mm-only debug patch...
[mmotm.git] / fs / namei.c
blobb5196c53717729b341df5a96e9c250af40694379
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/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <asm/uaccess.h>
38 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 static int __link_path_walk(const char *name, struct nameidata *nd);
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static int do_getname(const char __user *filename, char *page)
122 int retval;
123 unsigned long len = PATH_MAX;
125 if (!segment_eq(get_fs(), KERNEL_DS)) {
126 if ((unsigned long) filename >= TASK_SIZE)
127 return -EFAULT;
128 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
129 len = TASK_SIZE - (unsigned long) filename;
132 retval = strncpy_from_user(page, filename, len);
133 if (retval > 0) {
134 if (retval < len)
135 return 0;
136 return -ENAMETOOLONG;
137 } else if (!retval)
138 retval = -ENOENT;
139 return retval;
142 char * getname(const char __user * filename)
144 char *tmp, *result;
146 result = ERR_PTR(-ENOMEM);
147 tmp = __getname();
148 if (tmp) {
149 int retval = do_getname(filename, tmp);
151 result = tmp;
152 if (retval < 0) {
153 __putname(tmp);
154 result = ERR_PTR(retval);
157 audit_getname(result);
158 return result;
161 #ifdef CONFIG_AUDITSYSCALL
162 void putname(const char *name)
164 if (unlikely(!audit_dummy_context()))
165 audit_putname(name);
166 else
167 __putname(name);
169 EXPORT_SYMBOL(putname);
170 #endif
173 * This does basic POSIX ACL permission checking
175 static int acl_permission_check(struct inode *inode, int mask,
176 int (*check_acl)(struct inode *inode, int mask))
178 umode_t mode = inode->i_mode;
180 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
182 if (current_fsuid() == inode->i_uid)
183 mode >>= 6;
184 else {
185 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
188 return error;
191 if (in_group_p(inode->i_gid))
192 mode >>= 3;
196 * If the DACs are ok we don't need any capability check.
198 if ((mask & ~mode) == 0)
199 return 0;
200 return -EACCES;
204 * generic_permission - check for access rights on a Posix-like filesystem
205 * @inode: inode to check access rights for
206 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
207 * @check_acl: optional callback to check for Posix ACLs
209 * Used to check for read/write/execute permissions on a file.
210 * We use "fsuid" for this, letting us set arbitrary permissions
211 * for filesystem access without changing the "normal" uids which
212 * are used for other things..
214 int generic_permission(struct inode *inode, int mask,
215 int (*check_acl)(struct inode *inode, int mask))
217 int ret;
220 * Do the basic POSIX ACL permission checks.
222 ret = acl_permission_check(inode, mask, check_acl);
223 if (ret != -EACCES)
224 return ret;
227 * Read/write DACs are always overridable.
228 * Executable DACs are overridable if at least one exec bit is set.
230 if (!(mask & MAY_EXEC) || execute_ok(inode))
231 if (capable(CAP_DAC_OVERRIDE))
232 return 0;
235 * Searching includes executable on directories, else just read.
237 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
238 if (capable(CAP_DAC_READ_SEARCH))
239 return 0;
241 return -EACCES;
245 * inode_permission - check for access rights to a given inode
246 * @inode: inode to check permission on
247 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
249 * Used to check for read/write/execute permissions on an inode.
250 * We use "fsuid" for this, letting us set arbitrary permissions
251 * for filesystem access without changing the "normal" uids which
252 * are used for other things.
254 int inode_permission(struct inode *inode, int mask)
256 int retval;
258 if (mask & MAY_WRITE) {
259 umode_t mode = inode->i_mode;
262 * Nobody gets write access to a read-only fs.
264 if (IS_RDONLY(inode) &&
265 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
266 return -EROFS;
269 * Nobody gets write access to an immutable file.
271 if (IS_IMMUTABLE(inode))
272 return -EACCES;
275 if (inode->i_op->permission)
276 retval = inode->i_op->permission(inode, mask);
277 else
278 retval = generic_permission(inode, mask, inode->i_op->check_acl);
280 if (retval)
281 return retval;
283 retval = devcgroup_inode_permission(inode, mask);
284 if (retval)
285 return retval;
287 return security_inode_permission(inode,
288 mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND));
292 * file_permission - check for additional access rights to a given file
293 * @file: file to check access rights for
294 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
296 * Used to check for read/write/execute permissions on an already opened
297 * file.
299 * Note:
300 * Do not use this function in new code. All access checks should
301 * be done using inode_permission().
303 int file_permission(struct file *file, int mask)
305 return inode_permission(file->f_path.dentry->d_inode, mask);
309 * get_write_access() gets write permission for a file.
310 * put_write_access() releases this write permission.
311 * This is used for regular files.
312 * We cannot support write (and maybe mmap read-write shared) accesses and
313 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
314 * can have the following values:
315 * 0: no writers, no VM_DENYWRITE mappings
316 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
317 * > 0: (i_writecount) users are writing to the file.
319 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
320 * except for the cases where we don't hold i_writecount yet. Then we need to
321 * use {get,deny}_write_access() - these functions check the sign and refuse
322 * to do the change if sign is wrong. Exclusion between them is provided by
323 * the inode->i_lock spinlock.
326 int get_write_access(struct inode * inode)
328 spin_lock(&inode->i_lock);
329 if (atomic_read(&inode->i_writecount) < 0) {
330 spin_unlock(&inode->i_lock);
331 return -ETXTBSY;
333 atomic_inc(&inode->i_writecount);
334 spin_unlock(&inode->i_lock);
336 return 0;
339 int deny_write_access(struct file * file)
341 struct inode *inode = file->f_path.dentry->d_inode;
343 spin_lock(&inode->i_lock);
344 if (atomic_read(&inode->i_writecount) > 0) {
345 spin_unlock(&inode->i_lock);
346 return -ETXTBSY;
348 atomic_dec(&inode->i_writecount);
349 spin_unlock(&inode->i_lock);
351 return 0;
355 * path_get - get a reference to a path
356 * @path: path to get the reference to
358 * Given a path increment the reference count to the dentry and the vfsmount.
360 void path_get(struct path *path)
362 mntget(path->mnt);
363 dget(path->dentry);
365 EXPORT_SYMBOL(path_get);
368 * path_put - put a reference to a path
369 * @path: path to put the reference to
371 * Given a path decrement the reference count to the dentry and the vfsmount.
373 void path_put(struct path *path)
375 dput(path->dentry);
376 mntput(path->mnt);
378 EXPORT_SYMBOL(path_put);
381 * release_open_intent - free up open intent resources
382 * @nd: pointer to nameidata
384 void release_open_intent(struct nameidata *nd)
386 if (nd->intent.open.file->f_path.dentry == NULL)
387 put_filp(nd->intent.open.file);
388 else
389 fput(nd->intent.open.file);
392 static inline struct dentry *
393 do_revalidate(struct dentry *dentry, struct nameidata *nd)
395 int status = dentry->d_op->d_revalidate(dentry, nd);
396 if (unlikely(status <= 0)) {
398 * The dentry failed validation.
399 * If d_revalidate returned 0 attempt to invalidate
400 * the dentry otherwise d_revalidate is asking us
401 * to return a fail status.
403 if (!status) {
404 if (!d_invalidate(dentry)) {
405 dput(dentry);
406 dentry = NULL;
408 } else {
409 dput(dentry);
410 dentry = ERR_PTR(status);
413 return dentry;
417 * Internal lookup() using the new generic dcache.
418 * SMP-safe
420 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
422 struct dentry * dentry = __d_lookup(parent, name);
424 /* lockess __d_lookup may fail due to concurrent d_move()
425 * in some unrelated directory, so try with d_lookup
427 if (!dentry)
428 dentry = d_lookup(parent, name);
430 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
431 dentry = do_revalidate(dentry, nd);
433 return dentry;
437 * Short-cut version of permission(), for calling by
438 * path_walk(), when dcache lock is held. Combines parts
439 * of permission() and generic_permission(), and tests ONLY for
440 * MAY_EXEC permission.
442 * If appropriate, check DAC only. If not appropriate, or
443 * short-cut DAC fails, then call permission() to do more
444 * complete permission check.
446 static int exec_permission_lite(struct inode *inode)
448 int ret;
450 if (inode->i_op->permission) {
451 ret = inode->i_op->permission(inode, MAY_EXEC);
452 if (!ret)
453 goto ok;
454 return ret;
456 ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl);
457 if (!ret)
458 goto ok;
460 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
461 goto ok;
463 return ret;
465 return security_inode_permission(inode, MAY_EXEC);
469 * This is called when everything else fails, and we actually have
470 * to go to the low-level filesystem to find out what we should do..
472 * We get the directory mutex, and after getting that we also
473 * make sure that nobody added the entry to the dcache in the meantime..
474 * SMP-safe
476 static struct dentry *real_lookup(struct dentry *parent, struct qstr *name,
477 struct nameidata *nd)
479 struct dentry *result, *dentry;
480 struct inode *dir = parent->d_inode;
482 mutex_lock(&dir->i_mutex);
484 * First re-do the cached lookup just in case it was created
485 * while we waited for the directory mutex.
487 * FIXME! This could use version numbering or similar to
488 * avoid unnecessary cache lookups.
490 * The "dcache_lock" is purely to protect the RCU list walker
491 * from concurrent renames at this point (we mustn't get false
492 * negatives from the RCU list walk here, unlike the optimistic
493 * fast walk).
