Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / sunrpc / rpc_pipe.c
diff --git a/kernel/net/sunrpc/rpc_pipe.c b/kernel/net/sunrpc/rpc_pipe.c
new file mode 100644 (file)
index 0000000..d81186d
--- /dev/null
@@ -0,0 +1,1527 @@
+/*
+ * net/sunrpc/rpc_pipe.c
+ *
+ * Userland/kernel interface for rpcauth_gss.
+ * Code shamelessly plagiarized from fs/nfsd/nfsctl.c
+ * and fs/sysfs/inode.c
+ *
+ * Copyright (c) 2002, Trond Myklebust <trond.myklebust@fys.uio.no>
+ *
+ */
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/pagemap.h>
+#include <linux/mount.h>
+#include <linux/namei.h>
+#include <linux/fsnotify.h>
+#include <linux/kernel.h>
+#include <linux/rcupdate.h>
+#include <linux/utsname.h>
+
+#include <asm/ioctls.h>
+#include <linux/poll.h>
+#include <linux/wait.h>
+#include <linux/seq_file.h>
+
+#include <linux/sunrpc/clnt.h>
+#include <linux/workqueue.h>
+#include <linux/sunrpc/rpc_pipe_fs.h>
+#include <linux/sunrpc/cache.h>
+#include <linux/nsproxy.h>
+#include <linux/notifier.h>
+
+#include "netns.h"
+#include "sunrpc.h"
+
+#define RPCDBG_FACILITY RPCDBG_DEBUG
+
+#define NET_NAME(net)  ((net == &init_net) ? " (init_net)" : "")
+
+static struct file_system_type rpc_pipe_fs_type;
+static const struct rpc_pipe_ops gssd_dummy_pipe_ops;
+
+static struct kmem_cache *rpc_inode_cachep __read_mostly;
+
+#define RPC_UPCALL_TIMEOUT (30*HZ)
+
+static BLOCKING_NOTIFIER_HEAD(rpc_pipefs_notifier_list);
+
+int rpc_pipefs_notifier_register(struct notifier_block *nb)
+{
+       return blocking_notifier_chain_cond_register(&rpc_pipefs_notifier_list, nb);
+}
+EXPORT_SYMBOL_GPL(rpc_pipefs_notifier_register);
+
+void rpc_pipefs_notifier_unregister(struct notifier_block *nb)
+{
+       blocking_notifier_chain_unregister(&rpc_pipefs_notifier_list, nb);
+}
+EXPORT_SYMBOL_GPL(rpc_pipefs_notifier_unregister);
+
+static void rpc_purge_list(wait_queue_head_t *waitq, struct list_head *head,
+               void (*destroy_msg)(struct rpc_pipe_msg *), int err)
+{
+       struct rpc_pipe_msg *msg;
+
+       if (list_empty(head))
+               return;
+       do {
+               msg = list_entry(head->next, struct rpc_pipe_msg, list);
+               list_del_init(&msg->list);
+               msg->errno = err;
+               destroy_msg(msg);
+       } while (!list_empty(head));
+
+       if (waitq)
+               wake_up(waitq);
+}
+
+static void
+rpc_timeout_upcall_queue(struct work_struct *work)
+{
+       LIST_HEAD(free_list);
+       struct rpc_pipe *pipe =
+               container_of(work, struct rpc_pipe, queue_timeout.work);
+       void (*destroy_msg)(struct rpc_pipe_msg *);
+       struct dentry *dentry;
+
+       spin_lock(&pipe->lock);
+       destroy_msg = pipe->ops->destroy_msg;
+       if (pipe->nreaders == 0) {
+               list_splice_init(&pipe->pipe, &free_list);
+               pipe->pipelen = 0;
+       }
+       dentry = dget(pipe->dentry);
+       spin_unlock(&pipe->lock);
+       rpc_purge_list(dentry ? &RPC_I(d_inode(dentry))->waitq : NULL,
+                       &free_list, destroy_msg, -ETIMEDOUT);
+       dput(dentry);
+}
+
+ssize_t rpc_pipe_generic_upcall(struct file *filp, struct rpc_pipe_msg *msg,
+                               char __user *dst, size_t buflen)
+{
+       char *data = (char *)msg->data + msg->copied;
+       size_t mlen = min(msg->len - msg->copied, buflen);
+       unsigned long left;
+
+       left = copy_to_user(dst, data, mlen);
+       if (left == mlen) {
+               msg->errno = -EFAULT;
+               return -EFAULT;
+       }
+
+       mlen -= left;
+       msg->copied += mlen;
+       msg->errno = 0;
+       return mlen;
+}
+EXPORT_SYMBOL_GPL(rpc_pipe_generic_upcall);
+
+/**
+ * rpc_queue_upcall - queue an upcall message to userspace
+ * @pipe: upcall pipe on which to queue given message
+ * @msg: message to queue
+ *
+ * Call with an @inode created by rpc_mkpipe() to queue an upcall.
+ * A userspace process may then later read the upcall by performing a
+ * read on an open file for this inode.  It is up to the caller to
+ * initialize the fields of @msg (other than @msg->list) appropriately.
