2 * Copyright (C) Neil Brown 2002
3 * Copyright (C) Christoph Hellwig 2007
5 * This file contains the code mapping from inodes to NFS file handles,
6 * and for mapping back from file handles to dentries.
8 * For details on why we do all the strange and hairy things in here
9 * take a look at Documentation/filesystems/nfs/Exporting.
11 #include <linux/exportfs.h>
13 #include <linux/file.h>
14 #include <linux/module.h>
15 #include <linux/mount.h>
16 #include <linux/namei.h>
17 #include <linux/sched.h>
19 #define dprintk(fmt, args...) do{}while(0)
22 static int get_name(const struct path *path, char *name, struct dentry *child);
25 static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
26 char *name, struct dentry *child)
28 const struct export_operations *nop = dir->d_sb->s_export_op;
29 struct path path = {.mnt = mnt, .dentry = dir};
32 return nop->get_name(dir, name, child);
34 return get_name(&path, name, child);
38 * Check if the dentry or any of it's aliases is acceptable.
40 static struct dentry *
41 find_acceptable_alias(struct dentry *result,
42 int (*acceptable)(void *context, struct dentry *dentry),
45 struct dentry *dentry, *toput = NULL;
48 if (acceptable(context, result))
51 inode = result->d_inode;
52 spin_lock(&inode->i_lock);
53 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
55 spin_unlock(&inode->i_lock);
58 if (dentry != result && acceptable(context, dentry)) {
62 spin_lock(&inode->i_lock);
65 spin_unlock(&inode->i_lock);
73 * Find root of a disconnected subtree and return a reference to it.
75 static struct dentry *
76 find_disconnected_root(struct dentry *dentry)
79 while (!IS_ROOT(dentry)) {
80 struct dentry *parent = dget_parent(dentry);
82 if (!(parent->d_flags & DCACHE_DISCONNECTED)) {
93 static bool dentry_connected(struct dentry *dentry)
96 while (dentry->d_flags & DCACHE_DISCONNECTED) {
97 struct dentry *parent = dget_parent(dentry);
100 if (IS_ROOT(dentry)) {
110 static void clear_disconnected(struct dentry *dentry)
113 while (dentry->d_flags & DCACHE_DISCONNECTED) {
114 struct dentry *parent = dget_parent(dentry);
116 WARN_ON_ONCE(IS_ROOT(dentry));
118 spin_lock(&dentry->d_lock);
119 dentry->d_flags &= ~DCACHE_DISCONNECTED;
120 spin_unlock(&dentry->d_lock);
129 * Reconnect a directory dentry with its parent.
131 * This can return a dentry, or NULL, or an error.
133 * In the first case the returned dentry is the parent of the given
134 * dentry, and may itself need to be reconnected to its parent.
136 * In the NULL case, a concurrent VFS operation has either renamed or
137 * removed this directory. The concurrent operation has reconnected our
138 * dentry, so we no longer need to.
140 static struct dentry *reconnect_one(struct vfsmount *mnt,
141 struct dentry *dentry, char *nbuf)
143 struct dentry *parent;
147 parent = ERR_PTR(-EACCES);
148 mutex_lock(&dentry->d_inode->i_mutex);
149 if (mnt->mnt_sb->s_export_op->get_parent)
150 parent = mnt->mnt_sb->s_export_op->get_parent(dentry);
151 mutex_unlock(&dentry->d_inode->i_mutex);
153 if (IS_ERR(parent)) {
154 dprintk("%s: get_parent of %ld failed, err %d\n",
155 __func__, dentry->d_inode->i_ino, PTR_ERR(parent));
159 dprintk("%s: find name of %lu in %lu\n", __func__,
160 dentry->d_inode->i_ino, parent->d_inode->i_ino);
161 err = exportfs_get_name(mnt, parent, nbuf, dentry);
163 goto out_reconnected;
166 dprintk("%s: found name: %s\n", __func__, nbuf);
167 mutex_lock(&parent->d_inode->i_mutex);
168 tmp = lookup_one_len(nbuf, parent, strlen(nbuf));
169 mutex_unlock(&parent->d_inode->i_mutex);
171 dprintk("%s: lookup failed: %d\n", __func__, PTR_ERR(tmp));
176 goto out_reconnected;
179 if (IS_ROOT(dentry)) {
191 * Someone must have renamed our entry into another parent, in
192 * which case it has been reconnected by the rename.
