x86, xsave: xsave/xrstor specific routines
[linux-2.6/verdex.git] / fs / namei.c
bloba7b0a0b8012873dd1c8db6fe79c45b459096af41
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/personality.h>
26 #include <linux/security.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <asm/uaccess.h>
36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38 /* [Feb-1997 T. Schoebel-Theuer]
39 * Fundamental changes in the pathname lookup mechanisms (namei)
40 * were necessary because of omirr. The reason is that omirr needs
41 * to know the _real_ pathname, not the user-supplied one, in case
42 * of symlinks (and also when transname replacements occur).
44 * The new code replaces the old recursive symlink resolution with
45 * an iterative one (in case of non-nested symlink chains). It does
46 * this with calls to <fs>_follow_link().
47 * As a side effect, dir_namei(), _namei() and follow_link() are now
48 * replaced with a single function lookup_dentry() that can handle all
49 * the special cases of the former code.
51 * With the new dcache, the pathname is stored at each inode, at least as
52 * long as the refcount of the inode is positive. As a side effect, the
53 * size of the dcache depends on the inode cache and thus is dynamic.
55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
56 * resolution to correspond with current state of the code.
58 * Note that the symlink resolution is not *completely* iterative.
59 * There is still a significant amount of tail- and mid- recursion in
60 * the algorithm. Also, note that <fs>_readlink() is not used in
61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
62 * may return different results than <fs>_follow_link(). Many virtual
63 * filesystems (including /proc) exhibit this behavior.
66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
68 * and the name already exists in form of a symlink, try to create the new
69 * name indicated by the symlink. The old code always complained that the
70 * name already exists, due to not following the symlink even if its target
71 * is nonexistent. The new semantics affects also mknod() and link() when
72 * the name is a symlink pointing to a non-existant name.
74 * I don't know which semantics is the right one, since I have no access
75 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
77 * "old" one. Personally, I think the new semantics is much more logical.
78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
79 * file does succeed in both HP-UX and SunOs, but not in Solaris
80 * and in the old Linux semantics.
83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
84 * semantics. See the comments in "open_namei" and "do_link" below.
86 * [10-Sep-98 Alan Modra] Another symlink change.
89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
90 * inside the path - always follow.
91 * in the last component in creation/removal/renaming - never follow.
92 * if LOOKUP_FOLLOW passed - follow.
93 * if the pathname has trailing slashes - follow.
94 * otherwise - don't follow.
95 * (applied in that order).
97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
99 * During the 2.4 we need to fix the userland stuff depending on it -
100 * hopefully we will be able to get rid of that wart in 2.5. So far only
101 * XEmacs seems to be relying on it...
104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
106 * any extra contention...
109 static int __link_path_walk(const char *name, struct nameidata *nd);
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(!audit_dummy_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 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
189 if (current->fsuid == inode->i_uid)
190 mode >>= 6;
191 else {
192 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
193 int error = check_acl(inode, mask);
194 if (error == -EACCES)
195 goto check_capabilities;
196 else if (error != -EAGAIN)
197 return error;
200 if (in_group_p(inode->i_gid))
201 mode >>= 3;
205 * If the DACs are ok we don't need any capability check.
207 if ((mask & ~mode) == 0)
208 return 0;
210 check_capabilities:
212 * Read/write DACs are always overridable.
213 * Executable DACs are overridable if at least one exec bit is set.
215 if (!(mask & MAY_EXEC) ||
216 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
217 if (capable(CAP_DAC_OVERRIDE))
218 return 0;
221 * Searching includes executable on directories, else just read.
223 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
224 if (capable(CAP_DAC_READ_SEARCH))
225 return 0;
227 return -EACCES;
230 int inode_permission(struct inode *inode, int mask)
232 int retval;
234 if (mask & MAY_WRITE) {
235 umode_t mode = inode->i_mode;
238 * Nobody gets write access to a read-only fs.
240 if (IS_RDONLY(inode) &&
241 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
242 return -EROFS;
245 * Nobody gets write access to an immutable file.
247 if (IS_IMMUTABLE(inode))
248 return -EACCES;
251 /* Ordinary permission routines do not understand MAY_APPEND. */
252 if (inode->i_op && inode->i_op->permission) {
253 retval = inode->i_op->permission(inode, mask);
254 if (!retval) {
256 * Exec permission on a regular file is denied if none
257 * of the execute bits are set.
259 * This check should be done by the ->permission()
260 * method.
262 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
263 !(inode->i_mode & S_IXUGO))
264 return -EACCES;
266 } else {
267 retval = generic_permission(inode, mask, NULL);
269 if (retval)
270 return retval;
272 retval = devcgroup_inode_permission(inode, mask);
273 if (retval)
274 return retval;
276 return security_inode_permission(inode,
277 mask & (MAY_READ|MAY_WRITE|MAY_EXEC));
281 * vfs_permission - check for access rights to a given path
282 * @nd: lookup result that describes the path
283 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
285 * Used to check for read/write/execute permissions on a path.
286 * We use "fsuid" for this, letting us set arbitrary permissions
287 * for filesystem access without changing the "normal" uids which
288 * are used for other things.
290 int vfs_permission(struct nameidata *nd, int mask)
292 return inode_permission(nd->path.dentry->d_inode, mask);
296 * file_permission - check for additional access rights to a given file
297 * @file: file to check access rights for
298 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
300 * Used to check for read/write/execute permissions on an already opened
301 * file.
303 * Note:
304 * Do not use this function in new code. All access checks should
305 * be done using vfs_permission().
307 int file_permission(struct file *file, int mask)
309 return inode_permission(file->f_path.dentry->d_inode, mask);
313 * get_write_access() gets write permission for a file.
314 * put_write_access() releases this write permission.
315 * This is used for regular files.
316 * We cannot support write (and maybe mmap read-write shared) accesses and
317 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
318 * can have the following values:
319 * 0: no writers, no VM_DENYWRITE mappings
320 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
321 * > 0: (i_writecount) users are writing to the file.
323 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
324 * except for the cases where we don't hold i_writecount yet. Then we need to
325 * use {get,deny}_write_access() - these functions check the sign and refuse
326 * to do the change if sign is wrong. Exclusion between them is provided by
327 * the inode->i_lock spinlock.
330 int get_write_access(struct inode * inode)
332 spin_lock(&inode->i_lock);
333 if (atomic_read(&inode->i_writecount) < 0) {
334 spin_unlock(&inode->i_lock);
335 return -ETXTBSY;
337 atomic_inc(&inode->i_writecount);
338 spin_unlock(&inode->i_lock);
340 return 0;
343 int deny_write_access(struct file * file)
345 struct inode *inode = file->f_path.dentry->d_inode;
347 spin_lock(&inode->i_lock);
348 if (atomic_read(&inode->i_writecount) > 0) {
349 spin_unlock(&inode->i_lock);
350 return -ETXTBSY;
352 atomic_dec(&inode->i_writecount);
353 spin_unlock(&inode->i_lock);
355 return 0;
359 * path_get - get a reference to a path
360 * @path: path to get the reference to
362 * Given a path increment the reference count to the dentry and the vfsmount.
364 void path_get(struct path *path)
366 mntget(path->mnt);
367 dget(path->dentry);
369 EXPORT_SYMBOL(path_get);
372 * path_put - put a reference to a path
373 * @path: path to put the reference to
375 * Given a path decrement the reference count to the dentry and the vfsmount.
377 void path_put(struct path *path)
379 dput(path->dentry);
380 mntput(path->mnt);
382 EXPORT_SYMBOL(path_put);
385 * release_open_intent - free up open intent resources
386 * @nd: pointer to nameidata
388 void release_open_intent(struct nameidata *nd)
390 if (nd->intent.open.file->f_path.dentry == NULL)
391 put_filp(nd->intent.open.file);
392 else
393 fput(nd->intent.open.file);
396 static inline struct dentry *
397 do_revalidate(struct dentry *dentry, struct nameidata *nd)
399 int status = dentry->d_op->d_revalidate(dentry, nd);
400 if (unlikely(status <= 0)) {
402 * The dentry failed validation.
403 * If d_revalidate returned 0 attempt to invalidate
404 * the dentry otherwise d_revalidate is asking us
405 * to return a fail status.
407 if (!status) {
408 if (!d_invalidate(dentry)) {
409 dput(dentry);
410 dentry = NULL;
412 } else {
413 dput(dentry);
414 dentry = ERR_PTR(status);
417 return dentry;
421 * Internal lookup() using the new generic dcache.
422 * SMP-safe
424 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
426 struct dentry * dentry = __d_lookup(parent, name);
428 /* lockess __d_lookup may fail due to concurrent d_move()
429 * in some unrelated directory, so try with d_lookup
431 if (!dentry)
432 dentry = d_lookup(parent, name);
434 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
435 dentry = do_revalidate(dentry, nd);
437 return dentry;
441 * Short-cut version of permission(), for calling by
442 * path_walk(), when dcache lock is held. Combines parts
443 * of permission() and generic_permission(), and tests ONLY for
444 * MAY_EXEC permission.
446 * If appropriate, check DAC only. If not appropriate, or
447 * short-cut DAC fails, then call permission() to do more
448 * complete permission check.