+ */
+int
+rpc_queue_upcall(struct rpc_pipe *pipe, struct rpc_pipe_msg *msg)
+{
+       int res = -EPIPE;
+       struct dentry *dentry;
+
+       spin_lock(&pipe->lock);
+       if (pipe->nreaders) {
+               list_add_tail(&msg->list, &pipe->pipe);
+               pipe->pipelen += msg->len;
+               res = 0;
+       } else if (pipe->flags & RPC_PIPE_WAIT_FOR_OPEN) {
+               if (list_empty(&pipe->pipe))
+                       queue_delayed_work(rpciod_workqueue,
+                                       &pipe->queue_timeout,
+                                       RPC_UPCALL_TIMEOUT);
+               list_add_tail(&msg->list, &pipe->pipe);
+               pipe->pipelen += msg->len;
+               res = 0;
+       }
+       dentry = dget(pipe->dentry);
+       spin_unlock(&pipe->lock);
+       if (dentry) {
+               wake_up(&RPC_I(d_inode(dentry))->waitq);
+               dput(dentry);
+       }
+       return res;
+}
+EXPORT_SYMBOL_GPL(rpc_queue_upcall);
+
+static inline void
+rpc_inode_setowner(struct inode *inode, void *private)
+{
+       RPC_I(inode)->private = private;
+}
+
+static void
+rpc_close_pipes(struct inode *inode)
+{
+       struct rpc_pipe *pipe = RPC_I(inode)->pipe;
+       int need_release;
+       LIST_HEAD(free_list);
+
+       mutex_lock(&inode->i_mutex);
+       spin_lock(&pipe->lock);
+       need_release = pipe->nreaders != 0 || pipe->nwriters != 0;
+       pipe->nreaders = 0;
+       list_splice_init(&pipe->in_upcall, &free_list);
+       list_splice_init(&pipe->pipe, &free_list);
+       pipe->pipelen = 0;
+       pipe->dentry = NULL;
+       spin_unlock(&pipe->lock);
+       rpc_purge_list(&RPC_I(inode)->waitq, &free_list, pipe->ops->destroy_msg, -EPIPE);
+       pipe->nwriters = 0;
+       if (need_release && pipe->ops->release_pipe)
+               pipe->ops->release_pipe(inode);
+       cancel_delayed_work_sync(&pipe->queue_timeout);
+       rpc_inode_setowner(inode, NULL);
+       RPC_I(inode)->pipe = NULL;
+       mutex_unlock(&inode->i_mutex);
+}
+
+static struct inode *
+rpc_alloc_inode(struct super_block *sb)
+{
+       struct rpc_inode *rpci;
+       rpci = kmem_cache_alloc(rpc_inode_cachep, GFP_KERNEL);
+       if (!rpci)
+               return NULL;
+       return &rpci->vfs_inode;
+}
+
+static void
+rpc_i_callback(struct rcu_head *head)
+{
+       struct inode *inode = container_of(head, struct inode, i_rcu);
+       kmem_cache_free(rpc_inode_cachep, RPC_I(inode));
+}
+
+static void
+rpc_destroy_inode(struct inode *inode)
+{
+       call_rcu(&inode->i_rcu, rpc_i_callback);
+}
+
+static int
+rpc_pipe_open(struct inode *inode, struct file *filp)
+{
+       struct rpc_pipe *pipe;
+       int first_open;
+       int res = -ENXIO;
+
+       mutex_lock(&inode->i_mutex);
+       pipe = RPC_I(inode)->pipe;
+       if (pipe == NULL)
+               goto out;
+       first_open = pipe->nreaders == 0 && pipe->nwriters == 0;
+       if (first_open && pipe->ops->open_pipe) {
+               res = pipe->ops->open_pipe(inode);
+               if (res)
+                       goto out;
+       }
+       if (filp->f_mode & FMODE_READ)
+               pipe->nreaders++;
+       if (filp->f_mode & FMODE_WRITE)
+               pipe->nwriters++;
+       res = 0;
+out:
+       mutex_unlock(&inode->i_mutex);
+       return res;
+}
+
+static int
+rpc_pipe_release(struct inode *inode, struct file *filp)
+{
+       struct rpc_pipe *pipe;
+       struct rpc_pipe_msg *msg;
+       int last_close;
+
+       mutex_lock(&inode->i_mutex);
+       pipe = RPC_I(inode)->pipe;
+       if (pipe == NULL)
+               goto out;
+       msg = filp->private_data;
+       if (msg != NULL) {
+               spin_lock(&pipe->lock);
+               msg->errno = -EAGAIN;
+               list_del_init(&msg->list);
+               spin_unlock(&pipe->lock);
+               pipe->ops->destroy_msg(msg);
+       }
+       if (filp->f_mode & FMODE_WRITE)
+               pipe->nwriters --;
+       if (filp->f_mode & FMODE_READ) {
+               pipe->nreaders --;
+               if (pipe->nreaders == 0) {
+                       LIST_HEAD(free_list);
+                       spin_lock(&pipe->lock);
+                       list_splice_init(&pipe->pipe, &free_list);
+                       pipe->pipelen = 0;
+                       spin_unlock(&pipe->lock);
+                       rpc_purge_list(&RPC_I(inode)->waitq, &free_list,
+                                       pipe->ops->destroy_msg, -EAGAIN);
+               }
+       }
+       last_close = pipe->nwriters == 0 && pipe->nreaders == 0;
+       if (last_close && pipe->ops->release_pipe)
+               pipe->ops->release_pipe(inode);
+out:
+       mutex_unlock(&inode->i_mutex);
+       return 0;
+}
+
+static ssize_t
+rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
+{
+       struct inode *inode = file_inode(filp);
+       struct rpc_pipe *pipe;
+       struct rpc_pipe_msg *msg;
+       int res = 0;
+
+       mutex_lock(&inode->i_mutex);
+       pipe = RPC_I(inode)->pipe;
+       if (pipe == NULL) {
+               res = -EPIPE;
+               goto out_unlock;
+       }
+       msg = filp->private_data;
+       if (msg == NULL) {
+               spin_lock(&pipe->lock);
+               if (!list_empty(&pipe->pipe)) {
+                       msg = list_entry(pipe->pipe.next,
+                                       struct rpc_pipe_msg,
+                                       list);
+                       list_move(&msg->list, &pipe->in_upcall);
+                       pipe->pipelen -= msg->len;
+                       filp->private_data = msg;
+                       msg->copied = 0;
+               }
+               spin_unlock(&pipe->lock);
+               if (msg == NULL)
+                       goto out_unlock;
+       }
+       /* NOTE: it is up to the callback to update msg->copied */
+       res = pipe->ops->upcall(filp, msg, buf, len);
+       if (res < 0 || msg->len == msg->copied) {
+               filp->private_data = NULL;
+               spin_lock(&pipe->lock);
+               list_del_init(&msg->list);
+               spin_unlock(&pipe->lock);
+               pipe->ops->destroy_msg(msg);
+       }
+out_unlock:
+       mutex_unlock(&inode->i_mutex);
+       return res;
+}
+
+static ssize_t
+rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
+{
+       struct inode *inode = file_inode(filp);
+       int res;
+
+       mutex_lock(&inode->i_mutex);
+       res = -EPIPE;
+       if (RPC_I(inode)->pipe != NULL)
+               res = RPC_I(inode)->pipe->ops->downcall(filp, buf, len);
+       mutex_unlock(&inode->i_mutex);
+       return res;
+}
+
+static unsigned int
+rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
+{
+       struct inode *inode = file_inode(filp);
+       struct rpc_inode *rpci = RPC_I(inode);
+       unsigned int mask = POLLOUT | POLLWRNORM;
+
+       poll_wait(filp, &rpci->waitq, wait);
+
+       mutex_lock(&inode->i_mutex);
+       if (rpci->pipe == NULL)
+               mask |= POLLERR | POLLHUP;
+       else if (filp->private_data || !list_empty(&rpci->pipe->pipe))
