Commit ef739b21 authored by Chuck Lever's avatar Chuck Lever Committed by Anna Schumaker

xprtrdma: Reset credit grant properly after a disconnect

On a fresh connection, an RPC/RDMA client is supposed to send only
one RPC Call until it gets a credit grant in the first RPC Reply
from the server [RFC 8166, Section 3.3.3].

There is a bug in the Linux client's credit accounting mechanism
introduced by commit e7ce710a ("xprtrdma: Avoid deadlock when
credit window is reset"). On connect, it simply dumps all pending
RPC Calls onto the new connection.

Servers have been tolerant of this bad behavior. Currently no server
implementation ever changes its credit grant over reconnects, and
servers always repost enough Receives before connections are fully
established.

To correct this issue, ensure that the client resets both the credit
grant _and_ the congestion window when handling a reconnect.

Fixes: e7ce710a ("xprtrdma: Avoid deadlock when credit ... ")
Signed-off-by: default avatarChuck Lever <chuck.lever@oracle.com>
Cc: stable@kernel.org
Signed-off-by: default avatarAnna Schumaker <Anna.Schumaker@Netapp.com>
parent 91ca1866
...@@ -248,6 +248,7 @@ static void ...@@ -248,6 +248,7 @@ static void
xprt_rdma_bc_close(struct rpc_xprt *xprt) xprt_rdma_bc_close(struct rpc_xprt *xprt)
{ {
dprintk("svcrdma: %s: xprt %p\n", __func__, xprt); dprintk("svcrdma: %s: xprt %p\n", __func__, xprt);
xprt->cwnd = RPC_CWNDSHIFT;
} }
static void static void
......
...@@ -468,6 +468,12 @@ xprt_rdma_close(struct rpc_xprt *xprt) ...@@ -468,6 +468,12 @@ xprt_rdma_close(struct rpc_xprt *xprt)
xprt->reestablish_timeout = 0; xprt->reestablish_timeout = 0;
xprt_disconnect_done(xprt); xprt_disconnect_done(xprt);
rpcrdma_ep_disconnect(ep, ia); rpcrdma_ep_disconnect(ep, ia);
/* Prepare @xprt for the next connection by reinitializing
* its credit grant to one (see RFC 8166, Section 3.3.3).
*/
r_xprt->rx_buf.rb_credits = 1;
xprt->cwnd = RPC_CWNDSHIFT;
} }
/** /**
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment