation as after the first path_walk().
1706 * Well, almost - if the last component is normal we get its copy
1707 * stored in nd->last.name and we will have to putname() it when we
1708 * are done. Procfs-like symlinks just set LAST_BIND.
1710 nd->flags |= LOOKUP_PARENT;
1711 error = security_inode_follow_link(path.dentry, nd);
1712 if (error)
1713 goto exit_dput;
1714 error = __do_follow_link(&path, nd);
1715 if (error) {
1716 /* Does someone understand code flow here? Or it is only
1717 * me so stupid? Anathema to whoever designed this non-sense
1718 * with "intent.open".
1720 release_open_intent(nd);
1721 return error;
1723 nd->flags &= ~LOOKUP_PARENT;
1724 if (nd->last_type == LAST_BIND)
1725 goto ok;
1726 error = -EISDIR;
1727 if (nd->last_type != LAST_NORM)
1728 goto exit;
1729 if (nd->last.name[nd->last.len]) {
1730 __putname(nd->last.name);
1731 goto exit;
1733 error = -ELOOP;
1734 if (count++==32) {
1735 __putname(nd->last.name);
1736 goto exit;
1738 dir = nd->dentry;
1739 mutex_lock(&dir->d_inode->i_mutex);
1740 path.dentry = lookup_hash(nd);
1741 path.mnt = nd->mnt;
1742 __putname(nd->last.name);
1743 goto do_last;
1747 * lookup_create - lookup a dentry, creating it if it doesn't exist
1748 * @nd: nameidata info
1749 * @is_dir: directory flag
1751 * Simple function to lookup and return a dentry and create it
1752 * if it doesn't exist. Is SMP-safe.
1754 * Returns with nd->dentry->d_inode->i_mutex locked.
1756 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1758 struct dentry *dentry = ERR_PTR(-EEXIST);
1760 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1762 * Yucky last component or no last component at all?
1763 * (foo/., foo/.., /////)
1765 if (nd->last_type != LAST_NORM)
1766 goto fail;
1767 nd->flags &= ~LOOKUP_PARENT;
1770 * Do the final lookup.
1772 dentry = lookup_hash(nd);
1773 if (IS_ERR(dentry))
1774 goto fail;
1777 * Special case - lookup gave negative, but... we had foo/bar/
1778 * From the vfs_mknod() POV we just have a negative dentry -
1779 * all is fine. Let's be bastards - you had / on the end, you've
1780 * been asking for (non-existent) directory. -ENOENT for you.
1782 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1783 goto enoent;
1784 return dentry;
1785 enoent:
1786 dput(dentry);
1787 dentry = ERR_PTR(-ENOENT);
1788 fail:
1789 return dentry;
1791 EXPORT_SYMBOL_GPL(lookup_create);
1793 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1795 int error = may_create(dir, dentry, NULL);
1797 if (error)
1798 return error;
1800 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1801 return -EPERM;
1803 if (!dir->i_op || !dir->i_op->mknod)
1804 return -EPERM;
1806 error = security_inode_mknod(dir, dentry, mode, dev);
1807 if (error)
1808 return error;
1810 DQUOT_INIT(dir);
1811 error = dir->i_op->mknod(dir, dentry, mode, dev);
1812 if (!error)
1813 fsnotify_create(dir, dentry);
1814 return error;
1817 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1818 unsigned dev)
1820 int error = 0;
1821 char * tmp;
1822 struct dentry * dentry;
1823 struct nameidata nd;
1825 if (S_ISDIR(mode))
1826 return -EPERM;
1827 tmp = getname(filename);
1828 if (IS_ERR(tmp))
1829 return PTR_ERR(tmp);
1831 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1832 if (error)
1833 goto out;
1834 dentry = lookup_create(&nd, 0);
1835 error = PTR_ERR(dentry);
1837 if (!IS_POSIXACL(nd.dentry->d_inode))
1838 mode &= ~current->fs->umask;
1839 if (!IS_ERR(dentry)) {
1840 switch (mode & S_IFMT) {
1841 case 0: case S_IFREG:
1842 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1843 break;
1844 case S_IFCHR: case S_IFBLK:
1845 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1846 new_decode_dev(dev));
1847 break;
1848 case S_IFIFO: case S_IFSOCK:
1849 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1850 break;
1851 case S_IFDIR:
1852 error = -EPERM;
1853 break;
1854 default:
1855 error = -EINVAL;
1857 dput(dentry);
1859 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1860 path_release(&nd);
1861 out:
1862 putname(tmp);