+               mask |= POLLIN | POLLRDNORM;
+       mutex_unlock(&inode->i_mutex);
+       return mask;
+}
+
+static long
+rpc_pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+       struct inode *inode = file_inode(filp);
+       struct rpc_pipe *pipe;
+       int len;
+
+       switch (cmd) {
+       case FIONREAD:
+               mutex_lock(&inode->i_mutex);
+               pipe = RPC_I(inode)->pipe;
+               if (pipe == NULL) {
+                       mutex_unlock(&inode->i_mutex);
+                       return -EPIPE;
+               }
+               spin_lock(&pipe->lock);
+               len = pipe->pipelen;
+               if (filp->private_data) {
+                       struct rpc_pipe_msg *msg;
+                       msg = filp->private_data;
+                       len += msg->len - msg->copied;
+               }
+               spin_unlock(&pipe->lock);
+               mutex_unlock(&inode->i_mutex);
+               return put_user(len, (int __user *)arg);
+       default:
+               return -EINVAL;
+       }
+}
+
+static const struct file_operations rpc_pipe_fops = {
+       .owner          = THIS_MODULE,
+       .llseek         = no_llseek,
+       .read           = rpc_pipe_read,
+       .write          = rpc_pipe_write,
+       .poll           = rpc_pipe_poll,
+       .unlocked_ioctl = rpc_pipe_ioctl,
+       .open           = rpc_pipe_open,
+       .release        = rpc_pipe_release,
+};
+
+static int
+rpc_show_info(struct seq_file *m, void *v)
+{
+       struct rpc_clnt *clnt = m->private;
+
+       rcu_read_lock();
+       seq_printf(m, "RPC server: %s\n",
+                       rcu_dereference(clnt->cl_xprt)->servername);
+       seq_printf(m, "service: %s (%d) version %d\n", clnt->cl_program->name,
+                       clnt->cl_prog, clnt->cl_vers);
+       seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
+       seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
+       seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
+       rcu_read_unlock();
+       return 0;
+}
+
+static int
+rpc_info_open(struct inode *inode, struct file *file)
+{
+       struct rpc_clnt *clnt = NULL;
+       int ret = single_open(file, rpc_show_info, NULL);
+
+       if (!ret) {
+               struct seq_file *m = file->private_data;
+
+               spin_lock(&file->f_path.dentry->d_lock);
+               if (!d_unhashed(file->f_path.dentry))
+                       clnt = RPC_I(inode)->private;
+               if (clnt != NULL && atomic_inc_not_zero(&clnt->cl_count)) {
+                       spin_unlock(&file->f_path.dentry->d_lock);
+                       m->private = clnt;
+               } else {
+                       spin_unlock(&file->f_path.dentry->d_lock);
+                       single_release(inode, file);
+                       ret = -EINVAL;
+               }
+       }
+       return ret;
+}
+
+static int
+rpc_info_release(struct inode *inode, struct file *file)
+{
+       struct seq_file *m = file->private_data;
+       struct rpc_clnt *clnt = (struct rpc_clnt *)m->private;
+
+       if (clnt)
+               rpc_release_client(clnt);
+       return single_release(inode, file);
+}
+
+static const struct file_operations rpc_info_operations = {
+       .owner          = THIS_MODULE,
+       .open           = rpc_info_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = rpc_info_release,
+};
+
+
+/*
+ * Description of fs contents.
+ */
+struct rpc_filelist {
+       const char *name;
+       const struct file_operations *i_fop;
+       umode_t mode;
+};
+
+static struct inode *
+rpc_get_inode(struct super_block *sb, umode_t mode)
+{
+       struct inode *inode = new_inode(sb);
+       if (!inode)
+               return NULL;
+       inode->i_ino = get_next_ino();
+       inode->i_mode = mode;
+       inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+       switch (mode & S_IFMT) {
+       case S_IFDIR:
+               inode->i_fop = &simple_dir_operations;
+               inode->i_op = &simple_dir_inode_operations;
+               inc_nlink(inode);
+       default:
+               break;
+       }
+       return inode;
+}
+
+static int __rpc_create_common(struct inode *dir, struct dentry *dentry,
+                              umode_t mode,
+                              const struct file_operations *i_fop,
+                              void *private)
+{
+       struct inode *inode;
+
+       d_drop(dentry);
+       inode = rpc_get_inode(dir->i_sb, mode);
+       if (!inode)
+               goto out_err;
+       inode->i_ino = iunique(dir->i_sb, 100);
+       if (i_fop)
+               inode->i_fop = i_fop;
+       if (private)
+               rpc_inode_setowner(inode, private);
+       d_add(dentry, inode);
+       return 0;
+out_err:
+       printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %pd\n",
+                       __FILE__, __func__, dentry);
+       dput(dentry);
+       return -ENOMEM;
+}
+
+static int __rpc_create(struct inode *dir, struct dentry *dentry,
+                       umode_t mode,
+                       const struct file_operations *i_fop,
+                       void *private)
+{
+       int err;
+
+       err = __rpc_create_common(dir, dentry, S_IFREG | mode, i_fop, private);
+       if (err)
+               return err;
+       fsnotify_create(dir, dentry);
+       return 0;
+}
+
+static int __rpc_mkdir(struct inode *dir, struct dentry *dentry,
+                      umode_t mode,
+                      const struct file_operations *i_fop,
+                      void *private)
+{
+       int err;
+
+       err = __rpc_create_common(dir, dentry, S_IFDIR | mode, i_fop, private);
+       if (err)
+               return err;
+       inc_nlink(dir);
+       fsnotify_mkdir(dir, dentry);
+       return 0;
+}
+
+static void
+init_pipe(struct rpc_pipe *pipe)
+{
+       pipe->nreaders = 0;
+       pipe->nwriters = 0;
+       INIT_LIST_HEAD(&pipe->in_upcall);
+       INIT_LIST_HEAD(&pipe->in_downcall);
+       INIT_LIST_HEAD(&pipe->pipe);
+       pipe->pipelen = 0;
+       INIT_DELAYED_WORK(&pipe->queue_timeout,
+                           rpc_timeout_upcall_queue);
+       pipe->ops = NULL;
+       spin_lock_init(&pipe->lock);
+       pipe->dentry = NULL;
+}
+
+void rpc_destroy_pipe_data(struct rpc_pipe *pipe)
+{
+       kfree(pipe);
+}
+EXPORT_SYMBOL_GPL(rpc_destroy_pipe_data);
+
+struct rpc_pipe *rpc_mkpipe_data(const struct rpc_pipe_ops *ops, int flags)
+{
+       struct rpc_pipe *pipe;
+
+       pipe = kzalloc(sizeof(struct rpc_pipe), GFP_KERNEL);
+       if (!pipe)
+               return ERR_PTR(-ENOMEM);
+       init_pipe(pipe);
+       pipe->ops = ops;
+       pipe->flags = flags;
+       return pipe;
+}
+EXPORT_SYMBOL_GPL(rpc_mkpipe_data);
+
+static int __rpc_mkpipe_dentry(struct inode *dir, struct dentry *dentry,
+                              umode_t mode,
+                              const struct file_operations *i_fop,
+                              void *private,
+                              struct rpc_pipe *pipe)
+{
+       struct rpc_inode *rpci;
+       int err;
+
+       err = __rpc_create_common(dir, dentry, S_IFIFO | mode, i_fop, private);
+       if (err)
+               return err;
+       rpci = RPC_I(d_inode(dentry));
+       rpci->private = private;
+       rpci->pipe = pipe;
+       fsnotify_create(dir, dentry);
+       return 0;
+}
+
+static int __rpc_rmdir(struct inode *dir, struct dentry *dentry)
+{
+       int ret;
+
+       dget(dentry);
+       ret = simple_rmdir(dir, dentry);
+       d_delete(dentry);
+       dput(dentry);
+       return ret;
+}
+
+int rpc_rmdir(struct dentry *dentry)
+{
+       struct dentry *parent;
+       struct inode *dir;
+       int error;
+
+       parent = dget_parent(dentry);
+       dir = d_inode(parent);
+       mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
+       error = __rpc_rmdir(dir, dentry);
+       mutex_unlock(&dir->i_mutex);
+       dput(parent);
+       return error;
+}
+EXPORT_SYMBOL_GPL(rpc_rmdir);
+
+static int __rpc_unlink(struct inode *dir, struct dentry *dentry)
+{
+       int ret;
+
+       dget(dentry);
+       ret = simple_unlink(dir, dentry);
+       d_delete(dentry);
+       dput(dentry);
+       return ret;
+}
+
+static int __rpc_rmpipe(struct inode *dir, struct dentry *dentry)
+{
+       struct inode *inode = d_inode(dentry);
+
+       rpc_close_pipes(inode);
+       return __rpc_unlink(dir, dentry);
+}
+
+static struct dentry *__rpc_lookup_create_exclusive(struct dentry *parent,
+                                         const char *name)
+{
+       struct qstr q = QSTR_INIT(name, strlen(name));
+       struct dentry *dentry = d_hash_and_lookup(parent, &q);
+       if (!dentry) {
+               dentry = d_alloc(parent, &q);
+               if (!dentry)
+                       return ERR_PTR(-ENOMEM);
+       }
+       if (d_really_is_negative(dentry))
+               return dentry;
+       dput(dentry);
+       return ERR_PTR(-EEXIST);
+}
+
+/*
+ * FIXME: This probably has races.
+ */
+static void __rpc_depopulate(struct dentry *parent,
+                            const struct rpc_filelist *files,
+                            int start, int eof)
+{
+       struct inode *dir = d_inode(parent);
+       struct dentry *dentry;
+       struct qstr name;
+       int i;
+
+       for (i = start; i < eof; i++) {
+               name.name = files[i].name;
+               name.len = strlen(files[i].name);
+               dentry = d_hash_and_lookup(parent, &name);
+
+               if (dentry == NULL)
+                       continue;
+               if (d_really_is_negative(dentry))
+                       goto next;
+               switch (d_inode(dentry)->i_mode & S_IFMT) {
+                       default:
+                               BUG();
+                       case S_IFREG:
+                               __rpc_unlink(dir, dentry);
+                               break;
+                       case S_IFDIR:
+                               __rpc_rmdir(dir, dentry);
+               }
+next:
+               dput(dentry);
+       }
+}
+
+static void rpc_depopulate(struct dentry *parent,
+                          const struct rpc_filelist *files,
+                          int start, int eof)
+{
+       struct inode *dir = d_inode(parent);
+
+       mutex_lock_nested(&dir->i_mutex, I_MUTEX_CHILD);
+       __rpc_depopulate(parent, files, start, eof);
+       mutex_unlock(&dir->i_mutex);
+}
+
+static int rpc_populate(struct dentry *parent,
+                       const struct rpc_filelist *files,
+                       int start, int eof,
+                       void *private)
+{
+       struct inode *dir = d_inode(parent);
+       struct dentry *dentry;
+       int i, err;
+
+       mutex_lock(&dir->i_mutex);
+       for (i = start; i < eof; i++) {
+               dentry = __rpc_lookup_create_exclusive(parent, files[i].name);
+               err = PTR_ERR(dentry);
+               if (IS_ERR(dentry))
+                       goto out_bad;
+               switch (files[i].mode & S_IFMT) {
+                       default:
+                               BUG();
+                       case S_IFREG:
+                               err = __rpc_create(dir, dentry,
+                                               files[i].mode,
+                                               files[i].i_fop,
+                                               private);
+                               break;
+                       case S_IFDIR:
+                               err = __rpc_mkdir(dir, dentry,
+                                               files[i].mode,
+                                               NULL,
+                                               private);
+               }
+               if (err != 0)
+                       goto out_bad;
+       }
+       mutex_unlock(&dir->i_mutex);
+       return 0;
+out_bad:
+       __rpc_depopulate(parent, files, start, eof);
+       mutex_unlock(&dir->i_mutex);
+       printk(KERN_WARNING "%s: %s failed to populate directory %pd\n",
+                       __FILE__, __func__, parent);
+       return err;
+}
+
+static struct dentry *rpc_mkdir_populate(struct dentry *parent,
+               const char *name, umode_t mode, void *private,
+               int (*populate)(struct dentry *, void *), void *args_populate)
+{
+       struct dentry *dentry;
+       struct inode *dir = d_inode(parent);
+       int error;
+
+       mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
+       dentry = __rpc_lookup_create_exclusive(parent, name);
+       if (IS_ERR(dentry))
+               goto out;
+       error = __rpc_mkdir(dir, dentry, mode, NULL, private);
+       if (error != 0)
+               goto out_err;
+       if (populate != NULL) {
+               error = populate(dentry, args_populate);
+               if (error)
+                       goto err_rmdir;
+       }
+out:
+       mutex_unlock(&dir->i_mutex);
+       return dentry;
+err_rmdir:
+       __rpc_rmdir(dir, dentry);
+out_err:
+       dentry = ERR_PTR(error);
+       goto out;
+}
+
+static int rpc_rmdir_depopulate(struct dentry *dentry,
+               void (*depopulate)(struct dentry *))
+{
+       struct dentry *parent;
+       struct inode *dir;
+       int error;
+
+       parent = dget_parent(dentry);
+       dir = d_inode(parent);
+       mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
+       if (depopulate != NULL)
+               depopulate(dentry);
+       error = __rpc_rmdir(dir, dentry);
+       mutex_unlock(&dir->i_mutex);
+       dput(parent);
+       return error;
+}
+
+/**
+ * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
+ * @parent: dentry of directory to create new "pipe" in
+ * @name: name of pipe
+ * @private: private data to associate with the pipe, for the caller's use
+ * @pipe: &rpc_pipe containing input parameters
+ *
+ * Data is made available for userspace to read by calls to
+ * rpc_queue_upcall().  The actual reads will result in calls to
+ * @ops->upcall, which will be called with the file pointer,
+ * message, and userspace buffer to copy to.
+ *
+ * Writes can come at any time, and do not necessarily have to be
+ * responses to upcalls.  They will result in calls to @msg->downcall.
+ *
+ * The @private argument passed here will be available to all these methods
+ * from the file pointer, via RPC_I(file_inode(file))->private.
+ */
+struct dentry *rpc_mkpipe_dentry(struct dentry *parent, const char *name,
+                                void *private, struct rpc_pipe *pipe)
+{
+       struct dentry *dentry;
+       struct inode *dir = d_inode(parent);
+       umode_t umode = S_IFIFO | S_IRUSR | S_IWUSR;
+       int err;
+
+       if (pipe->ops->upcall == NULL)
+               umode &= ~S_IRUGO;
+       if (pipe->ops->downcall == NULL)
+               umode &= ~S_IWUGO;
+
+       mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
+       dentry = __rpc_lookup_create_exclusive(parent, name);
+       if (IS_ERR(dentry))
+               goto out;
+       err = __rpc_mkpipe_dentry(dir, dentry, umode, &rpc_pipe_fops,
+                                 private, pipe);
+       if (err)
+               goto out_err;
+out:
+       mutex_unlock(&dir->i_mutex);
+       return dentry;
+out_err:
+       dentry = ERR_PTR(err);
+       printk(KERN_WARNING "%s: %s() failed to create pipe %pd/%s (errno = %d)\n",
+                       __FILE__, __func__, parent, name,
+                       err);
+       goto out;
+}
+EXPORT_SYMBOL_GPL(rpc_mkpipe_dentry);
+
+/**
+ * rpc_unlink - remove a pipe
+ * @dentry: dentry for the pipe, as returned from rpc_mkpipe
+ *
+ * After this call, lookups will no longer find the pipe, and any
+ * attempts to read or write using preexisting opens of the pipe will
+ * return -EPIPE.
+ */
+int
+rpc_unlink(struct dentry *dentry)
+{
+       struct dentry *parent;
+       struct inode *dir;
+       int error = 0;
+
+       parent = dget_parent(dentry);
+       dir = d_inode(parent);
+       mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
+       error = __rpc_rmpipe(dir, dentry);
+       mutex_unlock(&dir->i_mutex);
+       dput(parent);
+       return error;
+}
+EXPORT_SYMBOL_GPL(rpc_unlink);
+
+/**
+ * rpc_init_pipe_dir_head - initialise a struct rpc_pipe_dir_head
+ * @pdh: pointer to struct rpc_pipe_dir_head
+ */
+void rpc_init_pipe_dir_head(struct rpc_pipe_dir_head *pdh)
+{
+       INIT_LIST_HEAD(&pdh->pdh_entries);
+       pdh->pdh_dentry = NULL;
+}
+EXPORT_SYMBOL_GPL(rpc_init_pipe_dir_head);
+
+/**
+ * rpc_init_pipe_dir_object - initialise a struct rpc_pipe_dir_object
+ * @pdo: pointer to struct rpc_pipe_dir_object
+ * @pdo_ops: pointer to const struct rpc_pipe_dir_object_ops
+ * @pdo_data: pointer to caller-defined data
+ */
+void rpc_init_pipe_dir_object(struct rpc_pipe_dir_object *pdo,
+               const struct rpc_pipe_dir_object_ops *pdo_ops,
+               void *pdo_data)
+{
+       INIT_LIST_HEAD(&pdo->pdo_head);
+       pdo->pdo_ops = pdo_ops;
+       pdo->pdo_data = pdo_data;
+}
+EXPORT_SYMBOL_GPL(rpc_init_pipe_dir_object);
+
+static int
+rpc_add_pipe_dir_object_locked(struct net *net,
+               struct rpc_pipe_dir_head *pdh,
+               struct rpc_pipe_dir_object *pdo)
+{
+       int ret = 0;
+
+       if (pdh->pdh_dentry)
+               ret = pdo->pdo_ops->create(pdh->pdh_dentry, pdo);
+       if (ret == 0)
+               list_add_tail(&pdo->pdo_head, &pdh->pdh_entries);
+       return ret;
+}
+
+static void
+rpc_remove_pipe_dir_object_locked(struct net *net,
+               struct rpc_pipe_dir_head *pdh,
+               struct rpc_pipe_dir_object *pdo)
+{
+       if (pdh->pdh_dentry)
+               pdo->pdo_ops->destroy(pdh->pdh_dentry, pdo);
+       list_del_init(&pdo->pdo_head);
+}
+
+/**
+ * rpc_add_pipe_dir_object - associate a rpc_pipe_dir_object to a directory
+ * @net: pointer to struct net
+ * @pdh: pointer to struct rpc_pipe_dir_head
+ * @pdo: pointer to struct rpc_pipe_dir_object
+ *
+ */
+int
+rpc_add_pipe_dir_object(struct net *net,
+               struct rpc_pipe_dir_head *pdh,
+               struct rpc_pipe_dir_object *pdo)
+{
+       int ret = 0;
+
+       if (list_empty(&pdo->pdo_head)) {
+               struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+               mutex_lock(&sn->pipefs_sb_lock);
+               ret = rpc_add_pipe_dir_object_locked(net, pdh, pdo);
+               mutex_unlock(&sn->pipefs_sb_lock);
+       }
+       return ret;
+}
+EXPORT_SYMBOL_GPL(rpc_add_pipe_dir_object);
+
+/**
+ * rpc_remove_pipe_dir_object - remove a rpc_pipe_dir_object from a directory
+ * @net: pointer to struct net
+ * @pdh: pointer to struct rpc_pipe_dir_head
+ * @pdo: pointer to struct rpc_pipe_dir_object
+ *
+ */
+void
+rpc_remove_pipe_dir_object(struct net *net,
+               struct rpc_pipe_dir_head *pdh,
+               struct rpc_pipe_dir_object *pdo)
+{
+       if (!list_empty(&pdo->pdo_head)) {
+               struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+               mutex_lock(&sn->pipefs_sb_lock);
+               rpc_remove_pipe_dir_object_locked(net, pdh, pdo);
+               mutex_unlock(&sn->pipefs_sb_lock);
+       }
+}
+EXPORT_SYMBOL_GPL(rpc_remove_pipe_dir_object);
+
+/**
+ * rpc_find_or_alloc_pipe_dir_object
+ * @net: pointer to struct net
+ * @pdh: pointer to struct rpc_pipe_dir_head
+ * @match: match struct rpc_pipe_dir_object to data
