Driver-core: Always create class directories for classses that support namespaces.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / namei.c
blob868d0cb9d473a92ccdcbe640ac44f4daf24cce62
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/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.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 <linux/fs_struct.h>
35 #include <asm/uaccess.h>
37 #include "internal.h"
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user *filename, char *page)
119 int retval;
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
139 char * getname(const char __user * filename)
141 char *tmp, *result;
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
155 return result;
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
162 audit_putname(name);
163 else
164 __putname(name);
166 EXPORT_SYMBOL(putname);
167 #endif
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask))
175 umode_t mode = inode->i_mode;
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179 if (current_fsuid() == inode->i_uid)
180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
183 int error = check_acl(inode, mask);
184 if (error != -EAGAIN)
185 return error;
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
193 * If the DACs are ok we don't need any capability check.
195 if ((mask & ~mode) == 0)
196 return 0;
197 return -EACCES;
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
206 * Used to check for read/write/execute permissions on a file.
207 * We use "fsuid" for this, letting us set arbitrary permissions
208 * for filesystem access without changing the "normal" uids which
209 * are used for other things..
211 int generic_permission(struct inode *inode, int mask,
212 int (*check_acl)(struct inode *inode, int mask))
214 int ret;
217 * Do the basic POSIX ACL permission checks.
219 ret = acl_permission_check(inode, mask, check_acl);
220 if (ret != -EACCES)
221 return ret;
224 * Read/write DACs are always overridable.
225 * Executable DACs are overridable if at least one exec bit is set.
227 if (!(mask & MAY_EXEC) || execute_ok(inode))
228 if (capable(CAP_DAC_OVERRIDE))
229 return 0;
232 * Searching includes executable on directories, else just read.
234 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
235 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
236 if (capable(CAP_DAC_READ_SEARCH))
237 return 0;
239 return -EACCES;
243 * inode_permission - check for access rights to a given inode
244 * @inode: inode to check permission on
245 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
247 * Used to check for read/write/execute permissions on an inode.
248 * We use "fsuid" for this, letting us set arbitrary permissions
249 * for filesystem access without changing the "normal" uids which
250 * are used for other things.
252 int inode_permission(struct inode *inode, int mask)
254 int retval;
256 if (mask & MAY_WRITE) {
257 umode_t mode = inode->i_mode;
260 * Nobody gets write access to a read-only fs.
262 if (IS_RDONLY(inode) &&
263 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
264 return -EROFS;
267 * Nobody gets write access to an immutable file.
269 if (IS_IMMUTABLE(inode))
270 return -EACCES;
273 if (inode->i_op->permission)
274 retval = inode->i_op->permission(inode, mask);
275 else
276 retval = generic_permission(inode, mask, inode->i_op->check_acl);
278 if (retval)
279 return retval;
281 retval = devcgroup_inode_permission(inode, mask);
282 if (retval)
283 return retval;
285 return security_inode_permission(inode,
286 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
290 * file_permission - check for additional access rights to a given file
291 * @file: file to check access rights for
292 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
294 * Used to check for read/write/execute permissions on an already opened
295 * file.
297 * Note:
298 * Do not use this function in new code. All access checks should
299 * be done using inode_permission().
301 int file_permission(struct file *file, int mask)
303 return inode_permission(file->f_path.dentry->d_inode, mask);
307 * get_write_access() gets write permission for a file.
308 * put_write_access() releases this write permission.
309 * This is used for regular files.
310 * We cannot support write (and maybe mmap read-write shared) accesses and
311 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
312 * can have the following values:
313 * 0: no writers, no VM_DENYWRITE mappings
314 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
315 * > 0: (i_writecount) users are writing to the file.
317 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
318 * except for the cases where we don't hold i_writecount yet. Then we need to
319 * use {get,deny}_write_access() - these functions check the sign and refuse
320 * to do the change if sign is wrong. Exclusion between them is provided by
321 * the inode->i_lock spinlock.
324 int get_write_access(struct inode * inode)
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) < 0) {
328 spin_unlock(&inode->i_lock);
329 return -ETXTBSY;
331 atomic_inc(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
334 return 0;
337 int deny_write_access(struct file * file)
339 struct inode *inode = file->f_path.dentry->d_inode;
341 spin_lock(&inode->i_lock);
342 if (atomic_read(&inode->i_writecount) > 0) {
343 spin_unlock(&inode->i_lock);
344 return -ETXTBSY;
346 atomic_dec(&inode->i_writecount);
347 spin_unlock(&inode->i_lock);
349 return 0;
353 * path_get - get a reference to a path
354 * @path: path to get the reference to
356 * Given a path increment the reference count to the dentry and the vfsmount.
358 void path_get(struct path *path)
360 mntget(path->mnt);
361 dget(path->dentry);
363 EXPORT_SYMBOL(path_get);
366 * path_put - put a reference to a path
367 * @path: path to put the reference to
369 * Given a path decrement the reference count to the dentry and the vfsmount.
371 void path_put(struct path *path)
373 dput(path->dentry);
374 mntput(path->mnt);
376 EXPORT_SYMBOL(path_put);
379 * release_open_intent - free up open intent resources
380 * @nd: pointer to nameidata
382 void release_open_intent(struct nameidata *nd)
384 if (nd->intent.open.file->f_path.dentry == NULL)
385 put_filp(nd->intent.open.file);
386 else
387 fput(nd->intent.open.file);
390 static inline struct dentry *
391 do_revalidate(struct dentry *dentry, struct nameidata *nd)
393 int status = dentry->d_op->d_revalidate(dentry, nd);
394 if (unlikely(status <= 0)) {
396 * The dentry failed validation.
397 * If d_revalidate returned 0 attempt to invalidate
398 * the dentry otherwise d_revalidate is asking us
399 * to return a fail status.
401 if (!status) {
402 if (!d_invalidate(dentry)) {
403 dput(dentry);
404 dentry = NULL;
406 } else {
407 dput(dentry);
408 dentry = ERR_PTR(status);
411 return dentry;
415 * force_reval_path - force revalidation of a dentry
417 * In some situations the path walking code will trust dentries without
418 * revalidating them. This causes problems for filesystems that depend on
419 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
420 * (which indicates that it's possible for the dentry to go stale), force
421 * a d_revalidate call before proceeding.
423 * Returns 0 if the revalidation was successful. If the revalidation fails,
424 * either return the error returned by d_revalidate or -ESTALE if the
425 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
426 * invalidate the dentry. It's up to the caller to handle putting references
427 * to the path if necessary.
429 static int
430 force_reval_path(struct path *path, struct nameidata *nd)
432 int status;
433 struct dentry *dentry = path->dentry;
436 * only check on filesystems where it's possible for the dentry to
437 * become stale. It's assumed that if this flag is set then the
438 * d_revalidate op will also be defined.
440 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
441 return 0;
443 status = dentry->d_op->d_revalidate(dentry, nd);
444 if (status > 0)
445 return 0;
447 if (!status) {
448 d_invalidate(dentry);
449 status = -ESTALE;
451 return status;
455 * Short-cut version of permission(), for calling on directories
456 * during pathname resolution. Combines parts of permission()
457 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
459 * If appropriate, check DAC only. If not appropriate, or
460 * short-cut DAC fails, then call ->permission() to do more
461 * complete permission check.
463 static int exec_permission(struct inode *inode)
465 int ret;
467 if (inode->i_op->permission) {
468 ret = inode->i_op->permission(inode, MAY_EXEC);
469 if (!ret)
470 goto ok;
471 return ret;
473 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
474 if (!ret)
475 goto ok;
477 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
478 goto ok;
480 return ret;
482 return security_inode_permission(inode, MAY_EXEC);
485 static __always_inline void set_root(struct nameidata *nd)
487 if (!nd->root.mnt) {
488 struct fs_struct *fs = current->fs;
489 read_lock(&fs->lock);
490 nd->root = fs->root;
491 path_get(&nd->root);
492 read_unlock(&fs->lock);
496 static int link_path_walk(const char *, struct nameidata *);
498 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
500 if (IS_ERR(link))
501 goto fail;
503 if (*link == '/') {
504 set_root(nd);
505 path_put(&nd->path);
506 nd->path = nd->root;
507 path_get(&nd->root);
510 return link_path_walk(link, nd);
511 fail:
512 path_put(&nd->path);
513 return PTR_ERR(link);
516 static void path_put_conditional(struct path *path, struct nameidata *nd)
518 dput(path->dentry);
519 if (path->mnt != nd->path.mnt)
520 mntput(path->mnt);
523 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
525 dput(nd->path.dentry);
526 if (nd->path.mnt != path->mnt) {
527 mntput(nd->path.mnt);
528 nd->path.mnt = path->mnt;
530 nd->path.dentry = path->dentry;
533 static __always_inline int
534 __do_follow_link(struct path *path, struct nameidata *nd, void **p)
536 int error;
537 struct dentry *dentry = path->dentry;
539 touch_atime(path->mnt, dentry);
540 nd_set_link(nd, NULL);
542 if (path->mnt != nd->path.mnt) {
543 path_to_nameidata(path, nd);
544 dget(dentry);
546 mntget(path->mnt);
547 nd->last_type = LAST_BIND;
548 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
549 error = PTR_ERR(*p);
550 if (!IS_ERR(*p)) {
551 char *s = nd_get_link(nd);
552 error = 0;
553 if (s)
554 error = __vfs_follow_link(nd, s);
555 else if (nd->last_type == LAST_BIND) {
556 error = force_reval_path(&nd->path, nd);
557 if (error)
558 path_put(&nd->path);
561 return error;
565 * This limits recursive symlink follows to 8, while
566 * limiting consecutive symlinks to 40.
