Linux-RDMA Archive on lore.kernel.org
 help / color / Atom feed
From: Chuck Lever <chuck.lever@oracle.com>
To: trondmy@hammerspace.com
Cc: linux-nfs@vger.kernel.org, linux-rdma@vger.kernel.org
Subject: [PATCH v3 16/26] xprtrdma: Do not wake RPC consumer on a failed LocalInv
Date: Mon, 19 Apr 2021 14:03:19 -0400
Message-ID: <161885539919.38598.16197910600306542657.stgit@manet.1015granger.net> (raw)
In-Reply-To: <161885481568.38598.16682844600209775665.stgit@manet.1015granger.net>

Throw away any reply where the LocalInv flushes or could not be
posted. The registered memory region is in an unknown state until
the disconnect completes.

rpcrdma_xprt_disconnect() will find and release the MR. No need to
put it back on the MR free list in this case.

The client retransmits pending RPC requests once it reestablishes a
fresh connection, so a replacement reply should be forthcoming on
the next connection instance.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 net/sunrpc/xprtrdma/frwr_ops.c  |   17 +++++++++++------
 net/sunrpc/xprtrdma/rpc_rdma.c  |   32 +++++++++++++++++++++++++++++---
 net/sunrpc/xprtrdma/xprt_rdma.h |    1 +
 3 files changed, 41 insertions(+), 9 deletions(-)

diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c
index 27087dc8ba3c..951ae20485f3 100644
--- a/net/sunrpc/xprtrdma/frwr_ops.c
+++ b/net/sunrpc/xprtrdma/frwr_ops.c
@@ -576,10 +576,14 @@ static void frwr_wc_localinv_done(struct ib_cq *cq, struct ib_wc *wc)
 	rep = mr->mr_req->rl_reply;
 	smp_rmb();
 
-	frwr_mr_done(wc, mr);
+	if (wc->status != IB_WC_SUCCESS) {
+		if (rep)
+			rpcrdma_unpin_rqst(rep);
+		rpcrdma_flush_disconnect(cq->cq_context, wc);
+		return;
+	}
+	frwr_mr_put(mr);
 	rpcrdma_complete_rqst(rep);
-
-	rpcrdma_flush_disconnect(cq->cq_context, wc);
 }
 
 /**
@@ -645,8 +649,9 @@ void frwr_unmap_async(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
 	trace_xprtrdma_post_linv_err(req, rc);
 
 	/* The final LOCAL_INV WR in the chain is supposed to
-	 * do the wake. If it was never posted, the wake will
-	 * not happen, so wake here in that case.
+	 * do the wake. If it was never posted, the wake does
+	 * not happen. Unpin the rqst in preparation for its
+	 * retransmission.
 	 */
-	rpcrdma_complete_rqst(req->rl_reply);
+	rpcrdma_unpin_rqst(req->rl_reply);
 }
diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
index be4e888e78a3..649f7d8b9733 100644
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -1326,9 +1326,35 @@ rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
 	return -EIO;
 }
 
-/* Perform XID lookup, reconstruction of the RPC reply, and
- * RPC completion while holding the transport lock to ensure
- * the rep, rqst, and rq_task pointers remain stable.
+/**
+ * rpcrdma_unpin_rqst - Release rqst without completing it
+ * @rep: RPC/RDMA Receive context
+ *
+ * This is done when a connection is lost so that a Reply
+ * can be dropped and its matching Call can be subsequently
+ * retransmitted on a new connection.
+ */
+void rpcrdma_unpin_rqst(struct rpcrdma_rep *rep)
+{
+	struct rpc_xprt *xprt = &rep->rr_rxprt->rx_xprt;
+	struct rpc_rqst *rqst = rep->rr_rqst;
+	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
+
+	req->rl_reply = NULL;
+	rep->rr_rqst = NULL;
+
+	spin_lock(&xprt->queue_lock);
+	xprt_unpin_rqst(rqst);
+	spin_unlock(&xprt->queue_lock);
+}
+
+/**
+ * rpcrdma_complete_rqst - Pass completed rqst back to RPC
+ * @rep: RPC/RDMA Receive context
+ *
+ * Reconstruct the RPC reply and complete the transaction
+ * while @rqst is still pinned to ensure the rep, rqst, and
+ * rq_task pointers remain stable.
  */
 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep)
 {
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index 1b187d1dee8a..bb8aba390b88 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -561,6 +561,7 @@ int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
 void rpcrdma_set_max_header_sizes(struct rpcrdma_ep *ep);
 void rpcrdma_reset_cwnd(struct rpcrdma_xprt *r_xprt);
 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep);
+void rpcrdma_unpin_rqst(struct rpcrdma_rep *rep);
 void rpcrdma_reply_handler(struct rpcrdma_rep *rep);
 