1864 return error;
1867 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1869 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1872 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1874 int error = may_create(dir, dentry, NULL);
1876 if (error)
1877 return error;
1879 if (!dir->i_op || !dir->i_op->mkdir)
1880 return -EPERM;
1882 mode &= (S_IRWXUGO|S_ISVTX);
1883 error = security_inode_mkdir(dir, dentry, mode);
1884 if (error)
1885 return error;
1887 DQUOT_INIT(dir);
1888 error = dir->i_op->mkdir(dir, dentry, mode);
1889 if (!error)
1890 fsnotify_mkdir(dir, dentry);
1891 return error;
1894 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1896 int error = 0;
1897 char * tmp;
1899 tmp = getname(pathname);
1900 error = PTR_ERR(tmp);
1901 if (!IS_ERR(tmp)) {
1902 struct dentry *dentry;
1903 struct nameidata nd;
1905 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1906 if (error)
1907 goto out;
1908 dentry = lookup_create(&nd, 1);
1909 error = PTR_ERR(dentry);
1910 if (!IS_ERR(dentry)) {
1911 if (!IS_POSIXACL(nd.dentry->d_inode))
1912 mode &= ~current->fs->umask;
1913 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1914 dput(dentry);
1916 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1917 path_release(&nd);
1918 out:
1919 putname(tmp);
1922 return error;
1925 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1927 return sys_mkdirat(AT_FDCWD, pathname, mode);
1931 * We try to drop the dentry early: we should have
1932 * a usage count of 2 if we're the only user of this
1933 * dentry, and if that is true (possibly after pruning
1934 * the dcache), then we drop the dentry now.
1936 * A low-level filesystem can, if it choses, legally
1937 * do a
1939 * if (!d_unhashed(dentry))
1940 * return -EBUSY;
1942 * if it cannot handle the case of removing a directory
1943 * that is still in use by something else..
1945 void dentry_unhash(struct dentry *dentry)
1947 dget(dentry);
1948 if (atomic_read(&dentry->d_count))
1949 shrink_dcache_parent(dentry);
1950 spin_lock(&dcache_lock);
1951 spin_lock(&dentry->d_lock);
1952 if (atomic_read(&dentry->d_count) == 2)
1953 __d_drop(dentry);
1954 spin_unlock(&dentry->d_lock);
1955 spin_unlock(&dcache_lock);
1958 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1960 int error = may_delete(dir, dentry, 1);
1962 if (error)
1963 return error;
1965 if (!dir->i_op || !dir->i_op->rmdir)
1966 return -EPERM;
1968 DQUOT_INIT(dir);
1970 mutex_lock(&dentry->d_inode->i_mutex);
1971 dentry_unhash(dentry);
1972 if (d_mountpoint(dentry))
1973 error = -EBUSY;
1974 else {
1975 error = security_inode_rmdir(dir, dentry);
1976 if (!error) {
1977 error = dir->i_op->rmdir(dir, dentry);
1978 if (!error)
1979 dentry->d_inode->i_flags |= S_DEAD;
1982 mutex_unlock(&dentry->d_inode->i_mutex);
1983 if (!error) {
1984 d_delete(dentry);
1986 dput(dentry);
1988 return error;
1991 static long do_rmdir(int dfd, const char __user *pathname)
1993 int error = 0;
1994 char * name;
1995 struct dentry *dentry;
1996 struct nameidata nd;
1998 name = getname(pathname);
1999 if(IS_ERR(name))
2000 return PTR_ERR(name);
2002 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2003 if (error)
2004 goto exit;
2006 switch(nd.last_type) {
2007 case LAST_DOTDOT:
2008 error = -ENOTEMPTY;
2009 goto exit1;
2010 case LAST_DOT:
2011 error = -EINVAL;
2012 goto exit1;
2013 case LAST_ROOT:
2014 error = -EBUSY;
2015 goto exit1;
2017 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2018 dentry = lookup_hash(&nd);
2019 error = PTR_ERR(dentry);
2020 if (!IS_ERR(dentry)) {
2021 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2022 dput(dentry);
2024 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2025 exit1:
2026 path_release(&nd);
2027 exit:
2028 putname(name);
2029 return error;
2032 asmlinkage long sys_rmdir(const char __user *pathname)
2034 return do_rmdir(AT_FDCWD, pathname);
2037 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2039 int error = may_delete(dir, dentry, 0);
2041 if (error)
2042 return error;
