4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/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>
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)
120 unsigned long len = PATH_MAX;
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
129 retval = strncpy_from_user(page, filename, len);
133 return -ENAMETOOLONG;
139 char * getname(const char __user * filename)
143 result = ERR_PTR(-ENOMEM);
146 int retval = do_getname(filename, tmp);
151 result = ERR_PTR(retval);
154 audit_getname(result);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
170 * This does basic POSIX ACL permission checking
172 static inline int __acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask), int rcu)
175 umode_t mode = inode->i_mode;
177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
179 if (current_fsuid() == inode->i_uid)
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
192 if (in_group_p(inode->i_gid))
197 * If the DACs are ok we don't need any capability check.
199 if ((mask & ~mode) == 0)
204 static inline int acl_permission_check(struct inode *inode, int mask,
205 int (*check_acl)(struct inode *inode, int mask))
207 return __acl_permission_check(inode, mask, check_acl, 0);
211 * generic_permission - check for access rights on a Posix-like filesystem
212 * @inode: inode to check access rights for
213 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
214 * @check_acl: optional callback to check for Posix ACLs
216 * Used to check for read/write/execute permissions on a file.
217 * We use "fsuid" for this, letting us set arbitrary permissions
218 * for filesystem access without changing the "normal" uids which
219 * are used for other things..
221 int generic_permission(struct inode *inode, int mask,
222 int (*check_acl)(struct inode *inode, int mask))
227 * Do the basic POSIX ACL permission checks.
229 ret = acl_permission_check(inode, mask, check_acl);
234 * Read/write DACs are always overridable.
235 * Executable DACs are overridable if at least one exec bit is set.
237 if (!(mask & MAY_EXEC) || execute_ok(inode))
238 if (capable(CAP_DAC_OVERRIDE))
242 * Searching includes executable on directories, else just read.
244 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
245 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
246 if (capable(CAP_DAC_READ_SEARCH))
253 * inode_permission - check for access rights to a given inode
254 * @inode: inode to check permission on
255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
257 * Used to check for read/write/execute permissions on an inode.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
260 * are used for other things.
262 int inode_permission(struct inode *inode, int mask)
266 if (mask & MAY_WRITE) {
267 umode_t mode = inode->i_mode;
270 * Nobody gets write access to a read-only fs.
272 if (IS_RDONLY(inode) &&
273 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
277 * Nobody gets write access to an immutable file.
279 if (IS_IMMUTABLE(inode))
283 if (inode->i_op->permission)
284 retval = inode->i_op->permission(inode, mask);
286 retval = generic_permission(inode, mask, inode->i_op->check_acl);
291 retval = devcgroup_inode_permission(inode, mask);
295 return security_inode_permission(inode, mask);
299 * file_permission - check for additional access rights to a given file
300 * @file: file to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
303 * Used to check for read/write/execute permissions on an already opened
307 * Do not use this function in new code. All access checks should
308 * be done using inode_permission().
310 int file_permission(struct file *file, int mask)
312 return inode_permission(file->f_path.dentry->d_inode, mask);
316 * get_write_access() gets write permission for a file.
317 * put_write_access() releases this write permission.
318 * This is used for regular files.
319 * We cannot support write (and maybe mmap read-write shared) accesses and
320 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
321 * can have the following values:
322 * 0: no writers, no VM_DENYWRITE mappings
323 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
324 * > 0: (i_writecount) users are writing to the file.
326 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
327 * except for the cases where we don't hold i_writecount yet. Then we need to
328 * use {get,deny}_write_access() - these functions check the sign and refuse
329 * to do the change if sign is wrong. Exclusion between them is provided by
330 * the inode->i_lock spinlock.
333 int get_write_access(struct inode * inode)
335 spin_lock(&inode->i_lock);
336 if (atomic_read(&inode->i_writecount) < 0) {
337 spin_unlock(&inode->i_lock);
340 atomic_inc(&inode->i_writecount);
341 spin_unlock(&inode->i_lock);
346 int deny_write_access(struct file * file)
348 struct inode *inode = file->f_path.dentry->d_inode;
350 spin_lock(&inode->i_lock);
351 if (atomic_read(&inode->i_writecount) > 0) {
352 spin_unlock(&inode->i_lock);
355 atomic_dec(&inode->i_writecount);
356 spin_unlock(&inode->i_lock);
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
365 * Given a path increment the reference count to the dentry and the vfsmount.
367 void path_get(struct path *path)
372 EXPORT_SYMBOL(path_get);
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
378 * Given a path decrement the reference count to the dentry and the vfsmount.
380 void path_put(struct path *path)
385 EXPORT_SYMBOL(path_put);
388 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
389 * @nd: nameidata pathwalk data to drop
390 * @Returns: 0 on success, -ECHLID on failure
392 * Path walking has 2 modes, rcu-walk and ref-walk (see
393 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
394 * to drop out of rcu-walk mode and take normal reference counts on dentries
395 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
396 * refcounts at the last known good point before rcu-walk got stuck, so
397 * ref-walk may continue from there. If this is not successful (eg. a seqcount
398 * has changed), then failure is returned and path walk restarts from the
399 * beginning in ref-walk mode.
401 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
402 * ref-walk. Must be called from rcu-walk context.
404 static int nameidata_drop_rcu(struct nameidata *nd)
406 struct fs_struct *fs = current->fs;
407 struct dentry *dentry = nd->path.dentry;
409 BUG_ON(!(nd->flags & LOOKUP_RCU));
411 spin_lock(&fs->lock);
412 if (nd->root.mnt != fs->root.mnt ||
413 nd->root.dentry != fs->root.dentry)
416 spin_lock(&dentry->d_lock);
417 if (!__d_rcu_to_refcount(dentry, nd->seq))
419 BUG_ON(nd->inode != dentry->d_inode);
420 spin_unlock(&dentry->d_lock);
423 spin_unlock(&fs->lock);
425 mntget(nd->path.mnt);
428 br_read_unlock(vfsmount_lock);
429 nd->flags &= ~LOOKUP_RCU;
432 spin_unlock(&dentry->d_lock);
435 spin_unlock(&fs->lock);
439 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
440 static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
442 if (nd->flags & LOOKUP_RCU)
443 return nameidata_drop_rcu(nd);
448 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
449 * @nd: nameidata pathwalk data to drop
450 * @dentry: dentry to drop
451 * @Returns: 0 on success, -ECHLID on failure
453 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
454 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
455 * @nd. Must be called from rcu-walk context.
457 static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
459 struct fs_struct *fs = current->fs;
460 struct dentry *parent = nd->path.dentry;
462 BUG_ON(!(nd->flags & LOOKUP_RCU));
464 spin_lock(&fs->lock);
465 if (nd->root.mnt != fs->root.mnt ||
466 nd->root.dentry != fs->root.dentry)
469 spin_lock(&parent->d_lock);
470 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
471 if (!__d_rcu_to_refcount(dentry, nd->seq))
474 * If the sequence check on the child dentry passed, then the child has
475 * not been removed from its parent. This means the parent dentry must
476 * be valid and able to take a reference at this point.
478 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
479 BUG_ON(!parent->d_count);
481 spin_unlock(&dentry->d_lock);
482 spin_unlock(&parent->d_lock);
485 spin_unlock(&fs->lock);
487 mntget(nd->path.mnt);
490 br_read_unlock(vfsmount_lock);
491 nd->flags &= ~LOOKUP_RCU;
494 spin_unlock(&dentry->d_lock);
495 spin_unlock(&parent->d_lock);
498 spin_unlock(&fs->lock);
502 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
503 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
505 if (nd->flags & LOOKUP_RCU)
506 return nameidata_dentry_drop_rcu(nd, dentry);
511 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
512 * @nd: nameidata pathwalk data to drop
513 * @Returns: 0 on success, -ECHLID on failure
515 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
516 * nd->path should be the final element of the lookup, so nd->root is discarded.
517 * Must be called from rcu-walk context.
519 static int nameidata_drop_rcu_last(struct nameidata *nd)
521 struct dentry *dentry = nd->path.dentry;
523 BUG_ON(!(nd->flags & LOOKUP_RCU));
524 nd->flags &= ~LOOKUP_RCU;
526 spin_lock(&dentry->d_lock);
527 if (!__d_rcu_to_refcount(dentry, nd->seq))
529 BUG_ON(nd->inode != dentry->d_inode);
530 spin_unlock(&dentry->d_lock);
532 mntget(nd->path.mnt);
535 br_read_unlock(vfsmount_lock);
540 spin_unlock(&dentry->d_lock);
542 br_read_unlock(vfsmount_lock);
546 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
547 static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
549 if (likely(nd->flags & LOOKUP_RCU))
550 return nameidata_drop_rcu_last(nd);
555 * release_open_intent - free up open intent resources
556 * @nd: pointer to nameidata
558 void release_open_intent(struct nameidata *nd)
560 if (nd->intent.open.file->f_path.dentry == NULL)
561 put_filp(nd->intent.open.file);
563 fput(nd->intent.open.file);
566 static inline struct dentry *
567 do_revalidate(struct dentry *dentry, struct nameidata *nd)
569 int status = dentry->d_op->d_revalidate(dentry, nd);
570 if (unlikely(status <= 0)) {
572 * The dentry failed validation.
573 * If d_revalidate returned 0 attempt to invalidate
574 * the dentry otherwise d_revalidate is asking us
575 * to return a fail status.
578 if (!d_invalidate(dentry)) {
584 dentry = ERR_PTR(status);
591 * force_reval_path - force revalidation of a dentry
593 * In some situations the path walking code will trust dentries without
594 * revalidating them. This causes problems for filesystems that depend on
595 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
596 * (which indicates that it's possible for the dentry to go stale), force
597 * a d_revalidate call before proceeding.
599 * Returns 0 if the revalidation was successful. If the revalidation fails,
600 * either return the error returned by d_revalidate or -ESTALE if the
601 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
602 * invalidate the dentry. It's up to the caller to handle putting references
603 * to the path if necessary.
606 force_reval_path(struct path *path, struct nameidata *nd)
609 struct dentry *dentry = path->dentry;
612 * only check on filesystems where it's possible for the dentry to
613 * become stale. It's assumed that if this flag is set then the
614 * d_revalidate op will also be defined.
616 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
619 status = dentry->d_op->d_revalidate(dentry, nd);
624 d_invalidate(dentry);
631 * Short-cut version of permission(), for calling on directories
632 * during pathname resolution. Combines parts of permission()
633 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
635 * If appropriate, check DAC only. If not appropriate, or
636 * short-cut DAC fails, then call ->permission() to do more
637 * complete permission check.
639 static inline int __exec_permission(struct inode *inode, int rcu)
643 if (inode->i_op->permission) {
646 ret = inode->i_op->permission(inode, MAY_EXEC);
651 ret = __acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl, rcu);
654 if (rcu && ret == -ECHILD)
657 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
662 return security_inode_exec_permission(inode, rcu);
665 static int exec_permission(struct inode *inode)
667 return __exec_permission(inode, 0);
670 static int exec_permission_rcu(struct inode *inode)
672 return __exec_permission(inode, 1);
675 static __always_inline void set_root(struct nameidata *nd)
678 get_fs_root(current->fs, &nd->root);
681 static int link_path_walk(const char *, struct nameidata *);
683 static __always_inline void set_root_rcu(struct nameidata *nd)
686 struct fs_struct *fs = current->fs;
687 spin_lock(&fs->lock);
689 spin_unlock(&fs->lock);
693 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
706 nd->inode = nd->path.dentry->d_inode;
708 ret = link_path_walk(link, nd);
712 return PTR_ERR(link);
715 static void path_put_conditional(struct path *path, struct nameidata *nd)
718 if (path->mnt != nd->path.mnt)
722 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
724 if (!(nd->flags & LOOKUP_RCU)) {
725 dput(nd->path.dentry);
726 if (nd->path.mnt != path->mnt)
727 mntput(nd->path.mnt);
729 nd->path.mnt = path->mnt;
730 nd->path.dentry = path->dentry;
733 static __always_inline int
734 __do_follow_link(struct path *path, struct nameidata *nd, void **p)
737 struct dentry *dentry = path->dentry;
739 touch_atime(path->mnt, dentry);
740 nd_set_link(nd, NULL);
742 if (path->mnt != nd->path.mnt) {
743 path_to_nameidata(path, nd);
744 nd->inode = nd->path.dentry->d_inode;
749 nd->last_type = LAST_BIND;
750 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
753 char *s = nd_get_link(nd);
756 error = __vfs_follow_link(nd, s);
757 else if (nd->last_type == LAST_BIND) {
758 error = force_reval_path(&nd->path, nd);
767 * This limits recursive symlink follows to 8, while
768 * limiting consecutive symlinks to 40.
770 * Without that kind of total limit, nasty chains of consecutive
771 * symlinks can cause almost arbitrarily long lookups.
773 static inline int do_follow_link(struct path *path, struct nameidata *nd)
777 if (current->link_count >= MAX_NESTED_LINKS)
779 if (current->total_link_count >= 40)
781 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
783 err = security_inode_follow_link(path->dentry, nd);
786 current->link_count++;
787 current->total_link_count++;
789 err = __do_follow_link(path, nd, &cookie);
790 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
791 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
793 current->link_count--;
797 path_put_conditional(path, nd);
802 static int follow_up_rcu(struct path *path)
804 struct vfsmount *parent;
805 struct dentry *mountpoint;
807 parent = path->mnt->mnt_parent;
808 if (parent == path->mnt)
810 mountpoint = path->mnt->mnt_mountpoint;
811 path->dentry = mountpoint;
816 int follow_up(struct path *path)
818 struct vfsmount *parent;
819 struct dentry *mountpoint;
821 br_read_lock(vfsmount_lock);
822 parent = path->mnt->mnt_parent;
823 if (parent == path->mnt) {
824 br_read_unlock(vfsmount_lock);
828 mountpoint = dget(path->mnt->mnt_mountpoint);
829 br_read_unlock(vfsmount_lock);
831 path->dentry = mountpoint;
838 * serialization is taken care of in namespace.c
840 static void __follow_mount_rcu(struct nameidata *nd, struct path *path,
841 struct inode **inode)
843 while (d_mountpoint(path->dentry)) {
844 struct vfsmount *mounted;
845 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
849 path->dentry = mounted->mnt_root;
850 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
851 *inode = path->dentry->d_inode;
855 static int __follow_mount(struct path *path)
858 while (d_mountpoint(path->dentry)) {
859 struct vfsmount *mounted = lookup_mnt(path);
866 path->dentry = dget(mounted->mnt_root);
872 static void follow_mount(struct path *path)
874 while (d_mountpoint(path->dentry)) {
875 struct vfsmount *mounted = lookup_mnt(path);
881 path->dentry = dget(mounted->mnt_root);
885 int follow_down(struct path *path)
887 struct vfsmount *mounted;
889 mounted = lookup_mnt(path);
894 path->dentry = dget(mounted->mnt_root);
900 static int follow_dotdot_rcu(struct nameidata *nd)
902 struct inode *inode = nd->inode;
907 if (nd->path.dentry == nd->root.dentry &&
908 nd->path.mnt == nd->root.mnt) {
911 if (nd->path.dentry != nd->path.mnt->mnt_root) {
912 struct dentry *old = nd->path.dentry;
913 struct dentry *parent = old->d_parent;
916 seq = read_seqcount_begin(&parent->d_seq);
917 if (read_seqcount_retry(&old->d_seq, nd->seq))
919 inode = parent->d_inode;
920 nd->path.dentry = parent;
924 if (!follow_up_rcu(&nd->path))
926 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
927 inode = nd->path.dentry->d_inode;
929 __follow_mount_rcu(nd, &nd->path, &inode);
935 static void follow_dotdot(struct nameidata *nd)
940 struct dentry *old = nd->path.dentry;
942 if (nd->path.dentry == nd->root.dentry &&
943 nd->path.mnt == nd->root.mnt) {
946 if (nd->path.dentry != nd->path.mnt->mnt_root) {
947 /* rare case of legitimate dget_parent()... */
948 nd->path.dentry = dget_parent(nd->path.dentry);
952 if (!follow_up(&nd->path))
955 follow_mount(&nd->path);
956 nd->inode = nd->path.dentry->d_inode;
960 * Allocate a dentry with name and parent, and perform a parent
961 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
962 * on error. parent->d_inode->i_mutex must be held. d_lookup must
963 * have verified that no child exists while under i_mutex.
