On 7/14/2023 11:13 AM, Chuck Lever wrote:
From: Chuck Lever <chuck.lever@xxxxxxxxxx>
On server-initiated disconnect, rpcrdma_xprt_disconnect() DMA-unmaps
the transport's Receive buffers, but rpcrdma_post_recvs() neglected
to remap them after a new connection had been established. The
result is immediate failure of the new connection with the Receives
flushing with LOCAL_PROT_ERR.
The fix is correct, my only comment is that the failure occurs when the
first message is received, and only the first CQE is a LOCAL_PROT_ERR.
The remainder of the posted receives are simply flushed.
Same result of course! The summary is ok as-is. Important fix.
Reviewed-by: Tom Talpey <tom@xxxxxxxxxx>
Fixes: 671c450b6fe0 ("xprtrdma: Fix oops in Receive handler after device removal")
Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
---
net/sunrpc/xprtrdma/verbs.c | 9 ++++-----
1 file changed, 4 insertions(+), 5 deletions(-)
Hi Anna-
Can you apply this for 6.5-rc ?
diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index b098fde373ab..28c0771c4e8c 100644
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -935,9 +935,6 @@ struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt,
if (!rep->rr_rdmabuf)
goto out_free;
- if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf))
- goto out_free_regbuf;
-
rep->rr_cid.ci_completion_id =
atomic_inc_return(&r_xprt->rx_ep->re_completion_ids);
@@ -956,8 +953,6 @@ struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt,
spin_unlock(&buf->rb_lock);
return rep;
-out_free_regbuf:
- rpcrdma_regbuf_free(rep->rr_rdmabuf);
out_free:
kfree(rep);
out:
@@ -1363,6 +1358,10 @@ void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, int needed, bool temp)
rep = rpcrdma_rep_create(r_xprt, temp);
if (!rep)
break;
+ if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf)) {
+ rpcrdma_rep_put(buf, rep);
+ break;
+ }
rep->rr_cid.ci_queue_id = ep->re_attr.recv_cq->res.id;
trace_xprtrdma_post_recv(rep);