+ * @alloc: allocate a new struct rpc_pipe_dir_object
+ * @data: user defined data for match() and alloc()
+ *
+ */
+struct rpc_pipe_dir_object *
+rpc_find_or_alloc_pipe_dir_object(struct net *net,
+               struct rpc_pipe_dir_head *pdh,
+               int (*match)(struct rpc_pipe_dir_object *, void *),
+               struct rpc_pipe_dir_object *(*alloc)(void *),
+               void *data)
+{
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+       struct rpc_pipe_dir_object *pdo;
+
+       mutex_lock(&sn->pipefs_sb_lock);
+       list_for_each_entry(pdo, &pdh->pdh_entries, pdo_head) {
+               if (!match(pdo, data))
+                       continue;
+               goto out;
+       }
+       pdo = alloc(data);
+       if (!pdo)
+               goto out;
+       rpc_add_pipe_dir_object_locked(net, pdh, pdo);
+out:
+       mutex_unlock(&sn->pipefs_sb_lock);
+       return pdo;
+}
+EXPORT_SYMBOL_GPL(rpc_find_or_alloc_pipe_dir_object);
+
+static void
+rpc_create_pipe_dir_objects(struct rpc_pipe_dir_head *pdh)
+{
+       struct rpc_pipe_dir_object *pdo;
+       struct dentry *dir = pdh->pdh_dentry;
+
+       list_for_each_entry(pdo, &pdh->pdh_entries, pdo_head)
+               pdo->pdo_ops->create(dir, pdo);
+}
+
+static void
+rpc_destroy_pipe_dir_objects(struct rpc_pipe_dir_head *pdh)
+{
+       struct rpc_pipe_dir_object *pdo;
+       struct dentry *dir = pdh->pdh_dentry;
+
+       list_for_each_entry(pdo, &pdh->pdh_entries, pdo_head)
+               pdo->pdo_ops->destroy(dir, pdo);
+}
+
+enum {
+       RPCAUTH_info,
+       RPCAUTH_EOF
+};
+
+static const struct rpc_filelist authfiles[] = {
+       [RPCAUTH_info] = {
+               .name = "info",
+               .i_fop = &rpc_info_operations,
+               .mode = S_IFREG | S_IRUSR,
+       },
+};
+
+static int rpc_clntdir_populate(struct dentry *dentry, void *private)
+{
+       return rpc_populate(dentry,
+                           authfiles, RPCAUTH_info, RPCAUTH_EOF,
+                           private);
+}
+
+static void rpc_clntdir_depopulate(struct dentry *dentry)
+{
+       rpc_depopulate(dentry, authfiles, RPCAUTH_info, RPCAUTH_EOF);
+}
+
+/**
+ * rpc_create_client_dir - Create a new rpc_client directory in rpc_pipefs
+ * @dentry: the parent of new directory
+ * @name: the name of new directory
+ * @rpc_client: rpc client to associate with this directory
+ *
+ * This creates a directory at the given @path associated with
+ * @rpc_clnt, which will contain a file named "info" with some basic
+ * information about the client, together with any "pipes" that may
+ * later be created using rpc_mkpipe().
+ */
+struct dentry *rpc_create_client_dir(struct dentry *dentry,
+                                  const char *name,
+                                  struct rpc_clnt *rpc_client)
+{
+       struct dentry *ret;
+
+       ret = rpc_mkdir_populate(dentry, name, S_IRUGO | S_IXUGO, NULL,
+                       rpc_clntdir_populate, rpc_client);
+       if (!IS_ERR(ret)) {
+               rpc_client->cl_pipedir_objects.pdh_dentry = ret;
+               rpc_create_pipe_dir_objects(&rpc_client->cl_pipedir_objects);
+       }
+       return ret;
+}
+
+/**
+ * rpc_remove_client_dir - Remove a directory created with rpc_create_client_dir()
+ * @rpc_client: rpc_client for the pipe
+ */
+int rpc_remove_client_dir(struct rpc_clnt *rpc_client)
+{
+       struct dentry *dentry = rpc_client->cl_pipedir_objects.pdh_dentry;
+
+       if (dentry == NULL)
+               return 0;
+       rpc_destroy_pipe_dir_objects(&rpc_client->cl_pipedir_objects);
+       rpc_client->cl_pipedir_objects.pdh_dentry = NULL;
+       return rpc_rmdir_depopulate(dentry, rpc_clntdir_depopulate);
+}
+
+static const struct rpc_filelist cache_pipefs_files[3] = {
+       [0] = {
+               .name = "channel",
+               .i_fop = &cache_file_operations_pipefs,
+               .mode = S_IFREG|S_IRUSR|S_IWUSR,
+       },
+       [1] = {
+               .name = "content",
+               .i_fop = &content_file_operations_pipefs,
+               .mode = S_IFREG|S_IRUSR,
+       },
+       [2] = {
+               .name = "flush",
+               .i_fop = &cache_flush_operations_pipefs,
+               .mode = S_IFREG|S_IRUSR|S_IWUSR,
+       },
+};
+
+static int rpc_cachedir_populate(struct dentry *dentry, void *private)
+{
+       return rpc_populate(dentry,
+                           cache_pipefs_files, 0, 3,
+                           private);
+}
+
+static void rpc_cachedir_depopulate(struct dentry *dentry)
+{
+       rpc_depopulate(dentry, cache_pipefs_files, 0, 3);
+}
+
+struct dentry *rpc_create_cache_dir(struct dentry *parent, const char *name,
+                                   umode_t umode, struct cache_detail *cd)
+{
+       return rpc_mkdir_populate(parent, name, umode, NULL,
+                       rpc_cachedir_populate, cd);
+}
+
+void rpc_remove_cache_dir(struct dentry *dentry)
+{
+       rpc_rmdir_depopulate(dentry, rpc_cachedir_depopulate);
+}
+
+/*
+ * populate the filesystem
+ */
+static const struct super_operations s_ops = {
+       .alloc_inode    = rpc_alloc_inode,
+       .destroy_inode  = rpc_destroy_inode,
+       .statfs         = simple_statfs,
+};
+
+#define RPCAUTH_GSSMAGIC 0x67596969
+
+/*
+ * We have a single directory with 1 node in it.