965 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
966 struct qstr *name, struct nameidata *nd)
968 struct inode *inode = parent->d_inode;
969 struct dentry *dentry;
972 /* Don't create child dentry for a dead directory. */
973 if (unlikely(IS_DEADDIR(inode)))
974 return ERR_PTR(-ENOENT);
976 dentry = d_alloc(parent, name);
977 if (unlikely(!dentry))
978 return ERR_PTR(-ENOMEM);
980 old = inode->i_op->lookup(inode, dentry, nd);
989 * It's more convoluted than I'd like it to be, but... it's still fairly
990 * small and for now I'd prefer to have fast path as straight as possible.
991 * It _is_ time-critical.
993 static int do_lookup(struct nameidata *nd, struct qstr *name,
994 struct path *path, struct inode **inode)
996 struct vfsmount *mnt = nd->path.mnt;
997 struct dentry *dentry, *parent = nd->path.dentry;
1000 * See if the low-level filesystem might want
1001 * to use its own hash..
1003 if (parent->d_op && parent->d_op->d_hash) {
1004 int err = parent->d_op->d_hash(parent, nd->inode, name);
1010 * Rename seqlock is not required here because in the off chance
1011 * of a false negative due to a concurrent rename, we're going to
1012 * do the non-racy lookup, below.
1014 if (nd->flags & LOOKUP_RCU) {
1018 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1020 if (nameidata_drop_rcu(nd))
1024 /* Memory barrier in read_seqcount_begin of child is enough */
1025 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1029 if (dentry->d_op && dentry->d_op->d_revalidate) {
1030 /* We commonly drop rcu-walk here */
1031 if (nameidata_dentry_drop_rcu(nd, dentry))
1033 goto need_revalidate;
1036 path->dentry = dentry;
1037 __follow_mount_rcu(nd, path, inode);
1039 dentry = __d_lookup(parent, name);
1043 if (dentry->d_op && dentry->d_op->d_revalidate)
1044 goto need_revalidate;
1047 path->dentry = dentry;
1048 __follow_mount(path);
1049 *inode = path->dentry->d_inode;
1054 dir = parent->d_inode;
1055 BUG_ON(nd->inode != dir);
1057 mutex_lock(&dir->i_mutex);
1059 * First re-do the cached lookup just in case it was created
1060 * while we waited for the directory semaphore, or the first
1061 * lookup failed due to an unrelated rename.
1063 * This could use version numbering or similar to avoid unnecessary
1064 * cache lookups, but then we'd have to do the first lookup in the
1065 * non-racy way. However in the common case here, everything should
1066 * be hot in cache, so would it be a big win?
1068 dentry = d_lookup(parent, name);
1069 if (likely(!dentry)) {
1070 dentry = d_alloc_and_lookup(parent, name, nd);
1071 mutex_unlock(&dir->i_mutex);
1077 * Uhhuh! Nasty case: the cache was re-populated while
1078 * we waited on the semaphore. Need to revalidate.
1080 mutex_unlock(&dir->i_mutex);
1084 dentry = do_revalidate(dentry, nd);
1092 return PTR_ERR(dentry);
1096 * This is a temporary kludge to deal with "automount" symlinks; proper
1097 * solution is to trigger them on follow_mount(), so that do_lookup()
1098 * would DTRT. To be killed before 2.6.34-final.
1100 static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
1102 return inode && unlikely(inode->i_op->follow_link) &&
1103 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
1108 * This is the basic name resolution function, turning a pathname into
1109 * the final dentry. We expect 'base' to be positive and a directory.
1111 * Returns 0 and nd will have valid dentry and mnt on success.
1112 * Returns error and drops reference to input namei data on failure.
1114 static int link_path_walk(const char *name, struct nameidata *nd)
1118 unsigned int lookup_flags = nd->flags;
1126 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1128 /* At this point we know we have a real path component. */
1130 struct inode *inode;
1135 nd->flags |= LOOKUP_CONTINUE;
1136 if (nd->flags & LOOKUP_RCU) {
1137 err = exec_permission_rcu(nd->inode);
1138 if (err == -ECHILD) {
1139 if (nameidata_drop_rcu(nd))
1145 err = exec_permission(nd->inode);
1151 c = *(const unsigned char *)name;
1153 hash = init_name_hash();
1156 hash = partial_name_hash(c, hash);
1157 c = *(const unsigned char *)name;
1158 } while (c && (c != '/'));
1159 this.len = name - (const char *) this.name;
1160 this.hash = end_name_hash(hash);
1162 /* remove trailing slashes? */
1164 goto last_component;
1165 while (*++name == '/');
1167 goto last_with_slashes;
1170 * "." and ".." are special - ".." especially so because it has
1171 * to be able to know about the current root directory and
1172 * parent relationships.
1174 if (this.name[0] == '.') switch (this.len) {
1178 if (this.name[1] != '.')
1180 if (nd->flags & LOOKUP_RCU) {
1181 if (follow_dotdot_rcu(nd))
1189 /* This does the actual lookups.. */
1190 err = do_lookup(nd, &this, &next, &inode);
1197 if (inode->i_op->follow_link) {
1198 /* We commonly drop rcu-walk here */
1199 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1201 BUG_ON(inode != next.dentry->d_inode);
1202 err = do_follow_link(&next, nd);
1205 nd->inode = nd->path.dentry->d_inode;
1210 path_to_nameidata(&next, nd);
1214 if (!nd->inode->i_op->lookup)
1217 /* here ends the main loop */
1220 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1222 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1223 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1224 if (lookup_flags & LOOKUP_PARENT)
1226 if (this.name[0] == '.') switch (this.len) {
1230 if (this.name[1] != '.')
1232 if (nd->flags & LOOKUP_RCU) {
1233 if (follow_dotdot_rcu(nd))
1241 err = do_lookup(nd, &this, &next, &inode);
1244 if (follow_on_final(inode, lookup_flags)) {
1245 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1247 BUG_ON(inode != next.dentry->d_inode);
1248 err = do_follow_link(&next, nd);
1251 nd->inode = nd->path.dentry->d_inode;
1253 path_to_nameidata(&next, nd);
1259 if (lookup_flags & LOOKUP_DIRECTORY) {
1261 if (!nd->inode->i_op->lookup)
1267 nd->last_type = LAST_NORM;
1268 if (this.name[0] != '.')
1271 nd->last_type = LAST_DOT;
1272 else if (this.len == 2 && this.name[1] == '.')
1273 nd->last_type = LAST_DOTDOT;
1278 * We bypassed the ordinary revalidation routines.
1279 * We may need to check the cached dentry for staleness.