495 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
497 result = d_lookup(parent, name);
498 if (result) {
500 * The cache was re-populated while we waited on the
501 * mutex. We need to revalidate, this time while
502 * holding i_mutex (to avoid another race).
504 if (result->d_op && result->d_op->d_revalidate) {
505 result = do_revalidate(result, nd);
506 if (result)
507 goto out_unlock;
509 * The dentry was left behind invalid. Just
510 * do the lookup.
512 } else {
513 goto out_unlock;
517 /* Don't create child dentry for a dead directory. */
518 result = ERR_PTR(-ENOENT);
519 if (IS_DEADDIR(dir))
520 goto out_unlock;
522 dentry = d_alloc(parent, name);
523 result = ERR_PTR(-ENOMEM);
524 if (dentry) {
525 result = dir->i_op->lookup(dir, dentry, nd);
526 if (result)
527 dput(dentry);
528 else
529 result = dentry;
531 out_unlock:
532 mutex_unlock(&dir->i_mutex);
533 return result;
537 * Wrapper to retry pathname resolution whenever the underlying
538 * file system returns an ESTALE.
540 * Retry the whole path once, forcing real lookup requests
541 * instead of relying on the dcache.
543 static __always_inline int link_path_walk(const char *name, struct nameidata *nd)
545 struct path save = nd->path;
546 int result;
548 /* make sure the stuff we saved doesn't go away */
549 path_get(&save);
551 result = __link_path_walk(name, nd);
552 if (result == -ESTALE) {
553 /* nd->path had been dropped */
554 nd->path = save;
555 path_get(&nd->path);
556 nd->flags |= LOOKUP_REVAL;
557 result = __link_path_walk(name, nd);
560 path_put(&save);
562 return result;
565 static __always_inline void set_root(struct nameidata *nd)
567 if (!nd->root.mnt) {
568 struct fs_struct *fs = current->fs;
569 read_lock(&fs->lock);
570 nd->root = fs->root;
571 path_get(&nd->root);
572 read_unlock(&fs->lock);
576 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
578 int res = 0;
579 char *name;
580 if (IS_ERR(link))
581 goto fail;
583 if (*link == '/') {
584 set_root(nd);
585 path_put(&nd->path);
586 nd->path = nd->root;
587 path_get(&nd->root);
590 res = link_path_walk(link, nd);
591 if (nd->depth || res || nd->last_type!=LAST_NORM)
592 return res;
594 * If it is an iterative symlinks resolution in open_namei() we
595 * have to copy the last component. And all that crap because of
596 * bloody create() on broken symlinks. Furrfu...
598 name = __getname();
599 if (unlikely(!name)) {
600 path_put(&nd->path);
601 return -ENOMEM;
603 strcpy(name, nd->last.name);
604 nd->last.name = name;
605 return 0;
606 fail:
607 path_put(&nd->path);
608 return PTR_ERR(link);
611 static void path_put_conditional(struct path *path, struct nameidata *nd)
613 dput(path->dentry);
614 if (path->mnt != nd->path.mnt)
615 mntput(path->mnt);
618 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
620 dput(nd->path.dentry);
621 if (nd->path.mnt != path->mnt)
622 mntput(nd->path.mnt);
623 nd->path.mnt = path->mnt;
624 nd->path.dentry = path->dentry;
627 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
629 int error;
630 void *cookie;
631 struct dentry *dentry = path->dentry;
633 touch_atime(path->mnt, dentry);
634 nd_set_link(nd, NULL);
636 if (path->mnt != nd->path.mnt) {
637 path_to_nameidata(path, nd);
638 dget(dentry);
640 mntget(path->mnt);
641 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
642 error = PTR_ERR(cookie);
643 if (!IS_ERR(cookie)) {
644 char *s = nd_get_link(nd);
645 error = 0;
646 if (s)
647 error = __vfs_follow_link(nd, s);
648 if (dentry->d_inode->i_op->put_link)
649 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
651 path_put(path);
653 return error;
657 * This limits recursive symlink follows to 8, while
658 * limiting consecutive symlinks to 40.
660 * Without that kind of total limit, nasty chains of consecutive
661 * symlinks can cause almost arbitrarily long lookups.
663 static inline int do_follow_link(struct path *path, struct nameidata *nd)
665 int err = -ELOOP;
666 if (current->link_count >= MAX_NESTED_LINKS)
667 goto loop;
668 if (current->total_link_count >= 40)
669 goto loop;
670 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
671 cond_resched();
672 err = security_inode_follow_link(path->dentry, nd);
673 if (err)
674 goto loop;
675 current->link_count++;
676 current->total_link_count++;
677 nd->depth++;
678 err = __do_follow_link(path, nd);
679 current->link_count--;
680 nd->depth--;
681 return err;
682 loop:
683 path_put_conditional(path, nd);
684 path_put(&nd->path);
685 return err;
688 int follow_up(struct path *path)
690 struct vfsmount *parent;
691 struct dentry *mountpoint;
692 spin_lock(&vfsmount_lock);
693 parent = path->mnt->mnt_parent;
694 if (parent == path->mnt) {
695 spin_unlock(&vfsmount_lock);
696 return 0;
698 mntget(parent);
699 mountpoint = dget(path->mnt->mnt_mountpoint);
700 spin_unlock(&vfsmount_lock);
701 dput(path->dentry);
702 path->dentry = mountpoint;
703 mntput(path->mnt);
704 path->mnt = parent;
705 return 1;
708 /* no need for dcache_lock, as serialization is taken care in
709 * namespace.c
711 static int __follow_mount(struct path *path)
713 int res = 0;
714 while (d_mountpoint(path->dentry)) {
715 struct vfsmount *mounted = lookup_mnt(path);
716 if (!mounted)
717 break;
718 dput(path->dentry);
719 if (res)
720 mntput(path->mnt);
721 path->mnt = mounted;
722 path->dentry = dget(mounted->mnt_root);
723 res = 1;
725 return res;
728 static void follow_mount(struct path *path)
730 while (d_mountpoint(path->dentry)) {
731 struct vfsmount *mounted = lookup_mnt(path);
732 if (!mounted)
733 break;
734 dput(path->dentry);
735 mntput(path->mnt);
736 path->mnt = mounted;
737 path->dentry = dget(mounted->mnt_root);
741 /* no need for dcache_lock, as serialization is taken care in
742 * namespace.c
744 int follow_down(struct path *path)
746 struct vfsmount *mounted;
748 mounted = lookup_mnt(path);
749 if (mounted) {
750 dput(path->dentry);
751 mntput(path->mnt);
752 path->mnt = mounted;
753 path->dentry = dget(mounted->mnt_root);
754 return 1;
756 return 0;
759 static __always_inline void follow_dotdot(struct nameidata *nd)
761 set_root(nd);
763 while(1) {
764 struct vfsmount *parent;
765 struct dentry *old = nd->path.dentry;
767 if (nd->path.dentry == nd->root.dentry &&
768 nd->path.mnt == nd->root.mnt) {
769 break;
771 spin_lock(&dcache_lock);
772 if (nd->path.dentry != nd->path.mnt->mnt_root) {
773 nd->path.dentry = dget(nd->path.dentry->d_parent);
774 spin_unlock(&dcache_lock);
775 dput(old);
776 break;
778 spin_unlock(&dcache_lock);
779 spin_lock(&vfsmount_lock);
780 parent = nd->path.mnt->mnt_parent;
781 if (parent == nd->path.mnt) {
782 spin_unlock(&vfsmount_lock);
783 break;
785 mntget(parent);
786 nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
787 spin_unlock(&vfsmount_lock);
788 dput(old);
789 mntput(nd->path.mnt);
790 nd->path.mnt = parent;
792 follow_mount(&nd->path);
796 * It's more convoluted than I'd like it to be, but... it's still fairly
797 * small and for now I'd prefer to have fast path as straight as possible.
798 * It _is_ time-critical.
800 static int do_lookup(struct nameidata *nd, struct qstr *name,
801 struct path *path)
803 struct vfsmount *mnt = nd->path.mnt;
804 struct dentry *dentry = __d_lookup(nd->path.dentry, name);
806 if (!dentry)
807 goto need_lookup;
808 if (dentry->d_op && dentry->d_op->d_revalidate)
809 goto need_revalidate;
810 done:
811 path->mnt = mnt;
812 path->dentry = dentry;
813 __follow_mount(path);
814 return 0;
816 need_lookup:
817 dentry = real_lookup(nd->path.dentry, name, nd);
818 if (IS_ERR(dentry))
819 goto fail;
820 goto done;
822 need_revalidate:
823 dentry = do_revalidate(dentry, nd);
824 if (!dentry)
825 goto need_lookup;
826 if (IS_ERR(dentry))
827 goto fail;
828 goto done;
830 fail:
831 return PTR_ERR(dentry);
835 * Name resolution.
836 * This is the basic name resolution function, turning a pathname into
837 * the final dentry. We expect 'base' to be positive and a directory.
839 * Returns 0 and nd will have valid dentry and mnt on success.
840 * Returns error and drops reference to input namei data on failure.
