nfs: fix a deadlock in nfs client initialization

The following deadlock can occur between a process waiting for a client
to initialize in while walking the client list during nfsv4 server trunking
detection and another process waiting for the nfs_clid_init_mutex so it
can initialize that client:

Process 1                               Process 2
---------                               ---------
spin_lock(&nn->nfs_client_lock);
list_add_tail(&CLIENTA->cl_share_link,
        &nn->nfs_client_list);
spin_unlock(&nn->nfs_client_lock);
                                        spin_lock(&nn->nfs_client_lock);
                                        list_add_tail(&CLIENTB->cl_share_link,
                                                &nn->nfs_client_list);
                                        spin_unlock(&nn->nfs_client_lock);
                                        mutex_lock(&nfs_clid_init_mutex);
                                        nfs41_walk_client_list(clp, result, cred);
                                        nfs_wait_client_init_complete(CLIENTA);
(waiting for nfs_clid_init_mutex)

Make sure nfs_match_client() only evaluates clients that have completed
initialization in order to prevent that deadlock.

This patch also fixes v4.0 trunking behavior by not marking the client
NFS_CS_READY until the clientid has been confirmed.

Signed-off-by: Scott Mayhew <smayhew@redhat.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

2 files changed, 24 insertions, 4 deletions

diff --git a/fs/nfs/client.c b/fs/nfs/client.c
index 0ac2fb1c6b63..b9129e2befea 100644
--- a/fs/nfs/client.c
+++ b/fs/nfs/client.c
@@ -291,12 +291,23 @@ static struct nfs_client *nfs_match_client(const struct nfs_client_initdata *dat
         const struct sockaddr *sap = data->addr;
         struct nfs_net *nn = net_generic(data->net, nfs_net_id);
 
+again:
         list_for_each_entry(clp, &nn->nfs_client_list, cl_share_link) {
                 const struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
                 /* Don't match clients that failed to initialise properly */
                 if (clp->cl_cons_state < 0)
                         continue;
 
+                /* If a client is still initializing then we need to wait */
+                if (clp->cl_cons_state > NFS_CS_READY) {
+                        refcount_inc(&clp->cl_count);
+                        spin_unlock(&nn->nfs_client_lock);
+                        nfs_wait_client_init_complete(clp);
+                        nfs_put_client(clp);
+                        spin_lock(&nn->nfs_client_lock);
+                        goto again;
+                }
+
                 /* Different NFS versions cannot share the same nfs_client */
                 if (clp->rpc_ops != data->nfs_mod->rpc_ops)
                         continue;
diff --git a/fs/nfs/nfs4client.c b/fs/nfs/nfs4client.c
index 12bbab0becb4..65a7e5da508c 100644
--- a/fs/nfs/nfs4client.c
+++ b/fs/nfs/nfs4client.c
@@ -404,15 +404,19 @@ struct nfs_client *nfs4_init_client(struct nfs_client *clp,
         if (error < 0)
                 goto error;
 
-        if (!nfs4_has_session(clp))
-                nfs_mark_client_ready(clp, NFS_CS_READY);
-
         error = nfs4_discover_server_trunking(clp, &old);
         if (error < 0)
                 goto error;
 
-        if (clp != old)
+        if (clp != old) {
                 clp->cl_preserve_clid = true;
+                /*
+                 * Mark the client as having failed initialization so other
+                 * processes walking the nfs_client_list in nfs_match_client()
+                 * won't try to use it.
+                 */
+                nfs_mark_client_ready(clp, -EPERM);
+        }
         nfs_put_client(clp);
         clear_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags);
         return old;
@@ -539,6 +543,9 @@ int nfs40_walk_client_list(struct nfs_client *new,
         spin_lock(&nn->nfs_client_lock);
         list_for_each_entry(pos, &nn->nfs_client_list, cl_share_link) {
 
+                if (pos == new)
+                        goto found;
+
                 status = nfs4_match_client(pos, new, &prev, nn);
                 if (status < 0)
                         goto out_unlock;
@@ -559,6 +566,7 @@ int nfs40_walk_client_list(struct nfs_client *new,
                  * way that a SETCLIENTID_CONFIRM to pos can succeed is
                  * if new and pos point to the same server:
                  */
+found:
                 refcount_inc(&pos->cl_count);
                 spin_unlock(&nn->nfs_client_lock);
 
@@ -572,6 +580,7 @@ int nfs40_walk_client_list(struct nfs_client *new,
                 case 0:
                         nfs4_swap_callback_idents(pos, new);
                         pos->cl_confirm = new->cl_confirm;
+                        nfs_mark_client_ready(pos, NFS_CS_READY);
 
                         prev = NULL;
                         *result = pos;