194 * Or someone removed it entirely, in which case filehandle
195 * lookup will succeed but the directory is now IS_DEAD and
196 * subsequent operations on it will fail.
198 * Alternatively, maybe there was no race at all, and the
199 * filesystem is just corrupt and gave us a parent that doesn't
200 * actually contain any entry pointing to this inode. So,
201 * double check that this worked and return -ESTALE if not:
203 if (!dentry_connected(dentry))
204 return ERR_PTR(-ESTALE);
209 * Make sure target_dir is fully connected to the dentry tree.
211 * On successful return, DCACHE_DISCONNECTED will be cleared on
212 * target_dir, and target_dir->d_parent->...->d_parent will reach the
213 * root of the filesystem.
215 * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
216 * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
217 * set but already be connected. In that case we'll verify the
218 * connection to root and then clear the flag.
220 * Note that target_dir could be removed by a concurrent operation. In
221 * that case reconnect_path may still succeed with target_dir fully
222 * connected, but further operations using the filehandle will fail when
223 * necessary (due to S_DEAD being set on the directory).
226 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
230 while (target_dir->d_flags & DCACHE_DISCONNECTED) {
231 struct dentry *pd = find_disconnected_root(target_dir);
233 BUG_ON(pd == mnt->mnt_sb->s_root);
236 /* must have found a connected parent - great */
237 clear_disconnected(target_dir);
241 struct dentry *parent;
243 * We have hit the top of a disconnected path, try to
244 * find parent and connect.
246 parent = reconnect_one(mnt, pd, nbuf);
248 goto out_reconnected;
249 if (IS_ERR(parent)) {
250 err = PTR_ERR(parent);
258 if (target_dir->d_flags & DCACHE_DISCONNECTED) {
259 /* something went wrong - oh-well */
267 clear_disconnected(target_dir);
271 struct getdents_callback {
272 struct dir_context ctx;
273 char *name; /* name that was found. It already points to a
274 buffer NAME_MAX+1 is size */
275 u64 ino; /* the inum we are looking for */
276 int found; /* inode matched? */
277 int sequence; /* sequence counter */
281 * A rather strange filldir function to capture
282 * the name matching the specified inode number.
284 static int filldir_one(void * __buf, const char * name, int len,
285 loff_t pos, u64 ino, unsigned int d_type)
287 struct getdents_callback *buf = __buf;
291 if (buf->ino == ino && len <= NAME_MAX) {
292 memcpy(buf->name, name, len);
293 buf->name[len] = '\0';
301 * get_name - default export_operations->get_name function
302 * @dentry: the directory in which to find a name
303 * @name: a pointer to a %NAME_MAX+1 char buffer to store the name
304 * @child: the dentry for the child directory.
306 * calls readdir on the parent until it finds an entry with
307 * the same inode number as the child, and returns that.
309 static int get_name(const struct path *path, char *name, struct dentry *child)
311 const struct cred *cred = current_cred();
312 struct inode *dir = path->dentry->d_inode;
316 struct path child_path = {
320 struct getdents_callback buffer = {
321 .ctx.actor = filldir_one,
326 if (!dir || !S_ISDIR(dir->i_mode))
332 * inode->i_ino is unsigned long, kstat->ino is u64, so the
333 * former would be insufficient on 32-bit hosts when the
334 * filesystem supports 64-bit inode numbers. So we need to
335 * actually call ->getattr, not just read i_ino:
337 error = vfs_getattr_nosec(&child_path, &stat);
340 buffer.ino = stat.ino;
342 * Open the directory ...
344 file = dentry_open(path, O_RDONLY, cred);
345 error = PTR_ERR(file);
350 if (!file->f_op->iterate)
355 int old_seq = buffer.sequence;
357 error = iterate_dir(file, &buffer.ctx);
367 if (old_seq == buffer.sequence)
378 * export_encode_fh - default export_operations->encode_fh function
379 * @inode: the object to encode
380 * @fh: where to store the file handle fragment
381 * @max_len: maximum length to store there
382 * @parent: parent directory inode, if wanted
384 * This default encode_fh function assumes that the 32 inode number
385 * is suitable for locating an inode, and that the generation number
386 * can be used to check that it is still valid. It places them in the
387 * filehandle fragment where export_decode_fh expects to find them.