842 static int __link_path_walk(const char *name, struct nameidata *nd)
844 struct path next;
845 struct inode *inode;
846 int err;
847 unsigned int lookup_flags = nd->flags;
849 while (*name=='/')
850 name++;
851 if (!*name)
852 goto return_reval;
854 inode = nd->path.dentry->d_inode;
855 if (nd->depth)
856 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
858 /* At this point we know we have a real path component. */
859 for(;;) {
860 unsigned long hash;
861 struct qstr this;
862 unsigned int c;
864 nd->flags |= LOOKUP_CONTINUE;
865 err = exec_permission_lite(inode);
866 if (err)
867 break;
869 this.name = name;
870 c = *(const unsigned char *)name;
872 hash = init_name_hash();
873 do {
874 name++;
875 hash = partial_name_hash(c, hash);
876 c = *(const unsigned char *)name;
877 } while (c && (c != '/'));
878 this.len = name - (const char *) this.name;
879 this.hash = end_name_hash(hash);
881 /* remove trailing slashes? */
882 if (!c)
883 goto last_component;
884 while (*++name == '/');
885 if (!*name)
886 goto last_with_slashes;
889 * "." and ".." are special - ".." especially so because it has
890 * to be able to know about the current root directory and
891 * parent relationships.
893 if (this.name[0] == '.') switch (this.len) {
894 default:
895 break;
896 case 2:
897 if (this.name[1] != '.')
898 break;
899 follow_dotdot(nd);
900 inode = nd->path.dentry->d_inode;
901 /* fallthrough */
902 case 1:
903 continue;
906 * See if the low-level filesystem might want
907 * to use its own hash..
909 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
910 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
911 &this);
912 if (err < 0)
913 break;
915 /* This does the actual lookups.. */
916 err = do_lookup(nd, &this, &next);
917 if (err)
918 break;
920 err = -ENOENT;
921 inode = next.dentry->d_inode;
922 if (!inode)
923 goto out_dput;
925 if (inode->i_op->follow_link) {
926 err = do_follow_link(&next, nd);
927 if (err)
928 goto return_err;
929 err = -ENOENT;
930 inode = nd->path.dentry->d_inode;
931 if (!inode)
932 break;
933 } else
934 path_to_nameidata(&next, nd);
935 err = -ENOTDIR;
936 if (!inode->i_op->lookup)
937 break;
938 continue;
939 /* here ends the main loop */
941 last_with_slashes:
942 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
943 last_component:
944 /* Clear LOOKUP_CONTINUE iff it was previously unset */
945 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
946 if (lookup_flags & LOOKUP_PARENT)
947 goto lookup_parent;
948 if (this.name[0] == '.') switch (this.len) {
949 default:
950 break;
951 case 2:
952 if (this.name[1] != '.')
953 break;
954 follow_dotdot(nd);
955 inode = nd->path.dentry->d_inode;
956 /* fallthrough */
957 case 1:
958 goto return_reval;
960 if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
961 err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
962 &this);
963 if (err < 0)
964 break;
966 err = do_lookup(nd, &this, &next);
967 if (err)
968 break;
969 inode = next.dentry->d_inode;
970 if ((lookup_flags & LOOKUP_FOLLOW)
971 && inode && inode->i_op->follow_link) {
972 err = do_follow_link(&next, nd);
973 if (err)
974 goto return_err;
975 inode = nd->path.dentry->d_inode;
976 } else
977 path_to_nameidata(&next, nd);
978 err = -ENOENT;
979 if (!inode)
980 break;
981 if (lookup_flags & LOOKUP_DIRECTORY) {
982 err = -ENOTDIR;
983 if (!inode->i_op->lookup)
984 break;
986 goto return_base;
987 lookup_parent:
988 nd->last = this;
989 nd->last_type = LAST_NORM;
990 if (this.name[0] != '.')
991 goto return_base;
992 if (this.len == 1)
993 nd->last_type = LAST_DOT;
994 else if (this.len == 2 && this.name[1] == '.')
995 nd->last_type = LAST_DOTDOT;
996 else
997 goto return_base;
998 return_reval:
1000 * We bypassed the ordinary revalidation routines.
1001 * We may need to check the cached dentry for staleness.
1003 if (nd->path.dentry && nd->path.dentry->d_sb &&
1004 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1005 err = -ESTALE;
1006 /* Note: we do not d_invalidate() */
1007 if (!nd->path.dentry->d_op->d_revalidate(
1008 nd->path.dentry, nd))
1009 break;
1011 return_base:
1012 return 0;
1013 out_dput:
1014 path_put_conditional(&next, nd);
1015 break;
1017 path_put(&nd->path);
1018 return_err:
1019 return err;
1022 static int path_walk(const char *name, struct nameidata *nd)
1024 current->total_link_count = 0;
1025 return link_path_walk(name, nd);
1028 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1030 int retval = 0;
1031 int fput_needed;
1032 struct file *file;
1034 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1035 nd->flags = flags;
1036 nd->depth = 0;
1037 nd->root.mnt = NULL;
1039 if (*name=='/') {
1040 set_root(nd);
1041 nd->path = nd->root;
1042 path_get(&nd->root);
1043 } else if (dfd == AT_FDCWD) {
1044 struct fs_struct *fs = current->fs;
1045 read_lock(&fs->lock);
1046 nd->path = fs->pwd;
1047 path_get(&fs->pwd);
1048 read_unlock(&fs->lock);
1049 } else {
1050 struct dentry *dentry;
1052 file = fget_light(dfd, &fput_needed);
1053 retval = -EBADF;
1054 if (!file)
1055 goto out_fail;
1057 dentry = file->f_path.dentry;
1059 retval = -ENOTDIR;
1060 if (!S_ISDIR(dentry->d_inode->i_mode))
1061 goto fput_fail;
1063 retval = file_permission(file, MAY_EXEC);
1064 if (retval)
1065 goto fput_fail;
1067 nd->path = file->f_path;
1068 path_get(&file->f_path);
1070 fput_light(file, fput_needed);
1072 return 0;
1074 fput_fail:
1075 fput_light(file, fput_needed);
1076 out_fail:
1077 return retval;
1080 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1081 static int do_path_lookup(int dfd, const char *name,
1082 unsigned int flags, struct nameidata *nd)
1084 int retval = path_init(dfd, name, flags, nd);
1085 if (!retval)
1086 retval = path_walk(name, nd);
1087 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1088 nd->path.dentry->d_inode))
1089 audit_inode(name, nd->path.dentry);
1090 if (nd->root.mnt) {
1091 path_put(&nd->root);
1092 nd->root.mnt = NULL;
1094 return retval;
1097 int path_lookup(const char *name, unsigned int flags,
1098 struct nameidata *nd)
1100 return do_path_lookup(AT_FDCWD, name, flags, nd);
1103 int kern_path(const char *name, unsigned int flags, struct path *path)
1105 struct nameidata nd;
1106 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1107 if (!res)
1108 *path = nd.path;
1109 return res;
1113 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1114 * @dentry: pointer to dentry of the base directory
1115 * @mnt: pointer to vfs mount of the base directory
1116 * @name: pointer to file name
1117 * @flags: lookup flags
1118 * @nd: pointer to nameidata
1120 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1121 const char *name, unsigned int flags,
1122 struct nameidata *nd)
1124 int retval;
1126 /* same as do_path_lookup */
1127 nd->last_type = LAST_ROOT;
1128 nd->flags = flags;
1129 nd->depth = 0;
1131 nd->path.dentry = dentry;
1132 nd->path.mnt = mnt;
1133 path_get(&nd->path);
1134 nd->root = nd->path;
1135 path_get(&nd->root);
1137 retval = path_walk(name, nd);
1138 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1139 nd->path.dentry->d_inode))
1140 audit_inode(name, nd->path.dentry);
1142 path_put(&nd->root);
1143 nd->root.mnt = NULL;
1145 return retval;
1149 * path_lookup_open - lookup a file path with open intent
1150 * @dfd: the directory to use as base, or AT_FDCWD
1151 * @name: pointer to file name
1152 * @lookup_flags: lookup intent flags
1153 * @nd: pointer to nameidata
1154 * @open_flags: open intent flags
1156 static int path_lookup_open(int dfd, const char *name,
1157 unsigned int lookup_flags, struct nameidata *nd, int open_flags)
1159 struct file *filp = get_empty_filp();
1160 int err;
1162 if (filp == NULL)
1163 return -ENFILE;
1164 nd->intent.open.file = filp;
1165 nd->intent.open.flags = open_flags;
1166 nd->intent.open.create_mode = 0;
1167 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1168 if (IS_ERR(nd->intent.open.file)) {
1169 if (err == 0) {
1170 err = PTR_ERR(nd->intent.open.file);
1171 path_put(&nd->path);
1173 } else if (err != 0)
1174 release_open_intent(nd);
1175 return err;
1178 static struct dentry *__lookup_hash(struct qstr *name,
1179 struct dentry *base, struct nameidata *nd)
1181 struct dentry *dentry;
1182 struct inode *inode;
1183 int err;
1185 inode = base->d_inode;
1188 * See if the low-level filesystem might want
1189 * to use its own hash..
1191 if (base->d_op && base->d_op->d_hash) {
1192 err = base->d_op->d_hash(base, name);
1193 dentry = ERR_PTR(err);
1194 if (err < 0)
1195 goto out;
1198 dentry = cached_lookup(base, name, nd);
1199 if (!dentry) {
1200 struct dentry *new;
1202 /* Don't create child dentry for a dead directory. */
1203 dentry = ERR_PTR(-ENOENT);
1204 if (IS_DEADDIR(inode))
1205 goto out;
1207 new = d_alloc(base, name);
1208 dentry = ERR_PTR(-ENOMEM);
1209 if (!new)
1210 goto out;
1211 dentry = inode->i_op->lookup(inode, new, nd);
1212 if (!dentry)
1213 dentry = new;
1214 else
1215 dput(new);
1217 out:
1218 return dentry;
1222 * Restricted form of lookup. Doesn't follow links, single-component only,
1223 * needs parent already locked. Doesn't follow mounts.
