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>
18 #include <linux/cred.h>
20 #define dprintk(fmt, args...) do{}while(0)
23 static int get_name(const struct path *path, char *name, struct dentry *child);
26 static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
27 char *name, struct dentry *child)
29 const struct export_operations *nop = dir->d_sb->s_export_op;
30 struct path path = {.mnt = mnt, .dentry = dir};
33 return nop->get_name(dir, name, child);
35 return get_name(&path, name, child);
39 * Check if the dentry or any of it's aliases is acceptable.
41 static struct dentry *
42 find_acceptable_alias(struct dentry *result,
43 int (*acceptable)(void *context, struct dentry *dentry),
46 struct dentry *dentry, *toput = NULL;
49 if (acceptable(context, result))
52 inode = result->d_inode;
53 spin_lock(&inode->i_lock);
54 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
56 spin_unlock(&inode->i_lock);
59 if (dentry != result && acceptable(context, dentry)) {
63 spin_lock(&inode->i_lock);
66 spin_unlock(&inode->i_lock);
73 static bool dentry_connected(struct dentry *dentry)
76 while (dentry->d_flags & DCACHE_DISCONNECTED) {
77 struct dentry *parent = dget_parent(dentry);
80 if (IS_ROOT(dentry)) {
90 static void clear_disconnected(struct dentry *dentry)
93 while (dentry->d_flags & DCACHE_DISCONNECTED) {
94 struct dentry *parent = dget_parent(dentry);
96 WARN_ON_ONCE(IS_ROOT(dentry));
98 spin_lock(&dentry->d_lock);
99 dentry->d_flags &= ~DCACHE_DISCONNECTED;
100 spin_unlock(&dentry->d_lock);
109 * Reconnect a directory dentry with its parent.
111 * This can return a dentry, or NULL, or an error.
113 * In the first case the returned dentry is the parent of the given
114 * dentry, and may itself need to be reconnected to its parent.
116 * In the NULL case, a concurrent VFS operation has either renamed or
117 * removed this directory. The concurrent operation has reconnected our
118 * dentry, so we no longer need to.
120 static struct dentry *reconnect_one(struct vfsmount *mnt,
121 struct dentry *dentry, char *nbuf)
123 struct dentry *parent;
127 parent = ERR_PTR(-EACCES);
128 inode_lock(dentry->d_inode);
129 if (mnt->mnt_sb->s_export_op->get_parent)
130 parent = mnt->mnt_sb->s_export_op->get_parent(dentry);
131 inode_unlock(dentry->d_inode);
133 if (IS_ERR(parent)) {
134 dprintk("%s: get_parent of %ld failed, err %d\n",
135 __func__, dentry->d_inode->i_ino, PTR_ERR(parent));
139 dprintk("%s: find name of %lu in %lu\n", __func__,
140 dentry->d_inode->i_ino, parent->d_inode->i_ino);
141 err = exportfs_get_name(mnt, parent, nbuf, dentry);
143 goto out_reconnected;
146 dprintk("%s: found name: %s\n", __func__, nbuf);
147 tmp = lookup_one_len_unlocked(nbuf, parent, strlen(nbuf));
149 dprintk("%s: lookup failed: %d\n", __func__, PTR_ERR(tmp));
154 * Somebody has renamed it since exportfs_get_name();
155 * great, since it could've only been renamed if it
156 * got looked up and thus connected, and it would
157 * remain connected afterwards. We are done.
160 goto out_reconnected;
163 if (IS_ROOT(dentry)) {
175 * Someone must have renamed our entry into another parent, in
176 * which case it has been reconnected by the rename.
178 * Or someone removed it entirely, in which case filehandle
179 * lookup will succeed but the directory is now IS_DEAD and
180 * subsequent operations on it will fail.
182 * Alternatively, maybe there was no race at all, and the
183 * filesystem is just corrupt and gave us a parent that doesn't
184 * actually contain any entry pointing to this inode. So,
185 * double check that this worked and return -ESTALE if not:
187 if (!dentry_connected(dentry))
188 return ERR_PTR(-ESTALE);
193 * Make sure target_dir is fully connected to the dentry tree.