+ */
+enum {
+       RPCAUTH_lockd,
+       RPCAUTH_mount,
+       RPCAUTH_nfs,
+       RPCAUTH_portmap,
+       RPCAUTH_statd,
+       RPCAUTH_nfsd4_cb,
+       RPCAUTH_cache,
+       RPCAUTH_nfsd,
+       RPCAUTH_gssd,
+       RPCAUTH_RootEOF
+};
+
+static const struct rpc_filelist files[] = {
+       [RPCAUTH_lockd] = {
+               .name = "lockd",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_mount] = {
+               .name = "mount",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_nfs] = {
+               .name = "nfs",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_portmap] = {
+               .name = "portmap",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_statd] = {
+               .name = "statd",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_nfsd4_cb] = {
+               .name = "nfsd4_cb",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_cache] = {
+               .name = "cache",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_nfsd] = {
+               .name = "nfsd",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+       [RPCAUTH_gssd] = {
+               .name = "gssd",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+};
+
+/*
+ * This call can be used only in RPC pipefs mount notification hooks.
+ */
+struct dentry *rpc_d_lookup_sb(const struct super_block *sb,
+                              const unsigned char *dir_name)
+{
+       struct qstr dir = QSTR_INIT(dir_name, strlen(dir_name));
+       return d_hash_and_lookup(sb->s_root, &dir);
+}
+EXPORT_SYMBOL_GPL(rpc_d_lookup_sb);
+
+int rpc_pipefs_init_net(struct net *net)
+{
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+       sn->gssd_dummy = rpc_mkpipe_data(&gssd_dummy_pipe_ops, 0);
+       if (IS_ERR(sn->gssd_dummy))
+               return PTR_ERR(sn->gssd_dummy);
+
+       mutex_init(&sn->pipefs_sb_lock);
+       sn->pipe_version = -1;
+       return 0;
+}
+
+void rpc_pipefs_exit_net(struct net *net)
+{
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+       rpc_destroy_pipe_data(sn->gssd_dummy);
+}
+
+/*
+ * This call will be used for per network namespace operations calls.
+ * Note: Function will be returned with pipefs_sb_lock taken if superblock was
+ * found. This lock have to be released by rpc_put_sb_net() when all operations
+ * will be completed.
+ */
+struct super_block *rpc_get_sb_net(const struct net *net)
+{
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+       mutex_lock(&sn->pipefs_sb_lock);
+       if (sn->pipefs_sb)
+               return sn->pipefs_sb;
+       mutex_unlock(&sn->pipefs_sb_lock);
+       return NULL;
+}
+EXPORT_SYMBOL_GPL(rpc_get_sb_net);
+
+void rpc_put_sb_net(const struct net *net)
+{
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+       WARN_ON(sn->pipefs_sb == NULL);
+       mutex_unlock(&sn->pipefs_sb_lock);
+}
+EXPORT_SYMBOL_GPL(rpc_put_sb_net);
+
+static const struct rpc_filelist gssd_dummy_clnt_dir[] = {
+       [0] = {
+               .name = "clntXX",
+               .mode = S_IFDIR | S_IRUGO | S_IXUGO,
+       },
+};
+
+static ssize_t
+dummy_downcall(struct file *filp, const char __user *src, size_t len)
+{
+       return -EINVAL;
+}
+
+static const struct rpc_pipe_ops gssd_dummy_pipe_ops = {
+       .upcall         = rpc_pipe_generic_upcall,
+       .downcall       = dummy_downcall,
+};
+
+/*
+ * Here we present a bogus "info" file to keep rpc.gssd happy. We don't expect
+ * that it will ever use this info to handle an upcall, but rpc.gssd expects
+ * that this file will be there and have a certain format.
+ */
+static int
+rpc_show_dummy_info(struct seq_file *m, void *v)
+{
+       seq_printf(m, "RPC server: %s\n", utsname()->nodename);
+       seq_printf(m, "service: foo (1) version 0\n");
+       seq_printf(m, "address: 127.0.0.1\n");
+       seq_printf(m, "protocol: tcp\n");
+       seq_printf(m, "port: 0\n");
+       return 0;
+}
+
+static int
+rpc_dummy_info_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, rpc_show_dummy_info, NULL);
+}
+
+static const struct file_operations rpc_dummy_info_operations = {
+       .owner          = THIS_MODULE,
+       .open           = rpc_dummy_info_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static const struct rpc_filelist gssd_dummy_info_file[] = {
+       [0] = {
+               .name = "info",
+               .i_fop = &rpc_dummy_info_operations,
+               .mode = S_IFREG | S_IRUSR,
+       },
+};
+
+/**
+ * rpc_gssd_dummy_populate - create a dummy gssd pipe
+ * @root:      root of the rpc_pipefs filesystem
+ * @pipe_data: pipe data created when netns is initialized
+ *
+ * Create a dummy set of directories and a pipe that gssd can hold open to
+ * indicate that it is up and running.