568 * Without that kind of total limit, nasty chains of consecutive
569 * symlinks can cause almost arbitrarily long lookups.
571 static inline int do_follow_link(struct path *path, struct nameidata *nd)
573 void *cookie;
574 int err = -ELOOP;
575 if (current->link_count >= MAX_NESTED_LINKS)
576 goto loop;
577 if (current->total_link_count >= 40)
578 goto loop;
579 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
580 cond_resched();
581 err = security_inode_follow_link(path->dentry, nd);
582 if (err)
583 goto loop;
584 current->link_count++;
585 current->total_link_count++;
586 nd->depth++;
587 err = __do_follow_link(path, nd, &cookie);
588 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
589 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
590 path_put(path);
591 current->link_count--;
592 nd->depth--;
593 return err;
594 loop:
595 path_put_conditional(path, nd);
596 path_put(&nd->path);
597 return err;
600 int follow_up(struct path *path)
602 struct vfsmount *parent;
603 struct dentry *mountpoint;
604 spin_lock(&vfsmount_lock);
605 parent = path->mnt->mnt_parent;
606 if (parent == path->mnt) {
607 spin_unlock(&vfsmount_lock);
608 return 0;
610 mntget(parent);
611 mountpoint = dget(path->mnt->mnt_mountpoint);
612 spin_unlock(&vfsmount_lock);
613 dput(path->dentry);
614 path->dentry = mountpoint;
615 mntput(path->mnt);
616 path->mnt = parent;
617 return 1;
620 /* no need for dcache_lock, as serialization is taken care in
621 * namespace.c
623 static int __follow_mount(struct path *path)
625 int res = 0;
626 while (d_mountpoint(path->dentry)) {
627 struct vfsmount *mounted = lookup_mnt(path);
628 if (!mounted)
629 break;
630 dput(path->dentry);
631 if (res)
632 mntput(path->mnt);
633 path->mnt = mounted;
634 path->dentry = dget(mounted->mnt_root);
635 res = 1;
637 return res;
640 static void follow_mount(struct path *path)
642 while (d_mountpoint(path->dentry)) {
643 struct vfsmount *mounted = lookup_mnt(path);
644 if (!mounted)
645 break;
646 dput(path->dentry);
647 mntput(path->mnt);
648 path->mnt = mounted;
649 path->dentry = dget(mounted->mnt_root);
653 /* no need for dcache_lock, as serialization is taken care in
654 * namespace.c
656 int follow_down(struct path *path)
658 struct vfsmount *mounted;
660 mounted = lookup_mnt(path);
661 if (mounted) {
662 dput(path->dentry);
663 mntput(path->mnt);
664 path->mnt = mounted;
665 path->dentry = dget(mounted->mnt_root);
666 return 1;
668 return 0;
671 static __always_inline void follow_dotdot(struct nameidata *nd)
673 set_root(nd);
675 while(1) {
676 struct dentry *old = nd->path.dentry;
678 if (nd->path.dentry == nd->root.dentry &&
679 nd->path.mnt == nd->root.mnt) {
680 break;
682 if (nd->path.dentry != nd->path.mnt->mnt_root) {
683 /* rare case of legitimate dget_parent()... */
684 nd->path.dentry = dget_parent(nd->path.dentry);
685 dput(old);
686 break;
688 if (!follow_up(&nd->path))
689 break;
691 follow_mount(&nd->path);
695 * It's more convoluted than I'd like it to be, but... it's still fairly
696 * small and for now I'd prefer to have fast path as straight as possible.
697 * It _is_ time-critical.
699 static int do_lookup(struct nameidata *nd, struct qstr *name,
700 struct path *path)
702 struct vfsmount *mnt = nd->path.mnt;
703 struct dentry *dentry, *parent;
704 struct inode *dir;
706 * See if the low-level filesystem might want
707 * to use its own hash..
709 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
710 int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name);
711 if (err < 0)
712 return err;
715 dentry = __d_lookup(nd->path.dentry, name);
716 if (!dentry)
717 goto need_lookup;
718 if (dentry->d_op && dentry->d_op->d_revalidate)
719 goto need_revalidate;
720 done:
721 path->mnt = mnt;
722 path->dentry = dentry;
723 __follow_mount(path);
724 return 0;
726 need_lookup:
727 parent = nd->path.dentry;
728 dir = parent->d_inode;
730 mutex_lock(&dir->i_mutex);
732 * First re-do the cached lookup just in case it was created
733 * while we waited for the directory semaphore..
735 * FIXME! This could use version numbering or similar to
736 * avoid unnecessary cache lookups.
738 * The "dcache_lock" is purely to protect the RCU list walker
739 * from concurrent renames at this point (we mustn't get false
740 * negatives from the RCU list walk here, unlike the optimistic
741 * fast walk).
743 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
745 dentry = d_lookup(parent, name);
746 if (!dentry) {
747 struct dentry *new;
749 /* Don't create child dentry for a dead directory. */
750 dentry = ERR_PTR(-ENOENT);
751 if (IS_DEADDIR(dir))
752 goto out_unlock;
754 new = d_alloc(parent, name);
755 dentry = ERR_PTR(-ENOMEM);
756 if (new) {
757 dentry = dir->i_op->lookup(dir, new, nd);
758 if (dentry)
759 dput(new);
760 else
761 dentry = new;
763 out_unlock:
764 mutex_unlock(&dir->i_mutex);
765 if (IS_ERR(dentry))
766 goto fail;
767 goto done;
771 * Uhhuh! Nasty case: the cache was re-populated while
772 * we waited on the semaphore. Need to revalidate.
774 mutex_unlock(&dir->i_mutex);
775 if (dentry->d_op && dentry->d_op->d_revalidate) {
776 dentry = do_revalidate(dentry, nd);
777 if (!dentry)
778 dentry = ERR_PTR(-ENOENT);
780 if (IS_ERR(dentry))
781 goto fail;
782 goto done;
784 need_revalidate:
785 dentry = do_revalidate(dentry, nd);
786 if (!dentry)
787 goto need_lookup;
788 if (IS_ERR(dentry))
789 goto fail;
790 goto done;
792 fail:
793 return PTR_ERR(dentry);
797 * This is a temporary kludge to deal with "automount" symlinks; proper
798 * solution is to trigger them on follow_mount(), so that do_lookup()
799 * would DTRT. To be killed before 2.6.34-final.
801 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
803 return inode && unlikely(inode->i_op->follow_link) &&
804 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
808 * Name resolution.
809 * This is the basic name resolution function, turning a pathname into
810 * the final dentry. We expect 'base' to be positive and a directory.
812 * Returns 0 and nd will have valid dentry and mnt on success.
813 * Returns error and drops reference to input namei data on failure.
815 static int link_path_walk(const char *name, struct nameidata *nd)
817 struct path next;
818 struct inode *inode;
819 int err;
820 unsigned int lookup_flags = nd->flags;
822 while (*name=='/')
823 name++;
824 if (!*name)
825 goto return_reval;
827 inode = nd->path.dentry->d_inode;
828 if (nd->depth)
829 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
831 /* At this point we know we have a real path component. */
832 for(;;) {
833 unsigned long hash;
834 struct qstr this;
835 unsigned int c;
837 nd->flags |= LOOKUP_CONTINUE;
838 err = exec_permission(inode);
839 if (err)
840 break;
842 this.name = name;
843 c = *(const unsigned char *)name;
845 hash = init_name_hash();
846 do {
847 name++;
848 hash = partial_name_hash(c, hash);
849 c = *(const unsigned char *)name;
850 } while (c && (c != '/'));
851 this.len = name - (const char *) this.name;
852 this.hash = end_name_hash(hash);
854 /* remove trailing slashes? */
855 if (!c)
856 goto last_component;
857 while (*++name == '/');
858 if (!*name)
859 goto last_with_slashes;
862 * "." and ".." are special - ".." especially so because it has
863 * to be able to know about the current root directory and
864 * parent relationships.
866 if (this.name[0] == '.') switch (this.len) {
867 default:
868 break;
869 case 2:
870 if (this.name[1] != '.')
871 break;
872 follow_dotdot(nd);
873 inode = nd->path.dentry->d_inode;
874 /* fallthrough */
875 case 1:
876 continue;
878 /* This does the actual lookups.. */
879 err = do_lookup(nd, &this, &next);
880 if (err)
881 break;
883 err = -ENOENT;
884 inode = next.dentry->d_inode;
885 if (!inode)
886 goto out_dput;
888 if (inode->i_op->follow_link) {
889 err = do_follow_link(&next, nd);
890 if (err)
891 goto return_err;
892 err = -ENOENT;
893 inode = nd->path.dentry->d_inode;
894 if (!inode)
895 break;
896 } else
897 path_to_nameidata(&next, nd);
898 err = -ENOTDIR;
899 if (!inode->i_op->lookup)
900 break;
901 continue;
902 /* here ends the main loop */
904 last_with_slashes:
905 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
906 last_component:
907 /* Clear LOOKUP_CONTINUE iff it was previously unset */
908 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
909 if (lookup_flags & LOOKUP_PARENT)
910 goto lookup_parent;
911 if (this.name[0] == '.') switch (this.len) {
912 default:
913 break;
914 case 2:
915 if (this.name[1] != '.')