1224 * SMP-safe.
1226 static struct dentry *lookup_hash(struct nameidata *nd)
1228 int err;
1230 err = inode_permission(nd->path.dentry->d_inode, MAY_EXEC);
1231 if (err)
1232 return ERR_PTR(err);
1233 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1236 static int __lookup_one_len(const char *name, struct qstr *this,
1237 struct dentry *base, int len)
1239 unsigned long hash;
1240 unsigned int c;
1242 this->name = name;
1243 this->len = len;
1244 if (!len)
1245 return -EACCES;
1247 hash = init_name_hash();
1248 while (len--) {
1249 c = *(const unsigned char *)name++;
1250 if (c == '/' || c == '\0')
1251 return -EACCES;
1252 hash = partial_name_hash(c, hash);
1254 this->hash = end_name_hash(hash);
1255 return 0;
1259 * lookup_one_len - filesystem helper to lookup single pathname component
1260 * @name: pathname component to lookup
1261 * @base: base directory to lookup from
1262 * @len: maximum length @len should be interpreted to
1264 * Note that this routine is purely a helper for filesystem usage and should
1265 * not be called by generic code. Also note that by using this function the
1266 * nameidata argument is passed to the filesystem methods and a filesystem
1267 * using this helper needs to be prepared for that.
1269 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1271 int err;
1272 struct qstr this;
1274 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1276 err = __lookup_one_len(name, &this, base, len);
1277 if (err)
1278 return ERR_PTR(err);
1280 err = inode_permission(base->d_inode, MAY_EXEC);
1281 if (err)
1282 return ERR_PTR(err);
1283 return __lookup_hash(&this, base, NULL);
1287 * lookup_one_noperm - bad hack for sysfs
1288 * @name: pathname component to lookup
1289 * @base: base directory to lookup from
1291 * This is a variant of lookup_one_len that doesn't perform any permission
1292 * checks. It's a horrible hack to work around the braindead sysfs
1293 * architecture and should not be used anywhere else.
1295 * DON'T USE THIS FUNCTION EVER, thanks.
1297 struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
1299 int err;
1300 struct qstr this;
1302 err = __lookup_one_len(name, &this, base, strlen(name));
1303 if (err)
1304 return ERR_PTR(err);
1305 return __lookup_hash(&this, base, NULL);
1308 int user_path_at(int dfd, const char __user *name, unsigned flags,
1309 struct path *path)
1311 struct nameidata nd;
1312 char *tmp = getname(name);
1313 int err = PTR_ERR(tmp);
1314 if (!IS_ERR(tmp)) {
1316 BUG_ON(flags & LOOKUP_PARENT);
1318 err = do_path_lookup(dfd, tmp, flags, &nd);
1319 putname(tmp);
1320 if (!err)
1321 *path = nd.path;
1323 return err;
1326 static int user_path_parent(int dfd, const char __user *path,
1327 struct nameidata *nd, char **name)
1329 char *s = getname(path);
1330 int error;
1332 if (IS_ERR(s))
1333 return PTR_ERR(s);
1335 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1336 if (error)
1337 putname(s);
1338 else
1339 *name = s;
1341 return error;
1345 * It's inline, so penalty for filesystems that don't use sticky bit is
1346 * minimal.
1348 static inline int check_sticky(struct inode *dir, struct inode *inode)
1350 uid_t fsuid = current_fsuid();
1352 if (!(dir->i_mode & S_ISVTX))
1353 return 0;
1354 if (inode->i_uid == fsuid)
1355 return 0;
1356 if (dir->i_uid == fsuid)
1357 return 0;
1358 return !capable(CAP_FOWNER);
1362 * Check whether we can remove a link victim from directory dir, check
1363 * whether the type of victim is right.
1364 * 1. We can't do it if dir is read-only (done in permission())
1365 * 2. We should have write and exec permissions on dir
1366 * 3. We can't remove anything from append-only dir
1367 * 4. We can't do anything with immutable dir (done in permission())
1368 * 5. If the sticky bit on dir is set we should either
1369 * a. be owner of dir, or
1370 * b. be owner of victim, or
1371 * c. have CAP_FOWNER capability
1372 * 6. If the victim is append-only or immutable we can't do antyhing with
1373 * links pointing to it.
1374 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1375 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1376 * 9. We can't remove a root or mountpoint.
1377 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1378 * nfs_async_unlink().
1380 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1382 int error;
1384 if (!victim->d_inode)
1385 return -ENOENT;
1387 BUG_ON(victim->d_parent->d_inode != dir);
1388 audit_inode_child(victim->d_name.name, victim, dir);
1390 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1391 if (error)
1392 return error;
1393 if (IS_APPEND(dir))
1394 return -EPERM;
1395 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1396 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1397 return -EPERM;
1398 if (isdir) {
1399 if (!S_ISDIR(victim->d_inode->i_mode))
1400 return -ENOTDIR;
1401 if (IS_ROOT(victim))
1402 return -EBUSY;
1403 } else if (S_ISDIR(victim->d_inode->i_mode))
1404 return -EISDIR;
1405 if (IS_DEADDIR(dir))
1406 return -ENOENT;
1407 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1408 return -EBUSY;
1409 return 0;
1412 /* Check whether we can create an object with dentry child in directory
1413 * dir.
1414 * 1. We can't do it if child already exists (open has special treatment for
1415 * this case, but since we are inlined it's OK)
1416 * 2. We can't do it if dir is read-only (done in permission())
1417 * 3. We should have write and exec permissions on dir
1418 * 4. We can't do it if dir is immutable (done in permission())
1420 static inline int may_create(struct inode *dir, struct dentry *child)
1422 if (child->d_inode)
1423 return -EEXIST;
1424 if (IS_DEADDIR(dir))
1425 return -ENOENT;
1426 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1430 * O_DIRECTORY translates into forcing a directory lookup.
1432 static inline int lookup_flags(unsigned int f)
1434 unsigned long retval = LOOKUP_FOLLOW;
1436 if (f & O_NOFOLLOW)
1437 retval &= ~LOOKUP_FOLLOW;
1439 if (f & O_DIRECTORY)
1440 retval |= LOOKUP_DIRECTORY;
1442 return retval;
1446 * p1 and p2 should be directories on the same fs.
1448 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1450 struct dentry *p;
1452 if (p1 == p2) {
1453 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1454 return NULL;
1457 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1459 p = d_ancestor(p2, p1);
1460 if (p) {
1461 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1462 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1463 return p;
1466 p = d_ancestor(p1, p2);
1467 if (p) {
1468 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1469 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1470 return p;
1473 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1474 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1475 return NULL;
1478 void unlock_rename(struct dentry *p1, struct dentry *p2)
1480 mutex_unlock(&p1->d_inode->i_mutex);
1481 if (p1 != p2) {
1482 mutex_unlock(&p2->d_inode->i_mutex);
1483 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1487 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1488 struct nameidata *nd)
1490 int error = may_create(dir, dentry);
1492 if (error)
1493 return error;
1495 if (!dir->i_op->create)
1496 return -EACCES; /* shouldn't it be ENOSYS? */
1497 mode &= S_IALLUGO;
1498 mode |= S_IFREG;
1499 error = security_inode_create(dir, dentry, mode);
1500 if (error)
1501 return error;
1502 vfs_dq_init(dir);
1503 error = dir->i_op->create(dir, dentry, mode, nd);
1504 if (!error)
1505 fsnotify_create(dir, dentry);
1506 return error;
1509 int may_open(struct path *path, int acc_mode, int flag)
1511 struct dentry *dentry = path->dentry;
1512 struct inode *inode = dentry->d_inode;
1513 int error;
1515 if (!inode)
1516 return -ENOENT;
1518 switch (inode->i_mode & S_IFMT) {
1519 case S_IFLNK:
1520 return -ELOOP;
1521 case S_IFDIR:
1522 if (acc_mode & MAY_WRITE)
1523 return -EISDIR;
1524 break;
1525 case S_IFBLK:
1526 case S_IFCHR:
1527 if (path->mnt->mnt_flags & MNT_NODEV)
1528 return -EACCES;
1529 /*FALLTHRU*/
1530 case S_IFIFO:
1531 case S_IFSOCK:
1532 flag &= ~O_TRUNC;
1533 break;
1536 error = inode_permission(inode, acc_mode);
1537 if (error)
1538 return error;
1540 error = ima_path_check(path, acc_mode ?
1541 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1542 ACC_MODE(flag) & (MAY_READ | MAY_WRITE),
1543 IMA_COUNT_UPDATE);
1545 if (error)
1546 return error;
1548 * An append-only file must be opened in append mode for writing.
1550 if (IS_APPEND(inode)) {
1551 error = -EPERM;
1552 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1553 goto err_out;
1554 if (flag & O_TRUNC)
1555 goto err_out;
1558 /* O_NOATIME can only be set by the owner or superuser */
1559 if (flag & O_NOATIME)
1560 if (!is_owner_or_cap(inode)) {
1561 error = -EPERM;
1562 goto err_out;
1566 * Ensure there are no outstanding leases on the file.
1568 error = break_lease(inode, flag);
1569 if (error)
1570 goto err_out;
1572 if (flag & O_TRUNC) {
1573 error = get_write_access(inode);
1574 if (error)
1575 goto err_out;
1578 * Refuse to truncate files with mandatory locks held on them.