+ */
+static struct dentry *
+rpc_gssd_dummy_populate(struct dentry *root, struct rpc_pipe *pipe_data)
+{
+       int ret = 0;
+       struct dentry *gssd_dentry;
+       struct dentry *clnt_dentry = NULL;
+       struct dentry *pipe_dentry = NULL;
+       struct qstr q = QSTR_INIT(files[RPCAUTH_gssd].name,
+                                 strlen(files[RPCAUTH_gssd].name));
+
+       /* We should never get this far if "gssd" doesn't exist */
+       gssd_dentry = d_hash_and_lookup(root, &q);
+       if (!gssd_dentry)
+               return ERR_PTR(-ENOENT);
+
+       ret = rpc_populate(gssd_dentry, gssd_dummy_clnt_dir, 0, 1, NULL);
+       if (ret) {
+               pipe_dentry = ERR_PTR(ret);
+               goto out;
+       }
+
+       q.name = gssd_dummy_clnt_dir[0].name;
+       q.len = strlen(gssd_dummy_clnt_dir[0].name);
+       clnt_dentry = d_hash_and_lookup(gssd_dentry, &q);
+       if (!clnt_dentry) {
+               pipe_dentry = ERR_PTR(-ENOENT);
+               goto out;
+       }
+
+       ret = rpc_populate(clnt_dentry, gssd_dummy_info_file, 0, 1, NULL);
+       if (ret) {
+               __rpc_depopulate(gssd_dentry, gssd_dummy_clnt_dir, 0, 1);
+               pipe_dentry = ERR_PTR(ret);
+               goto out;
+       }
+
+       pipe_dentry = rpc_mkpipe_dentry(clnt_dentry, "gssd", NULL, pipe_data);
+       if (IS_ERR(pipe_dentry)) {
+               __rpc_depopulate(clnt_dentry, gssd_dummy_info_file, 0, 1);
+               __rpc_depopulate(gssd_dentry, gssd_dummy_clnt_dir, 0, 1);
+       }
+out:
+       dput(clnt_dentry);
+       dput(gssd_dentry);
+       return pipe_dentry;
+}
+
+static void
+rpc_gssd_dummy_depopulate(struct dentry *pipe_dentry)
+{
+       struct dentry *clnt_dir = pipe_dentry->d_parent;
+       struct dentry *gssd_dir = clnt_dir->d_parent;
+
+       __rpc_rmpipe(d_inode(clnt_dir), pipe_dentry);
+       __rpc_depopulate(clnt_dir, gssd_dummy_info_file, 0, 1);
+       __rpc_depopulate(gssd_dir, gssd_dummy_clnt_dir, 0, 1);
+       dput(pipe_dentry);
+}
+
+static int
+rpc_fill_super(struct super_block *sb, void *data, int silent)
+{
+       struct inode *inode;
+       struct dentry *root, *gssd_dentry;
+       struct net *net = data;
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+       int err;
+
+       sb->s_blocksize = PAGE_CACHE_SIZE;
+       sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
+       sb->s_magic = RPCAUTH_GSSMAGIC;
+       sb->s_op = &s_ops;
+       sb->s_d_op = &simple_dentry_operations;
+       sb->s_time_gran = 1;
+
+       inode = rpc_get_inode(sb, S_IFDIR | S_IRUGO | S_IXUGO);
+       sb->s_root = root = d_make_root(inode);
+       if (!root)
+               return -ENOMEM;
+       if (rpc_populate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF, NULL))
+               return -ENOMEM;
+
+       gssd_dentry = rpc_gssd_dummy_populate(root, sn->gssd_dummy);
+       if (IS_ERR(gssd_dentry)) {
+               __rpc_depopulate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF);
+               return PTR_ERR(gssd_dentry);
+       }
+
+       dprintk("RPC:       sending pipefs MOUNT notification for net %p%s\n",
+               net, NET_NAME(net));
+       mutex_lock(&sn->pipefs_sb_lock);
+       sn->pipefs_sb = sb;
+       err = blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
+                                          RPC_PIPEFS_MOUNT,
+                                          sb);
+       if (err)
+               goto err_depopulate;
+       sb->s_fs_info = get_net(net);
+       mutex_unlock(&sn->pipefs_sb_lock);
+       return 0;
+
+err_depopulate:
+       rpc_gssd_dummy_depopulate(gssd_dentry);
+       blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
+                                          RPC_PIPEFS_UMOUNT,
+                                          sb);
+       sn->pipefs_sb = NULL;
+       __rpc_depopulate(root, files, RPCAUTH_lockd, RPCAUTH_RootEOF);
+       mutex_unlock(&sn->pipefs_sb_lock);
+       return err;
+}
+
+bool
+gssd_running(struct net *net)
+{
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+       struct rpc_pipe *pipe = sn->gssd_dummy;
+
+       return pipe->nreaders || pipe->nwriters;
+}
+EXPORT_SYMBOL_GPL(gssd_running);
+
+static struct dentry *
+rpc_mount(struct file_system_type *fs_type,
+               int flags, const char *dev_name, void *data)
+{
+       return mount_ns(fs_type, flags, current->nsproxy->net_ns, rpc_fill_super);
+}
+
+static void rpc_kill_sb(struct super_block *sb)
+{
+       struct net *net = sb->s_fs_info;
+       struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
+
+       mutex_lock(&sn->pipefs_sb_lock);
+       if (sn->pipefs_sb != sb) {
+               mutex_unlock(&sn->pipefs_sb_lock);
+               goto out;
+       }
+       sn->pipefs_sb = NULL;
+       dprintk("RPC:       sending pipefs UMOUNT notification for net %p%s\n",
+               net, NET_NAME(net));
+       blocking_notifier_call_chain(&rpc_pipefs_notifier_list,
+                                          RPC_PIPEFS_UMOUNT,
+                                          sb);
+       mutex_unlock(&sn->pipefs_sb_lock);
+       put_net(net);
+out:
+       kill_litter_super(sb);
+}
+
+static struct file_system_type rpc_pipe_fs_type = {
+       .owner          = THIS_MODULE,
+       .name           = "rpc_pipefs",
+       .mount          = rpc_mount,
+       .kill_sb        = rpc_kill_sb,
+};
+MODULE_ALIAS_FS("rpc_pipefs");
+MODULE_ALIAS("rpc_pipefs");
+
+static void
+init_once(void *foo)
+{
+       struct rpc_inode *rpci = (struct rpc_inode *) foo;
+
+       inode_init_once(&rpci->vfs_inode);
+       rpci->private = NULL;
+       rpci->pipe = NULL;
+       init_waitqueue_head(&rpci->waitq);
+}
+
+int register_rpc_pipefs(void)
+{
+       int err;
+
+       rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
+                               sizeof(struct rpc_inode),
+                               0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
+                                               SLAB_MEM_SPREAD),
+                               init_once);
+       if (!rpc_inode_cachep)
+               return -ENOMEM;
+       err = rpc_clients_notifier_register();
+       if (err)
+               goto err_notifier;
+       err = register_filesystem(&rpc_pipe_fs_type);
+       if (err)
+               goto err_register;
+       return 0;
+
+err_register:
+       rpc_clients_notifier_unregister();
+err_notifier:
+       kmem_cache_destroy(rpc_inode_cachep);
+       return err;
+}
+
+void unregister_rpc_pipefs(void)
+{
+       rpc_clients_notifier_unregister();
+       kmem_cache_destroy(rpc_inode_cachep);
+       unregister_filesystem(&rpc_pipe_fs_type);
+}