916 break;
917 follow_dotdot(nd);
918 inode = nd->path.dentry->d_inode;
919 /* fallthrough */
920 case 1:
921 goto return_reval;
923 err = do_lookup(nd, &this, &next);
924 if (err)
925 break;
926 inode = next.dentry->d_inode;
927 if (follow_on_final(inode, lookup_flags)) {
928 err = do_follow_link(&next, nd);
929 if (err)
930 goto return_err;
931 inode = nd->path.dentry->d_inode;
932 } else
933 path_to_nameidata(&next, nd);
934 err = -ENOENT;
935 if (!inode)
936 break;
937 if (lookup_flags & LOOKUP_DIRECTORY) {
938 err = -ENOTDIR;
939 if (!inode->i_op->lookup)
940 break;
942 goto return_base;
943 lookup_parent:
944 nd->last = this;
945 nd->last_type = LAST_NORM;
946 if (this.name[0] != '.')
947 goto return_base;
948 if (this.len == 1)
949 nd->last_type = LAST_DOT;
950 else if (this.len == 2 && this.name[1] == '.')
951 nd->last_type = LAST_DOTDOT;
952 else
953 goto return_base;
954 return_reval:
956 * We bypassed the ordinary revalidation routines.
957 * We may need to check the cached dentry for staleness.
959 if (nd->path.dentry && nd->path.dentry->d_sb &&
960 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
961 err = -ESTALE;
962 /* Note: we do not d_invalidate() */
963 if (!nd->path.dentry->d_op->d_revalidate(
964 nd->path.dentry, nd))
965 break;
967 return_base:
968 return 0;
969 out_dput:
970 path_put_conditional(&next, nd);
971 break;
973 path_put(&nd->path);
974 return_err:
975 return err;
978 static int path_walk(const char *name, struct nameidata *nd)
980 struct path save = nd->path;
981 int result;
983 current->total_link_count = 0;
985 /* make sure the stuff we saved doesn't go away */
986 path_get(&save);
988 result = link_path_walk(name, nd);
989 if (result == -ESTALE) {
990 /* nd->path had been dropped */
991 current->total_link_count = 0;
992 nd->path = save;
993 path_get(&nd->path);
994 nd->flags |= LOOKUP_REVAL;
995 result = link_path_walk(name, nd);
998 path_put(&save);
1000 return result;
1003 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1005 int retval = 0;
1006 int fput_needed;
1007 struct file *file;
1009 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1010 nd->flags = flags;
1011 nd->depth = 0;
1012 nd->root.mnt = NULL;
1014 if (*name=='/') {
1015 set_root(nd);
1016 nd->path = nd->root;
1017 path_get(&nd->root);
1018 } else if (dfd == AT_FDCWD) {
1019 struct fs_struct *fs = current->fs;
1020 read_lock(&fs->lock);
1021 nd->path = fs->pwd;
1022 path_get(&fs->pwd);
1023 read_unlock(&fs->lock);
1024 } else {
1025 struct dentry *dentry;
1027 file = fget_light(dfd, &fput_needed);
1028 retval = -EBADF;
1029 if (!file)
1030 goto out_fail;
1032 dentry = file->f_path.dentry;
1034 retval = -ENOTDIR;
1035 if (!S_ISDIR(dentry->d_inode->i_mode))
1036 goto fput_fail;
1038 retval = file_permission(file, MAY_EXEC);
1039 if (retval)
1040 goto fput_fail;
1042 nd->path = file->f_path;
1043 path_get(&file->f_path);
1045 fput_light(file, fput_needed);
1047 return 0;
1049 fput_fail:
1050 fput_light(file, fput_needed);
1051 out_fail:
1052 return retval;
1055 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1056 static int do_path_lookup(int dfd, const char *name,
1057 unsigned int flags, struct nameidata *nd)
1059 int retval = path_init(dfd, name, flags, nd);
1060 if (!retval)
1061 retval = path_walk(name, nd);
1062 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1063 nd->path.dentry->d_inode))
1064 audit_inode(name, nd->path.dentry);
1065 if (nd->root.mnt) {
1066 path_put(&nd->root);
1067 nd->root.mnt = NULL;
1069 return retval;
1072 int path_lookup(const char *name, unsigned int flags,
1073 struct nameidata *nd)
1075 return do_path_lookup(AT_FDCWD, name, flags, nd);
1078 int kern_path(const char *name, unsigned int flags, struct path *path)
1080 struct nameidata nd;
1081 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1082 if (!res)
1083 *path = nd.path;
1084 return res;
1088 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1089 * @dentry: pointer to dentry of the base directory
1090 * @mnt: pointer to vfs mount of the base directory
1091 * @name: pointer to file name
1092 * @flags: lookup flags
1093 * @nd: pointer to nameidata
1095 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1096 const char *name, unsigned int flags,
1097 struct nameidata *nd)
1099 int retval;
1101 /* same as do_path_lookup */
1102 nd->last_type = LAST_ROOT;
1103 nd->flags = flags;
1104 nd->depth = 0;
1106 nd->path.dentry = dentry;
1107 nd->path.mnt = mnt;
1108 path_get(&nd->path);
1109 nd->root = nd->path;
1110 path_get(&nd->root);
1112 retval = path_walk(name, nd);
1113 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1114 nd->path.dentry->d_inode))
1115 audit_inode(name, nd->path.dentry);
1117 path_put(&nd->root);
1118 nd->root.mnt = NULL;
1120 return retval;
1123 static struct dentry *__lookup_hash(struct qstr *name,
1124 struct dentry *base, struct nameidata *nd)
1126 struct dentry *dentry;
1127 struct inode *inode;
1128 int err;
1130 inode = base->d_inode;
1133 * See if the low-level filesystem might want
1134 * to use its own hash..
1136 if (base->d_op && base->d_op->d_hash) {
1137 err = base->d_op->d_hash(base, name);
1138 dentry = ERR_PTR(err);
1139 if (err < 0)
1140 goto out;
1143 dentry = __d_lookup(base, name);
1145 /* lockess __d_lookup may fail due to concurrent d_move()
1146 * in some unrelated directory, so try with d_lookup
1148 if (!dentry)
1149 dentry = d_lookup(base, name);
1151 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1152 dentry = do_revalidate(dentry, nd);
1154 if (!dentry) {
1155 struct dentry *new;
1157 /* Don't create child dentry for a dead directory. */
1158 dentry = ERR_PTR(-ENOENT);
1159 if (IS_DEADDIR(inode))
1160 goto out;
1162 new = d_alloc(base, name);
1163 dentry = ERR_PTR(-ENOMEM);
1164 if (!new)
1165 goto out;
1166 dentry = inode->i_op->lookup(inode, new, nd);
1167 if (!dentry)
1168 dentry = new;
1169 else
1170 dput(new);
1172 out:
1173 return dentry;
1177 * Restricted form of lookup. Doesn't follow links, single-component only,
1178 * needs parent already locked. Doesn't follow mounts.
1179 * SMP-safe.
1181 static struct dentry *lookup_hash(struct nameidata *nd)
1183 int err;
1185 err = exec_permission(nd->path.dentry->d_inode);
1186 if (err)
1187 return ERR_PTR(err);
1188 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1191 static int __lookup_one_len(const char *name, struct qstr *this,
1192 struct dentry *base, int len)
1194 unsigned long hash;
1195 unsigned int c;
1197 this->name = name;
1198 this->len = len;
1199 if (!len)
1200 return -EACCES;
1202 hash = init_name_hash();
1203 while (len--) {
1204 c = *(const unsigned char *)name++;
1205 if (c == '/' || c == '\0')
1206 return -EACCES;
1207 hash = partial_name_hash(c, hash);
1209 this->hash = end_name_hash(hash);
1210 return 0;
1214 * lookup_one_len - filesystem helper to lookup single pathname component
1215 * @name: pathname component to lookup
1216 * @base: base directory to lookup from
1217 * @len: maximum length @len should be interpreted to
1219 * Note that this routine is purely a helper for filesystem usage and should
1220 * not be called by generic code. Also note that by using this function the
1221 * nameidata argument is passed to the filesystem methods and a filesystem
1222 * using this helper needs to be prepared for that.
1224 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1226 int err;
1227 struct qstr this;
1229 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1231 err = __lookup_one_len(name, &this, base, len);
1232 if (err)
1233 return ERR_PTR(err);
1235 err = exec_permission(base->d_inode);
1236 if (err)
1237 return ERR_PTR(err);
1238 return __lookup_hash(&this, base, NULL);
1241 int user_path_at(int dfd, const char __user *name, unsigned flags,
1242 struct path *path)
1244 struct nameidata nd;
1245 char *tmp = getname(name);
1246 int err = PTR_ERR(tmp);
1247 if (!IS_ERR(tmp)) {
1249 BUG_ON(flags & LOOKUP_PARENT);
1251 err = do_path_lookup(dfd, tmp, flags, &nd);
1252 putname(tmp);
1253 if (!err)
1254 *path = nd.path;
1256 return err;
1259 static int user_path_parent(int dfd, const char __user *path,
1260 struct nameidata *nd, char **name)
1262 char *s = getname(path);
1263 int error;
1265 if (IS_ERR(s))
1266 return PTR_ERR(s);
1268 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1269 if (error)
1270 putname(s);
1271 else
1272 *name = s;
1274 return error;
1278 * It's inline, so penalty for filesystems that don't use sticky bit is
1279 * minimal.