450 static int exec_permission_lite(struct inode *inode)
452 umode_t mode = inode->i_mode;
454 if (inode->i_op && inode->i_op->permission)
455 return -EAGAIN;
457 if (current->fsuid == inode->i_uid)
458 mode >>= 6;
459 else if (in_group_p(inode->i_gid))
460 mode >>= 3;
462 if (mode & MAY_EXEC)
463 goto ok;
465 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
466 goto ok;
468 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
469 goto ok;
471 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
472 goto ok;
474 return -EACCES;
476 return security_inode_permission(inode, MAY_EXEC);
480 * This is called when everything else fails, and we actually have
481 * to go to the low-level filesystem to find out what we should do..
483 * We get the directory semaphore, and after getting that we also
484 * make sure that nobody added the entry to the dcache in the meantime..
485 * SMP-safe
487 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
489 struct dentry * result;
490 struct inode *dir = parent->d_inode;
492 mutex_lock(&dir->i_mutex);
494 * First re-do the cached lookup just in case it was created
495 * while we waited for the directory semaphore..
497 * FIXME! This could use version numbering or similar to
498 * avoid unnecessary cache lookups.
500 * The "dcache_lock" is purely to protect the RCU list walker
501 * from concurrent renames at this point (we mustn't get false
502 * negatives from the RCU list walk here, unlike the optimistic
503 * fast walk).
505 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
507 result = d_lookup(parent, name);
508 if (!result) {
509 struct dentry *dentry;
511 /* Don't create child dentry for a dead directory. */
512 result = ERR_PTR(-ENOENT);
513 if (IS_DEADDIR(dir))
514 goto out_unlock;
516 dentry = d_alloc(parent, name);
517 result = ERR_PTR(-ENOMEM);
518 if (dentry) {
519 result = dir->i_op->lookup(dir, dentry, nd);
520 if (result)
521 dput(dentry);
522 else
523 result = dentry;
525 out_unlock:
526 mutex_unlock(&dir->i_mutex);
527 return result;
531 * Uhhuh! Nasty case: the cache was re-populated while
532 * we waited on the semaphore. Need to revalidate.
534 mutex_unlock(&dir->i_mutex);
535 if (result->d_op && result->d_op->d_revalidate) {
536 result = do_revalidate(result, nd);
537 if (!result)
538 result = ERR_PTR(-ENOENT);
540 return result;
543 /* SMP-safe */
544 static __always_inline void
545 walk_init_root(const char *name, struct nameidata *nd)
547 struct fs_struct *fs = current->fs;
549 read_lock(&fs->lock);
550 nd->path = fs->root;
551 path_get(&fs->root);
552 read_unlock(&fs->lock);
556 * Wrapper to retry pathname resolution whenever the underlying
557 * file system returns an ESTALE.
559 * Retry the whole path once, forcing real lookup requests
560 * instead of relying on the dcache.
562 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
564 struct path save = nd->path;
565 int result;
567 /* make sure the stuff we saved doesn't go away */
568 path_get(&save);
570 result = __link_path_walk(name, nd);
571 if (result == -ESTALE) {
572 /* nd->path had been dropped */
573 nd->path = save;
574 path_get(&nd->path);
575 nd->flags |= LOOKUP_REVAL;
576 result = __link_path_walk(name, nd);
579 path_put(&save);
581 return result;
584 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
586 int res = 0;
587 char *name;
588 if (IS_ERR(link))
589 goto fail;
591 if (*link == '/') {
592 path_put(&nd->path);
593 walk_init_root(link, nd);
595 res = link_path_walk(link, nd);
596 if (nd->depth || res || nd->last_type!=LAST_NORM)
597 return res;
599 * If it is an iterative symlinks resolution in open_namei() we
600 * have to copy the last component. And all that crap because of
601 * bloody create() on broken symlinks. Furrfu...
603 name = __getname();
604 if (unlikely(!name)) {
605 path_put(&nd->path);
606 return -ENOMEM;
608 strcpy(name, nd->last.name);
609 nd->last.name = name;
610 return 0;
611 fail:
612 path_put(&nd->path);
613 return PTR_ERR(link);
616 static void path_put_conditional(struct path *path, struct nameidata *nd)
618 dput(path->dentry);
619 if (path->mnt != nd->path.mnt)
620 mntput(path->mnt);
623 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
625 dput(nd->path.dentry);
626 if (nd->path.mnt != path->mnt)
627 mntput(nd->path.mnt);
628 nd->path.mnt = path->mnt;
629 nd->path.dentry = path->dentry;
632 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
634 int error;
635 void *cookie;
636 struct dentry *dentry = path->dentry;
638 touch_atime(path->mnt, dentry);
639 nd_set_link(nd, NULL);
641 if (path->mnt != nd->path.mnt) {
642 path_to_nameidata(path, nd);
643 dget(dentry);
645 mntget(path->mnt);
646 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
647 error = PTR_ERR(cookie);
648 if (!IS_ERR(cookie)) {
649 char *s = nd_get_link(nd);
650 error = 0;
651 if (s)
652 error = __vfs_follow_link(nd, s);
653 if (dentry->d_inode->i_op->put_link)
654 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
656 path_put(path);
658 return error;
662 * This limits recursive symlink follows to 8, while
663 * limiting consecutive symlinks to 40.
665 * Without that kind of total limit, nasty chains of consecutive
666 * symlinks can cause almost arbitrarily long lookups.
668 static inline int do_follow_link(struct path *path, struct nameidata *nd)
670 int err = -ELOOP;
671 if (current->link_count >= MAX_NESTED_LINKS)
672 goto loop;
673 if (current->total_link_count >= 40)
674 goto loop;
675 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
676 cond_resched();
677 err = security_inode_follow_link(path->dentry, nd);
678 if (err)
679 goto loop;
680 current->link_count++;
681 current->total_link_count++;
682 nd->depth++;
683 err = __do_follow_link(path, nd);
684 current->link_count--;
685 nd->depth--;
686 return err;
687 loop:
688 path_put_conditional(path, nd);
689 path_put(&nd->path);
690 return err;
693 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
695 struct vfsmount *parent;
696 struct dentry *mountpoint;
697 spin_lock(&vfsmount_lock);
698 parent=(*mnt)->mnt_parent;
699 if (parent == *mnt) {
700 spin_unlock(&vfsmount_lock);
701 return 0;
703 mntget(parent);
704 mountpoint=dget((*mnt)->mnt_mountpoint);
705 spin_unlock(&vfsmount_lock);
706 dput(*dentry);
707 *dentry = mountpoint;
708 mntput(*mnt);
709 *mnt = parent;
710 return 1;
713 /* no need for dcache_lock, as serialization is taken care in
714 * namespace.c
716 static int __follow_mount(struct path *path)
718 int res = 0;
719 while (d_mountpoint(path->dentry)) {
720 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
721 if (!mounted)
722 break;
723 dput(path->dentry);
724 if (res)
725 mntput(path->mnt);
726 path->mnt = mounted;
727 path->dentry = dget(mounted->mnt_root);
728 res = 1;
730 return res;
733 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
735 while (d_mountpoint(*dentry)) {
736 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
737 if (!mounted)
738 break;
739 dput(*dentry);
740 mntput(*mnt);
741 *mnt = mounted;
742 *dentry = dget(mounted->mnt_root);
746 /* no need for dcache_lock, as serialization is taken care in
747 * namespace.c
749 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
751 struct vfsmount *mounted;
753 mounted = lookup_mnt(*mnt, *dentry);
754 if (mounted) {
755 dput(*dentry);
756 mntput(*mnt);
757 *mnt = mounted;
758 *dentry = dget(mounted->mnt_root);
759 return 1;
761 return 0;
764 static __always_inline void follow_dotdot(struct nameidata *nd)
766 struct fs_struct *fs = current->fs;
768 while(1) {
769 struct vfsmount *parent;
770 struct dentry *old = nd->path.dentry;
772 read_lock(&fs->lock);
773 if (nd->path.dentry == fs->root.dentry &&
774 nd->path.mnt == fs->root.mnt) {
775 read_unlock(&fs->lock);
776 break;
778 read_unlock(&fs->lock);
779 spin_lock(&dcache_lock);
780 if (nd->path.dentry != nd->path.mnt->mnt_root) {
781 nd->path.dentry = dget(nd->path.dentry->d_parent);
782 spin_unlock(&dcache_lock);
783 dput(old);
784 break;
786 spin_unlock(&dcache_lock);
787 spin_lock(&vfsmount_lock);
788 parent = nd->path.mnt->mnt_parent;
789 if (parent == nd->path.mnt) {
790 spin_unlock(&vfsmount_lock);
791 break;
793 mntget(parent);
794 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
795 spin_unlock(&vfsmount_lock);
796 dput(old);
797 mntput(nd->path.mnt);
798 nd->path.mnt = parent;
800 follow_mount(&nd->path.mnt, &nd->path.dentry);
804 * It's more convoluted than I'd like it to be, but... it's still fairly
805 * small and for now I'd prefer to have fast path as straight as possible.
806 * It _is_ time-critical.
