svcrdma: Fix another Receive buffer leak
authorChuck Lever <chuck.lever@oracle.com>
Tue, 30 Jun 2020 19:55:45 +0000 (15:55 -0400)
committerChuck Lever <chuck.lever@oracle.com>
Tue, 28 Jul 2020 14:18:13 +0000 (10:18 -0400)
commit64d26422516b2e347b32e6d9b1d40b3c19a62aae
tree246fe694a3ee7442335be356ca3c7a908e4a2713
parenta68d5a502bbacfbd31f98371f777d574b3a91baf
svcrdma: Fix another Receive buffer leak

During a connection tear down, the Receive queue is flushed before
the device resources are freed. Typically, all the Receives flush
with IB_WR_FLUSH_ERR.

However, any pending successful Receives flush with IB_WR_SUCCESS,
and the server automatically posts a fresh Receive to replace the
completing one. This happens even after the connection has closed
and the RQ is drained. Receives that are posted after the RQ is
drained appear never to complete, causing a Receive resource leak.
The leaked Receive buffer is left DMA-mapped.

To prevent these late-posted recv_ctxt's from leaking, block new
Receive posting after XPT_CLOSE is set.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
net/sunrpc/xprtrdma/svc_rdma_recvfrom.c