On Fri, 2018-11-30 at 12:08 -0500, Chuck Lever wrote:
> Hi-
>
> I'm testing krb5/krb5i/krb5p with stock v4.20-rc4 and NFS/TCP.
> The workload is synthetic:
>
> /home/cel/bin/iozone -M -+u -i0 -i1 -s1g -r8k -t16 -c
>
> The client is a 12-core Xeon system.
>
>
> I'm seeing all kinds of symptoms:
>
> - memory leaks: bvec and enc_pages are leaking
>
> - EBADMSG is reported to user space
>
> - data corruption
>
> - connect deadlocks resulting in a mount hang
>
> - invalid soft IRQ receive buffer warnings
>
Does the following patch help?
8<----------------------------------------------
>From 8ff4cd9f0f6912e14f657371b6b7eecf6d2091ee Mon Sep 17 00:00:00 2001
From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
Date: Fri, 30 Nov 2018 12:48:47 -0500
Subject: [PATCH] SUNRPC: call_connect_status() must handle tasks that got
transmitted
If a task failed to get the write lock in the call to xprt_connect(), then
it will be queued on xprt->sending. In that case, it is possible for it
to get transmitted before the call to call_connect_status(), in which
case it needs to be handled by call_transmit_status() instead.
Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
---
net/sunrpc/clnt.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index ae3b8145da35..e35d642558e7 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1915,6 +1915,13 @@ call_connect_status(struct rpc_task *task)
struct rpc_clnt *clnt = task->tk_client;
int status = task->tk_status;
+ /* Check if the task was already transmitted */
+ if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate)) {
+ xprt_end_transmit(task);
+ task->tk_action = call_transmit_status;
+ return;
+ }
+
dprint_status(task);
trace_rpc_connect_status(task);
--
2.19.2
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@xxxxxxxxxxxxxxx