808 static int do_lookup(struct nameidata *nd, struct qstr *name,
809 struct path *path)
811 struct vfsmount *mnt = nd->path.mnt;
812 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
814 if (!dentry)
815 goto need_lookup;
816 if (dentry->d_op && dentry->d_op->d_revalidate)
817 goto need_revalidate;
818 done:
819 path->mnt = mnt;
820 path->dentry = dentry;
821 __follow_mount(path);
822 return 0;
824 need_lookup:
825 dentry = real_lookup(nd->path.dentry, name, nd);
826 if (IS_ERR(dentry))
827 goto fail;
828 goto done;
830 need_revalidate:
831 dentry = do_revalidate(dentry, nd);
832 if (!dentry)
833 goto need_lookup;
834 if (IS_ERR(dentry))
835 goto fail;
836 goto done;
838 fail:
839 return PTR_ERR(dentry);
843 * Name resolution.
844 * This is the basic name resolution function, turning a pathname into
845 * the final dentry. We expect 'base' to be positive and a directory.
847 * Returns 0 and nd will have valid dentry and mnt on success.
848 * Returns error and drops reference to input namei data on failure.
850 static int __link_path_walk(const char *name, struct nameidata *nd)
852 struct path next;
853 struct inode *inode;
854 int err;
855 unsigned int lookup_flags = nd->flags;
857 while (*name=='/')
858 name++;
859 if (!*name)
860 goto return_reval;
862 inode = nd->path.dentry->d_inode;
863 if (nd->depth)
864 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
866 /* At this point we know we have a real path component. */
867 for(;;) {
868 unsigned long hash;
869 struct qstr this;
870 unsigned int c;
872 nd->flags |= LOOKUP_CONTINUE;
873 err = exec_permission_lite(inode);
874 if (err == -EAGAIN)
875 err = vfs_permission(nd, MAY_EXEC);
876 if (err)
877 break;
879 this.name = name;
880 c = *(const unsigned char *)name;
882 hash = init_name_hash();
883 do {
884 name++;
885 hash = partial_name_hash(c, hash);
886 c = *(const unsigned char *)name;
887 } while (c && (c != '/'));
888 this.len = name - (const char *) this.name;
889 this.hash = end_name_hash(hash);
891 /* remove trailing slashes? */
892 if (!c)
893 goto last_component;
894 while (*++name == '/');
895 if (!*name)
896 goto last_with_slashes;
899 * "." and ".." are special - ".." especially so because it has
900 * to be able to know about the current root directory and
901 * parent relationships.
903 if (this.name[0] == '.') switch (this.len) {
904 default:
905 break;
906 case 2:
907 if (this.name[1] != '.')
908 break;
909 follow_dotdot(nd);
910 inode = nd->path.dentry->d_inode;
911 /* fallthrough */
912 case 1:
913 continue;
916 * See if the low-level filesystem might want
917 * to use its own hash..
919 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
920 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
921 &this);
922 if (err < 0)
923 break;
925 /* This does the actual lookups.. */
926 err = do_lookup(nd, &this, &next);
927 if (err)
928 break;
930 err = -ENOENT;
931 inode = next.dentry->d_inode;
932 if (!inode)
933 goto out_dput;
934 err = -ENOTDIR;
935 if (!inode->i_op)
936 goto out_dput;
938 if (inode->i_op->follow_link) {
939 err = do_follow_link(&next, nd);
940 if (err)
941 goto return_err;
942 err = -ENOENT;
943 inode = nd->path.dentry->d_inode;
944 if (!inode)
945 break;
946 err = -ENOTDIR;
947 if (!inode->i_op)
948 break;
949 } else
950 path_to_nameidata(&next, nd);
951 err = -ENOTDIR;
952 if (!inode->i_op->lookup)
953 break;
954 continue;
955 /* here ends the main loop */
957 last_with_slashes:
958 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
959 last_component:
960 /* Clear LOOKUP_CONTINUE iff it was previously unset */
961 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
962 if (lookup_flags & LOOKUP_PARENT)
963 goto lookup_parent;
964 if (this.name[0] == '.') switch (this.len) {
965 default:
966 break;
967 case 2:
968 if (this.name[1] != '.')
969 break;
970 follow_dotdot(nd);
971 inode = nd->path.dentry->d_inode;
972 /* fallthrough */
973 case 1:
974 goto return_reval;
976 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
977 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
978 &this);
979 if (err < 0)
980 break;
982 err = do_lookup(nd, &this, &next);
983 if (err)
984 break;
985 inode = next.dentry->d_inode;
986 if ((lookup_flags & LOOKUP_FOLLOW)
987 && inode && inode->i_op && inode->i_op->follow_link) {
988 err = do_follow_link(&next, nd);
989 if (err)
990 goto return_err;
991 inode = nd->path.dentry->d_inode;
992 } else
993 path_to_nameidata(&next, nd);
994 err = -ENOENT;
995 if (!inode)
996 break;
997 if (lookup_flags & LOOKUP_DIRECTORY) {
998 err = -ENOTDIR;
999 if (!inode->i_op || !inode->i_op->lookup)
1000 break;
1002 goto return_base;
1003 lookup_parent:
1004 nd->last = this;
1005 nd->last_type = LAST_NORM;
1006 if (this.name[0] != '.')
1007 goto return_base;
1008 if (this.len == 1)
1009 nd->last_type = LAST_DOT;
1010 else if (this.len == 2 && this.name[1] == '.')
1011 nd->last_type = LAST_DOTDOT;
1012 else
1013 goto return_base;
1014 return_reval:
1016 * We bypassed the ordinary revalidation routines.
1017 * We may need to check the cached dentry for staleness.
1019 if (nd->path.dentry && nd->path.dentry->d_sb &&
1020 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1021 err = -ESTALE;
1022 /* Note: we do not d_invalidate() */
1023 if (!nd->path.dentry->d_op->d_revalidate(
1024 nd->path.dentry, nd))
1025 break;
1027 return_base:
1028 return 0;
1029 out_dput:
1030 path_put_conditional(&next, nd);
1031 break;
1033 path_put(&nd->path);
1034 return_err:
1035 return err;
1038 static int path_walk(const char *name, struct nameidata *nd)
1040 current->total_link_count = 0;
1041 return link_path_walk(name, nd);
1044 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1045 static int do_path_lookup(int dfd, const char *name,
1046 unsigned int flags, struct nameidata *nd)
1048 int retval = 0;
1049 int fput_needed;
1050 struct file *file;
1051 struct fs_struct *fs = current->fs;
1053 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1054 nd->flags = flags;
1055 nd->depth = 0;
1057 if (*name=='/') {
1058 read_lock(&fs->lock);
1059 nd->path = fs->root;
1060 path_get(&fs->root);
1061 read_unlock(&fs->lock);
1062 } else if (dfd == AT_FDCWD) {
1063 read_lock(&fs->lock);
1064 nd->path = fs->pwd;
1065 path_get(&fs->pwd);
1066 read_unlock(&fs->lock);
1067 } else {
1068 struct dentry *dentry;
1070 file = fget_light(dfd, &fput_needed);
1071 retval = -EBADF;
1072 if (!file)
1073 goto out_fail;
1075 dentry = file->f_path.dentry;
1077 retval = -ENOTDIR;
1078 if (!S_ISDIR(dentry->d_inode->i_mode))
1079 goto fput_fail;
1081 retval = file_permission(file, MAY_EXEC);
1082 if (retval)
1083 goto fput_fail;
1085 nd->path = file->f_path;
1086 path_get(&file->f_path);
1088 fput_light(file, fput_needed);
1091 retval = path_walk(name, nd);
1092 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1093 nd->path.dentry->d_inode))
1094 audit_inode(name, nd->path.dentry);
1095 out_fail:
1096 return retval;
1098 fput_fail:
1099 fput_light(file, fput_needed);
1100 goto out_fail;
1103 int path_lookup(const char *name, unsigned int flags,
1104 struct nameidata *nd)
1106 return do_path_lookup(AT_FDCWD, name, flags, nd);
1110 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1111 * @dentry: pointer to dentry of the base directory
1112 * @mnt: pointer to vfs mount of the base directory
1113 * @name: pointer to file name
1114 * @flags: lookup flags
1115 * @nd: pointer to nameidata
1117 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1118 const char *name, unsigned int flags,
1119 struct nameidata *nd)
1121 int retval;
1123 /* same as do_path_lookup */
1124 nd->last_type = LAST_ROOT;
1125 nd->flags = flags;
1126 nd->depth = 0;
1128 nd->path.dentry = dentry;
1129 nd->path.mnt = mnt;
1130 path_get(&nd->path);
1132 retval = path_walk(name, nd);
1133 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1134 nd->path.dentry->d_inode))
1135 audit_inode(name, nd->path.dentry);
1137 return retval;
1141 static int __path_lookup_intent_open(int dfd, const char *name,
1142 unsigned int lookup_flags, struct nameidata *nd,
1143 int open_flags, int create_mode)
1145 struct file *filp = get_empty_filp();
1146 int err;
1148 if (filp == NULL)
1149 return -ENFILE;
1150 nd->intent.open.file = filp;
1151 nd->intent.open.flags = open_flags;
1152 nd->intent.open.create_mode = create_mode;
1153 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1154 if (IS_ERR(nd->intent.open.file)) {
1155 if (err == 0) {
1156 err = PTR_ERR(nd->intent.open.file);
1157 path_put(&nd->path);
1159 } else if (err != 0)
1160 release_open_intent(nd);
1161 return err;
1165 * path_lookup_open - lookup a file path with open intent
1166 * @dfd: the directory to use as base, or AT_FDCWD
1167 * @name: pointer to file name
1168 * @lookup_flags: lookup intent flags
1169 * @nd: pointer to nameidata
1170 * @open_flags: open intent flags
1172 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1173 struct nameidata *nd, int open_flags)
1175 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1176 open_flags, 0);
1180 * path_lookup_create - lookup a file path with open + create intent
1181 * @dfd: the directory to use as base, or AT_FDCWD
1182 * @name: pointer to file name
1183 * @lookup_flags: lookup intent flags
1184 * @nd: pointer to nameidata
1185 * @open_flags: open intent flags
1186 * @create_mode: create intent flags
1188 static int path_lookup_create(int dfd, const char *name,
1189 unsigned int lookup_flags, struct nameidata *nd,
1190 int open_flags, int create_mode)
1192 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1193 nd, open_flags, create_mode);
1196 static struct dentry *__lookup_hash(struct qstr *name,
1197 struct dentry *base, struct nameidata *nd)
1199 struct dentry *dentry;
1200 struct inode *inode;
1201 int err;
1203 inode = base->d_inode;
1206 * See if the low-level filesystem might want
1207 * to use its own hash..