1281 static inline int check_sticky(struct inode *dir, struct inode *inode)
1283 uid_t fsuid = current_fsuid();
1285 if (!(dir->i_mode & S_ISVTX))
1286 return 0;
1287 if (inode->i_uid == fsuid)
1288 return 0;
1289 if (dir->i_uid == fsuid)
1290 return 0;
1291 return !capable(CAP_FOWNER);
1295 * Check whether we can remove a link victim from directory dir, check
1296 * whether the type of victim is right.
1297 * 1. We can't do it if dir is read-only (done in permission())
1298 * 2. We should have write and exec permissions on dir
1299 * 3. We can't remove anything from append-only dir
1300 * 4. We can't do anything with immutable dir (done in permission())
1301 * 5. If the sticky bit on dir is set we should either
1302 * a. be owner of dir, or
1303 * b. be owner of victim, or
1304 * c. have CAP_FOWNER capability
1305 * 6. If the victim is append-only or immutable we can't do antyhing with
1306 * links pointing to it.
1307 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1308 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1309 * 9. We can't remove a root or mountpoint.
1310 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1311 * nfs_async_unlink().
1313 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1315 int error;
1317 if (!victim->d_inode)
1318 return -ENOENT;
1320 BUG_ON(victim->d_parent->d_inode != dir);
1321 audit_inode_child(victim, dir);
1323 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1324 if (error)
1325 return error;
1326 if (IS_APPEND(dir))
1327 return -EPERM;
1328 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1329 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1330 return -EPERM;
1331 if (isdir) {
1332 if (!S_ISDIR(victim->d_inode->i_mode))
1333 return -ENOTDIR;
1334 if (IS_ROOT(victim))
1335 return -EBUSY;
1336 } else if (S_ISDIR(victim->d_inode->i_mode))
1337 return -EISDIR;
1338 if (IS_DEADDIR(dir))
1339 return -ENOENT;
1340 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1341 return -EBUSY;
1342 return 0;
1345 /* Check whether we can create an object with dentry child in directory
1346 * dir.
1347 * 1. We can't do it if child already exists (open has special treatment for
1348 * this case, but since we are inlined it's OK)
1349 * 2. We can't do it if dir is read-only (done in permission())
1350 * 3. We should have write and exec permissions on dir
1351 * 4. We can't do it if dir is immutable (done in permission())
1353 static inline int may_create(struct inode *dir, struct dentry *child)
1355 if (child->d_inode)
1356 return -EEXIST;
1357 if (IS_DEADDIR(dir))
1358 return -ENOENT;
1359 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1363 * p1 and p2 should be directories on the same fs.
1365 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1367 struct dentry *p;
1369 if (p1 == p2) {
1370 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1371 return NULL;
1374 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1376 p = d_ancestor(p2, p1);
1377 if (p) {
1378 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1379 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1380 return p;
1383 p = d_ancestor(p1, p2);
1384 if (p) {
1385 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1386 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1387 return p;
1390 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1391 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1392 return NULL;
1395 void unlock_rename(struct dentry *p1, struct dentry *p2)
1397 mutex_unlock(&p1->d_inode->i_mutex);
1398 if (p1 != p2) {
1399 mutex_unlock(&p2->d_inode->i_mutex);
1400 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1404 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1405 struct nameidata *nd)
1407 int error = may_create(dir, dentry);
1409 if (error)
1410 return error;
1412 if (!dir->i_op->create)
1413 return -EACCES; /* shouldn't it be ENOSYS? */
1414 mode &= S_IALLUGO;
1415 mode |= S_IFREG;
1416 error = security_inode_create(dir, dentry, mode);
1417 if (error)
1418 return error;
1419 error = dir->i_op->create(dir, dentry, mode, nd);
1420 if (!error)
1421 fsnotify_create(dir, dentry);
1422 return error;
1425 int may_open(struct path *path, int acc_mode, int flag)
1427 struct dentry *dentry = path->dentry;
1428 struct inode *inode = dentry->d_inode;
1429 int error;
1431 if (!inode)
1432 return -ENOENT;
1434 switch (inode->i_mode & S_IFMT) {
1435 case S_IFLNK:
1436 return -ELOOP;
1437 case S_IFDIR:
1438 if (acc_mode & MAY_WRITE)
1439 return -EISDIR;
1440 break;
1441 case S_IFBLK:
1442 case S_IFCHR:
1443 if (path->mnt->mnt_flags & MNT_NODEV)
1444 return -EACCES;
1445 /*FALLTHRU*/
1446 case S_IFIFO:
1447 case S_IFSOCK:
1448 flag &= ~O_TRUNC;
1449 break;
1452 error = inode_permission(inode, acc_mode);
1453 if (error)
1454 return error;
1457 * An append-only file must be opened in append mode for writing.
1459 if (IS_APPEND(inode)) {
1460 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1461 return -EPERM;
1462 if (flag & O_TRUNC)
1463 return -EPERM;
1466 /* O_NOATIME can only be set by the owner or superuser */
1467 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1468 return -EPERM;
1471 * Ensure there are no outstanding leases on the file.
1473 return break_lease(inode, flag);
1476 static int handle_truncate(struct path *path)
1478 struct inode *inode = path->dentry->d_inode;
1479 int error = get_write_access(inode);
1480 if (error)
1481 return error;
1483 * Refuse to truncate files with mandatory locks held on them.
1485 error = locks_verify_locked(inode);
1486 if (!error)
1487 error = security_path_truncate(path, 0,
1488 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1489 if (!error) {
1490 error = do_truncate(path->dentry, 0,
1491 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1492 NULL);
1494 put_write_access(inode);
1495 return error;
1499 * Be careful about ever adding any more callers of this
1500 * function. Its flags must be in the namei format, not
1501 * what get passed to sys_open().
1503 static int __open_namei_create(struct nameidata *nd, struct path *path,
1504 int open_flag, int mode)
1506 int error;
1507 struct dentry *dir = nd->path.dentry;
1509 if (!IS_POSIXACL(dir->d_inode))
1510 mode &= ~current_umask();
1511 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1512 if (error)
1513 goto out_unlock;
1514 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1515 out_unlock:
1516 mutex_unlock(&dir->d_inode->i_mutex);
1517 dput(nd->path.dentry);
1518 nd->path.dentry = path->dentry;
1519 if (error)
1520 return error;
1521 /* Don't check for write permission, don't truncate */
1522 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
1526 * Note that while the flag value (low two bits) for sys_open means:
1527 * 00 - read-only
1528 * 01 - write-only
1529 * 10 - read-write
1530 * 11 - special
1531 * it is changed into
1532 * 00 - no permissions needed
1533 * 01 - read-permission
1534 * 10 - write-permission
1535 * 11 - read-write
1536 * for the internal routines (ie open_namei()/follow_link() etc)
1537 * This is more logical, and also allows the 00 "no perm needed"
1538 * to be used for symlinks (where the permissions are checked
1539 * later).
1542 static inline int open_to_namei_flags(int flag)
1544 if ((flag+1) & O_ACCMODE)
1545 flag++;
1546 return flag;
1549 static int open_will_truncate(int flag, struct inode *inode)
1552 * We'll never write to the fs underlying
1553 * a device file.
1555 if (special_file(inode->i_mode))
1556 return 0;
1557 return (flag & O_TRUNC);
1560 static struct file *finish_open(struct nameidata *nd,
1561 int open_flag, int acc_mode)
1563 struct file *filp;
1564 int will_truncate;
1565 int error;
1567 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
1568 if (will_truncate) {
1569 error = mnt_want_write(nd->path.mnt);
1570 if (error)
1571 goto exit;
1573 error = may_open(&nd->path, acc_mode, open_flag);
1574 if (error) {
1575 if (will_truncate)
1576 mnt_drop_write(nd->path.mnt);
1577 goto exit;
1579 filp = nameidata_to_filp(nd);
1580 if (!IS_ERR(filp)) {
1581 error = ima_file_check(filp, acc_mode);
1582 if (error) {
1583 fput(filp);
1584 filp = ERR_PTR(error);
1587 if (!IS_ERR(filp)) {
1588 if (will_truncate) {
1589 error = handle_truncate(&nd->path);
1590 if (error) {
1591 fput(filp);
1592 filp = ERR_PTR(error);
1597 * It is now safe to drop the mnt write
1598 * because the filp has had a write taken
1599 * on its behalf.