1281 if (nd->path.dentry && nd->path.dentry->d_sb &&
1282 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1283 if (nameidata_drop_rcu_maybe(nd))
1286 /* Note: we do not d_invalidate() */
1287 if (!nd->path.dentry->d_op->d_revalidate(
1288 nd->path.dentry, nd))
1292 if (nameidata_drop_rcu_last_maybe(nd))
1296 if (!(nd->flags & LOOKUP_RCU))
1297 path_put_conditional(&next, nd);
1300 if (!(nd->flags & LOOKUP_RCU))
1301 path_put(&nd->path);
1306 static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1308 current->total_link_count = 0;
1310 return link_path_walk(name, nd);
1313 static inline int path_walk_simple(const char *name, struct nameidata *nd)
1315 current->total_link_count = 0;
1317 return link_path_walk(name, nd);
1320 static int path_walk(const char *name, struct nameidata *nd)
1322 struct path save = nd->path;
1325 current->total_link_count = 0;
1327 /* make sure the stuff we saved doesn't go away */
1330 result = link_path_walk(name, nd);
1331 if (result == -ESTALE) {
1332 /* nd->path had been dropped */
1333 current->total_link_count = 0;
1335 path_get(&nd->path);
1336 nd->flags |= LOOKUP_REVAL;
1337 result = link_path_walk(name, nd);
1345 static void path_finish_rcu(struct nameidata *nd)
1347 if (nd->flags & LOOKUP_RCU) {
1348 /* RCU dangling. Cancel it. */
1349 nd->flags &= ~LOOKUP_RCU;
1350 nd->root.mnt = NULL;
1352 br_read_unlock(vfsmount_lock);
1358 static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1364 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1365 nd->flags = flags | LOOKUP_RCU;
1367 nd->root.mnt = NULL;
1371 struct fs_struct *fs = current->fs;
1373 br_read_lock(vfsmount_lock);
1376 spin_lock(&fs->lock);
1377 nd->root = fs->root;
1378 nd->path = nd->root;
1379 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1380 spin_unlock(&fs->lock);
1382 } else if (dfd == AT_FDCWD) {
1383 struct fs_struct *fs = current->fs;
1385 br_read_lock(vfsmount_lock);
1388 spin_lock(&fs->lock);
1390 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1391 spin_unlock(&fs->lock);
1393 struct dentry *dentry;
1395 file = fget_light(dfd, &fput_needed);
1400 dentry = file->f_path.dentry;
1403 if (!S_ISDIR(dentry->d_inode->i_mode))
1406 retval = file_permission(file, MAY_EXEC);
1410 nd->path = file->f_path;
1414 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1415 br_read_lock(vfsmount_lock);
1418 nd->inode = nd->path.dentry->d_inode;
1422 fput_light(file, fput_needed);
1427 static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1433 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1436 nd->root.mnt = NULL;
1440 nd->path = nd->root;
1441 path_get(&nd->root);
1442 } else if (dfd == AT_FDCWD) {
1443 get_fs_pwd(current->fs, &nd->path);
1445 struct dentry *dentry;
1447 file = fget_light(dfd, &fput_needed);
1452 dentry = file->f_path.dentry;
1455 if (!S_ISDIR(dentry->d_inode->i_mode))
1458 retval = file_permission(file, MAY_EXEC);
1462 nd->path = file->f_path;
1463 path_get(&file->f_path);
1465 fput_light(file, fput_needed);
1467 nd->inode = nd->path.dentry->d_inode;
1471 fput_light(file, fput_needed);
1476 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1477 static int do_path_lookup(int dfd, const char *name,
1478 unsigned int flags, struct nameidata *nd)
1483 * Path walking is largely split up into 2 different synchronisation
1484 * schemes, rcu-walk and ref-walk (explained in
1485 * Documentation/filesystems/path-lookup.txt). These share much of the
1486 * path walk code, but some things particularly setup, cleanup, and
1487 * following mounts are sufficiently divergent that functions are
1488 * duplicated. Typically there is a function foo(), and its RCU
1489 * analogue, foo_rcu().
1491 * -ECHILD is the error number of choice (just to avoid clashes) that
1492 * is returned if some aspect of an rcu-walk fails. Such an error must
1493 * be handled by restarting a traditional ref-walk (which will always
1494 * be able to complete).
1496 retval = path_init_rcu(dfd, name, flags, nd);
1497 if (unlikely(retval))
1499 retval = path_walk_rcu(name, nd);
1500 path_finish_rcu(nd);
1502 path_put(&nd->root);
1503 nd->root.mnt = NULL;
1506 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1507 /* slower, locked walk */
1508 if (retval == -ESTALE)
1509 flags |= LOOKUP_REVAL;
1510 retval = path_init(dfd, name, flags, nd);
1511 if (unlikely(retval))
1513 retval = path_walk(name, nd);
1515 path_put(&nd->root);
1516 nd->root.mnt = NULL;
1520 if (likely(!retval)) {
1521 if (unlikely(!audit_dummy_context())) {
1522 if (nd->path.dentry && nd->inode)
1523 audit_inode(name, nd->path.dentry);
1530 int path_lookup(const char *name, unsigned int flags,
1531 struct nameidata *nd)
1533 return do_path_lookup(AT_FDCWD, name, flags, nd);
1536 int kern_path(const char *name, unsigned int flags, struct path *path)
1538 struct nameidata nd;
1539 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1546 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1547 * @dentry: pointer to dentry of the base directory
1548 * @mnt: pointer to vfs mount of the base directory
1549 * @name: pointer to file name
1550 * @flags: lookup flags
1551 * @nd: pointer to nameidata
1553 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1554 const char *name, unsigned int flags,
1555 struct nameidata *nd)
1559 /* same as do_path_lookup */
1560 nd->last_type = LAST_ROOT;
1564 nd->path.dentry = dentry;
1566 path_get(&nd->path);
1567 nd->root = nd->path;
1568 path_get(&nd->root);
1569 nd->inode = nd->path.dentry->d_inode;
1571 retval = path_walk(name, nd);
1572 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
1574 audit_inode(name, nd->path.dentry);
1576 path_put(&nd->root);
1577 nd->root.mnt = NULL;
1582 static struct dentry *__lookup_hash(struct qstr *name,
1583 struct dentry *base, struct nameidata *nd)
1585 struct inode *inode = base->d_inode;
1586 struct dentry *dentry;
1589 err = exec_permission(inode);
1591 return ERR_PTR(err);
1594 * See if the low-level filesystem might want
1595 * to use its own hash..
1597 if (base->d_op && base->d_op->d_hash) {
1598 err = base->d_op->d_hash(base, inode, name);
1599 dentry = ERR_PTR(err);
1605 * Don't bother with __d_lookup: callers are for creat as
1606 * well as unlink, so a lot of the time it would cost
1609 dentry = d_lookup(base, name);
1611 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1612 dentry = do_revalidate(dentry, nd);
1615 dentry = d_alloc_and_lookup(base, name, nd);
1621 * Restricted form of lookup. Doesn't follow links, single-component only,
1622 * needs parent already locked. Doesn't follow mounts.
1625 static struct dentry *lookup_hash(struct nameidata *nd)
1627 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1630 static int __lookup_one_len(const char *name, struct qstr *this,
1631 struct dentry *base, int len)
1641 hash = init_name_hash();
1643 c = *(const unsigned char *)name++;
1644 if (c == '/' || c == '\0')
1646 hash = partial_name_hash(c, hash);
1648 this->hash = end_name_hash(hash);
1653 * lookup_one_len - filesystem helper to lookup single pathname component
1654 * @name: pathname component to lookup
1655 * @base: base directory to lookup from
1656 * @len: maximum length @len should be interpreted to
1658 * Note that this routine is purely a helper for filesystem usage and should
1659 * not be called by generic code. Also note that by using this function the
1660 * nameidata argument is passed to the filesystem methods and a filesystem
1661 * using this helper needs to be prepared for that.
1663 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1668 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1670 err = __lookup_one_len(name, &this, base, len);
1672 return ERR_PTR(err);
1674 return __lookup_hash(&this, base, NULL);
1677 int user_path_at(int dfd, const char __user *name, unsigned flags,
1680 struct nameidata nd;
1681 char *tmp = getname(name);
1682 int err = PTR_ERR(tmp);
1685 BUG_ON(flags & LOOKUP_PARENT);
1687 err = do_path_lookup(dfd, tmp, flags, &nd);
1695 static int user_path_parent(int dfd, const char __user *path,
1696 struct nameidata *nd, char **name)
1698 char *s = getname(path);
1704 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1714 * It's inline, so penalty for filesystems that don't use sticky bit is
1717 static inline int check_sticky(struct inode *dir, struct inode *inode)
1719 uid_t fsuid = current_fsuid();
1721 if (!(dir->i_mode & S_ISVTX))
1723 if (inode->i_uid == fsuid)
1725 if (dir->i_uid == fsuid)
1727 return !capable(CAP_FOWNER);
1731 * Check whether we can remove a link victim from directory dir, check
1732 * whether the type of victim is right.