1209 if (base->d_op && base->d_op->d_hash) {
1210 err = base->d_op->d_hash(base, name);
1211 dentry = ERR_PTR(err);
1212 if (err < 0)
1213 goto out;
1216 dentry = cached_lookup(base, name, nd);
1217 if (!dentry) {
1218 struct dentry *new;
1220 /* Don't create child dentry for a dead directory. */
1221 dentry = ERR_PTR(-ENOENT);
1222 if (IS_DEADDIR(inode))
1223 goto out;
1225 new = d_alloc(base, name);
1226 dentry = ERR_PTR(-ENOMEM);
1227 if (!new)
1228 goto out;
1229 dentry = inode->i_op->lookup(inode, new, nd);
1230 if (!dentry)
1231 dentry = new;
1232 else
1233 dput(new);
1235 out:
1236 return dentry;
1240 * Restricted form of lookup. Doesn't follow links, single-component only,
1241 * needs parent already locked. Doesn't follow mounts.
1242 * SMP-safe.
1244 static struct dentry *lookup_hash(struct nameidata *nd)
1246 int err;
1248 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1249 if (err)
1250 return ERR_PTR(err);
1251 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1254 static int __lookup_one_len(const char *name, struct qstr *this,
1255 struct dentry *base, int len)
1257 unsigned long hash;
1258 unsigned int c;
1260 this->name = name;
1261 this->len = len;
1262 if (!len)
1263 return -EACCES;
1265 hash = init_name_hash();
1266 while (len--) {
1267 c = *(const unsigned char *)name++;
1268 if (c == '/' || c == '\0')
1269 return -EACCES;
1270 hash = partial_name_hash(c, hash);
1272 this->hash = end_name_hash(hash);
1273 return 0;
1277 * lookup_one_len - filesystem helper to lookup single pathname component
1278 * @name: pathname component to lookup
1279 * @base: base directory to lookup from
1280 * @len: maximum length @len should be interpreted to
1282 * Note that this routine is purely a helper for filesystem usage and should
1283 * not be called by generic code. Also note that by using this function the
1284 * nameidata argument is passed to the filesystem methods and a filesystem
1285 * using this helper needs to be prepared for that.
1287 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1289 int err;
1290 struct qstr this;
1292 err = __lookup_one_len(name, &this, base, len);
1293 if (err)
1294 return ERR_PTR(err);
1296 err = inode_permission(base->d_inode, MAY_EXEC);
1297 if (err)
1298 return ERR_PTR(err);
1299 return __lookup_hash(&this, base, NULL);
1303 * lookup_one_noperm - bad hack for sysfs
1304 * @name: pathname component to lookup
1305 * @base: base directory to lookup from
1307 * This is a variant of lookup_one_len that doesn't perform any permission
1308 * checks. It's a horrible hack to work around the braindead sysfs
1309 * architecture and should not be used anywhere else.
1311 * DON'T USE THIS FUNCTION EVER, thanks.
1313 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1315 int err;
1316 struct qstr this;
1318 err = __lookup_one_len(name, &this, base, strlen(name));
1319 if (err)
1320 return ERR_PTR(err);
1321 return __lookup_hash(&this, base, NULL);
1324 int user_path_at(int dfd, const char __user *name, unsigned flags,
1325 struct path *path)
1327 struct nameidata nd;
1328 char *tmp = getname(name);
1329 int err = PTR_ERR(tmp);
1330 if (!IS_ERR(tmp)) {
1332 BUG_ON(flags & LOOKUP_PARENT);
1334 err = do_path_lookup(dfd, tmp, flags, &nd);
1335 putname(tmp);
1336 if (!err)
1337 *path = nd.path;
1339 return err;
1342 static int user_path_parent(int dfd, const char __user *path,
1343 struct nameidata *nd, char **name)
1345 char *s = getname(path);
1346 int error;
1348 if (IS_ERR(s))
1349 return PTR_ERR(s);
1351 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1352 if (error)
1353 putname(s);
1354 else
1355 *name = s;
1357 return error;
1361 * It's inline, so penalty for filesystems that don't use sticky bit is
1362 * minimal.
1364 static inline int check_sticky(struct inode *dir, struct inode *inode)
1366 if (!(dir->i_mode & S_ISVTX))
1367 return 0;
1368 if (inode->i_uid == current->fsuid)
1369 return 0;
1370 if (dir->i_uid == current->fsuid)
1371 return 0;
1372 return !capable(CAP_FOWNER);
1376 * Check whether we can remove a link victim from directory dir, check
1377 * whether the type of victim is right.
1378 * 1. We can't do it if dir is read-only (done in permission())
1379 * 2. We should have write and exec permissions on dir
1380 * 3. We can't remove anything from append-only dir
1381 * 4. We can't do anything with immutable dir (done in permission())
1382 * 5. If the sticky bit on dir is set we should either
1383 * a. be owner of dir, or
1384 * b. be owner of victim, or
1385 * c. have CAP_FOWNER capability
1386 * 6. If the victim is append-only or immutable we can't do antyhing with
1387 * links pointing to it.
1388 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1389 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1390 * 9. We can't remove a root or mountpoint.
1391 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1392 * nfs_async_unlink().
1394 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1396 int error;
1398 if (!victim->d_inode)
1399 return -ENOENT;
1401 BUG_ON(victim->d_parent->d_inode != dir);
1402 audit_inode_child(victim->d_name.name, victim, dir);
1404 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1405 if (error)
1406 return error;
1407 if (IS_APPEND(dir))
1408 return -EPERM;
1409 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1410 IS_IMMUTABLE(victim->d_inode))
1411 return -EPERM;
1412 if (isdir) {
1413 if (!S_ISDIR(victim->d_inode->i_mode))
1414 return -ENOTDIR;
1415 if (IS_ROOT(victim))
1416 return -EBUSY;
1417 } else if (S_ISDIR(victim->d_inode->i_mode))
1418 return -EISDIR;
1419 if (IS_DEADDIR(dir))
1420 return -ENOENT;
1421 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1422 return -EBUSY;
1423 return 0;
1426 /* Check whether we can create an object with dentry child in directory
1427 * dir.
1428 * 1. We can't do it if child already exists (open has special treatment for
1429 * this case, but since we are inlined it's OK)
1430 * 2. We can't do it if dir is read-only (done in permission())
1431 * 3. We should have write and exec permissions on dir
1432 * 4. We can't do it if dir is immutable (done in permission())
1434 static inline int may_create(struct inode *dir, struct dentry *child,
1435 struct nameidata *nd)
1437 if (child->d_inode)
1438 return -EEXIST;
1439 if (IS_DEADDIR(dir))
1440 return -ENOENT;
1441 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1445 * O_DIRECTORY translates into forcing a directory lookup.
1447 static inline int lookup_flags(unsigned int f)
1449 unsigned long retval = LOOKUP_FOLLOW;
1451 if (f & O_NOFOLLOW)
1452 retval &= ~LOOKUP_FOLLOW;
1454 if (f & O_DIRECTORY)
1455 retval |= LOOKUP_DIRECTORY;
1457 return retval;
1461 * p1 and p2 should be directories on the same fs.
1463 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1465 struct dentry *p;
1467 if (p1 == p2) {
1468 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1469 return NULL;
1472 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1474 for (p = p1; p->d_parent != p; p = p->d_parent) {
1475 if (p->d_parent == p2) {
1476 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1477 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1478 return p;
1482 for (p = p2; p->d_parent != p; p = p->d_parent) {
1483 if (p->d_parent == p1) {
1484 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1485 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1486 return p;
1490 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1491 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1492 return NULL;
1495 void unlock_rename(struct dentry *p1, struct dentry *p2)
1497 mutex_unlock(&p1->d_inode->i_mutex);
1498 if (p1 != p2) {
1499 mutex_unlock(&p2->d_inode->i_mutex);
1500 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1504 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1505 struct nameidata *nd)
1507 int error = may_create(dir, dentry, nd);
1509 if (error)
1510 return error;
1512 if (!dir->i_op || !dir->i_op->create)
1513 return -EACCES; /* shouldn't it be ENOSYS? */
1514 mode &= S_IALLUGO;
1515 mode |= S_IFREG;
1516 error = security_inode_create(dir, dentry, mode);
1517 if (error)
1518 return error;
1519 DQUOT_INIT(dir);
1520 error = dir->i_op->create(dir, dentry, mode, nd);
1521 if (!error)
1522 fsnotify_create(dir, dentry);
1523 return error;
1526 int may_open(struct nameidata *nd, int acc_mode, int flag)
1528 struct dentry *dentry = nd->path.dentry;
1529 struct inode *inode = dentry->d_inode;
1530 int error;
1532 if (!inode)
1533 return -ENOENT;
1535 if (S_ISLNK(inode->i_mode))
1536 return -ELOOP;
1538 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
1539 return -EISDIR;
1542 * FIFO's, sockets and device files are special: they don't
1543 * actually live on the filesystem itself, and as such you
1544 * can write to them even if the filesystem is read-only.