1580 error = locks_verify_locked(inode);
1581 if (!error)
1582 error = security_path_truncate(path, 0,
1583 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN);
1584 if (!error) {
1585 vfs_dq_init(inode);
1587 error = do_truncate(dentry, 0,
1588 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1589 NULL);
1591 put_write_access(inode);
1592 if (error)
1593 goto err_out;
1594 } else
1595 if (flag & FMODE_WRITE)
1596 vfs_dq_init(inode);
1598 return 0;
1599 err_out:
1600 ima_counts_put(path, acc_mode ?
1601 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC) :
1602 ACC_MODE(flag) & (MAY_READ | MAY_WRITE));
1603 return error;
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_umask();
1619 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1620 if (error)
1621 goto out_unlock;
1622 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1623 out_unlock:
1624 mutex_unlock(&dir->d_inode->i_mutex);
1625 dput(nd->path.dentry);
1626 nd->path.dentry = path->dentry;
1627 if (error)
1628 return error;
1629 /* Don't check for write permission, don't truncate */
1630 return may_open(&nd->path, 0, flag & ~O_TRUNC);
1634 * Note that while the flag value (low two bits) for sys_open means:
1635 * 00 - read-only
1636 * 01 - write-only
1637 * 10 - read-write
1638 * 11 - special
1639 * it is changed into
1640 * 00 - no permissions needed
1641 * 01 - read-permission
1642 * 10 - write-permission
1643 * 11 - read-write
1644 * for the internal routines (ie open_namei()/follow_link() etc)
1645 * This is more logical, and also allows the 00 "no perm needed"
1646 * to be used for symlinks (where the permissions are checked
1647 * later).
1650 static inline int open_to_namei_flags(int flag)
1652 if ((flag+1) & O_ACCMODE)
1653 flag++;
1654 return flag;
1657 static int open_will_write_to_fs(int flag, struct inode *inode)
1660 * We'll never write to the fs underlying
1661 * a device file.
1663 if (special_file(inode->i_mode))
1664 return 0;
1665 return (flag & O_TRUNC);
1669 * Note that the low bits of the passed in "open_flag"
1670 * are not the same as in the local variable "flag". See
1671 * open_to_namei_flags() for more details.
1673 struct file *do_filp_open(int dfd, const char *pathname,
1674 int open_flag, int mode, int acc_mode)
1676 struct file *filp;
1677 struct nameidata nd;
1678 int error;
1679 struct path path;
1680 struct dentry *dir;
1681 int count = 0;
1682 int will_write;
1683 int flag = open_to_namei_flags(open_flag);
1686 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1687 * check for O_DSYNC if the need any syncing at all we enforce it's
1688 * always set instead of having to deal with possibly weird behaviour
1689 * for malicious applications setting only __O_SYNC.
1691 if (open_flag & __O_SYNC)
1692 open_flag |= O_DSYNC;
1694 if (!acc_mode)
1695 acc_mode = MAY_OPEN | ACC_MODE(flag);
1697 /* O_TRUNC implies we need access checks for write permissions */
1698 if (flag & O_TRUNC)
1699 acc_mode |= MAY_WRITE;
1701 /* Allow the LSM permission hook to distinguish append
1702 access from general write access. */
1703 if (flag & O_APPEND)
1704 acc_mode |= MAY_APPEND;
1707 * The simplest case - just a plain lookup.
1709 if (!(flag & O_CREAT)) {
1710 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1711 &nd, flag);
1712 if (error)
1713 return ERR_PTR(error);
1714 goto ok;
1718 * Create - we need to know the parent.
1720 error = path_init(dfd, pathname, LOOKUP_PARENT, &nd);
1721 if (error)
1722 return ERR_PTR(error);
1723 error = path_walk(pathname, &nd);
1724 if (error) {
1725 if (nd.root.mnt)
1726 path_put(&nd.root);
1727 return ERR_PTR(error);
1729 if (unlikely(!audit_dummy_context()))
1730 audit_inode(pathname, nd.path.dentry);
1733 * We have the parent and last component. First of all, check
1734 * that we are not asked to creat(2) an obvious directory - that
1735 * will not do.
1737 error = -EISDIR;
1738 if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len])
1739 goto exit_parent;
1741 error = -ENFILE;
1742 filp = get_empty_filp();
1743 if (filp == NULL)
1744 goto exit_parent;
1745 nd.intent.open.file = filp;
1746 nd.intent.open.flags = flag;
1747 nd.intent.open.create_mode = mode;
1748 dir = nd.path.dentry;
1749 nd.flags &= ~LOOKUP_PARENT;
1750 nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN;
1751 if (flag & O_EXCL)
1752 nd.flags |= LOOKUP_EXCL;
1753 mutex_lock(&dir->d_inode->i_mutex);
1754 path.dentry = lookup_hash(&nd);
1755 path.mnt = nd.path.mnt;
1757 do_last:
1758 error = PTR_ERR(path.dentry);
1759 if (IS_ERR(path.dentry)) {
1760 mutex_unlock(&dir->d_inode->i_mutex);
1761 goto exit;
1764 if (IS_ERR(nd.intent.open.file)) {
1765 error = PTR_ERR(nd.intent.open.file);
1766 goto exit_mutex_unlock;
1769 /* Negative dentry, just create the file */
1770 if (!path.dentry->d_inode) {
1772 * This write is needed to ensure that a
1773 * ro->rw transition does not occur between
1774 * the time when the file is created and when
1775 * a permanent write count is taken through
1776 * the 'struct file' in nameidata_to_filp().
1778 error = mnt_want_write(nd.path.mnt);
1779 if (error)
1780 goto exit_mutex_unlock;
1781 error = __open_namei_create(&nd, &path, flag, mode);
1782 if (error) {
1783 mnt_drop_write(nd.path.mnt);
1784 goto exit;
1786 filp = nameidata_to_filp(&nd, open_flag);
1787 if (IS_ERR(filp))
1788 ima_counts_put(&nd.path,
1789 acc_mode & (MAY_READ | MAY_WRITE |
1790 MAY_EXEC));
1791 mnt_drop_write(nd.path.mnt);
1792 if (nd.root.mnt)
1793 path_put(&nd.root);
1794 return filp;
1798 * It already exists.
1800 mutex_unlock(&dir->d_inode->i_mutex);
1801 audit_inode(pathname, path.dentry);
1803 error = -EEXIST;
1804 if (flag & O_EXCL)
1805 goto exit_dput;
1807 if (__follow_mount(&path)) {
1808 error = -ELOOP;
1809 if (flag & O_NOFOLLOW)
1810 goto exit_dput;
1813 error = -ENOENT;
1814 if (!path.dentry->d_inode)
1815 goto exit_dput;
1816 if (path.dentry->d_inode->i_op->follow_link)
1817 goto do_link;
1819 path_to_nameidata(&path, &nd);
1820 error = -EISDIR;
1821 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1822 goto exit;
1825 * Consider:
1826 * 1. may_open() truncates a file
1827 * 2. a rw->ro mount transition occurs
1828 * 3. nameidata_to_filp() fails due to
1829 * the ro mount.
1830 * That would be inconsistent, and should
1831 * be avoided. Taking this mnt write here
1832 * ensures that (2) can not occur.
1834 will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
1835 if (will_write) {
1836 error = mnt_want_write(nd.path.mnt);
1837 if (error)
1838 goto exit;
1840 error = may_open(&nd.path, acc_mode, flag);
1841 if (error) {
1842 if (will_write)
1843 mnt_drop_write(nd.path.mnt);
1844 goto exit;
1846 filp = nameidata_to_filp(&nd, open_flag);
1847 if (IS_ERR(filp))
1848 ima_counts_put(&nd.path,
1849 acc_mode & (MAY_READ | MAY_WRITE | MAY_EXEC));
1851 * It is now safe to drop the mnt write
1852 * because the filp has had a write taken
1853 * on its behalf.
1855 if (will_write)
1856 mnt_drop_write(nd.path.mnt);
1857 if (nd.root.mnt)
1858 path_put(&nd.root);
1859 return filp;
1861 exit_mutex_unlock:
1862 mutex_unlock(&dir->d_inode->i_mutex);
1863 exit_dput:
1864 path_put_conditional(&path, &nd);
1865 exit:
1866 if (!IS_ERR(nd.intent.open.file))
1867 release_open_intent(&nd);
1868 exit_parent:
1869 if (nd.root.mnt)
1870 path_put(&nd.root);
1871 path_put(&nd.path);
1872 return ERR_PTR(error);
1874 do_link:
1875 error = -ELOOP;
1876 if (flag & O_NOFOLLOW)
1877 goto exit_dput;
1879 * This is subtle. Instead of calling do_follow_link() we do the
1880 * thing by hands. The reason is that this way we have zero link_count
1881 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1882 * After that we have the parent and last component, i.e.
1883 * we are in the same situation as after the first path_walk().
1884 * Well, almost - if the last component is normal we get its copy
1885 * stored in nd->last.name and we will have to putname() it when we
1886 * are done. Procfs-like symlinks just set LAST_BIND.
1888 nd.flags |= LOOKUP_PARENT;
1889 error = security_inode_follow_link(path.dentry, &nd);
1890 if (error)
1891 goto exit_dput;
1892 error = __do_follow_link(&path, &nd);
1893 if (error) {
1894 /* Does someone understand code flow here? Or it is only
1895 * me so stupid? Anathema to whoever designed this non-sense
1896 * with "intent.open".