2044 if (!dir->i_op || !dir->i_op->unlink)
2045 return -EPERM;
2047 DQUOT_INIT(dir);
2049 mutex_lock(&dentry->d_inode->i_mutex);
2050 if (d_mountpoint(dentry))
2051 error = -EBUSY;
2052 else {
2053 error = security_inode_unlink(dir, dentry);
2054 if (!error)
2055 error = dir->i_op->unlink(dir, dentry);
2057 mutex_unlock(&dentry->d_inode->i_mutex);
2059 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2060 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2061 d_delete(dentry);
2064 return error;
2068 * Make sure that the actual truncation of the file will occur outside its
2069 * directory's i_mutex. Truncate can take a long time if there is a lot of
2070 * writeout happening, and we don't want to prevent access to the directory
2071 * while waiting on the I/O.
2073 static long do_unlinkat(int dfd, const char __user *pathname)
2075 int error = 0;
2076 char * name;
2077 struct dentry *dentry;
2078 struct nameidata nd;
2079 struct inode *inode = NULL;
2081 name = getname(pathname);
2082 if(IS_ERR(name))
2083 return PTR_ERR(name);
2085 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2086 if (error)
2087 goto exit;
2088 error = -EISDIR;
2089 if (nd.last_type != LAST_NORM)
2090 goto exit1;
2091 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2092 dentry = lookup_hash(&nd);
2093 error = PTR_ERR(dentry);
2094 if (!IS_ERR(dentry)) {
2095 /* Why not before? Because we want correct error value */
2096 if (nd.last.name[nd.last.len])
2097 goto slashes;
2098 inode = dentry->d_inode;
2099 if (inode)
2100 atomic_inc(&inode->i_count);
2101 error = vfs_unlink(nd.dentry->d_inode, dentry);
2102 exit2:
2103 dput(dentry);
2105 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2106 if (inode)
2107 iput(inode); /* truncate the inode here */
2108 exit1:
2109 path_release(&nd);
2110 exit:
2111 putname(name);
2112 return error;
2114 slashes:
2115 error = !dentry->d_inode ? -ENOENT :
2116 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2117 goto exit2;
2120 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2122 if ((flag & ~AT_REMOVEDIR) != 0)
2123 return -EINVAL;
2125 if (flag & AT_REMOVEDIR)
2126 return do_rmdir(dfd, pathname);
2128 return do_unlinkat(dfd, pathname);
2131 asmlinkage long sys_unlink(const char __user *pathname)
2133 return do_unlinkat(AT_FDCWD, pathname);
2136 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2138 int error = may_create(dir, dentry, NULL);
2140 if (error)
2141 return error;
2143 if (!dir->i_op || !dir->i_op->symlink)
2144 return -EPERM;
2146 error = security_inode_symlink(dir, dentry, oldname);
2147 if (error)
2148 return error;
2150 DQUOT_INIT(dir);
2151 error = dir->i_op->symlink(dir, dentry, oldname);
2152 if (!error)
2153 fsnotify_create(dir, dentry);
2154 return error;
2157 asmlinkage long sys_symlinkat(const char __user *oldname,
2158 int newdfd, const char __user *newname)
2160 int error = 0;
2161 char * from;
2162 char * to;
2164 from = getname(oldname);
2165 if(IS_ERR(from))
2166 return PTR_ERR(from);
2167 to = getname(newname);
2168 error = PTR_ERR(to);
2169 if (!IS_ERR(to)) {
2170 struct dentry *dentry;
2171 struct nameidata nd;
2173 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2174 if (error)
2175 goto out;
2176 dentry = lookup_create(&nd, 0);
2177 error = PTR_ERR(dentry);
2178 if (!IS_ERR(dentry)) {
2179 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2180 dput(dentry);
2182 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2183 path_release(&nd);
2184 out:
2185 putname(to);
2187 putname(from);
2188 return error;
2191 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2193 return sys_symlinkat(oldname, AT_FDCWD, newname);
2196 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2198 struct inode *inode = old_dentry->d_inode;
2199 int error;
2201 if (!inode)
2202 return -ENOENT;
2204 error = may_create(dir, new_dentry, NULL);
2205 if (error)
2206 return error;
2208 if (dir->i_sb != inode->i_sb)
2209 return -EXDEV;
2212 * A link to an append-only or immutable file cannot be created.