1546 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1547 flag &= ~O_TRUNC;
1548 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1549 if (nd->path.mnt->mnt_flags & MNT_NODEV)
1550 return -EACCES;
1552 flag &= ~O_TRUNC;
1555 error = vfs_permission(nd, acc_mode);
1556 if (error)
1557 return error;
1559 * An append-only file must be opened in append mode for writing.
1561 if (IS_APPEND(inode)) {
1562 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1563 return -EPERM;
1564 if (flag & O_TRUNC)
1565 return -EPERM;
1568 /* O_NOATIME can only be set by the owner or superuser */
1569 if (flag & O_NOATIME)
1570 if (!is_owner_or_cap(inode))
1571 return -EPERM;
1574 * Ensure there are no outstanding leases on the file.
1576 error = break_lease(inode, flag);
1577 if (error)
1578 return error;
1580 if (flag & O_TRUNC) {
1581 error = get_write_access(inode);
1582 if (error)
1583 return error;
1586 * Refuse to truncate files with mandatory locks held on them.
1588 error = locks_verify_locked(inode);
1589 if (!error) {
1590 DQUOT_INIT(inode);
1592 error = do_truncate(dentry, 0,
1593 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1594 NULL);
1596 put_write_access(inode);
1597 if (error)
1598 return error;
1599 } else
1600 if (flag & FMODE_WRITE)
1601 DQUOT_INIT(inode);
1603 return 0;
1607 * Be careful about ever adding any more callers of this
1608 * function. Its flags must be in the namei format, not
1609 * what get passed to sys_open().
1611 static int __open_namei_create(struct nameidata *nd, struct path *path,
1612 int flag, int mode)
1614 int error;
1615 struct dentry *dir = nd->path.dentry;
1617 if (!IS_POSIXACL(dir->d_inode))
1618 mode &= ~current->fs->umask;
1619 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1620 mutex_unlock(&dir->d_inode->i_mutex);
1621 dput(nd->path.dentry);
1622 nd->path.dentry = path->dentry;
1623 if (error)
1624 return error;
1625 /* Don't check for write permission, don't truncate */
1626 return may_open(nd, 0, flag & ~O_TRUNC);
1630 * Note that while the flag value (low two bits) for sys_open means:
1631 * 00 - read-only
1632 * 01 - write-only
1633 * 10 - read-write
1634 * 11 - special
1635 * it is changed into
1636 * 00 - no permissions needed
1637 * 01 - read-permission
1638 * 10 - write-permission
1639 * 11 - read-write
1640 * for the internal routines (ie open_namei()/follow_link() etc)
1641 * This is more logical, and also allows the 00 "no perm needed"
1642 * to be used for symlinks (where the permissions are checked
1643 * later).
1646 static inline int open_to_namei_flags(int flag)
1648 if ((flag+1) & O_ACCMODE)
1649 flag++;
1650 return flag;
1653 static int open_will_write_to_fs(int flag, struct inode *inode)
1656 * We'll never write to the fs underlying
1657 * a device file.
1659 if (special_file(inode->i_mode))
1660 return 0;
1661 return (flag & O_TRUNC);
1665 * Note that the low bits of the passed in "open_flag"
1666 * are not the same as in the local variable "flag". See
1667 * open_to_namei_flags() for more details.
1669 struct file *do_filp_open(int dfd, const char *pathname,
1670 int open_flag, int mode)
1672 struct file *filp;
1673 struct nameidata nd;
1674 int acc_mode, error;
1675 struct path path;
1676 struct dentry *dir;
1677 int count = 0;
1678 int will_write;
1679 int flag = open_to_namei_flags(open_flag);
1681 acc_mode = MAY_OPEN | ACC_MODE(flag);
1683 /* O_TRUNC implies we need access checks for write permissions */
1684 if (flag & O_TRUNC)
1685 acc_mode |= MAY_WRITE;
1687 /* Allow the LSM permission hook to distinguish append
1688 access from general write access. */
1689 if (flag & O_APPEND)
1690 acc_mode |= MAY_APPEND;
1693 * The simplest case - just a plain lookup.
1695 if (!(flag & O_CREAT)) {
1696 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1697 &nd, flag);
1698 if (error)
1699 return ERR_PTR(error);
1700 goto ok;
1704 * Create - we need to know the parent.
1706 error = path_lookup_create(dfd, pathname, LOOKUP_PARENT,
1707 &nd, flag, mode);
1708 if (error)
1709 return ERR_PTR(error);
1712 * We have the parent and last component. First of all, check
1713 * that we are not asked to creat(2) an obvious directory - that
1714 * will not do.
1716 error = -EISDIR;
1717 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1718 goto exit;
1720 dir = nd.path.dentry;
1721 nd.flags &= ~LOOKUP_PARENT;
1722 mutex_lock(&dir->d_inode->i_mutex);
1723 path.dentry = lookup_hash(&nd);
1724 path.mnt = nd.path.mnt;
1726 do_last:
1727 error = PTR_ERR(path.dentry);
1728 if (IS_ERR(path.dentry)) {
1729 mutex_unlock(&dir->d_inode->i_mutex);
1730 goto exit;
1733 if (IS_ERR(nd.intent.open.file)) {
1734 error = PTR_ERR(nd.intent.open.file);
1735 goto exit_mutex_unlock;
1738 /* Negative dentry, just create the file */
1739 if (!path.dentry->d_inode) {
1741 * This write is needed to ensure that a
1742 * ro->rw transition does not occur between
1743 * the time when the file is created and when
1744 * a permanent write count is taken through
1745 * the 'struct file' in nameidata_to_filp().
1747 error = mnt_want_write(nd.path.mnt);
1748 if (error)
1749 goto exit_mutex_unlock;
1750 error = __open_namei_create(&nd, &path, flag, mode);
1751 if (error) {
1752 mnt_drop_write(nd.path.mnt);
1753 goto exit;
1755 filp = nameidata_to_filp(&nd, open_flag);
1756 mnt_drop_write(nd.path.mnt);
1757 return filp;
1761 * It already exists.
1763 mutex_unlock(&dir->d_inode->i_mutex);
1764 audit_inode(pathname, path.dentry);
1766 error = -EEXIST;
1767 if (flag & O_EXCL)
1768 goto exit_dput;
1770 if (__follow_mount(&path)) {
1771 error = -ELOOP;
1772 if (flag & O_NOFOLLOW)
1773 goto exit_dput;
1776 error = -ENOENT;
1777 if (!path.dentry->d_inode)
1778 goto exit_dput;
1779 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1780 goto do_link;
1782 path_to_nameidata(&path, &nd);
1783 error = -EISDIR;
1784 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1785 goto exit;
1788 * Consider:
1789 * 1. may_open() truncates a file
1790 * 2. a rw->ro mount transition occurs
1791 * 3. nameidata_to_filp() fails due to
1792 * the ro mount.
1793 * That would be inconsistent, and should
1794 * be avoided. Taking this mnt write here
1795 * ensures that (2) can not occur.
1797 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1798 if (will_write) {
1799 error = mnt_want_write(nd.path.mnt);
1800 if (error)
1801 goto exit;
1803 error = may_open(&nd, acc_mode, flag);
1804 if (error) {
1805 if (will_write)
1806 mnt_drop_write(nd.path.mnt);
1807 goto exit;
1809 filp = nameidata_to_filp(&nd, open_flag);
1811 * It is now safe to drop the mnt write
1812 * because the filp has had a write taken
1813 * on its behalf.
1815 if (will_write)
1816 mnt_drop_write(nd.path.mnt);
1817 return filp;
1819 exit_mutex_unlock:
1820 mutex_unlock(&dir->d_inode->i_mutex);
1821 exit_dput:
1822 path_put_conditional(&path, &nd);
1823 exit:
1824 if (!IS_ERR(nd.intent.open.file))
1825 release_open_intent(&nd);
1826 path_put(&nd.path);
1827 return ERR_PTR(error);
1829 do_link:
1830 error = -ELOOP;
1831 if (flag & O_NOFOLLOW)
1832 goto exit_dput;
1834 * This is subtle. Instead of calling do_follow_link() we do the
1835 * thing by hands. The reason is that this way we have zero link_count
1836 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1837 * After that we have the parent and last component, i.e.
1838 * we are in the same situation as after the first path_walk().
1839 * Well, almost - if the last component is normal we get its copy
1840 * stored in nd->last.name and we will have to putname() it when we
1841 * are done. Procfs-like symlinks just set LAST_BIND.