1601 if (will_truncate)
1602 mnt_drop_write(nd->path.mnt);
1603 return filp;
1605 exit:
1606 if (!IS_ERR(nd->intent.open.file))
1607 release_open_intent(nd);
1608 path_put(&nd->path);
1609 return ERR_PTR(error);
1612 static struct file *do_last(struct nameidata *nd, struct path *path,
1613 int open_flag, int acc_mode,
1614 int mode, const char *pathname)
1616 struct dentry *dir = nd->path.dentry;
1617 struct file *filp;
1618 int error = -EISDIR;
1620 switch (nd->last_type) {
1621 case LAST_DOTDOT:
1622 follow_dotdot(nd);
1623 dir = nd->path.dentry;
1624 case LAST_DOT:
1625 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
1626 if (!dir->d_op->d_revalidate(dir, nd)) {
1627 error = -ESTALE;
1628 goto exit;
1631 /* fallthrough */
1632 case LAST_ROOT:
1633 if (open_flag & O_CREAT)
1634 goto exit;
1635 /* fallthrough */
1636 case LAST_BIND:
1637 audit_inode(pathname, dir);
1638 goto ok;
1641 /* trailing slashes? */
1642 if (nd->last.name[nd->last.len]) {
1643 if (open_flag & O_CREAT)
1644 goto exit;
1645 nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW;
1648 /* just plain open? */
1649 if (!(open_flag & O_CREAT)) {
1650 error = do_lookup(nd, &nd->last, path);
1651 if (error)
1652 goto exit;
1653 error = -ENOENT;
1654 if (!path->dentry->d_inode)
1655 goto exit_dput;
1656 if (path->dentry->d_inode->i_op->follow_link)
1657 return NULL;
1658 error = -ENOTDIR;
1659 if (nd->flags & LOOKUP_DIRECTORY) {
1660 if (!path->dentry->d_inode->i_op->lookup)
1661 goto exit_dput;
1663 path_to_nameidata(path, nd);
1664 audit_inode(pathname, nd->path.dentry);
1665 goto ok;
1668 /* OK, it's O_CREAT */
1669 mutex_lock(&dir->d_inode->i_mutex);
1671 path->dentry = lookup_hash(nd);
1672 path->mnt = nd->path.mnt;
1674 error = PTR_ERR(path->dentry);
1675 if (IS_ERR(path->dentry)) {
1676 mutex_unlock(&dir->d_inode->i_mutex);
1677 goto exit;
1680 if (IS_ERR(nd->intent.open.file)) {
1681 error = PTR_ERR(nd->intent.open.file);
1682 goto exit_mutex_unlock;
1685 /* Negative dentry, just create the file */
1686 if (!path->dentry->d_inode) {
1688 * This write is needed to ensure that a
1689 * ro->rw transition does not occur between
1690 * the time when the file is created and when
1691 * a permanent write count is taken through
1692 * the 'struct file' in nameidata_to_filp().
1694 error = mnt_want_write(nd->path.mnt);
1695 if (error)
1696 goto exit_mutex_unlock;
1697 error = __open_namei_create(nd, path, open_flag, mode);
1698 if (error) {
1699 mnt_drop_write(nd->path.mnt);
1700 goto exit;
1702 filp = nameidata_to_filp(nd);
1703 mnt_drop_write(nd->path.mnt);
1704 if (!IS_ERR(filp)) {
1705 error = ima_file_check(filp, acc_mode);
1706 if (error) {
1707 fput(filp);
1708 filp = ERR_PTR(error);
1711 return filp;
1715 * It already exists.
1717 mutex_unlock(&dir->d_inode->i_mutex);
1718 audit_inode(pathname, path->dentry);
1720 error = -EEXIST;
1721 if (open_flag & O_EXCL)
1722 goto exit_dput;
1724 if (__follow_mount(path)) {
1725 error = -ELOOP;
1726 if (open_flag & O_NOFOLLOW)
1727 goto exit_dput;
1730 error = -ENOENT;
1731 if (!path->dentry->d_inode)
1732 goto exit_dput;
1734 if (path->dentry->d_inode->i_op->follow_link)
1735 return NULL;
1737 path_to_nameidata(path, nd);
1738 error = -EISDIR;
1739 if (S_ISDIR(path->dentry->d_inode->i_mode))
1740 goto exit;
1742 filp = finish_open(nd, open_flag, acc_mode);
1743 return filp;
1745 exit_mutex_unlock:
1746 mutex_unlock(&dir->d_inode->i_mutex);
1747 exit_dput:
1748 path_put_conditional(path, nd);
1749 exit:
1750 if (!IS_ERR(nd->intent.open.file))
1751 release_open_intent(nd);
1752 path_put(&nd->path);
1753 return ERR_PTR(error);
1757 * Note that the low bits of the passed in "open_flag"
1758 * are not the same as in the local variable "flag". See
1759 * open_to_namei_flags() for more details.
1761 struct file *do_filp_open(int dfd, const char *pathname,
1762 int open_flag, int mode, int acc_mode)
1764 struct file *filp;
1765 struct nameidata nd;
1766 int error;
1767 struct path path;
1768 int count = 0;
1769 int flag = open_to_namei_flags(open_flag);
1770 int force_reval = 0;
1772 if (!(open_flag & O_CREAT))
1773 mode = 0;
1776 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1777 * check for O_DSYNC if the need any syncing at all we enforce it's
1778 * always set instead of having to deal with possibly weird behaviour
1779 * for malicious applications setting only __O_SYNC.
1781 if (open_flag & __O_SYNC)
1782 open_flag |= O_DSYNC;
1784 if (!acc_mode)
1785 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1787 /* O_TRUNC implies we need access checks for write permissions */
1788 if (open_flag & O_TRUNC)
1789 acc_mode |= MAY_WRITE;
1791 /* Allow the LSM permission hook to distinguish append
1792 access from general write access. */
1793 if (open_flag & O_APPEND)
1794 acc_mode |= MAY_APPEND;
1796 /* find the parent */
1797 reval:
1798 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1799 if (error)
1800 return ERR_PTR(error);
1801 if (force_reval)
1802 nd.flags |= LOOKUP_REVAL;
1804 current->total_link_count = 0;
1805 error = link_path_walk(pathname, &nd);
1806 if (error) {
1807 filp = ERR_PTR(error);
1808 goto out;
1810 if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT))
1811 audit_inode(pathname, nd.path.dentry);
1814 * We have the parent and last component.
1817 error = -ENFILE;
1818 filp = get_empty_filp();
1819 if (filp == NULL)
1820 goto exit_parent;
1821 nd.intent.open.file = filp;
1822 filp->f_flags = open_flag;
1823 nd.intent.open.flags = flag;
1824 nd.intent.open.create_mode = mode;
1825 nd.flags &= ~LOOKUP_PARENT;
1826 nd.flags |= LOOKUP_OPEN;
1827 if (open_flag & O_CREAT) {
1828 nd.flags |= LOOKUP_CREATE;
1829 if (open_flag & O_EXCL)
1830 nd.flags |= LOOKUP_EXCL;
1832 if (open_flag & O_DIRECTORY)
1833 nd.flags |= LOOKUP_DIRECTORY;
1834 if (!(open_flag & O_NOFOLLOW))
1835 nd.flags |= LOOKUP_FOLLOW;
1836 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1837 while (unlikely(!filp)) { /* trailing symlink */
1838 struct path holder;
1839 struct inode *inode = path.dentry->d_inode;
1840 void *cookie;
1841 error = -ELOOP;
1842 /* S_ISDIR part is a temporary automount kludge */
1843 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode))
1844 goto exit_dput;
1845 if (count++ == 32)
1846 goto exit_dput;
1848 * This is subtle. Instead of calling do_follow_link() we do
1849 * the thing by hands. The reason is that this way we have zero
1850 * link_count and path_walk() (called from ->follow_link)
1851 * honoring LOOKUP_PARENT. After that we have the parent and
1852 * last component, i.e. we are in the same situation as after
1853 * the first path_walk(). Well, almost - if the last component
1854 * is normal we get its copy stored in nd->last.name and we will
1855 * have to putname() it when we are done. Procfs-like symlinks
1856 * just set LAST_BIND.
1858 nd.flags |= LOOKUP_PARENT;
1859 error = security_inode_follow_link(path.dentry, &nd);
1860 if (error)
1861 goto exit_dput;
1862 error = __do_follow_link(&path, &nd, &cookie);
1863 if (unlikely(error)) {
1864 /* nd.path had been dropped */
1865 if (!IS_ERR(cookie) && inode->i_op->put_link)
1866 inode->i_op->put_link(path.dentry, &nd, cookie);
1867 path_put(&path);
1868 release_open_intent(&nd);
1869 filp = ERR_PTR(error);
1870 goto out;
1872 holder = path;
1873 nd.flags &= ~LOOKUP_PARENT;
1874 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
1875 if (inode->i_op->put_link)
1876 inode->i_op->put_link(holder.dentry, &nd, cookie);
1877 path_put(&holder);
1879 out:
1880 if (nd.root.mnt)
1881 path_put(&nd.root);
1882 if (filp == ERR_PTR(-ESTALE) && !force_reval) {
1883 force_reval = 1;
1884 goto reval;
1886 return filp;
1888 exit_dput:
1889 path_put_conditional(&path, &nd);
1890 if (!IS_ERR(nd.intent.open.file))
1891 release_open_intent(&nd);
1892 exit_parent:
1893 path_put(&nd.path);
1894 filp = ERR_PTR(error);
1895 goto out;
1899 * filp_open - open file and return file pointer
1901 * @filename: path to open
1902 * @flags: open flags as per the open(2) second argument
1903 * @mode: mode for the new file if O_CREAT is set, else ignored
1905 * This is the helper to open a file from kernelspace if you really
1906 * have to. But in generally you should not do this, so please move
1907 * along, nothing to see here..