2214 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2215 return -EPERM;
2216 if (!dir->i_op || !dir->i_op->link)
2217 return -EPERM;
2218 if (S_ISDIR(old_dentry->d_inode->i_mode))
2219 return -EPERM;
2221 error = security_inode_link(old_dentry, dir, new_dentry);
2222 if (error)
2223 return error;
2225 mutex_lock(&old_dentry->d_inode->i_mutex);
2226 DQUOT_INIT(dir);
2227 error = dir->i_op->link(old_dentry, dir, new_dentry);
2228 mutex_unlock(&old_dentry->d_inode->i_mutex);
2229 if (!error)
2230 fsnotify_create(dir, new_dentry);
2231 return error;
2235 * Hardlinks are often used in delicate situations. We avoid
2236 * security-related surprises by not following symlinks on the
2237 * newname. --KAB
2239 * We don't follow them on the oldname either to be compatible
2240 * with linux 2.0, and to avoid hard-linking to directories
2241 * and other special files. --ADM
2243 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2244 int newdfd, const char __user *newname,
2245 int flags)
2247 struct dentry *new_dentry;
2248 struct nameidata nd, old_nd;
2249 int error;
2250 char * to;
2252 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2253 return -EINVAL;
2255 to = getname(newname);
2256 if (IS_ERR(to))
2257 return PTR_ERR(to);
2259 error = __user_walk_fd(olddfd, oldname,
2260 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2261 &old_nd);
2262 if (error)
2263 goto exit;
2264 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2265 if (error)
2266 goto out;
2267 error = -EXDEV;
2268 if (old_nd.mnt != nd.mnt)
2269 goto out_release;
2270 new_dentry = lookup_create(&nd, 0);
2271 error = PTR_ERR(new_dentry);
2272 if (!IS_ERR(new_dentry)) {
2273 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2274 dput(new_dentry);
2276 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2277 out_release:
2278 path_release(&nd);
2279 out:
2280 path_release(&old_nd);
2281 exit:
2282 putname(to);
2284 return error;
2287 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2289 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2293 * The worst of all namespace operations - renaming directory. "Perverted"
2294 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2295 * Problems:
2296 * a) we can get into loop creation. Check is done in is_subdir().
2297 * b) race potential - two innocent renames can create a loop together.
2298 * That's where 4.4 screws up. Current fix: serialization on
2299 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2300 * story.
2301 * c) we have to lock _three_ objects - parents and victim (if it exists).
2302 * And that - after we got ->i_mutex on parents (until then we don't know
2303 * whether the target exists). Solution: try to be smart with locking
2304 * order for inodes. We rely on the fact that tree topology may change
2305 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2306 * move will be locked. Thus we can rank directories by the tree
2307 * (ancestors first) and rank all non-directories after them.
2308 * That works since everybody except rename does "lock parent, lookup,
2309 * lock child" and rename is under ->s_vfs_rename_mutex.
2310 * HOWEVER, it relies on the assumption that any object with ->lookup()
2311 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2312 * we'd better make sure that there's no link(2) for them.