1733 * 1. We can't do it if dir is read-only (done in permission())
1734 * 2. We should have write and exec permissions on dir
1735 * 3. We can't remove anything from append-only dir
1736 * 4. We can't do anything with immutable dir (done in permission())
1737 * 5. If the sticky bit on dir is set we should either
1738 * a. be owner of dir, or
1739 * b. be owner of victim, or
1740 * c. have CAP_FOWNER capability
1741 * 6. If the victim is append-only or immutable we can't do antyhing with
1742 * links pointing to it.
1743 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1744 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1745 * 9. We can't remove a root or mountpoint.
1746 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1747 * nfs_async_unlink().
1749 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1753 if (!victim->d_inode)
1756 BUG_ON(victim->d_parent->d_inode != dir);
1757 audit_inode_child(victim, dir);
1759 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1764 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1765 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1768 if (!S_ISDIR(victim->d_inode->i_mode))
1770 if (IS_ROOT(victim))
1772 } else if (S_ISDIR(victim->d_inode->i_mode))
1774 if (IS_DEADDIR(dir))
1776 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1781 /* Check whether we can create an object with dentry child in directory
1783 * 1. We can't do it if child already exists (open has special treatment for
1784 * this case, but since we are inlined it's OK)
1785 * 2. We can't do it if dir is read-only (done in permission())
1786 * 3. We should have write and exec permissions on dir
1787 * 4. We can't do it if dir is immutable (done in permission())
1789 static inline int may_create(struct inode *dir, struct dentry *child)
1793 if (IS_DEADDIR(dir))
1795 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1799 * p1 and p2 should be directories on the same fs.
1801 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1806 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1810 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1812 p = d_ancestor(p2, p1);
1814 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1815 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1819 p = d_ancestor(p1, p2);
1821 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1822 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1826 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1827 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1831 void unlock_rename(struct dentry *p1, struct dentry *p2)
1833 mutex_unlock(&p1->d_inode->i_mutex);
1835 mutex_unlock(&p2->d_inode->i_mutex);
1836 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1840 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1841 struct nameidata *nd)
1843 int error = may_create(dir, dentry);
1848 if (!dir->i_op->create)
1849 return -EACCES; /* shouldn't it be ENOSYS? */
1852 error = security_inode_create(dir, dentry, mode);
1855 error = dir->i_op->create(dir, dentry, mode, nd);
1857 fsnotify_create(dir, dentry);
1861 int may_open(struct path *path, int acc_mode, int flag)
1863 struct dentry *dentry = path->dentry;
1864 struct inode *inode = dentry->d_inode;
1870 switch (inode->i_mode & S_IFMT) {
1874 if (acc_mode & MAY_WRITE)
1879 if (path->mnt->mnt_flags & MNT_NODEV)
1888 error = inode_permission(inode, acc_mode);
1893 * An append-only file must be opened in append mode for writing.
1895 if (IS_APPEND(inode)) {
1896 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1902 /* O_NOATIME can only be set by the owner or superuser */
1903 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1907 * Ensure there are no outstanding leases on the file.
1909 return break_lease(inode, flag);
1912 static int handle_truncate(struct path *path)
1914 struct inode *inode = path->dentry->d_inode;
1915 int error = get_write_access(inode);
1919 * Refuse to truncate files with mandatory locks held on them.
1921 error = locks_verify_locked(inode);
1923 error = security_path_truncate(path);
1925 error = do_truncate(path->dentry, 0,
1926 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1929 put_write_access(inode);
1934 * Be careful about ever adding any more callers of this
1935 * function. Its flags must be in the namei format, not
1936 * what get passed to sys_open().
1938 static int __open_namei_create(struct nameidata *nd, struct path *path,
1939 int open_flag, int mode)
1942 struct dentry *dir = nd->path.dentry;
1944 if (!IS_POSIXACL(dir->d_inode))
1945 mode &= ~current_umask();
1946 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1949 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1951 mutex_unlock(&dir->d_inode->i_mutex);
1952 dput(nd->path.dentry);
1953 nd->path.dentry = path->dentry;
1957 /* Don't check for write permission, don't truncate */
1958 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
1962 * Note that while the flag value (low two bits) for sys_open means:
1967 * it is changed into
1968 * 00 - no permissions needed
1969 * 01 - read-permission
1970 * 10 - write-permission
1972 * for the internal routines (ie open_namei()/follow_link() etc)
1973 * This is more logical, and also allows the 00 "no perm needed"
1974 * to be used for symlinks (where the permissions are checked
1978 static inline int open_to_namei_flags(int flag)
1980 if ((flag+1) & O_ACCMODE)
1985 static int open_will_truncate(int flag, struct inode *inode)
1988 * We'll never write to the fs underlying
1991 if (special_file(inode->i_mode))
1993 return (flag & O_TRUNC);
1996 static struct file *finish_open(struct nameidata *nd,
1997 int open_flag, int acc_mode)
2003 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
2004 if (will_truncate) {
2005 error = mnt_want_write(nd->path.mnt);
2009 error = may_open(&nd->path, acc_mode, open_flag);
2012 mnt_drop_write(nd->path.mnt);
2015 filp = nameidata_to_filp(nd);
2016 if (!IS_ERR(filp)) {
2017 error = ima_file_check(filp, acc_mode);
2020 filp = ERR_PTR(error);
2023 if (!IS_ERR(filp)) {
2024 if (will_truncate) {
2025 error = handle_truncate(&nd->path);
2028 filp = ERR_PTR(error);
2033 * It is now safe to drop the mnt write
2034 * because the filp has had a write taken
2038 mnt_drop_write(nd->path.mnt);
2039 path_put(&nd->path);
2043 if (!IS_ERR(nd->intent.open.file))
2044 release_open_intent(nd);
2045 path_put(&nd->path);
2046 return ERR_PTR(error);
2050 * Handle O_CREAT case for do_filp_open
2052 static struct file *do_last(struct nameidata *nd, struct path *path,
2053 int open_flag, int acc_mode,
2054 int mode, const char *pathname)
2056 struct dentry *dir = nd->path.dentry;
2058 int error = -EISDIR;
2060 switch (nd->last_type) {
2063 dir = nd->path.dentry;
2065 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
2066 if (!dir->d_op->d_revalidate(dir, nd)) {
2075 audit_inode(pathname, dir);
2079 /* trailing slashes? */
2080 if (nd->last.name[nd->last.len])
2083 mutex_lock(&dir->d_inode->i_mutex);
2085 path->dentry = lookup_hash(nd);
2086 path->mnt = nd->path.mnt;
2088 error = PTR_ERR(path->dentry);
2089 if (IS_ERR(path->dentry)) {
2090 mutex_unlock(&dir->d_inode->i_mutex);
2094 if (IS_ERR(nd->intent.open.file)) {
2095 error = PTR_ERR(nd->intent.open.file);
2096 goto exit_mutex_unlock;
2099 /* Negative dentry, just create the file */
2100 if (!path->dentry->d_inode) {
2102 * This write is needed to ensure that a
2103 * ro->rw transition does not occur between
2104 * the time when the file is created and when
2105 * a permanent write count is taken through
2106 * the 'struct file' in nameidata_to_filp().
2108 error = mnt_want_write(nd->path.mnt);
2110 goto exit_mutex_unlock;
2111 error = __open_namei_create(nd, path, open_flag, mode);
2113 mnt_drop_write(nd->path.mnt);
2116 filp = nameidata_to_filp(nd);
2117 mnt_drop_write(nd->path.mnt);
2118 path_put(&nd->path);
2119 if (!IS_ERR(filp)) {
2120 error = ima_file_check(filp, acc_mode);
2123 filp = ERR_PTR(error);
2130 * It already exists.
2132 mutex_unlock(&dir->d_inode->i_mutex);
2133 audit_inode(pathname, path->dentry);
2136 if (open_flag & O_EXCL)
2139 if (__follow_mount(path)) {
2141 if (open_flag & O_NOFOLLOW)
2146 if (!path->dentry->d_inode)
2149 if (path->dentry->d_inode->i_op->follow_link)
2152 path_to_nameidata(path, nd);
2153 nd->inode = path->dentry->d_inode;
2155 if (S_ISDIR(nd->inode->i_mode))
2158 filp = finish_open(nd, open_flag, acc_mode);
2162 mutex_unlock(&dir->d_inode->i_mutex);
2164 path_put_conditional(path, nd);
2166 if (!IS_ERR(nd->intent.open.file))
2167 release_open_intent(nd);
2168 path_put(&nd->path);
2169 return ERR_PTR(error);
2173 * Note that the low bits of the passed in "open_flag"
2174 * are not the same as in the local variable "flag". See
2175 * open_to_namei_flags() for more details.