1898 release_open_intent(&nd);
1899 if (nd.root.mnt)
1900 path_put(&nd.root);
1901 return ERR_PTR(error);
1903 nd.flags &= ~LOOKUP_PARENT;
1904 if (nd.last_type == LAST_BIND)
1905 goto ok;
1906 error = -EISDIR;
1907 if (nd.last_type != LAST_NORM)
1908 goto exit;
1909 if (nd.last.name[nd.last.len]) {
1910 __putname(nd.last.name);
1911 goto exit;
1913 error = -ELOOP;
1914 if (count++==32) {
1915 __putname(nd.last.name);
1916 goto exit;
1918 dir = nd.path.dentry;
1919 mutex_lock(&dir->d_inode->i_mutex);
1920 path.dentry = lookup_hash(&nd);
1921 path.mnt = nd.path.mnt;
1922 __putname(nd.last.name);
1923 goto do_last;
1927 * filp_open - open file and return file pointer
1929 * @filename: path to open
1930 * @flags: open flags as per the open(2) second argument
1931 * @mode: mode for the new file if O_CREAT is set, else ignored
1933 * This is the helper to open a file from kernelspace if you really
1934 * have to. But in generally you should not do this, so please move
1935 * along, nothing to see here..
1937 struct file *filp_open(const char *filename, int flags, int mode)
1939 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
1941 EXPORT_SYMBOL(filp_open);
1944 * lookup_create - lookup a dentry, creating it if it doesn't exist
1945 * @nd: nameidata info
1946 * @is_dir: directory flag
1948 * Simple function to lookup and return a dentry and create it
1949 * if it doesn't exist. Is SMP-safe.
1951 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1953 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1955 struct dentry *dentry = ERR_PTR(-EEXIST);
1957 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1959 * Yucky last component or no last component at all?
1960 * (foo/., foo/.., /////)
1962 if (nd->last_type != LAST_NORM)
1963 goto fail;
1964 nd->flags &= ~LOOKUP_PARENT;
1965 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
1966 nd->intent.open.flags = O_EXCL;
1969 * Do the final lookup.
1971 dentry = lookup_hash(nd);
1972 if (IS_ERR(dentry))
1973 goto fail;
1975 if (dentry->d_inode)
1976 goto eexist;
1978 * Special case - lookup gave negative, but... we had foo/bar/
1979 * From the vfs_mknod() POV we just have a negative dentry -
1980 * all is fine. Let's be bastards - you had / on the end, you've
1981 * been asking for (non-existent) directory. -ENOENT for you.
1983 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
1984 dput(dentry);
1985 dentry = ERR_PTR(-ENOENT);
1987 return dentry;
1988 eexist:
1989 dput(dentry);
1990 dentry = ERR_PTR(-EEXIST);
1991 fail:
1992 return dentry;
1994 EXPORT_SYMBOL_GPL(lookup_create);
1996 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1998 int error = may_create(dir, dentry);
2000 if (error)
2001 return error;
2003 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2004 return -EPERM;
2006 if (!dir->i_op->mknod)
2007 return -EPERM;
2009 error = devcgroup_inode_mknod(mode, dev);
2010 if (error)
2011 return error;
2013 error = security_inode_mknod(dir, dentry, mode, dev);
2014 if (error)
2015 return error;
2017 vfs_dq_init(dir);
2018 error = dir->i_op->mknod(dir, dentry, mode, dev);
2019 if (!error)
2020 fsnotify_create(dir, dentry);
2021 return error;
2024 static int may_mknod(mode_t mode)
2026 switch (mode & S_IFMT) {
2027 case S_IFREG:
2028 case S_IFCHR:
2029 case S_IFBLK:
2030 case S_IFIFO:
2031 case S_IFSOCK:
2032 case 0: /* zero mode translates to S_IFREG */
2033 return 0;
2034 case S_IFDIR:
2035 return -EPERM;
2036 default:
2037 return -EINVAL;
2041 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2042 unsigned, dev)
2044 int error;
2045 char *tmp;
2046 struct dentry *dentry;
2047 struct nameidata nd;
2049 if (S_ISDIR(mode))
2050 return -EPERM;
2052 error = user_path_parent(dfd, filename, &nd, &tmp);
2053 if (error)
2054 return error;
2056 dentry = lookup_create(&nd, 0);
2057 if (IS_ERR(dentry)) {
2058 error = PTR_ERR(dentry);
2059 goto out_unlock;
2061 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2062 mode &= ~current_umask();
2063 error = may_mknod(mode);
2064 if (error)
2065 goto out_dput;
2066 error = mnt_want_write(nd.path.mnt);
2067 if (error)
2068 goto out_dput;
2069 error = security_path_mknod(&nd.path, dentry, mode, dev);
2070 if (error)
2071 goto out_drop_write;
2072 switch (mode & S_IFMT) {
2073 case 0: case S_IFREG:
2074 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2075 break;
2076 case S_IFCHR: case S_IFBLK:
2077 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2078 new_decode_dev(dev));
2079 break;
2080 case S_IFIFO: case S_IFSOCK:
2081 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2082 break;
2084 out_drop_write:
2085 mnt_drop_write(nd.path.mnt);
2086 out_dput:
2087 dput(dentry);
2088 out_unlock:
2089 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2090 path_put(&nd.path);
2091 putname(tmp);
2093 return error;
2096 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2098 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2101 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2103 int error = may_create(dir, dentry);
2105 if (error)
2106 return error;
2108 if (!dir->i_op->mkdir)
2109 return -EPERM;
2111 mode &= (S_IRWXUGO|S_ISVTX);
2112 error = security_inode_mkdir(dir, dentry, mode);
2113 if (error)
2114 return error;
2116 vfs_dq_init(dir);
2117 error = dir->i_op->mkdir(dir, dentry, mode);
2118 if (!error)
2119 fsnotify_mkdir(dir, dentry);
2120 return error;
2123 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2125 int error = 0;
2126 char * tmp;
2127 struct dentry *dentry;
2128 struct nameidata nd;
2130 error = user_path_parent(dfd, pathname, &nd, &tmp);
2131 if (error)
2132 goto out_err;
2134 dentry = lookup_create(&nd, 1);
2135 error = PTR_ERR(dentry);
2136 if (IS_ERR(dentry))
2137 goto out_unlock;
2139 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2140 mode &= ~current_umask();
2141 error = mnt_want_write(nd.path.mnt);
2142 if (error)
2143 goto out_dput;
2144 error = security_path_mkdir(&nd.path, dentry, mode);
2145 if (error)
2146 goto out_drop_write;
2147 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2148 out_drop_write:
2149 mnt_drop_write(nd.path.mnt);
2150 out_dput:
2151 dput(dentry);
2152 out_unlock:
2153 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2154 path_put(&nd.path);
2155 putname(tmp);
2156 out_err:
2157 return error;
2160 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2162 return sys_mkdirat(AT_FDCWD, pathname, mode);
2166 * We try to drop the dentry early: we should have
2167 * a usage count of 2 if we're the only user of this
2168 * dentry, and if that is true (possibly after pruning
2169 * the dcache), then we drop the dentry now.
2171 * A low-level filesystem can, if it choses, legally
2172 * do a
2174 * if (!d_unhashed(dentry))
2175 * return -EBUSY;
2177 * if it cannot handle the case of removing a directory
2178 * that is still in use by something else..
2180 void dentry_unhash(struct dentry *dentry)
2182 dget(dentry);
2183 shrink_dcache_parent(dentry);
2184 spin_lock(&dcache_lock);
2185 spin_lock(&dentry->d_lock);
2186 if (atomic_read(&dentry->d_count) == 2)
2187 __d_drop(dentry);
2188 spin_unlock(&dentry->d_lock);
2189 spin_unlock(&dcache_lock);
2192 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2194 int error = may_delete(dir, dentry, 1);
2196 if (error)
2197 return error;
2199 if (!dir->i_op->rmdir)
2200 return -EPERM;
2202 vfs_dq_init(dir);
2204 mutex_lock(&dentry->d_inode->i_mutex);
2205 dentry_unhash(dentry);
2206 if (d_mountpoint(dentry))
2207 error = -EBUSY;
2208 else {
2209 error = security_inode_rmdir(dir, dentry);
2210 if (!error) {
2211 error = dir->i_op->rmdir(dir, dentry);
2212 if (!error)
2213 dentry->d_inode->i_flags |= S_DEAD;
2216 mutex_unlock(&dentry->d_inode->i_mutex);
2217 if (!error) {
2218 d_delete(dentry);
2220 dput(dentry);
2222 return error;
2225 static long do_rmdir(int dfd, const char __user *pathname)
2227 int error = 0;
2228 char * name;
2229 struct dentry *dentry;
2230 struct nameidata nd;
2232 error = user_path_parent(dfd, pathname, &nd, &name);
2233 if (error)
2234 return error;
2236 switch(nd.last_type) {
2237 case LAST_DOTDOT:
2238 error = -ENOTEMPTY;
2239 goto exit1;
2240 case LAST_DOT:
2241 error = -EINVAL;
2242 goto exit1;
2243 case LAST_ROOT:
2244 error = -EBUSY;
2245 goto exit1;
2248 nd.flags &= ~LOOKUP_PARENT;
2250 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2251 dentry = lookup_hash(&nd);
2252 error = PTR_ERR(dentry);
2253 if (IS_ERR(dentry))
2254 goto exit2;
2255 error = mnt_want_write(nd.path.mnt);
2256 if (error)
2257 goto exit3;
2258 error = security_path_rmdir(&nd.path, dentry);
2259 if (error)
2260 goto exit4;
2261 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2262 exit4:
2263 mnt_drop_write(nd.path.mnt);
2264 exit3:
2265 dput(dentry);
2266 exit2:
2267 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2268 exit1:
2269 path_put(&nd.path);
2270 putname(name);
2271 return error;
2274 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2276 return do_rmdir(AT_FDCWD, pathname);
2279 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2281 int error = may_delete(dir, dentry, 0);
2283 if (error)
2284 return error;
2286 if (!dir->i_op->unlink)
2287 return -EPERM;
2289 vfs_dq_init(dir);
2291 mutex_lock(&dentry->d_inode->i_mutex);
2292 if (d_mountpoint(dentry))
2293 error = -EBUSY;
2294 else {
2295 error = security_inode_unlink(dir, dentry);
2296 if (!error)
2297 error = dir->i_op->unlink(dir, dentry);
2299 mutex_unlock(&dentry->d_inode->i_mutex);
2301 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2302 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2303 fsnotify_link_count(dentry->d_inode);
2304 d_delete(dentry);
2307 return error;
2311 * Make sure that the actual truncation of the file will occur outside its
2312 * directory's i_mutex. Truncate can take a long time if there is a lot of
2313 * writeout happening, and we don't want to prevent access to the directory
2314 * while waiting on the I/O.