1909 struct file *filp_open(const char *filename, int flags, int mode)
1911 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1913 EXPORT_SYMBOL(filp_open);
1916 * lookup_create - lookup a dentry, creating it if it doesn't exist
1917 * @nd: nameidata info
1918 * @is_dir: directory flag
1920 * Simple function to lookup and return a dentry and create it
1921 * if it doesn't exist. Is SMP-safe.
1923 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1925 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1927 struct dentry *dentry = ERR_PTR(-EEXIST);
1929 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1931 * Yucky last component or no last component at all?
1932 * (foo/., foo/.., /////)
1934 if (nd->last_type != LAST_NORM)
1935 goto fail;
1936 nd->flags &= ~LOOKUP_PARENT;
1937 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1938 nd->intent.open.flags = O_EXCL;
1941 * Do the final lookup.
1943 dentry = lookup_hash(nd);
1944 if (IS_ERR(dentry))
1945 goto fail;
1947 if (dentry->d_inode)
1948 goto eexist;
1950 * Special case - lookup gave negative, but... we had foo/bar/
1951 * From the vfs_mknod() POV we just have a negative dentry -
1952 * all is fine. Let's be bastards - you had / on the end, you've
1953 * been asking for (non-existent) directory. -ENOENT for you.
1955 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1956 dput(dentry);
1957 dentry = ERR_PTR(-ENOENT);
1959 return dentry;
1960 eexist:
1961 dput(dentry);
1962 dentry = ERR_PTR(-EEXIST);
1963 fail:
1964 return dentry;
1966 EXPORT_SYMBOL_GPL(lookup_create);
1968 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1970 int error = may_create(dir, dentry);
1972 if (error)
1973 return error;
1975 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1976 return -EPERM;
1978 if (!dir->i_op->mknod)
1979 return -EPERM;
1981 error = devcgroup_inode_mknod(mode, dev);
1982 if (error)
1983 return error;
1985 error = security_inode_mknod(dir, dentry, mode, dev);
1986 if (error)
1987 return error;
1989 error = dir->i_op->mknod(dir, dentry, mode, dev);
1990 if (!error)
1991 fsnotify_create(dir, dentry);
1992 return error;
1995 static int may_mknod(mode_t mode)
1997 switch (mode & S_IFMT) {
1998 case S_IFREG:
1999 case S_IFCHR:
2000 case S_IFBLK:
2001 case S_IFIFO:
2002 case S_IFSOCK:
2003 case 0: /* zero mode translates to S_IFREG */
2004 return 0;
2005 case S_IFDIR:
2006 return -EPERM;
2007 default:
2008 return -EINVAL;
2012 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2013 unsigned, dev)
2015 int error;
2016 char *tmp;
2017 struct dentry *dentry;
2018 struct nameidata nd;
2020 if (S_ISDIR(mode))
2021 return -EPERM;
2023 error = user_path_parent(dfd, filename, &nd, &tmp);
2024 if (error)
2025 return error;
2027 dentry = lookup_create(&nd, 0);
2028 if (IS_ERR(dentry)) {
2029 error = PTR_ERR(dentry);
2030 goto out_unlock;
2032 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2033 mode &= ~current_umask();
2034 error = may_mknod(mode);
2035 if (error)
2036 goto out_dput;
2037 error = mnt_want_write(nd.path.mnt);
2038 if (error)
2039 goto out_dput;
2040 error = security_path_mknod(&nd.path, dentry, mode, dev);
2041 if (error)
2042 goto out_drop_write;
2043 switch (mode & S_IFMT) {
2044 case 0: case S_IFREG:
2045 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2046 break;
2047 case S_IFCHR: case S_IFBLK:
2048 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2049 new_decode_dev(dev));
2050 break;
2051 case S_IFIFO: case S_IFSOCK:
2052 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2053 break;
2055 out_drop_write:
2056 mnt_drop_write(nd.path.mnt);
2057 out_dput:
2058 dput(dentry);
2059 out_unlock:
2060 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2061 path_put(&nd.path);
2062 putname(tmp);
2064 return error;
2067 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2069 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2072 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2074 int error = may_create(dir, dentry);
2076 if (error)
2077 return error;
2079 if (!dir->i_op->mkdir)
2080 return -EPERM;
2082 mode &= (S_IRWXUGO|S_ISVTX);
2083 error = security_inode_mkdir(dir, dentry, mode);
2084 if (error)
2085 return error;
2087 error = dir->i_op->mkdir(dir, dentry, mode);
2088 if (!error)
2089 fsnotify_mkdir(dir, dentry);
2090 return error;
2093 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2095 int error = 0;
2096 char * tmp;
2097 struct dentry *dentry;
2098 struct nameidata nd;
2100 error = user_path_parent(dfd, pathname, &nd, &tmp);
2101 if (error)
2102 goto out_err;
2104 dentry = lookup_create(&nd, 1);
2105 error = PTR_ERR(dentry);
2106 if (IS_ERR(dentry))
2107 goto out_unlock;
2109 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2110 mode &= ~current_umask();
2111 error = mnt_want_write(nd.path.mnt);
2112 if (error)
2113 goto out_dput;
2114 error = security_path_mkdir(&nd.path, dentry, mode);
2115 if (error)
2116 goto out_drop_write;
2117 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2118 out_drop_write:
2119 mnt_drop_write(nd.path.mnt);
2120 out_dput:
2121 dput(dentry);
2122 out_unlock:
2123 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2124 path_put(&nd.path);
2125 putname(tmp);
2126 out_err:
2127 return error;
2130 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2132 return sys_mkdirat(AT_FDCWD, pathname, mode);
2136 * We try to drop the dentry early: we should have
2137 * a usage count of 2 if we're the only user of this
2138 * dentry, and if that is true (possibly after pruning
2139 * the dcache), then we drop the dentry now.
2141 * A low-level filesystem can, if it choses, legally
2142 * do a
2144 * if (!d_unhashed(dentry))
2145 * return -EBUSY;
2147 * if it cannot handle the case of removing a directory
2148 * that is still in use by something else..
2150 void dentry_unhash(struct dentry *dentry)
2152 dget(dentry);
2153 shrink_dcache_parent(dentry);
2154 spin_lock(&dcache_lock);
2155 spin_lock(&dentry->d_lock);
2156 if (atomic_read(&dentry->d_count) == 2)
2157 __d_drop(dentry);
2158 spin_unlock(&dentry->d_lock);
2159 spin_unlock(&dcache_lock);
2162 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2164 int error = may_delete(dir, dentry, 1);
2166 if (error)
2167 return error;
2169 if (!dir->i_op->rmdir)
2170 return -EPERM;
2172 mutex_lock(&dentry->d_inode->i_mutex);
2173 dentry_unhash(dentry);
2174 if (d_mountpoint(dentry))
2175 error = -EBUSY;
2176 else {
2177 error = security_inode_rmdir(dir, dentry);
2178 if (!error) {
2179 error = dir->i_op->rmdir(dir, dentry);
2180 if (!error) {
2181 dentry->d_inode->i_flags |= S_DEAD;
2182 dont_mount(dentry);
2186 mutex_unlock(&dentry->d_inode->i_mutex);
2187 if (!error) {
2188 d_delete(dentry);
2190 dput(dentry);
2192 return error;
2195 static long do_rmdir(int dfd, const char __user *pathname)
2197 int error = 0;
2198 char * name;
2199 struct dentry *dentry;
2200 struct nameidata nd;
2202 error = user_path_parent(dfd, pathname, &nd, &name);
2203 if (error)
2204 return error;
2206 switch(nd.last_type) {
2207 case LAST_DOTDOT:
2208 error = -ENOTEMPTY;
2209 goto exit1;
2210 case LAST_DOT:
2211 error = -EINVAL;
2212 goto exit1;
2213 case LAST_ROOT:
2214 error = -EBUSY;
2215 goto exit1;
2218 nd.flags &= ~LOOKUP_PARENT;
2220 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2221 dentry = lookup_hash(&nd);
2222 error = PTR_ERR(dentry);
2223 if (IS_ERR(dentry))
2224 goto exit2;
2225 error = mnt_want_write(nd.path.mnt);
2226 if (error)
2227 goto exit3;
2228 error = security_path_rmdir(&nd.path, dentry);
2229 if (error)
2230 goto exit4;
2231 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2232 exit4:
2233 mnt_drop_write(nd.path.mnt);
2234 exit3:
2235 dput(dentry);
2236 exit2:
2237 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2238 exit1:
2239 path_put(&nd.path);
2240 putname(name);
2241 return error;
2244 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2246 return do_rmdir(AT_FDCWD, pathname);
2249 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2251 int error = may_delete(dir, dentry, 0);
2253 if (error)
2254 return error;
2256 if (!dir->i_op->unlink)
2257 return -EPERM;
2259 mutex_lock(&dentry->d_inode->i_mutex);
2260 if (d_mountpoint(dentry))
2261 error = -EBUSY;
2262 else {
2263 error = security_inode_unlink(dir, dentry);
2264 if (!error) {
2265 error = dir->i_op->unlink(dir, dentry);
2266 if (!error)
2267 dont_mount(dentry);
2270 mutex_unlock(&dentry->d_inode->i_mutex);
2272 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2273 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2274 fsnotify_link_count(dentry->d_inode);
2275 d_delete(dentry);
2278 return error;
2282 * Make sure that the actual truncation of the file will occur outside its
2283 * directory's i_mutex. Truncate can take a long time if there is a lot of
2284 * writeout happening, and we don't want to prevent access to the directory
2285 * while waiting on the I/O.