2313 * d) some filesystems don't support opened-but-unlinked directories,
2314 * either because of layout or because they are not ready to deal with
2315 * all cases correctly. The latter will be fixed (taking this sort of
2316 * stuff into VFS), but the former is not going away. Solution: the same
2317 * trick as in rmdir().
2318 * e) conversion from fhandle to dentry may come in the wrong moment - when
2319 * we are removing the target. Solution: we will have to grab ->i_mutex
2320 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2321 * ->i_mutex on parents, which works but leads to some truely excessive
2322 * locking].
2324 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2325 struct inode *new_dir, struct dentry *new_dentry)
2327 int error = 0;
2328 struct inode *target;
2331 * If we are going to change the parent - check write permissions,
2332 * we'll need to flip '..'.
2334 if (new_dir != old_dir) {
2335 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2336 if (error)
2337 return error;
2340 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2341 if (error)
2342 return error;
2344 target = new_dentry->d_inode;
2345 if (target) {
2346 mutex_lock(&target->i_mutex);
2347 dentry_unhash(new_dentry);
2349 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2350 error = -EBUSY;
2351 else
2352 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2353 if (target) {
2354 if (!error)
2355 target->i_flags |= S_DEAD;
2356 mutex_unlock(&target->i_mutex);
2357 if (d_unhashed(new_dentry))
2358 d_rehash(new_dentry);
2359 dput(new_dentry);
2361 if (!error)
2362 d_move(old_dentry,new_dentry);
2363 return error;
2366 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2367 struct inode *new_dir, struct dentry *new_dentry)
2369 struct inode *target;
2370 int error;
2372 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2373 if (error)
2374 return error;
2376 dget(new_dentry);
2377 target = new_dentry->d_inode;
2378 if (target)
2379 mutex_lock(&target->i_mutex);
2380 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2381 error = -EBUSY;
2382 else
2383 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2384 if (!error) {
2385 /* The following d_move() should become unconditional */
2386 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2387 d_move(old_dentry, new_dentry);
2389 if (target)
2390 mutex_unlock(&target->i_mutex);
2391 dput(new_dentry);
2392 return error;
2395 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2396 struct inode *new_dir, struct dentry *new_dentry)
2398 int error;
2399 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2400 const char *old_name;
2402 if (old_dentry->d_inode == new_dentry->d_inode)
2403 return 0;
2405 error = may_delete(old_dir, old_dentry, is_dir);
2406 if (error)
2407 return error;
2409 if (!new_dentry->d_inode)
2410 error = may_create(new_dir, new_dentry, NULL);
2411 else
2412 error = may_delete(new_dir, new_dentry, is_dir);
2413 if (error)
2414 return error;
2416 if (!old_dir->i_op || !old_dir->i_op->rename)
2417 return -EPERM;
2419 DQUOT_INIT(old_dir);
2420 DQUOT_INIT(new_dir);
2422 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2424 if (is_dir)
2425 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2426 else
2427 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2428 if (!error) {
2429 const char *new_name = old_dentry->d_name.name;
2430 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2431 new_dentry->d_inode, old_dentry->d_inode);
2433 fsnotify_oldname_free(old_name);
2435 return error;
2438 static int do_rename(int olddfd, const char *oldname,
2439 int newdfd, const char *newname)
2441 int error = 0;
2442 struct dentry * old_dir, * new_dir;
2443 struct dentry * old_dentry, *new_dentry;
2444 struct dentry * trap;
2445 struct nameidata oldnd, newnd;
2447 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2448 if (error)
2449 goto exit;
2451 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2452 if (error)
2453 goto exit1;
2455 error = -EXDEV;
2456 if (oldnd.mnt != newnd.mnt)
2457 goto exit2;
2459 old_dir = oldnd.dentry;
2460 error = -EBUSY;
2461 if (oldnd.last_type != LAST_NORM)
2462 goto exit2;
2464 new_dir = newnd.dentry;
2465 if (newnd.last_type != LAST_NORM)
2466 goto exit2;
2468 trap = lock_rename(new_dir, old_dir);
2470 old_dentry = lookup_hash(&oldnd);
2471 error = PTR_ERR(old_dentry);
2472 if (IS_ERR(old_dentry))
2473 goto exit3;
2474 /* source must exist */
2475 error = -ENOENT;
2476 if (!old_dentry->d_inode)
2477 goto exit4;
2478 /* unless the source is a directory trailing slashes give -ENOTDIR */