2177 struct file *do_filp_open(int dfd, const char *pathname,
2178 int open_flag, int mode, int acc_mode)
2181 struct nameidata nd;
2185 int flag = open_to_namei_flags(open_flag);
2188 if (!(open_flag & O_CREAT))
2191 /* Must never be set by userspace */
2192 open_flag &= ~FMODE_NONOTIFY;
2195 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2196 * check for O_DSYNC if the need any syncing at all we enforce it's
2197 * always set instead of having to deal with possibly weird behaviour
2198 * for malicious applications setting only __O_SYNC.
2200 if (open_flag & __O_SYNC)
2201 open_flag |= O_DSYNC;
2204 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
2206 /* O_TRUNC implies we need access checks for write permissions */
2207 if (open_flag & O_TRUNC)
2208 acc_mode |= MAY_WRITE;
2210 /* Allow the LSM permission hook to distinguish append
2211 access from general write access. */
2212 if (open_flag & O_APPEND)
2213 acc_mode |= MAY_APPEND;
2215 flags = LOOKUP_OPEN;
2216 if (open_flag & O_CREAT) {
2217 flags |= LOOKUP_CREATE;
2218 if (open_flag & O_EXCL)
2219 flags |= LOOKUP_EXCL;
2221 if (open_flag & O_DIRECTORY)
2222 flags |= LOOKUP_DIRECTORY;
2223 if (!(open_flag & O_NOFOLLOW))
2224 flags |= LOOKUP_FOLLOW;
2226 filp = get_empty_filp();
2228 return ERR_PTR(-ENFILE);
2230 filp->f_flags = open_flag;
2231 nd.intent.open.file = filp;
2232 nd.intent.open.flags = flag;
2233 nd.intent.open.create_mode = mode;
2235 if (open_flag & O_CREAT)
2238 /* !O_CREAT, simple open */
2239 error = do_path_lookup(dfd, pathname, flags, &nd);
2240 if (unlikely(error))
2243 if (!(nd.flags & LOOKUP_FOLLOW)) {
2244 if (nd.inode->i_op->follow_link)
2248 if (nd.flags & LOOKUP_DIRECTORY) {
2249 if (!nd.inode->i_op->lookup)
2252 audit_inode(pathname, nd.path.dentry);
2253 filp = finish_open(&nd, open_flag, acc_mode);
2257 /* OK, have to create the file. Find the parent. */
2258 error = path_init_rcu(dfd, pathname,
2259 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2262 error = path_walk_rcu(pathname, &nd);
2263 path_finish_rcu(&nd);
2264 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2265 /* slower, locked walk */
2266 if (error == -ESTALE) {
2268 flags |= LOOKUP_REVAL;
2270 error = path_init(dfd, pathname,
2271 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2275 error = path_walk_simple(pathname, &nd);
2277 if (unlikely(error))
2279 if (unlikely(!audit_dummy_context()))
2280 audit_inode(pathname, nd.path.dentry);
2283 * We have the parent and last component.
2286 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2287 while (unlikely(!filp)) { /* trailing symlink */
2291 /* S_ISDIR part is a temporary automount kludge */
2292 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(nd.inode->i_mode))
2297 * This is subtle. Instead of calling do_follow_link() we do
2298 * the thing by hands. The reason is that this way we have zero
2299 * link_count and path_walk() (called from ->follow_link)
2300 * honoring LOOKUP_PARENT. After that we have the parent and
2301 * last component, i.e. we are in the same situation as after
2302 * the first path_walk(). Well, almost - if the last component
2303 * is normal we get its copy stored in nd->last.name and we will
2304 * have to putname() it when we are done. Procfs-like symlinks
2305 * just set LAST_BIND.
2307 nd.flags |= LOOKUP_PARENT;
2308 error = security_inode_follow_link(path.dentry, &nd);
2311 error = __do_follow_link(&path, &nd, &cookie);
2312 if (unlikely(error)) {
2313 if (!IS_ERR(cookie) && nd.inode->i_op->put_link)
2314 nd.inode->i_op->put_link(path.dentry, &nd, cookie);
2315 /* nd.path had been dropped */
2320 nd.flags &= ~LOOKUP_PARENT;
2321 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
2322 if (nd.inode->i_op->put_link)
2323 nd.inode->i_op->put_link(holder.dentry, &nd, cookie);
2329 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
2334 path_put_conditional(&path, &nd);
2338 if (!IS_ERR(nd.intent.open.file))
2339 release_open_intent(&nd);
2340 filp = ERR_PTR(error);
2345 * filp_open - open file and return file pointer
2347 * @filename: path to open
2348 * @flags: open flags as per the open(2) second argument
2349 * @mode: mode for the new file if O_CREAT is set, else ignored
2351 * This is the helper to open a file from kernelspace if you really
2352 * have to. But in generally you should not do this, so please move
2353 * along, nothing to see here..
2355 struct file *filp_open(const char *filename, int flags, int mode)
2357 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
2359 EXPORT_SYMBOL(filp_open);
2362 * lookup_create - lookup a dentry, creating it if it doesn't exist
2363 * @nd: nameidata info
2364 * @is_dir: directory flag
2366 * Simple function to lookup and return a dentry and create it
2367 * if it doesn't exist. Is SMP-safe.
2369 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2371 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2373 struct dentry *dentry = ERR_PTR(-EEXIST);
2375 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2377 * Yucky last component or no last component at all?
2378 * (foo/., foo/.., /////)
2380 if (nd->last_type != LAST_NORM)
2382 nd->flags &= ~LOOKUP_PARENT;
2383 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2384 nd->intent.open.flags = O_EXCL;
2387 * Do the final lookup.
2389 dentry = lookup_hash(nd);
2393 if (dentry->d_inode)
2396 * Special case - lookup gave negative, but... we had foo/bar/
2397 * From the vfs_mknod() POV we just have a negative dentry -
2398 * all is fine. Let's be bastards - you had / on the end, you've
2399 * been asking for (non-existent) directory. -ENOENT for you.
2401 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2403 dentry = ERR_PTR(-ENOENT);
2408 dentry = ERR_PTR(-EEXIST);
2412 EXPORT_SYMBOL_GPL(lookup_create);
2414 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2416 int error = may_create(dir, dentry);
2421 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2424 if (!dir->i_op->mknod)
2427 error = devcgroup_inode_mknod(mode, dev);
2431 error = security_inode_mknod(dir, dentry, mode, dev);
2435 error = dir->i_op->mknod(dir, dentry, mode, dev);
2437 fsnotify_create(dir, dentry);
2441 static int may_mknod(mode_t mode)
2443 switch (mode & S_IFMT) {
2449 case 0: /* zero mode translates to S_IFREG */
2458 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2463 struct dentry *dentry;
2464 struct nameidata nd;
2469 error = user_path_parent(dfd, filename, &nd, &tmp);
2473 dentry = lookup_create(&nd, 0);
2474 if (IS_ERR(dentry)) {
2475 error = PTR_ERR(dentry);
2478 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2479 mode &= ~current_umask();
2480 error = may_mknod(mode);
2483 error = mnt_want_write(nd.path.mnt);
2486 error = security_path_mknod(&nd.path, dentry, mode, dev);
2488 goto out_drop_write;
2489 switch (mode & S_IFMT) {
2490 case 0: case S_IFREG:
2491 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
2493 case S_IFCHR: case S_IFBLK:
2494 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
2495 new_decode_dev(dev));
2497 case S_IFIFO: case S_IFSOCK:
2498 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
2502 mnt_drop_write(nd.path.mnt);
2506 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2513 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2515 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2518 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2520 int error = may_create(dir, dentry);
2525 if (!dir->i_op->mkdir)
2528 mode &= (S_IRWXUGO|S_ISVTX);
2529 error = security_inode_mkdir(dir, dentry, mode);
2533 error = dir->i_op->mkdir(dir, dentry, mode);
2535 fsnotify_mkdir(dir, dentry);
2539 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2543 struct dentry *dentry;
2544 struct nameidata nd;
2546 error = user_path_parent(dfd, pathname, &nd, &tmp);
2550 dentry = lookup_create(&nd, 1);
2551 error = PTR_ERR(dentry);
2555 if (!IS_POSIXACL(nd.path.dentry->d_inode))
2556 mode &= ~current_umask();
2557 error = mnt_want_write(nd.path.mnt);
2560 error = security_path_mkdir(&nd.path, dentry, mode);
2562 goto out_drop_write;
2563 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
2565 mnt_drop_write(nd.path.mnt);
2569 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2576 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2578 return sys_mkdirat(AT_FDCWD, pathname, mode);
2582 * We try to drop the dentry early: we should have
2583 * a usage count of 2 if we're the only user of this
2584 * dentry, and if that is true (possibly after pruning
2585 * the dcache), then we drop the dentry now.