2316 static long do_unlinkat(int dfd, const char __user *pathname)
2318 int error;
2319 char *name;
2320 struct dentry *dentry;
2321 struct nameidata nd;
2322 struct inode *inode = NULL;
2324 error = user_path_parent(dfd, pathname, &nd, &name);
2325 if (error)
2326 return error;
2328 error = -EISDIR;
2329 if (nd.last_type != LAST_NORM)
2330 goto exit1;
2332 nd.flags &= ~LOOKUP_PARENT;
2334 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2335 dentry = lookup_hash(&nd);
2336 error = PTR_ERR(dentry);
2337 if (!IS_ERR(dentry)) {
2338 /* Why not before? Because we want correct error value */
2339 if (nd.last.name[nd.last.len])
2340 goto slashes;
2341 inode = dentry->d_inode;
2342 if (inode)
2343 atomic_inc(&inode->i_count);
2344 error = mnt_want_write(nd.path.mnt);
2345 if (error)
2346 goto exit2;
2347 error = security_path_unlink(&nd.path, dentry);
2348 if (error)
2349 goto exit3;
2350 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2351 exit3:
2352 mnt_drop_write(nd.path.mnt);
2353 exit2:
2354 dput(dentry);
2356 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2357 if (inode)
2358 iput(inode); /* truncate the inode here */
2359 exit1:
2360 path_put(&nd.path);
2361 putname(name);
2362 return error;
2364 slashes:
2365 error = !dentry->d_inode ? -ENOENT :
2366 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2367 goto exit2;
2370 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2372 if ((flag & ~AT_REMOVEDIR) != 0)
2373 return -EINVAL;
2375 if (flag & AT_REMOVEDIR)
2376 return do_rmdir(dfd, pathname);
2378 return do_unlinkat(dfd, pathname);
2381 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2383 return do_unlinkat(AT_FDCWD, pathname);
2386 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2388 int error = may_create(dir, dentry);
2390 if (error)
2391 return error;
2393 if (!dir->i_op->symlink)
2394 return -EPERM;
2396 error = security_inode_symlink(dir, dentry, oldname);
2397 if (error)
2398 return error;
2400 vfs_dq_init(dir);
2401 error = dir->i_op->symlink(dir, dentry, oldname);
2402 if (!error)
2403 fsnotify_create(dir, dentry);
2404 return error;
2407 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2408 int, newdfd, const char __user *, newname)
2410 int error;
2411 char *from;
2412 char *to;
2413 struct dentry *dentry;
2414 struct nameidata nd;
2416 from = getname(oldname);
2417 if (IS_ERR(from))
2418 return PTR_ERR(from);
2420 error = user_path_parent(newdfd, newname, &nd, &to);
2421 if (error)
2422 goto out_putname;
2424 dentry = lookup_create(&nd, 0);
2425 error = PTR_ERR(dentry);
2426 if (IS_ERR(dentry))
2427 goto out_unlock;
2429 error = mnt_want_write(nd.path.mnt);
2430 if (error)
2431 goto out_dput;
2432 error = security_path_symlink(&nd.path, dentry, from);
2433 if (error)
2434 goto out_drop_write;
2435 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2436 out_drop_write:
2437 mnt_drop_write(nd.path.mnt);
2438 out_dput:
2439 dput(dentry);
2440 out_unlock:
2441 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2442 path_put(&nd.path);
2443 putname(to);
2444 out_putname:
2445 putname(from);
2446 return error;
2449 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2451 return sys_symlinkat(oldname, AT_FDCWD, newname);
2454 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2456 struct inode *inode = old_dentry->d_inode;
2457 int error;
2459 if (!inode)
2460 return -ENOENT;
2462 error = may_create(dir, new_dentry);
2463 if (error)
2464 return error;
2466 if (dir->i_sb != inode->i_sb)
2467 return -EXDEV;
2470 * A link to an append-only or immutable file cannot be created.
2472 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2473 return -EPERM;
2474 if (!dir->i_op->link)
2475 return -EPERM;
2476 if (S_ISDIR(inode->i_mode))
2477 return -EPERM;
2479 error = security_inode_link(old_dentry, dir, new_dentry);
2480 if (error)
2481 return error;
2483 mutex_lock(&inode->i_mutex);
2484 vfs_dq_init(dir);
2485 error = dir->i_op->link(old_dentry, dir, new_dentry);
2486 mutex_unlock(&inode->i_mutex);
2487 if (!error)
2488 fsnotify_link(dir, inode, new_dentry);
2489 return error;
2493 * Hardlinks are often used in delicate situations. We avoid
2494 * security-related surprises by not following symlinks on the
2495 * newname. --KAB
2497 * We don't follow them on the oldname either to be compatible
2498 * with linux 2.0, and to avoid hard-linking to directories
2499 * and other special files. --ADM
2501 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2502 int, newdfd, const char __user *, newname, int, flags)
2504 struct dentry *new_dentry;
2505 struct nameidata nd;
2506 struct path old_path;
2507 int error;
2508 char *to;
2510 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2511 return -EINVAL;
2513 error = user_path_at(olddfd, oldname,
2514 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2515 &old_path);
2516 if (error)
2517 return error;
2519 error = user_path_parent(newdfd, newname, &nd, &to);
2520 if (error)
2521 goto out;
2522 error = -EXDEV;
2523 if (old_path.mnt != nd.path.mnt)
2524 goto out_release;
2525 new_dentry = lookup_create(&nd, 0);
2526 error = PTR_ERR(new_dentry);
2527 if (IS_ERR(new_dentry))
2528 goto out_unlock;
2529 error = mnt_want_write(nd.path.mnt);
2530 if (error)
2531 goto out_dput;
2532 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2533 if (error)
2534 goto out_drop_write;
2535 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2536 out_drop_write:
2537 mnt_drop_write(nd.path.mnt);
2538 out_dput:
2539 dput(new_dentry);
2540 out_unlock:
2541 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2542 out_release:
2543 path_put(&nd.path);
2544 putname(to);
2545 out:
2546 path_put(&old_path);
2548 return error;
2551 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2553 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2557 * The worst of all namespace operations - renaming directory. "Perverted"
2558 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2559 * Problems:
2560 * a) we can get into loop creation. Check is done in is_subdir().
2561 * b) race potential - two innocent renames can create a loop together.
2562 * That's where 4.4 screws up. Current fix: serialization on
2563 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2564 * story.
2565 * c) we have to lock _three_ objects - parents and victim (if it exists).
2566 * And that - after we got ->i_mutex on parents (until then we don't know
2567 * whether the target exists). Solution: try to be smart with locking
2568 * order for inodes. We rely on the fact that tree topology may change
2569 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2570 * move will be locked. Thus we can rank directories by the tree
2571 * (ancestors first) and rank all non-directories after them.
2572 * That works since everybody except rename does "lock parent, lookup,
2573 * lock child" and rename is under ->s_vfs_rename_mutex.
2574 * HOWEVER, it relies on the assumption that any object with ->lookup()
2575 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2576 * we'd better make sure that there's no link(2) for them.
2577 * d) some filesystems don't support opened-but-unlinked directories,
2578 * either because of layout or because they are not ready to deal with
2579 * all cases correctly. The latter will be fixed (taking this sort of
2580 * stuff into VFS), but the former is not going away. Solution: the same
2581 * trick as in rmdir().
2582 * e) conversion from fhandle to dentry may come in the wrong moment - when
2583 * we are removing the target. Solution: we will have to grab ->i_mutex
2584 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2585 * ->i_mutex on parents, which works but leads to some truely excessive
2586 * locking].
2588 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2589 struct inode *new_dir, struct dentry *new_dentry)
2591 int error = 0;
2592 struct inode *target;
2595 * If we are going to change the parent - check write permissions,
2596 * we'll need to flip '..'.