2479 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2480 error = -ENOTDIR;
2481 if (oldnd.last.name[oldnd.last.len])
2482 goto exit4;
2483 if (newnd.last.name[newnd.last.len])
2484 goto exit4;
2486 /* source should not be ancestor of target */
2487 error = -EINVAL;
2488 if (old_dentry == trap)
2489 goto exit4;
2490 new_dentry = lookup_hash(&newnd);
2491 error = PTR_ERR(new_dentry);
2492 if (IS_ERR(new_dentry))
2493 goto exit4;
2494 /* target should not be an ancestor of source */
2495 error = -ENOTEMPTY;
2496 if (new_dentry == trap)
2497 goto exit5;
2499 error = vfs_rename(old_dir->d_inode, old_dentry,
2500 new_dir->d_inode, new_dentry);
2501 exit5:
2502 dput(new_dentry);
2503 exit4:
2504 dput(old_dentry);
2505 exit3:
2506 unlock_rename(new_dir, old_dir);
2507 exit2:
2508 path_release(&newnd);
2509 exit1:
2510 path_release(&oldnd);
2511 exit:
2512 return error;
2515 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2516 int newdfd, const char __user *newname)
2518 int error;
2519 char * from;
2520 char * to;
2522 from = getname(oldname);
2523 if(IS_ERR(from))
2524 return PTR_ERR(from);
2525 to = getname(newname);
2526 error = PTR_ERR(to);
2527 if (!IS_ERR(to)) {
2528 error = do_rename(olddfd, from, newdfd, to);
2529 putname(to);
2531 putname(from);
2532 return error;
2535 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2537 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2540 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2542 int len;
2544 len = PTR_ERR(link);
2545 if (IS_ERR(link))
2546 goto out;
2548 len = strlen(link);
2549 if (len > (unsigned) buflen)
2550 len = buflen;
2551 if (copy_to_user(buffer, link, len))
2552 len = -EFAULT;
2553 out:
2554 return len;
2558 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2559 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2560 * using) it for any given inode is up to filesystem.
2562 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2564 struct nameidata nd;
2565 void *cookie;
2567 nd.depth = 0;
2568 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2569 if (!IS_ERR(cookie)) {
2570 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2571 if (dentry->d_inode->i_op->put_link)
2572 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2573 cookie = ERR_PTR(res);
2575 return PTR_ERR(cookie);
2578 int vfs_follow_link(struct nameidata *nd, const char *link)
2580 return __vfs_follow_link(nd, link);
2583 /* get the link contents into pagecache */
2584 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2586 struct page * page;
2587 struct address_space *mapping = dentry->d_inode->i_mapping;
2588 page = read_mapping_page(mapping, 0, NULL);
2589 if (IS_ERR(page))
2590 goto sync_fail;
2591 wait_on_page_locked(page);
2592 if (!PageUptodate(page))
2593 goto async_fail;
2594 *ppage = page;
2595 return kmap(page);
2597 async_fail:
2598 page_cache_release(page);
2599 return ERR_PTR(-EIO);
2601 sync_fail:
2602 return (char*)page;
2605 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2607 struct page *page = NULL;
2608 char *s = page_getlink(dentry, &page);
2609 int res = vfs_readlink(dentry,buffer,buflen,s);
2610 if (page) {
2611 kunmap(page);
2612 page_cache_release(page);
2614 return res;
2617 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2619 struct page *page = NULL;
2620 nd_set_link(nd, page_getlink(dentry, &page));
2621 return page;
2624 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2626 struct page *page = cookie;
2628 if (page) {
2629 kunmap(page);
2630 page_cache_release(page);
2634 int __page_symlink(struct inode *inode, const char *symname, int len,
2635 gfp_t gfp_mask)
2637 struct address_space *mapping = inode->i_mapping;
2638 struct page *page;
2639 int err = -ENOMEM;
2640 char *kaddr;
2642 retry:
2643 page = find_or_create_page(mapping, 0, gfp_mask);
2644 if (!page)
2645 goto fail;
2646 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2647 if (err == AOP_TRUNCATED_PAGE) {
2648 page_cache_release(page);
2649 goto retry;
2651 if (err)
2652 goto fail_map;
2653 kaddr = kmap_atomic(page, KM_USER0);
2654 memcpy(kaddr, symname, len-1);
2655 kunmap_atomic(kaddr, KM_USER0);
2656 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2657 if (err == AOP_TRUNCATED_PAGE) {
2658 page_cache_release(page);
2659 goto retry;
2661 if (err)
2662 goto fail_map;
2664 * Notice that we are _not_ going to block here - end of page is
2665 * unmapped, so this will only try to map the rest of page, see
2666 * that it is unmapped (typically even will not look into inode -
2667 * ->i_size will be enough for everything) and zero it out.