2587 * A low-level filesystem can, if it choses, legally
2590 * if (!d_unhashed(dentry))
2593 * if it cannot handle the case of removing a directory
2594 * that is still in use by something else..
2596 void dentry_unhash(struct dentry *dentry)
2599 shrink_dcache_parent(dentry);
2600 spin_lock(&dentry->d_lock);
2601 if (dentry->d_count == 2)
2603 spin_unlock(&dentry->d_lock);
2606 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2608 int error = may_delete(dir, dentry, 1);
2613 if (!dir->i_op->rmdir)
2616 mutex_lock(&dentry->d_inode->i_mutex);
2617 dentry_unhash(dentry);
2618 if (d_mountpoint(dentry))
2621 error = security_inode_rmdir(dir, dentry);
2623 error = dir->i_op->rmdir(dir, dentry);
2625 dentry->d_inode->i_flags |= S_DEAD;
2630 mutex_unlock(&dentry->d_inode->i_mutex);
2639 static long do_rmdir(int dfd, const char __user *pathname)
2643 struct dentry *dentry;
2644 struct nameidata nd;
2646 error = user_path_parent(dfd, pathname, &nd, &name);
2650 switch(nd.last_type) {
2662 nd.flags &= ~LOOKUP_PARENT;
2664 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2665 dentry = lookup_hash(&nd);
2666 error = PTR_ERR(dentry);
2669 error = mnt_want_write(nd.path.mnt);
2672 error = security_path_rmdir(&nd.path, dentry);
2675 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2677 mnt_drop_write(nd.path.mnt);
2681 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2688 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2690 return do_rmdir(AT_FDCWD, pathname);
2693 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2695 int error = may_delete(dir, dentry, 0);
2700 if (!dir->i_op->unlink)
2703 mutex_lock(&dentry->d_inode->i_mutex);
2704 if (d_mountpoint(dentry))
2707 error = security_inode_unlink(dir, dentry);
2709 error = dir->i_op->unlink(dir, dentry);
2714 mutex_unlock(&dentry->d_inode->i_mutex);
2716 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2717 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2718 fsnotify_link_count(dentry->d_inode);
2726 * Make sure that the actual truncation of the file will occur outside its
2727 * directory's i_mutex. Truncate can take a long time if there is a lot of
2728 * writeout happening, and we don't want to prevent access to the directory
2729 * while waiting on the I/O.
2731 static long do_unlinkat(int dfd, const char __user *pathname)
2735 struct dentry *dentry;
2736 struct nameidata nd;
2737 struct inode *inode = NULL;
2739 error = user_path_parent(dfd, pathname, &nd, &name);
2744 if (nd.last_type != LAST_NORM)
2747 nd.flags &= ~LOOKUP_PARENT;
2749 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2750 dentry = lookup_hash(&nd);
2751 error = PTR_ERR(dentry);
2752 if (!IS_ERR(dentry)) {
2753 /* Why not before? Because we want correct error value */
2754 if (nd.last.name[nd.last.len])
2756 inode = dentry->d_inode;
2759 error = mnt_want_write(nd.path.mnt);
2762 error = security_path_unlink(&nd.path, dentry);
2765 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2767 mnt_drop_write(nd.path.mnt);
2771 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2773 iput(inode); /* truncate the inode here */
2780 error = !dentry->d_inode ? -ENOENT :
2781 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2785 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2787 if ((flag & ~AT_REMOVEDIR) != 0)
2790 if (flag & AT_REMOVEDIR)
2791 return do_rmdir(dfd, pathname);
2793 return do_unlinkat(dfd, pathname);
2796 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2798 return do_unlinkat(AT_FDCWD, pathname);
2801 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2803 int error = may_create(dir, dentry);
2808 if (!dir->i_op->symlink)
2811 error = security_inode_symlink(dir, dentry, oldname);
2815 error = dir->i_op->symlink(dir, dentry, oldname);
2817 fsnotify_create(dir, dentry);
2821 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2822 int, newdfd, const char __user *, newname)
2827 struct dentry *dentry;
2828 struct nameidata nd;
2830 from = getname(oldname);
2832 return PTR_ERR(from);
2834 error = user_path_parent(newdfd, newname, &nd, &to);
2838 dentry = lookup_create(&nd, 0);
2839 error = PTR_ERR(dentry);
2843 error = mnt_want_write(nd.path.mnt);
2846 error = security_path_symlink(&nd.path, dentry, from);
2848 goto out_drop_write;
2849 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
2851 mnt_drop_write(nd.path.mnt);
2855 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2863 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2865 return sys_symlinkat(oldname, AT_FDCWD, newname);
2868 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2870 struct inode *inode = old_dentry->d_inode;
2876 error = may_create(dir, new_dentry);
2880 if (dir->i_sb != inode->i_sb)
2884 * A link to an append-only or immutable file cannot be created.
2886 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2888 if (!dir->i_op->link)
2890 if (S_ISDIR(inode->i_mode))
2893 error = security_inode_link(old_dentry, dir, new_dentry);
2897 mutex_lock(&inode->i_mutex);
2898 error = dir->i_op->link(old_dentry, dir, new_dentry);
2899 mutex_unlock(&inode->i_mutex);
2901 fsnotify_link(dir, inode, new_dentry);
2906 * Hardlinks are often used in delicate situations. We avoid
2907 * security-related surprises by not following symlinks on the
2910 * We don't follow them on the oldname either to be compatible
2911 * with linux 2.0, and to avoid hard-linking to directories
2912 * and other special files. --ADM
2914 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2915 int, newdfd, const char __user *, newname, int, flags)
2917 struct dentry *new_dentry;
2918 struct nameidata nd;
2919 struct path old_path;
2923 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2926 error = user_path_at(olddfd, oldname,
2927 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2932 error = user_path_parent(newdfd, newname, &nd, &to);
2936 if (old_path.mnt != nd.path.mnt)
2938 new_dentry = lookup_create(&nd, 0);
2939 error = PTR_ERR(new_dentry);
2940 if (IS_ERR(new_dentry))
2942 error = mnt_want_write(nd.path.mnt);
2945 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2947 goto out_drop_write;
2948 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
2950 mnt_drop_write(nd.path.mnt);
2954 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2959 path_put(&old_path);
2964 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2966 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2970 * The worst of all namespace operations - renaming directory. "Perverted"
2971 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2973 * a) we can get into loop creation. Check is done in is_subdir().
2974 * b) race potential - two innocent renames can create a loop together.
2975 * That's where 4.4 screws up. Current fix: serialization on
2976 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2978 * c) we have to lock _three_ objects - parents and victim (if it exists).
2979 * And that - after we got ->i_mutex on parents (until then we don't know
2980 * whether the target exists). Solution: try to be smart with locking
2981 * order for inodes. We rely on the fact that tree topology may change
2982 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2983 * move will be locked. Thus we can rank directories by the tree
2984 * (ancestors first) and rank all non-directories after them.
2985 * That works since everybody except rename does "lock parent, lookup,
2986 * lock child" and rename is under ->s_vfs_rename_mutex.