195 * On successful return, DCACHE_DISCONNECTED will be cleared on
196 * target_dir, and target_dir->d_parent->...->d_parent will reach the
197 * root of the filesystem.
199 * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
200 * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
201 * set but already be connected. In that case we'll verify the
202 * connection to root and then clear the flag.
204 * Note that target_dir could be removed by a concurrent operation. In
205 * that case reconnect_path may still succeed with target_dir fully
206 * connected, but further operations using the filehandle will fail when
207 * necessary (due to S_DEAD being set on the directory).
210 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
212 struct dentry *dentry, *parent;
214 dentry = dget(target_dir);
216 while (dentry->d_flags & DCACHE_DISCONNECTED) {
217 BUG_ON(dentry == mnt->mnt_sb->s_root);
220 parent = reconnect_one(mnt, dentry, nbuf);
222 parent = dget_parent(dentry);
228 return PTR_ERR(parent);
232 clear_disconnected(target_dir);
236 struct getdents_callback {
237 struct dir_context ctx;
238 char *name; /* name that was found. It already points to a
239 buffer NAME_MAX+1 is size */
240 u64 ino; /* the inum we are looking for */
241 int found; /* inode matched? */
242 int sequence; /* sequence counter */
246 * A rather strange filldir function to capture
247 * the name matching the specified inode number.
249 static int filldir_one(struct dir_context *ctx, const char *name, int len,
250 loff_t pos, u64 ino, unsigned int d_type)
252 struct getdents_callback *buf =
253 container_of(ctx, struct getdents_callback, ctx);
257 if (buf->ino == ino && len <= NAME_MAX) {
258 memcpy(buf->name, name, len);
259 buf->name[len] = '\0';
267 * get_name - default export_operations->get_name function
268 * @path: the directory in which to find a name
269 * @name: a pointer to a %NAME_MAX+1 char buffer to store the name
270 * @child: the dentry for the child directory.
272 * calls readdir on the parent until it finds an entry with
273 * the same inode number as the child, and returns that.
275 static int get_name(const struct path *path, char *name, struct dentry *child)
277 const struct cred *cred = current_cred();
278 struct inode *dir = path->dentry->d_inode;
282 struct path child_path = {
286 struct getdents_callback buffer = {
287 .ctx.actor = filldir_one,
292 if (!dir || !S_ISDIR(dir->i_mode))
298 * inode->i_ino is unsigned long, kstat->ino is u64, so the
299 * former would be insufficient on 32-bit hosts when the
300 * filesystem supports 64-bit inode numbers. So we need to
301 * actually call ->getattr, not just read i_ino:
303 error = vfs_getattr_nosec(&child_path, &stat,
304 STATX_INO, AT_STATX_SYNC_AS_STAT);
307 buffer.ino = stat.ino;
309 * Open the directory ...
311 file = dentry_open(path, O_RDONLY, cred);
312 error = PTR_ERR(file);
317 if (!file->f_op->iterate && !file->f_op->iterate_shared)
322 int old_seq = buffer.sequence;
324 error = iterate_dir(file, &buffer.ctx);
334 if (old_seq == buffer.sequence)
345 * export_encode_fh - default export_operations->encode_fh function
346 * @inode: the object to encode
347 * @fid: where to store the file handle fragment
348 * @max_len: maximum length to store there
349 * @parent: parent directory inode, if wanted
351 * This default encode_fh function assumes that the 32 inode number
352 * is suitable for locating an inode, and that the generation number
353 * can be used to check that it is still valid. It places them in the
354 * filehandle fragment where export_decode_fh expects to find them.