2668 * OTOH it's obviously correct and should make the page up-to-date.
2670 if (!PageUptodate(page)) {
2671 err = mapping->a_ops->readpage(NULL, page);
2672 if (err != AOP_TRUNCATED_PAGE)
2673 wait_on_page_locked(page);
2674 } else {
2675 unlock_page(page);
2677 page_cache_release(page);
2678 if (err < 0)
2679 goto fail;
2680 mark_inode_dirty(inode);
2681 return 0;
2682 fail_map:
2683 unlock_page(page);
2684 page_cache_release(page);
2685 fail:
2686 return err;
2689 int page_symlink(struct inode *inode, const char *symname, int len)
2691 return __page_symlink(inode, symname, len,
2692 mapping_gfp_mask(inode->i_mapping));
2695 struct inode_operations page_symlink_inode_operations = {
2696 .readlink = generic_readlink,
2697 .follow_link = page_follow_link_light,
2698 .put_link = page_put_link,
2701 EXPORT_SYMBOL(__user_walk);
2702 EXPORT_SYMBOL(__user_walk_fd);
2703 EXPORT_SYMBOL(follow_down);
2704 EXPORT_SYMBOL(follow_up);
2705 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2706 EXPORT_SYMBOL(getname);
2707 EXPORT_SYMBOL(lock_rename);
2708 EXPORT_SYMBOL(lookup_one_len);
2709 EXPORT_SYMBOL(page_follow_link_light);
2710 EXPORT_SYMBOL(page_put_link);
2711 EXPORT_SYMBOL(page_readlink);
2712 EXPORT_SYMBOL(__page_symlink);
2713 EXPORT_SYMBOL(page_symlink);
2714 EXPORT_SYMBOL(page_symlink_inode_operations);
2715 EXPORT_SYMBOL(path_lookup);
2716 EXPORT_SYMBOL(path_release);
2717 EXPORT_SYMBOL(path_walk);
2718 EXPORT_SYMBOL(permission);
2719 EXPORT_SYMBOL(vfs_permission);
2720 EXPORT_SYMBOL(file_permission);
2721 EXPORT_SYMBOL(unlock_rename);
2722 EXPORT_SYMBOL(vfs_create);
2723 EXPORT_SYMBOL(vfs_follow_link);
2724 EXPORT_SYMBOL(vfs_link);
2725 EXPORT_SYMBOL(vfs_mkdir);
2726 EXPORT_SYMBOL(vfs_mknod);
2727 EXPORT_SYMBOL(generic_permission);
2728 EXPORT_SYMBOL(vfs_readlink);
2729 EXPORT_SYMBOL(vfs_rename);
2730 EXPORT_SYMBOL(vfs_rmdir);
2731 EXPORT_SYMBOL(vfs_symlink);
2732 EXPORT_SYMBOL(vfs_unlink);
2733 EXPORT_SYMBOL(dentry_unhash);
2734 EXPORT_SYMBOL(generic_readlink);