1843 nd.flags |= LOOKUP_PARENT;
1844 error = security_inode_follow_link(path.dentry, &nd);
1845 if (error)
1846 goto exit_dput;
1847 error = __do_follow_link(&path, &nd);
1848 if (error) {
1849 /* Does someone understand code flow here? Or it is only
1850 * me so stupid? Anathema to whoever designed this non-sense
1851 * with "intent.open".
1853 release_open_intent(&nd);
1854 return ERR_PTR(error);
1856 nd.flags &= ~LOOKUP_PARENT;
1857 if (nd.last_type == LAST_BIND)
1858 goto ok;
1859 error = -EISDIR;
1860 if (nd.last_type != LAST_NORM)
1861 goto exit;
1862 if (nd.last.name[nd.last.len]) {
1863 __putname(nd.last.name);
1864 goto exit;
1866 error = -ELOOP;
1867 if (count++==32) {
1868 __putname(nd.last.name);
1869 goto exit;
1871 dir = nd.path.dentry;
1872 mutex_lock(&dir->d_inode->i_mutex);
1873 path.dentry = lookup_hash(&nd);
1874 path.mnt = nd.path.mnt;
1875 __putname(nd.last.name);
1876 goto do_last;
1880 * filp_open - open file and return file pointer
1882 * @filename: path to open
1883 * @flags: open flags as per the open(2) second argument
1884 * @mode: mode for the new file if O_CREAT is set, else ignored
1886 * This is the helper to open a file from kernelspace if you really
1887 * have to. But in generally you should not do this, so please move
1888 * along, nothing to see here..
1890 struct file *filp_open(const char *filename, int flags, int mode)
1892 return do_filp_open(AT_FDCWD, filename, flags, mode);
1894 EXPORT_SYMBOL(filp_open);
1897 * lookup_create - lookup a dentry, creating it if it doesn't exist
1898 * @nd: nameidata info
1899 * @is_dir: directory flag
1901 * Simple function to lookup and return a dentry and create it
1902 * if it doesn't exist. Is SMP-safe.
1904 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1906 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1908 struct dentry *dentry = ERR_PTR(-EEXIST);
1910 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1912 * Yucky last component or no last component at all?
1913 * (foo/., foo/.., /////)
1915 if (nd->last_type != LAST_NORM)
1916 goto fail;
1917 nd->flags &= ~LOOKUP_PARENT;
1918 nd->flags |= LOOKUP_CREATE;
1919 nd->intent.open.flags = O_EXCL;
1922 * Do the final lookup.
1924 dentry = lookup_hash(nd);
1925 if (IS_ERR(dentry))
1926 goto fail;
1928 if (dentry->d_inode)
1929 goto eexist;
1931 * Special case - lookup gave negative, but... we had foo/bar/
1932 * From the vfs_mknod() POV we just have a negative dentry -
1933 * all is fine. Let's be bastards - you had / on the end, you've
1934 * been asking for (non-existent) directory. -ENOENT for you.
1936 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1937 dput(dentry);
1938 dentry = ERR_PTR(-ENOENT);
1940 return dentry;
1941 eexist:
1942 dput(dentry);
1943 dentry = ERR_PTR(-EEXIST);
1944 fail:
1945 return dentry;
1947 EXPORT_SYMBOL_GPL(lookup_create);
1949 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1951 int error = may_create(dir, dentry, NULL);
1953 if (error)
1954 return error;
1956 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1957 return -EPERM;
1959 if (!dir->i_op || !dir->i_op->mknod)
1960 return -EPERM;
1962 error = devcgroup_inode_mknod(mode, dev);
1963 if (error)
1964 return error;
1966 error = security_inode_mknod(dir, dentry, mode, dev);
1967 if (error)
1968 return error;
1970 DQUOT_INIT(dir);
1971 error = dir->i_op->mknod(dir, dentry, mode, dev);
1972 if (!error)
1973 fsnotify_create(dir, dentry);
1974 return error;
1977 static int may_mknod(mode_t mode)
1979 switch (mode & S_IFMT) {
1980 case S_IFREG:
1981 case S_IFCHR:
1982 case S_IFBLK:
1983 case S_IFIFO:
1984 case S_IFSOCK:
1985 case 0: /* zero mode translates to S_IFREG */
1986 return 0;
1987 case S_IFDIR:
1988 return -EPERM;
1989 default:
1990 return -EINVAL;
1994 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1995 unsigned dev)
1997 int error;
1998 char *tmp;
1999 struct dentry *dentry;
2000 struct nameidata nd;
2002 if (S_ISDIR(mode))
2003 return -EPERM;
2005 error = user_path_parent(dfd, filename, &nd, &tmp);
2006 if (error)
2007 return error;
2009 dentry = lookup_create(&nd, 0);
2010 if (IS_ERR(dentry)) {
2011 error = PTR_ERR(dentry);
2012 goto out_unlock;
2014 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2015 mode &= ~current->fs->umask;
2016 error = may_mknod(mode);
2017 if (error)
2018 goto out_dput;
2019 error = mnt_want_write(nd.path.mnt);
2020 if (error)
2021 goto out_dput;
2022 switch (mode & S_IFMT) {
2023 case 0: case S_IFREG:
2024 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2025 break;
2026 case S_IFCHR: case S_IFBLK:
2027 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2028 new_decode_dev(dev));
2029 break;
2030 case S_IFIFO: case S_IFSOCK:
2031 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2032 break;
2034 mnt_drop_write(nd.path.mnt);
2035 out_dput:
2036 dput(dentry);
2037 out_unlock:
2038 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2039 path_put(&nd.path);
2040 putname(tmp);
2042 return error;
2045 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
2047 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2050 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2052 int error = may_create(dir, dentry, NULL);
2054 if (error)
2055 return error;
2057 if (!dir->i_op || !dir->i_op->mkdir)
2058 return -EPERM;
2060 mode &= (S_IRWXUGO|S_ISVTX);
2061 error = security_inode_mkdir(dir, dentry, mode);
2062 if (error)
2063 return error;
2065 DQUOT_INIT(dir);
2066 error = dir->i_op->mkdir(dir, dentry, mode);
2067 if (!error)
2068 fsnotify_mkdir(dir, dentry);
2069 return error;
2072 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2074 int error = 0;
2075 char * tmp;
2076 struct dentry *dentry;
2077 struct nameidata nd;
2079 error = user_path_parent(dfd, pathname, &nd, &tmp);
2080 if (error)
2081 goto out_err;
2083 dentry = lookup_create(&nd, 1);
2084 error = PTR_ERR(dentry);
2085 if (IS_ERR(dentry))
2086 goto out_unlock;
2088 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2089 mode &= ~current->fs->umask;
2090 error = mnt_want_write(nd.path.mnt);
2091 if (error)
2092 goto out_dput;
2093 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2094 mnt_drop_write(nd.path.mnt);
2095 out_dput:
2096 dput(dentry);
2097 out_unlock:
2098 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2099 path_put(&nd.path);
2100 putname(tmp);
2101 out_err:
2102 return error;
2105 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2107 return sys_mkdirat(AT_FDCWD, pathname, mode);
2111 * We try to drop the dentry early: we should have
2112 * a usage count of 2 if we're the only user of this
2113 * dentry, and if that is true (possibly after pruning
2114 * the dcache), then we drop the dentry now.
2116 * A low-level filesystem can, if it choses, legally
2117 * do a
2119 * if (!d_unhashed(dentry))
2120 * return -EBUSY;
2122 * if it cannot handle the case of removing a directory
2123 * that is still in use by something else..
2125 void dentry_unhash(struct dentry *dentry)
2127 dget(dentry);
2128 shrink_dcache_parent(dentry);
2129 spin_lock(&dcache_lock);
2130 spin_lock(&dentry->d_lock);
2131 if (atomic_read(&dentry->d_count) == 2)
2132 __d_drop(dentry);
2133 spin_unlock(&dentry->d_lock);
2134 spin_unlock(&dcache_lock);
2137 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2139 int error = may_delete(dir, dentry, 1);
2141 if (error)
2142 return error;
2144 if (!dir->i_op || !dir->i_op->rmdir)
2145 return -EPERM;
2147 DQUOT_INIT(dir);
2149 mutex_lock(&dentry->d_inode->i_mutex);
2150 dentry_unhash(dentry);
2151 if (d_mountpoint(dentry))
2152 error = -EBUSY;
2153 else {
2154 error = security_inode_rmdir(dir, dentry);
2155 if (!error) {
2156 error = dir->i_op->rmdir(dir, dentry);
2157 if (!error)
2158 dentry->d_inode->i_flags |= S_DEAD;
2161 mutex_unlock(&dentry->d_inode->i_mutex);
2162 if (!error) {
2163 d_delete(dentry);
2165 dput(dentry);
2167 return error;
2170 static long do_rmdir(int dfd, const char __user *pathname)
2172 int error = 0;
2173 char * name;
2174 struct dentry *dentry;
2175 struct nameidata nd;
2177 error = user_path_parent(dfd, pathname, &nd, &name);
2178 if (error)
2179 return error;
2181 switch(nd.last_type) {
2182 case LAST_DOTDOT:
2183 error = -ENOTEMPTY;
2184 goto exit1;
2185 case LAST_DOT:
2186 error = -EINVAL;
2187 goto exit1;
2188 case LAST_ROOT:
2189 error = -EBUSY;
2190 goto exit1;
2192 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2193 dentry = lookup_hash(&nd);
2194 error = PTR_ERR(dentry);
2195 if (IS_ERR(dentry))
2196 goto exit2;
2197 error = mnt_want_write(nd.path.mnt);
2198 if (error)
2199 goto exit3;
2200 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2201 mnt_drop_write(nd.path.mnt);
2202 exit3:
2203 dput(dentry);
2204 exit2:
2205 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2206 exit1:
2207 path_put(&nd.path);
2208 putname(name);
2209 return error;
2212 asmlinkage long sys_rmdir(const char __user *pathname)
2214 return do_rmdir(AT_FDCWD, pathname);
2217 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2219 int error = may_delete(dir, dentry, 0);
2221 if (error)
2222 return error;
2224 if (!dir->i_op || !dir->i_op->unlink)
2225 return -EPERM;
2227 DQUOT_INIT(dir);
2229 mutex_lock(&dentry->d_inode->i_mutex);
2230 if (d_mountpoint(dentry))
2231 error = -EBUSY;
2232 else {
2233 error = security_inode_unlink(dir, dentry);
2234 if (!error)
2235 error = dir->i_op->unlink(dir, dentry);
2237 mutex_unlock(&dentry->d_inode->i_mutex);
2239 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2240 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2241 fsnotify_link_count(dentry->d_inode);
2242 d_delete(dentry);
2245 return error;
2249 * Make sure that the actual truncation of the file will occur outside its
2250 * directory's i_mutex. Truncate can take a long time if there is a lot of
2251 * writeout happening, and we don't want to prevent access to the directory
2252 * while waiting on the I/O.