389 static int export_encode_fh(struct inode *inode, struct fid *fid,
390 int *max_len, struct inode *parent)
393 int type = FILEID_INO32_GEN;
395 if (parent && (len < 4)) {
397 return FILEID_INVALID;
398 } else if (len < 2) {
400 return FILEID_INVALID;
404 fid->i32.ino = inode->i_ino;
405 fid->i32.gen = inode->i_generation;
407 fid->i32.parent_ino = parent->i_ino;
408 fid->i32.parent_gen = parent->i_generation;
410 type = FILEID_INO32_GEN_PARENT;
416 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
417 int *max_len, struct inode *parent)
419 const struct export_operations *nop = inode->i_sb->s_export_op;
421 if (nop && nop->encode_fh)
422 return nop->encode_fh(inode, fid->raw, max_len, parent);
424 return export_encode_fh(inode, fid, max_len, parent);
426 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
428 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
432 struct dentry *p = NULL;
433 struct inode *inode = dentry->d_inode, *parent = NULL;
435 if (connectable && !S_ISDIR(inode->i_mode)) {
436 p = dget_parent(dentry);
438 * note that while p might've ceased to be our parent already,
439 * it's still pinned by and still positive.
444 error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
449 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
451 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
452 int fh_len, int fileid_type,
453 int (*acceptable)(void *, struct dentry *), void *context)
455 const struct export_operations *nop = mnt->mnt_sb->s_export_op;
456 struct dentry *result, *alias;
457 char nbuf[NAME_MAX+1];
461 * Try to get any dentry for the given file handle from the filesystem.
463 if (!nop || !nop->fh_to_dentry)
464 return ERR_PTR(-ESTALE);
465 result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
467 result = ERR_PTR(-ESTALE);
471 if (S_ISDIR(result->d_inode->i_mode)) {
473 * This request is for a directory.
475 * On the positive side there is only one dentry for each
476 * directory inode. On the negative side this implies that we
477 * to ensure our dentry is connected all the way up to the
480 if (result->d_flags & DCACHE_DISCONNECTED) {
481 err = reconnect_path(mnt, result, nbuf);
486 if (!acceptable(context, result)) {
494 * It's not a directory. Life is a little more complicated.
496 struct dentry *target_dir, *nresult;
499 * See if either the dentry we just got from the filesystem
500 * or any alias for it is acceptable. This is always true
501 * if this filesystem is exported without the subtreecheck
502 * option. If the filesystem is exported with the subtree
503 * check option there's a fair chance we need to look at
504 * the parent directory in the file handle and make sure
505 * it's connected to the filesystem root.
507 alias = find_acceptable_alias(result, acceptable, context);
512 * Try to extract a dentry for the parent directory from the
513 * file handle. If this fails we'll have to give up.
516 if (!nop->fh_to_parent)
519 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
520 fh_len, fileid_type);
523 err = PTR_ERR(target_dir);
524 if (IS_ERR(target_dir))
528 * And as usual we need to make sure the parent directory is
529 * connected to the filesystem root. The VFS really doesn't
530 * like disconnected directories..
532 err = reconnect_path(mnt, target_dir, nbuf);
539 * Now that we've got both a well-connected parent and a
540 * dentry for the inode we're after, make sure that our
541 * inode is actually connected to the parent.
543 err = exportfs_get_name(mnt, target_dir, nbuf, result);
545 mutex_lock(&target_dir->d_inode->i_mutex);
546 nresult = lookup_one_len(nbuf, target_dir,
548 mutex_unlock(&target_dir->d_inode->i_mutex);
549 if (!IS_ERR(nresult)) {
550 if (nresult->d_inode) {
559 * At this point we are done with the parent, but it's pinned
560 * by the child dentry anyway.
565 * And finally make sure the dentry is actually acceptable
568 alias = find_acceptable_alias(result, acceptable, context);
581 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
583 MODULE_LICENSE("GPL");