Add a CLSET_ASYNC command, which allows us to (ab)use the clnt_dg transport

to make asynchronous RPCs. This is needed to help fix ypbind, which can no
longer override the clnt_dg_call() method (formerly the clntudp_call()
method) due to all the internal descriptor locking code in TI-RPC. Turning
on this flag allows us to send an RPC request, then return immediately,
and handle a reply later, rather than being forced to do the request
and reply in a single function call.

Also fix a byte ordering bug: when clnt_dg_call() increments the XID
prior to transmitting a request, it uses the raw value, which is wrong.
The XID is stored in network byte order, i.e. big-endian. The CLSET_XID
and CLGET_XID commands in clnt_dg_control() use ntohl()/htonl() to get
the byte ordering right, but because clnt_dg_call() does not do this,
using CLSET_XID/CLGET_XID doesn't actually work, unless you're on a
big endian host, which we aren't (yet). Fix clnt_dg_call() to byte swap
properly when doing the increment.

1 files changed, 1 insertions, 0 deletions

diff --git a/include/rpc/clnt.h b/include/rpc/clnt.h
index 2b0587b169ef..f845e8e346b5 100644
--- a/include/rpc/clnt.h
+++ b/include/rpc/clnt.h
@@ -228,6 +228,7 @@ struct rpc_timers {
  */
 #define CLSET_RETRY_TIMEOUT 4   /* set retry timeout (timeval) */
 #define CLGET_RETRY_TIMEOUT 5   /* get retry timeout (timeval) */
+#define CLSET_ASYNC		19
 
 /*
  * void