2598 if (new_dir != old_dir) {
2599 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2600 if (error)
2601 return error;
2604 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2605 if (error)
2606 return error;
2608 target = new_dentry->d_inode;
2609 if (target) {
2610 mutex_lock(&target->i_mutex);
2611 dentry_unhash(new_dentry);
2613 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2614 error = -EBUSY;
2615 else
2616 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2617 if (target) {
2618 if (!error)
2619 target->i_flags |= S_DEAD;
2620 mutex_unlock(&target->i_mutex);
2621 dput(new_dentry);
2623 if (!error)
2624 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2625 d_move(old_dentry,new_dentry);
2626 return error;
2629 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2630 struct inode *new_dir, struct dentry *new_dentry)
2632 struct inode *target;
2633 int error;
2635 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2636 if (error)
2637 return error;
2639 dget(new_dentry);
2640 target = new_dentry->d_inode;
2641 if (target)
2642 mutex_lock(&target->i_mutex);
2643 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2644 error = -EBUSY;
2645 else
2646 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2647 if (!error) {
2648 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2649 d_move(old_dentry, new_dentry);
2651 if (target)
2652 mutex_unlock(&target->i_mutex);
2653 dput(new_dentry);
2654 return error;
2657 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2658 struct inode *new_dir, struct dentry *new_dentry)
2660 int error;
2661 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2662 const char *old_name;
2664 if (old_dentry->d_inode == new_dentry->d_inode)
2665 return 0;
2667 error = may_delete(old_dir, old_dentry, is_dir);
2668 if (error)
2669 return error;
2671 if (!new_dentry->d_inode)
2672 error = may_create(new_dir, new_dentry);
2673 else
2674 error = may_delete(new_dir, new_dentry, is_dir);
2675 if (error)
2676 return error;
2678 if (!old_dir->i_op->rename)
2679 return -EPERM;
2681 vfs_dq_init(old_dir);
2682 vfs_dq_init(new_dir);
2684 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2686 if (is_dir)
2687 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2688 else
2689 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2690 if (!error) {
2691 const char *new_name = old_dentry->d_name.name;
2692 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2693 new_dentry->d_inode, old_dentry);
2695 fsnotify_oldname_free(old_name);
2697 return error;
2700 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
2701 int, newdfd, const char __user *, newname)
2703 struct dentry *old_dir, *new_dir;
2704 struct dentry *old_dentry, *new_dentry;
2705 struct dentry *trap;
2706 struct nameidata oldnd, newnd;
2707 char *from;
2708 char *to;
2709 int error;
2711 error = user_path_parent(olddfd, oldname, &oldnd, &from);
2712 if (error)
2713 goto exit;
2715 error = user_path_parent(newdfd, newname, &newnd, &to);
2716 if (error)
2717 goto exit1;
2719 error = -EXDEV;
2720 if (oldnd.path.mnt != newnd.path.mnt)
2721 goto exit2;
2723 old_dir = oldnd.path.dentry;
2724 error = -EBUSY;
2725 if (oldnd.last_type != LAST_NORM)
2726 goto exit2;
2728 new_dir = newnd.path.dentry;
2729 if (newnd.last_type != LAST_NORM)
2730 goto exit2;
2732 oldnd.flags &= ~LOOKUP_PARENT;
2733 newnd.flags &= ~LOOKUP_PARENT;
2734 newnd.flags |= LOOKUP_RENAME_TARGET;
2736 trap = lock_rename(new_dir, old_dir);
2738 old_dentry = lookup_hash(&oldnd);
2739 error = PTR_ERR(old_dentry);
2740 if (IS_ERR(old_dentry))
2741 goto exit3;
2742 /* source must exist */
2743 error = -ENOENT;
2744 if (!old_dentry->d_inode)
2745 goto exit4;
2746 /* unless the source is a directory trailing slashes give -ENOTDIR */
2747 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2748 error = -ENOTDIR;
2749 if (oldnd.last.name[oldnd.last.len])
2750 goto exit4;
2751 if (newnd.last.name[newnd.last.len])
2752 goto exit4;
2754 /* source should not be ancestor of target */
2755 error = -EINVAL;
2756 if (old_dentry == trap)
2757 goto exit4;
2758 new_dentry = lookup_hash(&newnd);
2759 error = PTR_ERR(new_dentry);
2760 if (IS_ERR(new_dentry))
2761 goto exit4;
2762 /* target should not be an ancestor of source */
2763 error = -ENOTEMPTY;
2764 if (new_dentry == trap)
2765 goto exit5;
2767 error = mnt_want_write(oldnd.path.mnt);
2768 if (error)
2769 goto exit5;
2770 error = security_path_rename(&oldnd.path, old_dentry,
2771 &newnd.path, new_dentry);
2772 if (error)
2773 goto exit6;
2774 error = vfs_rename(old_dir->d_inode, old_dentry,
2775 new_dir->d_inode, new_dentry);
2776 exit6:
2777 mnt_drop_write(oldnd.path.mnt);
2778 exit5:
2779 dput(new_dentry);
2780 exit4:
2781 dput(old_dentry);
2782 exit3:
2783 unlock_rename(new_dir, old_dir);
2784 exit2:
2785 path_put(&newnd.path);
2786 putname(to);
2787 exit1:
2788 path_put(&oldnd.path);
2789 putname(from);
2790 exit:
2791 return error;
2794 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
2796 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2799 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2801 int len;
2803 len = PTR_ERR(link);
2804 if (IS_ERR(link))
2805 goto out;
2807 len = strlen(link);
2808 if (len > (unsigned) buflen)
2809 len = buflen;
2810 if (copy_to_user(buffer, link, len))
2811 len = -EFAULT;
2812 out:
2813 return len;
2817 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2818 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2819 * using) it for any given inode is up to filesystem.
2821 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2823 struct nameidata nd;
2824 void *cookie;
2825 int res;
2827 nd.depth = 0;
2828 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2829 if (IS_ERR(cookie))
2830 return PTR_ERR(cookie);
2832 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2833 if (dentry->d_inode->i_op->put_link)
2834 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2835 return res;
2838 int vfs_follow_link(struct nameidata *nd, const char *link)
2840 return __vfs_follow_link(nd, link);
2843 /* get the link contents into pagecache */
2844 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2846 char *kaddr;
2847 struct page *page;
2848 struct address_space *mapping = dentry->d_inode->i_mapping;
2849 page = read_mapping_page(mapping, 0, NULL);
2850 if (IS_ERR(page))
2851 return (char*)page;
2852 *ppage = page;
2853 kaddr = kmap(page);
2854 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
2855 return kaddr;
2858 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2860 struct page *page = NULL;
2861 char *s = page_getlink(dentry, &page);
2862 int res = vfs_readlink(dentry,buffer,buflen,s);
2863 if (page) {
2864 kunmap(page);
2865 page_cache_release(page);
2867 return res;
2870 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2872 struct page *page = NULL;
2873 nd_set_link(nd, page_getlink(dentry, &page));
2874 return page;
2877 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2879 struct page *page = cookie;
2881 if (page) {
2882 kunmap(page);
2883 page_cache_release(page);
2888 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
2890 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
2892 struct address_space *mapping = inode->i_mapping;
2893 struct page *page;
2894 void *fsdata;
2895 int err;
2896 char *kaddr;
2897 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
2898 if (nofs)
2899 flags |= AOP_FLAG_NOFS;
2901 retry:
2902 err = pagecache_write_begin(NULL, mapping, 0, len-1,
2903 flags, &page, &fsdata);
2904 if (err)
2905 goto fail;
2907 kaddr = kmap_atomic(page, KM_USER0);
2908 memcpy(kaddr, symname, len-1);
2909 kunmap_atomic(kaddr, KM_USER0);
2911 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
2912 page, fsdata);
2913 if (err < 0)
2914 goto fail;
2915 if (err < len-1)
2916 goto retry;
2918 mark_inode_dirty(inode);
2919 return 0;
2920 fail:
2921 return err;
2924 int page_symlink(struct inode *inode, const char *symname, int len)
2926 return __page_symlink(inode, symname, len,
2927 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
2930 const struct inode_operations page_symlink_inode_operations = {
2931 .readlink = generic_readlink,
2932 .follow_link = page_follow_link_light,
2933 .put_link = page_put_link,
2936 EXPORT_SYMBOL(user_path_at);
2937 EXPORT_SYMBOL(follow_down);
2938 EXPORT_SYMBOL(follow_up);
2939 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2940 EXPORT_SYMBOL(getname);
2941 EXPORT_SYMBOL(lock_rename);
2942 EXPORT_SYMBOL(lookup_one_len);
2943 EXPORT_SYMBOL(page_follow_link_light);
2944 EXPORT_SYMBOL(page_put_link);
2945 EXPORT_SYMBOL(page_readlink);
2946 EXPORT_SYMBOL(__page_symlink);
2947 EXPORT_SYMBOL(page_symlink);
2948 EXPORT_SYMBOL(page_symlink_inode_operations);
2949 EXPORT_SYMBOL(path_lookup);
2950 EXPORT_SYMBOL(kern_path);
2951 EXPORT_SYMBOL(vfs_path_lookup);
2952 EXPORT_SYMBOL(inode_permission);
2953 EXPORT_SYMBOL(file_permission);
2954 EXPORT_SYMBOL(unlock_rename);
2955 EXPORT_SYMBOL(vfs_create);
2956 EXPORT_SYMBOL(vfs_follow_link);
2957 EXPORT_SYMBOL(vfs_link);
2958 EXPORT_SYMBOL(vfs_mkdir);
2959 EXPORT_SYMBOL(vfs_mknod);
2960 EXPORT_SYMBOL(generic_permission);
2961 EXPORT_SYMBOL(vfs_readlink);
2962 EXPORT_SYMBOL(vfs_rename);
2963 EXPORT_SYMBOL(vfs_rmdir);
2964 EXPORT_SYMBOL(vfs_symlink);
2965 EXPORT_SYMBOL(vfs_unlink);
2966 EXPORT_SYMBOL(dentry_unhash);
2967 EXPORT_SYMBOL(generic_readlink);