356 static int export_encode_fh(struct inode *inode, struct fid *fid,
357 int *max_len, struct inode *parent)
360 int type = FILEID_INO32_GEN;
362 if (parent && (len < 4)) {
364 return FILEID_INVALID;
365 } else if (len < 2) {
367 return FILEID_INVALID;
371 fid->i32.ino = inode->i_ino;
372 fid->i32.gen = inode->i_generation;
374 fid->i32.parent_ino = parent->i_ino;
375 fid->i32.parent_gen = parent->i_generation;
377 type = FILEID_INO32_GEN_PARENT;
383 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
384 int *max_len, struct inode *parent)
386 const struct export_operations *nop = inode->i_sb->s_export_op;
388 if (nop && nop->encode_fh)
389 return nop->encode_fh(inode, fid->raw, max_len, parent);
391 return export_encode_fh(inode, fid, max_len, parent);
393 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
395 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
399 struct dentry *p = NULL;
400 struct inode *inode = dentry->d_inode, *parent = NULL;
402 if (connectable && !S_ISDIR(inode->i_mode)) {
403 p = dget_parent(dentry);
405 * note that while p might've ceased to be our parent already,
406 * it's still pinned by and still positive.
411 error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
416 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
418 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
419 int fh_len, int fileid_type,
420 int (*acceptable)(void *, struct dentry *), void *context)
422 const struct export_operations *nop = mnt->mnt_sb->s_export_op;
423 struct dentry *result, *alias;
424 char nbuf[NAME_MAX+1];
428 * Try to get any dentry for the given file handle from the filesystem.
430 if (!nop || !nop->fh_to_dentry)
431 return ERR_PTR(-ESTALE);
432 result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
433 if (PTR_ERR(result) == -ENOMEM)
434 return ERR_CAST(result);
435 if (IS_ERR_OR_NULL(result))
436 return ERR_PTR(-ESTALE);
439 * If no acceptance criteria was specified by caller, a disconnected
440 * dentry is also accepatable. Callers may use this mode to query if
441 * file handle is stale or to get a reference to an inode without
442 * risking the high overhead caused by directory reconnect.
447 if (d_is_dir(result)) {
449 * This request is for a directory.
451 * On the positive side there is only one dentry for each
452 * directory inode. On the negative side this implies that we
453 * to ensure our dentry is connected all the way up to the
456 if (result->d_flags & DCACHE_DISCONNECTED) {
457 err = reconnect_path(mnt, result, nbuf);
462 if (!acceptable(context, result)) {
470 * It's not a directory. Life is a little more complicated.
472 struct dentry *target_dir, *nresult;
475 * See if either the dentry we just got from the filesystem
476 * or any alias for it is acceptable. This is always true
477 * if this filesystem is exported without the subtreecheck
478 * option. If the filesystem is exported with the subtree
479 * check option there's a fair chance we need to look at
480 * the parent directory in the file handle and make sure
481 * it's connected to the filesystem root.
483 alias = find_acceptable_alias(result, acceptable, context);
488 * Try to extract a dentry for the parent directory from the
489 * file handle. If this fails we'll have to give up.
492 if (!nop->fh_to_parent)
495 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
496 fh_len, fileid_type);
499 err = PTR_ERR(target_dir);
500 if (IS_ERR(target_dir))
504 * And as usual we need to make sure the parent directory is
505 * connected to the filesystem root. The VFS really doesn't
506 * like disconnected directories..
508 err = reconnect_path(mnt, target_dir, nbuf);
515 * Now that we've got both a well-connected parent and a
516 * dentry for the inode we're after, make sure that our
517 * inode is actually connected to the parent.
519 err = exportfs_get_name(mnt, target_dir, nbuf, result);
521 inode_lock(target_dir->d_inode);
522 nresult = lookup_one_len(nbuf, target_dir,
524 inode_unlock(target_dir->d_inode);
525 if (!IS_ERR(nresult)) {
526 if (nresult->d_inode) {
535 * At this point we are done with the parent, but it's pinned
536 * by the child dentry anyway.
541 * And finally make sure the dentry is actually acceptable
544 alias = find_acceptable_alias(result, acceptable, context);
559 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
561 MODULE_LICENSE("GPL");