2254 static long do_unlinkat(int dfd, const char __user *pathname)
2256 int error;
2257 char *name;
2258 struct dentry *dentry;
2259 struct nameidata nd;
2260 struct inode *inode = NULL;
2262 error = user_path_parent(dfd, pathname, &nd, &name);
2263 if (error)
2264 return error;
2266 error = -EISDIR;
2267 if (nd.last_type != LAST_NORM)
2268 goto exit1;
2269 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2270 dentry = lookup_hash(&nd);
2271 error = PTR_ERR(dentry);
2272 if (!IS_ERR(dentry)) {
2273 /* Why not before? Because we want correct error value */
2274 if (nd.last.name[nd.last.len])
2275 goto slashes;
2276 inode = dentry->d_inode;
2277 if (inode)
2278 atomic_inc(&inode->i_count);
2279 error = mnt_want_write(nd.path.mnt);
2280 if (error)
2281 goto exit2;
2282 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2283 mnt_drop_write(nd.path.mnt);
2284 exit2:
2285 dput(dentry);
2287 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2288 if (inode)
2289 iput(inode); /* truncate the inode here */
2290 exit1:
2291 path_put(&nd.path);
2292 putname(name);
2293 return error;
2295 slashes:
2296 error = !dentry->d_inode ? -ENOENT :
2297 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2298 goto exit2;
2301 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2303 if ((flag & ~AT_REMOVEDIR) != 0)
2304 return -EINVAL;
2306 if (flag & AT_REMOVEDIR)
2307 return do_rmdir(dfd, pathname);
2309 return do_unlinkat(dfd, pathname);
2312 asmlinkage long sys_unlink(const char __user *pathname)
2314 return do_unlinkat(AT_FDCWD, pathname);
2317 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2319 int error = may_create(dir, dentry, NULL);
2321 if (error)
2322 return error;
2324 if (!dir->i_op || !dir->i_op->symlink)
2325 return -EPERM;
2327 error = security_inode_symlink(dir, dentry, oldname);
2328 if (error)
2329 return error;
2331 DQUOT_INIT(dir);
2332 error = dir->i_op->symlink(dir, dentry, oldname);
2333 if (!error)
2334 fsnotify_create(dir, dentry);
2335 return error;
2338 asmlinkage long sys_symlinkat(const char __user *oldname,
2339 int newdfd, const char __user *newname)
2341 int error;
2342 char *from;
2343 char *to;
2344 struct dentry *dentry;
2345 struct nameidata nd;
2347 from = getname(oldname);
2348 if (IS_ERR(from))
2349 return PTR_ERR(from);
2351 error = user_path_parent(newdfd, newname, &nd, &to);
2352 if (error)
2353 goto out_putname;
2355 dentry = lookup_create(&nd, 0);
2356 error = PTR_ERR(dentry);
2357 if (IS_ERR(dentry))
2358 goto out_unlock;
2360 error = mnt_want_write(nd.path.mnt);
2361 if (error)
2362 goto out_dput;
2363 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2364 mnt_drop_write(nd.path.mnt);
2365 out_dput:
2366 dput(dentry);
2367 out_unlock:
2368 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2369 path_put(&nd.path);
2370 putname(to);
2371 out_putname:
2372 putname(from);
2373 return error;
2376 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2378 return sys_symlinkat(oldname, AT_FDCWD, newname);
2381 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2383 struct inode *inode = old_dentry->d_inode;
2384 int error;
2386 if (!inode)
2387 return -ENOENT;
2389 error = may_create(dir, new_dentry, NULL);
2390 if (error)
2391 return error;
2393 if (dir->i_sb != inode->i_sb)
2394 return -EXDEV;
2397 * A link to an append-only or immutable file cannot be created.
2399 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2400 return -EPERM;
2401 if (!dir->i_op || !dir->i_op->link)
2402 return -EPERM;
2403 if (S_ISDIR(inode->i_mode))
2404 return -EPERM;
2406 error = security_inode_link(old_dentry, dir, new_dentry);
2407 if (error)
2408 return error;
2410 mutex_lock(&inode->i_mutex);
2411 DQUOT_INIT(dir);
2412 error = dir->i_op->link(old_dentry, dir, new_dentry);
2413 mutex_unlock(&inode->i_mutex);
2414 if (!error)
2415 fsnotify_link(dir, inode, new_dentry);
2416 return error;
2420 * Hardlinks are often used in delicate situations. We avoid
2421 * security-related surprises by not following symlinks on the
2422 * newname. --KAB
2424 * We don't follow them on the oldname either to be compatible
2425 * with linux 2.0, and to avoid hard-linking to directories
2426 * and other special files. --ADM
2428 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2429 int newdfd, const char __user *newname,
2430 int flags)
2432 struct dentry *new_dentry;
2433 struct nameidata nd;
2434 struct path old_path;
2435 int error;
2436 char *to;
2438 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2439 return -EINVAL;
2441 error = user_path_at(olddfd, oldname,
2442 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2443 &old_path);
2444 if (error)
2445 return error;
2447 error = user_path_parent(newdfd, newname, &nd, &to);
2448 if (error)
2449 goto out;
2450 error = -EXDEV;
2451 if (old_path.mnt != nd.path.mnt)
2452 goto out_release;
2453 new_dentry = lookup_create(&nd, 0);
2454 error = PTR_ERR(new_dentry);
2455 if (IS_ERR(new_dentry))
2456 goto out_unlock;
2457 error = mnt_want_write(nd.path.mnt);
2458 if (error)
2459 goto out_dput;
2460 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2461 mnt_drop_write(nd.path.mnt);
2462 out_dput:
2463 dput(new_dentry);
2464 out_unlock:
2465 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2466 out_release:
2467 path_put(&nd.path);
2468 putname(to);
2469 out:
2470 path_put(&old_path);
2472 return error;
2475 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2477 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2481 * The worst of all namespace operations - renaming directory. "Perverted"
2482 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2483 * Problems:
2484 * a) we can get into loop creation. Check is done in is_subdir().
2485 * b) race potential - two innocent renames can create a loop together.
2486 * That's where 4.4 screws up. Current fix: serialization on
2487 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2488 * story.
2489 * c) we have to lock _three_ objects - parents and victim (if it exists).
2490 * And that - after we got ->i_mutex on parents (until then we don't know
2491 * whether the target exists). Solution: try to be smart with locking
2492 * order for inodes. We rely on the fact that tree topology may change
2493 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2494 * move will be locked. Thus we can rank directories by the tree
2495 * (ancestors first) and rank all non-directories after them.
2496 * That works since everybody except rename does "lock parent, lookup,
2497 * lock child" and rename is under ->s_vfs_rename_mutex.
2498 * HOWEVER, it relies on the assumption that any object with ->lookup()
2499 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2500 * we'd better make sure that there's no link(2) for them.
2501 * d) some filesystems don't support opened-but-unlinked directories,
2502 * either because of layout or because they are not ready to deal with
2503 * all cases correctly. The latter will be fixed (taking this sort of
2504 * stuff into VFS), but the former is not going away. Solution: the same
2505 * trick as in rmdir().
2506 * e) conversion from fhandle to dentry may come in the wrong moment - when
2507 * we are removing the target. Solution: we will have to grab ->i_mutex
2508 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2509 * ->i_mutex on parents, which works but leads to some truely excessive
2510 * locking].
2512 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2513 struct inode *new_dir, struct dentry *new_dentry)
2515 int error = 0;
2516 struct inode *target;
2519 * If we are going to change the parent - check write permissions,
2520 * we'll need to flip '..'.
2522 if (new_dir != old_dir) {
2523 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2524 if (error)
2525 return error;
2528 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2529 if (error)
2530 return error;
2532 target = new_dentry->d_inode;
2533 if (target) {
2534 mutex_lock(&target->i_mutex);
2535 dentry_unhash(new_dentry);
2537 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2538 error = -EBUSY;
2539 else
2540 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2541 if (target) {
2542 if (!error)
2543 target->i_flags |= S_DEAD;
2544 mutex_unlock(&target->i_mutex);
2545 if (d_unhashed(new_dentry))
2546 d_rehash(new_dentry);
2547 dput(new_dentry);
2549 if (!error)
2550 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2551 d_move(old_dentry,new_dentry);
2552 return error;
2555 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2556 struct inode *new_dir, struct dentry *new_dentry)
2558 struct inode *target;
2559 int error;
2561 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2562 if (error)
2563 return error;
2565 dget(new_dentry);
2566 target = new_dentry->d_inode;
2567 if (target)
2568 mutex_lock(&target->i_mutex);
2569 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2570 error = -EBUSY;
2571 else
2572 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2573 if (!error) {
2574 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2575 d_move(old_dentry, new_dentry);
2577 if (target)
2578 mutex_unlock(&target->i_mutex);
2579 dput(new_dentry);
2580 return error;
2583 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2584 struct inode *new_dir, struct dentry *new_dentry)
2586 int error;
2587 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2588 const char *old_name;
2590 if (old_dentry->d_inode == new_dentry->d_inode)
2591 return 0;
2593 error = may_delete(old_dir, old_dentry, is_dir);
2594 if (error)
2595 return error;
2597 if (!new_dentry->d_inode)
2598 error = may_create(new_dir, new_dentry, NULL);
2599 else
2600 error = may_delete(new_dir, new_dentry, is_dir);
2601 if (error)
2602 return error;
2604 if (!old_dir->i_op || !old_dir->i_op->rename)
2605 return -EPERM;
2607 DQUOT_INIT(old_dir);
2608 DQUOT_INIT(new_dir);
2610 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2612 if (is_dir)
2613 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2614 else
2615 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2616 if (!error) {
2617 const char *new_name = old_dentry->d_name.name;
2618 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2619 new_dentry->d_inode, old_dentry);
2621 fsnotify_oldname_free(old_name);
2623 return error;
2626 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2627 int newdfd, const char __user *newname)
2629 struct dentry *old_dir, *new_dir;
2630 struct dentry *old_dentry, *new_dentry;
2631 struct dentry *trap;
2632 struct nameidata oldnd, newnd;
2633 char *from;
2634 char *to;
2635 int error;
2637 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2638 if (error)
2639 goto exit;
2641 error = user_path_parent(newdfd, newname, &newnd, &to);
2642 if (error)
2643 goto exit1;
2645 error = -EXDEV;
2646 if (oldnd.path.mnt != newnd.path.mnt)
2647 goto exit2;
2649 old_dir = oldnd.path.dentry;
2650 error = -EBUSY;
2651 if (oldnd.last_type != LAST_NORM)
2652 goto exit2;
2654 new_dir = newnd.path.dentry;
2655 if (newnd.last_type != LAST_NORM)
2656 goto exit2;
2658 trap = lock_rename(new_dir, old_dir);
2660 old_dentry = lookup_hash(&oldnd);
2661 error = PTR_ERR(old_dentry);
2662 if (IS_ERR(old_dentry))
2663 goto exit3;
2664 /* source must exist */
2665 error = -ENOENT;
2666 if (!old_dentry->d_inode)
2667 goto exit4;
2668 /* unless the source is a directory trailing slashes give -ENOTDIR */
2669 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2670 error = -ENOTDIR;
2671 if (oldnd.last.name[oldnd.last.len])
2672 goto exit4;
2673 if (newnd.last.name[newnd.last.len])
2674 goto exit4;
2676 /* source should not be ancestor of target */
2677 error = -EINVAL;
2678 if (old_dentry == trap)
2679 goto exit4;
2680 new_dentry = lookup_hash(&newnd);
2681 error = PTR_ERR(new_dentry);
2682 if (IS_ERR(new_dentry))
2683 goto exit4;
2684 /* target should not be an ancestor of source */
2685 error = -ENOTEMPTY;
2686 if (new_dentry == trap)
2687 goto exit5;
2689 error = mnt_want_write(oldnd.path.mnt);
2690 if (error)
2691 goto exit5;
2692 error = vfs_rename(old_dir->d_inode, old_dentry,
2693 new_dir->d_inode, new_dentry);
2694 mnt_drop_write(oldnd.path.mnt);
2695 exit5:
2696 dput(new_dentry);
2697 exit4:
2698 dput(old_dentry);
2699 exit3:
2700 unlock_rename(new_dir, old_dir);
2701 exit2:
2702 path_put(&newnd.path);
2703 putname(to);
2704 exit1:
2705 path_put(&oldnd.path);
2706 putname(from);
2707 exit:
2708 return error;
2711 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2713 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2716 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2718 int len;
2720 len = PTR_ERR(link);
2721 if (IS_ERR(link))
2722 goto out;
2724 len = strlen(link);
2725 if (len > (unsigned) buflen)
2726 len = buflen;
2727 if (copy_to_user(buffer, link, len))
2728 len = -EFAULT;
2729 out:
2730 return len;
2734 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2735 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2736 * using) it for any given inode is up to filesystem.
2738 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2740 struct nameidata nd;
2741 void *cookie;
2742 int res;
2744 nd.depth = 0;
2745 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2746 if (IS_ERR(cookie))
2747 return PTR_ERR(cookie);
2749 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2750 if (dentry->d_inode->i_op->put_link)
2751 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2752 return res;
2755 int vfs_follow_link(struct nameidata *nd, const char *link)
2757 return __vfs_follow_link(nd, link);
2760 /* get the link contents into pagecache */
2761 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2763 struct page * page;
2764 struct address_space *mapping = dentry->d_inode->i_mapping;
2765 page = read_mapping_page(mapping, 0, NULL);
2766 if (IS_ERR(page))
2767 return (char*)page;
2768 *ppage = page;
2769 return kmap(page);
2772 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2774 struct page *page = NULL;
2775 char *s = page_getlink(dentry, &page);
2776 int res = vfs_readlink(dentry,buffer,buflen,s);
2777 if (page) {
2778 kunmap(page);
2779 page_cache_release(page);
2781 return res;
2784 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2786 struct page *page = NULL;
2787 nd_set_link(nd, page_getlink(dentry, &page));
2788 return page;
2791 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2793 struct page *page = cookie;
2795 if (page) {
2796 kunmap(page);
2797 page_cache_release(page);
2801 int __page_symlink(struct inode *inode, const char *symname, int len,
2802 gfp_t gfp_mask)
2804 struct address_space *mapping = inode->i_mapping;
2805 struct page *page;
2806 void *fsdata;
2807 int err;
2808 char *kaddr;
2810 retry:
2811 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2812 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
2813 if (err)
2814 goto fail;
2816 kaddr = kmap_atomic(page, KM_USER0);
2817 memcpy(kaddr, symname, len-1);
2818 kunmap_atomic(kaddr, KM_USER0);
2820 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2821 page, fsdata);
2822 if (err < 0)
2823 goto fail;
2824 if (err < len-1)
2825 goto retry;
2827 mark_inode_dirty(inode);
2828 return 0;
2829 fail:
2830 return err;
2833 int page_symlink(struct inode *inode, const char *symname, int len)
2835 return __page_symlink(inode, symname, len,
2836 mapping_gfp_mask(inode->i_mapping));
2839 const struct inode_operations page_symlink_inode_operations = {
2840 .readlink = generic_readlink,
2841 .follow_link = page_follow_link_light,
2842 .put_link = page_put_link,
2845 EXPORT_SYMBOL(user_path_at);
2846 EXPORT_SYMBOL(follow_down);
2847 EXPORT_SYMBOL(follow_up);
2848 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2849 EXPORT_SYMBOL(getname);
2850 EXPORT_SYMBOL(lock_rename);
2851 EXPORT_SYMBOL(lookup_one_len);
2852 EXPORT_SYMBOL(page_follow_link_light);
2853 EXPORT_SYMBOL(page_put_link);
2854 EXPORT_SYMBOL(page_readlink);
2855 EXPORT_SYMBOL(__page_symlink);
2856 EXPORT_SYMBOL(page_symlink);
2857 EXPORT_SYMBOL(page_symlink_inode_operations);
2858 EXPORT_SYMBOL(path_lookup);
2859 EXPORT_SYMBOL(vfs_path_lookup);
2860 EXPORT_SYMBOL(inode_permission);
2861 EXPORT_SYMBOL(vfs_permission);
2862 EXPORT_SYMBOL(file_permission);
2863 EXPORT_SYMBOL(unlock_rename);
2864 EXPORT_SYMBOL(vfs_create);
2865 EXPORT_SYMBOL(vfs_follow_link);
2866 EXPORT_SYMBOL(vfs_link);
2867 EXPORT_SYMBOL(vfs_mkdir);
2868 EXPORT_SYMBOL(vfs_mknod);
2869 EXPORT_SYMBOL(generic_permission);
2870 EXPORT_SYMBOL(vfs_readlink);
2871 EXPORT_SYMBOL(vfs_rename);
2872 EXPORT_SYMBOL(vfs_rmdir);
2873 EXPORT_SYMBOL(vfs_symlink);
2874 EXPORT_SYMBOL(vfs_unlink);
2875 EXPORT_SYMBOL(dentry_unhash);
2876 EXPORT_SYMBOL(generic_readlink);