2987 * HOWEVER, it relies on the assumption that any object with ->lookup()
2988 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2989 * we'd better make sure that there's no link(2) for them.
2990 * d) some filesystems don't support opened-but-unlinked directories,
2991 * either because of layout or because they are not ready to deal with
2992 * all cases correctly. The latter will be fixed (taking this sort of
2993 * stuff into VFS), but the former is not going away. Solution: the same
2994 * trick as in rmdir().
2995 * e) conversion from fhandle to dentry may come in the wrong moment - when
2996 * we are removing the target. Solution: we will have to grab ->i_mutex
2997 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2998 * ->i_mutex on parents, which works but leads to some truly excessive
3001 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3002 struct inode *new_dir, struct dentry *new_dentry)
3005 struct inode *target;
3008 * If we are going to change the parent - check write permissions,
3009 * we'll need to flip '..'.
3011 if (new_dir != old_dir) {
3012 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3017 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3021 target = new_dentry->d_inode;
3023 mutex_lock(&target->i_mutex);
3024 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3028 dentry_unhash(new_dentry);
3029 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3033 target->i_flags |= S_DEAD;
3034 dont_mount(new_dentry);
3036 mutex_unlock(&target->i_mutex);
3037 if (d_unhashed(new_dentry))
3038 d_rehash(new_dentry);
3042 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3043 d_move(old_dentry,new_dentry);
3047 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3048 struct inode *new_dir, struct dentry *new_dentry)
3050 struct inode *target;
3053 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3058 target = new_dentry->d_inode;
3060 mutex_lock(&target->i_mutex);
3061 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3064 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3067 dont_mount(new_dentry);
3068 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3069 d_move(old_dentry, new_dentry);
3072 mutex_unlock(&target->i_mutex);
3077 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3078 struct inode *new_dir, struct dentry *new_dentry)
3081 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3082 const unsigned char *old_name;
3084 if (old_dentry->d_inode == new_dentry->d_inode)
3087 error = may_delete(old_dir, old_dentry, is_dir);
3091 if (!new_dentry->d_inode)
3092 error = may_create(new_dir, new_dentry);
3094 error = may_delete(new_dir, new_dentry, is_dir);
3098 if (!old_dir->i_op->rename)
3101 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3104 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3106 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3108 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3109 new_dentry->d_inode, old_dentry);
3110 fsnotify_oldname_free(old_name);
3115 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3116 int, newdfd, const char __user *, newname)
3118 struct dentry *old_dir, *new_dir;
3119 struct dentry *old_dentry, *new_dentry;
3120 struct dentry *trap;
3121 struct nameidata oldnd, newnd;
3126 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3130 error = user_path_parent(newdfd, newname, &newnd, &to);
3135 if (oldnd.path.mnt != newnd.path.mnt)
3138 old_dir = oldnd.path.dentry;
3140 if (oldnd.last_type != LAST_NORM)
3143 new_dir = newnd.path.dentry;
3144 if (newnd.last_type != LAST_NORM)
3147 oldnd.flags &= ~LOOKUP_PARENT;
3148 newnd.flags &= ~LOOKUP_PARENT;
3149 newnd.flags |= LOOKUP_RENAME_TARGET;
3151 trap = lock_rename(new_dir, old_dir);
3153 old_dentry = lookup_hash(&oldnd);
3154 error = PTR_ERR(old_dentry);
3155 if (IS_ERR(old_dentry))
3157 /* source must exist */
3159 if (!old_dentry->d_inode)
3161 /* unless the source is a directory trailing slashes give -ENOTDIR */
3162 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3164 if (oldnd.last.name[oldnd.last.len])
3166 if (newnd.last.name[newnd.last.len])
3169 /* source should not be ancestor of target */
3171 if (old_dentry == trap)
3173 new_dentry = lookup_hash(&newnd);
3174 error = PTR_ERR(new_dentry);
3175 if (IS_ERR(new_dentry))
3177 /* target should not be an ancestor of source */
3179 if (new_dentry == trap)
3182 error = mnt_want_write(oldnd.path.mnt);
3185 error = security_path_rename(&oldnd.path, old_dentry,
3186 &newnd.path, new_dentry);
3189 error = vfs_rename(old_dir->d_inode, old_dentry,
3190 new_dir->d_inode, new_dentry);
3192 mnt_drop_write(oldnd.path.mnt);
3198 unlock_rename(new_dir, old_dir);
3200 path_put(&newnd.path);
3203 path_put(&oldnd.path);
3209 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3211 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3214 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3218 len = PTR_ERR(link);
3223 if (len > (unsigned) buflen)
3225 if (copy_to_user(buffer, link, len))
3232 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3233 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3234 * using) it for any given inode is up to filesystem.
3236 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3238 struct nameidata nd;
3243 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3245 return PTR_ERR(cookie);
3247 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3248 if (dentry->d_inode->i_op->put_link)
3249 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3253 int vfs_follow_link(struct nameidata *nd, const char *link)
3255 return __vfs_follow_link(nd, link);
3258 /* get the link contents into pagecache */
3259 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3263 struct address_space *mapping = dentry->d_inode->i_mapping;
3264 page = read_mapping_page(mapping, 0, NULL);
3269 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3273 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3275 struct page *page = NULL;
3276 char *s = page_getlink(dentry, &page);
3277 int res = vfs_readlink(dentry,buffer,buflen,s);
3280 page_cache_release(page);
3285 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3287 struct page *page = NULL;
3288 nd_set_link(nd, page_getlink(dentry, &page));
3292 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3294 struct page *page = cookie;
3298 page_cache_release(page);
3303 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3305 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3307 struct address_space *mapping = inode->i_mapping;
3312 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3314 flags |= AOP_FLAG_NOFS;
3317 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3318 flags, &page, &fsdata);
3322 kaddr = kmap_atomic(page, KM_USER0);
3323 memcpy(kaddr, symname, len-1);
3324 kunmap_atomic(kaddr, KM_USER0);
3326 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3333 mark_inode_dirty(inode);
3339 int page_symlink(struct inode *inode, const char *symname, int len)
3341 return __page_symlink(inode, symname, len,
3342 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3345 const struct inode_operations page_symlink_inode_operations = {
3346 .readlink = generic_readlink,
3347 .follow_link = page_follow_link_light,
3348 .put_link = page_put_link,
3351 EXPORT_SYMBOL(user_path_at);
3352 EXPORT_SYMBOL(follow_down);
3353 EXPORT_SYMBOL(follow_up);
3354 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3355 EXPORT_SYMBOL(getname);
3356 EXPORT_SYMBOL(lock_rename);
3357 EXPORT_SYMBOL(lookup_one_len);
3358 EXPORT_SYMBOL(page_follow_link_light);
3359 EXPORT_SYMBOL(page_put_link);
3360 EXPORT_SYMBOL(page_readlink);
3361 EXPORT_SYMBOL(__page_symlink);
3362 EXPORT_SYMBOL(page_symlink);
3363 EXPORT_SYMBOL(page_symlink_inode_operations);
3364 EXPORT_SYMBOL(path_lookup);
3365 EXPORT_SYMBOL(kern_path);
3366 EXPORT_SYMBOL(vfs_path_lookup);
3367 EXPORT_SYMBOL(inode_permission);
3368 EXPORT_SYMBOL(file_permission);
3369 EXPORT_SYMBOL(unlock_rename);
3370 EXPORT_SYMBOL(vfs_create);
3371 EXPORT_SYMBOL(vfs_follow_link);
3372 EXPORT_SYMBOL(vfs_link);
3373 EXPORT_SYMBOL(vfs_mkdir);
3374 EXPORT_SYMBOL(vfs_mknod);
3375 EXPORT_SYMBOL(generic_permission);
3376 EXPORT_SYMBOL(vfs_readlink);
3377 EXPORT_SYMBOL(vfs_rename);
3378 EXPORT_SYMBOL(vfs_rmdir);
3379 EXPORT_SYMBOL(vfs_symlink);
3380 EXPORT_SYMBOL(vfs_unlink);
3381 EXPORT_SYMBOL(dentry_unhash);
3382 EXPORT_SYMBOL(generic_readlink);