 static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)



  parent reply index

Thread overview: 34+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-04-19 18:01 [PATCH v3 00/26] NFS/RDMA client patches for next Chuck Lever
2021-04-19 18:01 ` [PATCH v3 01/26] SUNRPC: Move fault injection call sites Chuck Lever
2021-04-19 18:05   ` Chuck Lever III
2021-04-19 18:01 ` [PATCH v3 02/26] SUNRPC: Remove trace_xprt_transmit_queued Chuck Lever
2021-04-19 18:01 ` [PATCH v3 03/26] SUNRPC: Add tracepoint that fires when an RPC is retransmitted Chuck Lever
2021-04-19 18:02 ` [PATCH v3 04/26] xprtrdma: Avoid Receive Queue wrapping Chuck Lever
2021-04-19 18:02 ` [PATCH v3 05/26] xprtrdma: Do not refresh Receive Queue while it is draining Chuck Lever
2021-04-19 18:02 ` [PATCH v3 06/26] xprtrdma: Put flushed Receives on free list instead of destroying them Chuck Lever
2021-04-19 18:02 ` [PATCH v3 07/26] xprtrdma: Improve locking around rpcrdma_rep destruction Chuck Lever
2021-04-23 21:06   ` Trond Myklebust
2021-04-24 17:39     ` Chuck Lever III
2021-04-24 17:59       ` Trond Myklebust
2021-04-19 18:02 ` [PATCH v3 08/26] xprtrdma: Improve commentary around rpcrdma_reps_unmap() Chuck Lever
2021-04-19 18:02 ` [PATCH v3 09/26] xprtrdma: Improve locking around rpcrdma_rep creation Chuck Lever
2021-04-19 18:02 ` [PATCH v3 10/26] xprtrdma: Fix cwnd update ordering Chuck Lever
2021-04-19 18:02 ` [PATCH v3 11/26] xprtrdma: Delete rpcrdma_recv_buffer_put() Chuck Lever
2021-04-19 18:02 ` [PATCH v3 12/26] xprtrdma: rpcrdma_mr_pop() already does list_del_init() Chuck Lever
2021-04-19 18:03 ` [PATCH v3 13/26] xprtrdma: Rename frwr_release_mr() Chuck Lever
2021-04-19 18:03 ` [PATCH v3 14/26] xprtrdma: Clarify use of barrier in frwr_wc_localinv_done() Chuck Lever
2021-04-19 18:03 ` [PATCH v3 15/26] xprtrdma: Do not recycle MR after FastReg/LocalInv flushes Chuck Lever
2021-04-25 14:19   ` Dan Aloni
2021-04-25 16:21     ` Chuck Lever III
2021-04-25 17:00       ` Dan Aloni
2021-04-19 18:03 ` Chuck Lever [this message]
2021-04-19 18:03 ` [PATCH v3 17/26] xprtrdma: Avoid Send Queue wrapping Chuck Lever
2021-04-19 18:03 ` [PATCH v3 18/26] xprtrdma: Add tracepoints showing FastReg WRs and remote invalidation Chuck Lever
2021-04-19 18:03 ` [PATCH v3 19/26] xprtrdma: Add an rpcrdma_mr_completion_class Chuck Lever
2021-04-19 18:03 ` [PATCH v3 20/26] xprtrdma: Don't display r_xprt memory addresses in tracepoints Chuck Lever
2021-04-19 18:03 ` [PATCH v3 21/26] xprtrdma: Remove the RPC/RDMA QP event handler Chuck Lever
2021-04-19 18:03 ` [PATCH v3 22/26] xprtrdma: Move fr_cid to struct rpcrdma_mr Chuck Lever
2021-04-19 18:04 ` [PATCH v3 23/26] xprtrdma: Move cqe " Chuck Lever
2021-04-19 18:04 ` [PATCH v3 24/26] xprtrdma: Move fr_linv_done field " Chuck Lever
2021-04-19 18:04 ` [PATCH v3 25/26] xprtrdma: Move the Work Request union " Chuck Lever
2021-04-19 18:04 ` [PATCH v3 26/26] xprtrdma: Move fr_mr field " Chuck Lever

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=161885539919.38598.16197910600306542657.stgit@manet.1015granger.net \
    --to=chuck.lever@oracle.com \
    --cc=linux-nfs@vger.kernel.org \
    --cc=linux-rdma@vger.kernel.org \
    --cc=trondmy@hammerspace.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Linux-RDMA Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-rdma/0 linux-rdma/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-rdma linux-rdma/ https://lore.kernel.org/linux-rdma \
		linux-rdma@vger.kernel.org
	public-inbox-index linux-rdma

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-rdma


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git