2287 static long do_unlinkat(int dfd, const char __user *pathname)
2289 int error;
2290 char *name;
2291 struct dentry *dentry;
2292 struct nameidata nd;
2293 struct inode *inode = NULL;
2295 error = user_path_parent(dfd, pathname, &nd, &name);
2296 if (error)
2297 return error;
2299 error = -EISDIR;
2300 if (nd.last_type != LAST_NORM)
2301 goto exit1;
2303 nd.flags &= ~LOOKUP_PARENT;
2305 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2306 dentry = lookup_hash(&nd);
2307 error = PTR_ERR(dentry);
2308 if (!IS_ERR(dentry)) {
2309 /* Why not before? Because we want correct error value */
2310 if (nd.last.name[nd.last.len])
2311 goto slashes;
2312 inode = dentry->d_inode;
2313 if (inode)
2314 atomic_inc(&inode->i_count);
2315 error = mnt_want_write(nd.path.mnt);
2316 if (error)
2317 goto exit2;
2318 error = security_path_unlink(&nd.path, dentry);
2319 if (error)
2320 goto exit3;
2321 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2322 exit3:
2323 mnt_drop_write(nd.path.mnt);
2324 exit2:
2325 dput(dentry);
2327 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2328 if (inode)
2329 iput(inode); /* truncate the inode here */
2330 exit1:
2331 path_put(&nd.path);
2332 putname(name);
2333 return error;
2335 slashes:
2336 error = !dentry->d_inode ? -ENOENT :
2337 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2338 goto exit2;
2341 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2343 if ((flag & ~AT_REMOVEDIR) != 0)
2344 return -EINVAL;
2346 if (flag & AT_REMOVEDIR)
2347 return do_rmdir(dfd, pathname);
2349 return do_unlinkat(dfd, pathname);
2352 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2354 return do_unlinkat(AT_FDCWD, pathname);
2357 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2359 int error = may_create(dir, dentry);
2361 if (error)
2362 return error;
2364 if (!dir->i_op->symlink)
2365 return -EPERM;
2367 error = security_inode_symlink(dir, dentry, oldname);
2368 if (error)
2369 return error;
2371 error = dir->i_op->symlink(dir, dentry, oldname);
2372 if (!error)
2373 fsnotify_create(dir, dentry);
2374 return error;
2377 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2378 int, newdfd, const char __user *, newname)
2380 int error;
2381 char *from;
2382 char *to;
2383 struct dentry *dentry;
2384 struct nameidata nd;
2386 from = getname(oldname);
2387 if (IS_ERR(from))
2388 return PTR_ERR(from);
2390 error = user_path_parent(newdfd, newname, &nd, &to);
2391 if (error)
2392 goto out_putname;
2394 dentry = lookup_create(&nd, 0);
2395 error = PTR_ERR(dentry);
2396 if (IS_ERR(dentry))
2397 goto out_unlock;
2399 error = mnt_want_write(nd.path.mnt);
2400 if (error)
2401 goto out_dput;
2402 error = security_path_symlink(&nd.path, dentry, from);
2403 if (error)
2404 goto out_drop_write;
2405 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2406 out_drop_write:
2407 mnt_drop_write(nd.path.mnt);
2408 out_dput:
2409 dput(dentry);
2410 out_unlock:
2411 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2412 path_put(&nd.path);
2413 putname(to);
2414 out_putname:
2415 putname(from);
2416 return error;
2419 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2421 return sys_symlinkat(oldname, AT_FDCWD, newname);
2424 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2426 struct inode *inode = old_dentry->d_inode;
2427 int error;
2429 if (!inode)
2430 return -ENOENT;
2432 error = may_create(dir, new_dentry);
2433 if (error)
2434 return error;
2436 if (dir->i_sb != inode->i_sb)
2437 return -EXDEV;
2440 * A link to an append-only or immutable file cannot be created.
2442 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2443 return -EPERM;
2444 if (!dir->i_op->link)
2445 return -EPERM;
2446 if (S_ISDIR(inode->i_mode))
2447 return -EPERM;
2449 error = security_inode_link(old_dentry, dir, new_dentry);
2450 if (error)
2451 return error;
2453 mutex_lock(&inode->i_mutex);
2454 error = dir->i_op->link(old_dentry, dir, new_dentry);
2455 mutex_unlock(&inode->i_mutex);
2456 if (!error)
2457 fsnotify_link(dir, inode, new_dentry);
2458 return error;
2462 * Hardlinks are often used in delicate situations. We avoid
2463 * security-related surprises by not following symlinks on the
2464 * newname. --KAB
2466 * We don't follow them on the oldname either to be compatible
2467 * with linux 2.0, and to avoid hard-linking to directories
2468 * and other special files. --ADM
2470 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2471 int, newdfd, const char __user *, newname, int, flags)
2473 struct dentry *new_dentry;
2474 struct nameidata nd;
2475 struct path old_path;
2476 int error;
2477 char *to;
2479 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2480 return -EINVAL;
2482 error = user_path_at(olddfd, oldname,
2483 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2484 &old_path);
2485 if (error)
2486 return error;
2488 error = user_path_parent(newdfd, newname, &nd, &to);
2489 if (error)
2490 goto out;
2491 error = -EXDEV;
2492 if (old_path.mnt != nd.path.mnt)
2493 goto out_release;
2494 new_dentry = lookup_create(&nd, 0);
2495 error = PTR_ERR(new_dentry);
2496 if (IS_ERR(new_dentry))
2497 goto out_unlock;
2498 error = mnt_want_write(nd.path.mnt);
2499 if (error)
2500 goto out_dput;
2501 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2502 if (error)
2503 goto out_drop_write;
2504 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2505 out_drop_write:
2506 mnt_drop_write(nd.path.mnt);
2507 out_dput:
2508 dput(new_dentry);
2509 out_unlock:
2510 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2511 out_release:
2512 path_put(&nd.path);
2513 putname(to);
2514 out:
2515 path_put(&old_path);
2517 return error;
2520 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2522 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2526 * The worst of all namespace operations - renaming directory. "Perverted"
2527 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2528 * Problems:
2529 * a) we can get into loop creation. Check is done in is_subdir().
2530 * b) race potential - two innocent renames can create a loop together.
2531 * That's where 4.4 screws up. Current fix: serialization on
2532 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2533 * story.
2534 * c) we have to lock _three_ objects - parents and victim (if it exists).
2535 * And that - after we got ->i_mutex on parents (until then we don't know
2536 * whether the target exists). Solution: try to be smart with locking
2537 * order for inodes. We rely on the fact that tree topology may change
2538 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2539 * move will be locked. Thus we can rank directories by the tree
2540 * (ancestors first) and rank all non-directories after them.
2541 * That works since everybody except rename does "lock parent, lookup,
2542 * lock child" and rename is under ->s_vfs_rename_mutex.
2543 * HOWEVER, it relies on the assumption that any object with ->lookup()
2544 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2545 * we'd better make sure that there's no link(2) for them.
2546 * d) some filesystems don't support opened-but-unlinked directories,
2547 * either because of layout or because they are not ready to deal with
2548 * all cases correctly. The latter will be fixed (taking this sort of
2549 * stuff into VFS), but the former is not going away. Solution: the same
2550 * trick as in rmdir().
2551 * e) conversion from fhandle to dentry may come in the wrong moment - when
2552 * we are removing the target. Solution: we will have to grab ->i_mutex
2553 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2554 * ->i_mutex on parents, which works but leads to some truly excessive
2555 * locking].
2557 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2558 struct inode *new_dir, struct dentry *new_dentry)
2560 int error = 0;
2561 struct inode *target;
2564 * If we are going to change the parent - check write permissions,
2565 * we'll need to flip '..'.
2567 if (new_dir != old_dir) {
2568 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2569 if (error)
2570 return error;
2573 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2574 if (error)
2575 return error;
2577 target = new_dentry->d_inode;
2578 if (target)
2579 mutex_lock(&target->i_mutex);
2580 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2581 error = -EBUSY;
2582 else {
2583 if (target)
2584 dentry_unhash(new_dentry);
2585 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2587 if (target) {
2588 if (!error) {
2589 target->i_flags |= S_DEAD;
2590 dont_mount(new_dentry);
2592 mutex_unlock(&target->i_mutex);
2593 if (d_unhashed(new_dentry))
2594 d_rehash(new_dentry);
2595 dput(new_dentry);
2597 if (!error)
2598 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2599 d_move(old_dentry,new_dentry);
2600 return error;
2603 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2604 struct inode *new_dir, struct dentry *new_dentry)
2606 struct inode *target;
2607 int error;
2609 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2610 if (error)
2611 return error;
2613 dget(new_dentry);
2614 target = new_dentry->d_inode;
2615 if (target)
2616 mutex_lock(&target->i_mutex);
2617 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2618 error = -EBUSY;
2619 else
2620 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2621 if (!error) {
2622 if (target)
2623 dont_mount(new_dentry);
2624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2625 d_move(old_dentry, new_dentry);
2627 if (target)
2628 mutex_unlock(&target->i_mutex);
2629 dput(new_dentry);
2630 return error;
2633 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2634 struct inode *new_dir, struct dentry *new_dentry)
2636 int error;
2637 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2638 const char *old_name;
2640 if (old_dentry->d_inode == new_dentry->d_inode)
2641 return 0;
2643 error = may_delete(old_dir, old_dentry, is_dir);
2644 if (error)
2645 return error;
2647 if (!new_dentry->d_inode)
2648 error = may_create(new_dir, new_dentry);
2649 else
2650 error = may_delete(new_dir, new_dentry, is_dir);
2651 if (error)
2652 return error;
2654 if (!old_dir->i_op->rename)
2655 return -EPERM;
2657 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2659 if (is_dir)
2660 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2661 else
2662 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2663 if (!error)
2664 fsnotify_move(old_dir, new_dir, old_name, is_dir,
2665 new_dentry->d_inode, old_dentry);
2666 fsnotify_oldname_free(old_name);
2668 return error;
2671 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2672 int, newdfd, const char __user *, newname)
2674 struct dentry *old_dir, *new_dir;
2675 struct dentry *old_dentry, *new_dentry;
2676 struct dentry *trap;
2677 struct nameidata oldnd, newnd;
2678 char *from;
2679 char *to;
2680 int error;
2682 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2683 if (error)
2684 goto exit;
2686 error = user_path_parent(newdfd, newname, &newnd, &to);
2687 if (error)
2688 goto exit1;
2690 error = -EXDEV;
2691 if (oldnd.path.mnt != newnd.path.mnt)
2692 goto exit2;
2694 old_dir = oldnd.path.dentry;
2695 error = -EBUSY;
2696 if (oldnd.last_type != LAST_NORM)
2697 goto exit2;
2699 new_dir = newnd.path.dentry;
2700 if (newnd.last_type != LAST_NORM)
2701 goto exit2;
2703 oldnd.flags &= ~LOOKUP_PARENT;
2704 newnd.flags &= ~LOOKUP_PARENT;
2705 newnd.flags |= LOOKUP_RENAME_TARGET;
2707 trap = lock_rename(new_dir, old_dir);
2709 old_dentry = lookup_hash(&oldnd);
2710 error = PTR_ERR(old_dentry);
2711 if (IS_ERR(old_dentry))
2712 goto exit3;
2713 /* source must exist */
2714 error = -ENOENT;
2715 if (!old_dentry->d_inode)
2716 goto exit4;
2717 /* unless the source is a directory trailing slashes give -ENOTDIR */
2718 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2719 error = -ENOTDIR;
2720 if (oldnd.last.name[oldnd.last.len])
2721 goto exit4;
2722 if (newnd.last.name[newnd.last.len])
2723 goto exit4;
2725 /* source should not be ancestor of target */
2726 error = -EINVAL;
2727 if (old_dentry == trap)
2728 goto exit4;
2729 new_dentry = lookup_hash(&newnd);
2730 error = PTR_ERR(new_dentry);
2731 if (IS_ERR(new_dentry))
2732 goto exit4;
2733 /* target should not be an ancestor of source */
2734 error = -ENOTEMPTY;
2735 if (new_dentry == trap)
2736 goto exit5;
2738 error = mnt_want_write(oldnd.path.mnt);
2739 if (error)
2740 goto exit5;
2741 error = security_path_rename(&oldnd.path, old_dentry,
2742 &newnd.path, new_dentry);
2743 if (error)
2744 goto exit6;
2745 error = vfs_rename(old_dir->d_inode, old_dentry,
2746 new_dir->d_inode, new_dentry);
2747 exit6:
2748 mnt_drop_write(oldnd.path.mnt);
2749 exit5:
2750 dput(new_dentry);
2751 exit4:
2752 dput(old_dentry);
2753 exit3:
2754 unlock_rename(new_dir, old_dir);
2755 exit2:
2756 path_put(&newnd.path);
2757 putname(to);
2758 exit1:
2759 path_put(&oldnd.path);
2760 putname(from);
2761 exit:
2762 return error;
2765 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2767 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2770 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2772 int len;
2774 len = PTR_ERR(link);
2775 if (IS_ERR(link))
2776 goto out;
2778 len = strlen(link);
2779 if (len > (unsigned) buflen)
2780 len = buflen;
2781 if (copy_to_user(buffer, link, len))
2782 len = -EFAULT;
2783 out:
2784 return len;
2788 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2789 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2790 * using) it for any given inode is up to filesystem.
2792 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2794 struct nameidata nd;
2795 void *cookie;
2796 int res;
2798 nd.depth = 0;
2799 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2800 if (IS_ERR(cookie))
2801 return PTR_ERR(cookie);
2803 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2804 if (dentry->d_inode->i_op->put_link)
2805 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2806 return res;
2809 int vfs_follow_link(struct nameidata *nd, const char *link)
2811 return __vfs_follow_link(nd, link);
2814 /* get the link contents into pagecache */
2815 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2817 char *kaddr;
2818 struct page *page;
2819 struct address_space *mapping = dentry->d_inode->i_mapping;
2820 page = read_mapping_page(mapping, 0, NULL);
2821 if (IS_ERR(page))
2822 return (char*)page;
2823 *ppage = page;
2824 kaddr = kmap(page);
2825 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2826 return kaddr;
2829 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2831 struct page *page = NULL;
2832 char *s = page_getlink(dentry, &page);
2833 int res = vfs_readlink(dentry,buffer,buflen,s);
2834 if (page) {
2835 kunmap(page);
2836 page_cache_release(page);
2838 return res;
2841 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2843 struct page *page = NULL;
2844 nd_set_link(nd, page_getlink(dentry, &page));
2845 return page;
2848 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2850 struct page *page = cookie;
2852 if (page) {
2853 kunmap(page);
2854 page_cache_release(page);
2859 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2861 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2863 struct address_space *mapping = inode->i_mapping;
2864 struct page *page;
2865 void *fsdata;
2866 int err;
2867 char *kaddr;
2868 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2869 if (nofs)
2870 flags |= AOP_FLAG_NOFS;
2872 retry:
2873 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2874 flags, &page, &fsdata);
2875 if (err)
2876 goto fail;
2878 kaddr = kmap_atomic(page, KM_USER0);
2879 memcpy(kaddr, symname, len-1);
2880 kunmap_atomic(kaddr, KM_USER0);
2882 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2883 page, fsdata);
2884 if (err < 0)
2885 goto fail;
2886 if (err < len-1)
2887 goto retry;
2889 mark_inode_dirty(inode);
2890 return 0;
2891 fail:
2892 return err;
2895 int page_symlink(struct inode *inode, const char *symname, int len)
2897 return __page_symlink(inode, symname, len,
2898 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2901 const struct inode_operations page_symlink_inode_operations = {
2902 .readlink = generic_readlink,
2903 .follow_link = page_follow_link_light,
2904 .put_link = page_put_link,
2907 EXPORT_SYMBOL(user_path_at);
2908 EXPORT_SYMBOL(follow_down);
2909 EXPORT_SYMBOL(follow_up);
2910 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2911 EXPORT_SYMBOL(getname);
2912 EXPORT_SYMBOL(lock_rename);
2913 EXPORT_SYMBOL(lookup_one_len);
2914 EXPORT_SYMBOL(page_follow_link_light);
2915 EXPORT_SYMBOL(page_put_link);
2916 EXPORT_SYMBOL(page_readlink);
2917 EXPORT_SYMBOL(__page_symlink);
2918 EXPORT_SYMBOL(page_symlink);
2919 EXPORT_SYMBOL(page_symlink_inode_operations);
2920 EXPORT_SYMBOL(path_lookup);
2921 EXPORT_SYMBOL(kern_path);
2922 EXPORT_SYMBOL(vfs_path_lookup);
2923 EXPORT_SYMBOL(inode_permission);
2924 EXPORT_SYMBOL(file_permission);
2925 EXPORT_SYMBOL(unlock_rename);
2926 EXPORT_SYMBOL(vfs_create);
2927 EXPORT_SYMBOL(vfs_follow_link);
2928 EXPORT_SYMBOL(vfs_link);
2929 EXPORT_SYMBOL(vfs_mkdir);
2930 EXPORT_SYMBOL(vfs_mknod);
2931 EXPORT_SYMBOL(generic_permission);
2932 EXPORT_SYMBOL(vfs_readlink);
2933 EXPORT_SYMBOL(vfs_rename);
2934 EXPORT_SYMBOL(vfs_rmdir);
2935 EXPORT_SYMBOL(vfs_symlink);
2936 EXPORT_SYMBOL(vfs_unlink);
2937 EXPORT_SYMBOL(dentry_unhash);
2938 EXPORT_SYMBOL(generic_readlink);