RDMA/rxe: Replace mr by rkey in responder resources

Currently rxe saves a copy of MR in responder resources for RDMA reads.
Since the responder resources are never freed just over written if more
are needed this MR may not have a reference freed until the QP is
destroyed. This patch uses the rkey instead of the MR and on subsequent
packets of a multipacket read reply message it looks up the MR from the
rkey for each packet. This makes it possible for a user to deregister an
MR or unbind a MW on the fly and get correct behaviour.

Link: https://lore.kernel.org/r/20220304000808.225811-3-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

1 files changed, 85 insertions, 38 deletions

diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c
index b5ebe853748a..b1ec003f0bb8 100644
--- a/drivers/infiniband/sw/rxe/rxe_resp.c
+++ b/drivers/infiniband/sw/rxe/rxe_resp.c
@@ -642,6 +642,78 @@ static struct sk_buff *prepare_ack_packet(struct rxe_qp *qp,
 	return skb;
 }
 
+static struct resp_res *rxe_prepare_read_res(struct rxe_qp *qp,
+					struct rxe_pkt_info *pkt)
+{
+	struct resp_res *res;
+	u32 pkts;
+
+	res = &qp->resp.resources[qp->resp.res_head];
+	rxe_advance_resp_resource(qp);
+	free_rd_atomic_resource(qp, res);
+
+	res->type = RXE_READ_MASK;
+	res->replay = 0;
+	res->read.va = qp->resp.va + qp->resp.offset;
+	res->read.va_org = qp->resp.va + qp->resp.offset;
+	res->read.resid = qp->resp.resid;
+	res->read.length = qp->resp.resid;
+	res->read.rkey = qp->resp.rkey;
+
+	pkts = max_t(u32, (reth_len(pkt) + qp->mtu - 1)/qp->mtu, 1);
+	res->first_psn = pkt->psn;
+	res->cur_psn = pkt->psn;
+	res->last_psn = (pkt->psn + pkts - 1) & BTH_PSN_MASK;
+
+	res->state = rdatm_res_state_new;
+
+	return res;
+}
+
+/**
+ * rxe_recheck_mr - revalidate MR from rkey and get a reference
+ * @qp: the qp
+ * @rkey: the rkey
+ *
+ * This code allows the MR to be invalidated or deregistered or
+ * the MW if one was used to be invalidated or deallocated.
+ * It is assumed that the access permissions if originally good
+ * are OK and the mappings to be unchanged.
+ *
+ * Return: mr on success else NULL
+ */
+static struct rxe_mr *rxe_recheck_mr(struct rxe_qp *qp, u32 rkey)
+{
+	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
+	struct rxe_mr *mr;
+	struct rxe_mw *mw;
+
+	if (rkey_is_mw(rkey)) {
+		mw = rxe_pool_get_index(&rxe->mw_pool, rkey >> 8);
+		if (!mw || mw->rkey != rkey)
+			return NULL;
+
+		if (mw->state != RXE_MW_STATE_VALID) {
+			rxe_drop_ref(mw);
+			return NULL;
+		}
+
+		mr = mw->mr;
+		rxe_drop_ref(mw);
+	} else {
+		mr = rxe_pool_get_index(&rxe->mr_pool, rkey >> 8);
+		if (!mr || mr->rkey != rkey)
+			return NULL;
+	}
+
+	if (mr->state != RXE_MR_STATE_VALID) {
+		rxe_drop_ref(mr);
+		return NULL;
+	}
+
+	return mr;
+}
+
 /* RDMA read response. If res is not NULL, then we have a current RDMA request
  * being processed or replayed.
  */
@@ -656,53 +728,26 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 	int opcode;
 	int err;
 	struct resp_res *res = qp->resp.res;
+	struct rxe_mr *mr;
 
 	if (!res) {
-		/* This is the first time we process that request. Get a
-		 * resource
-		 */
-		res = &qp->resp.resources[qp->resp.res_head];
-
-		free_rd_atomic_resource(qp, res);
-		rxe_advance_resp_resource(qp);
-
-		res->type		= RXE_READ_MASK;
-		res->replay		= 0;
-
-		res->read.va		= qp->resp.va +
-					  qp->resp.offset;
-		res->read.va_org	= qp->resp.va +
-					  qp->resp.offset;
-
-		res->first_psn		= req_pkt->psn;
-
-		if (reth_len(req_pkt)) {
-			res->last_psn	= (req_pkt->psn +
-					   (reth_len(req_pkt) + mtu - 1) /
-					   mtu - 1) & BTH_PSN_MASK;
-		} else {
-			res->last_psn	= res->first_psn;
-		}
-		res->cur_psn		= req_pkt->psn;
-
-		res->read.resid		= qp->resp.resid;
-		res->read.length	= qp->resp.resid;
-		res->read.rkey		= qp->resp.rkey;
-
-		/* note res inherits the reference to mr from qp */
-		res->read.mr		= qp->resp.mr;
-		qp->resp.mr		= NULL;
-
-		qp->resp.res		= res;
-		res->state		= rdatm_res_state_new;
+		res = rxe_prepare_read_res(qp, req_pkt);
+		qp->resp.res = res;
 	}
 
 	if (res->state == rdatm_res_state_new) {
+		mr = qp->resp.mr;
+		qp->resp.mr = NULL;
+
 		if (res->read.resid <= mtu)
 			opcode = IB_OPCODE_RC_RDMA_READ_RESPONSE_ONLY;
 		else
 			opcode = IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST;
 	} else {
+		mr = rxe_recheck_mr(qp, res->read.rkey);
+		if (!mr)
+			return RESPST_ERR_RKEY_VIOLATION;
+
 		if (res->read.resid > mtu)
 			opcode = IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE;
 		else
@@ -718,10 +763,12 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 	if (!skb)
 		return RESPST_ERR_RNR;
 
-	err = rxe_mr_copy(res->read.mr, res->read.va, payload_addr(&ack_pkt),
+	err = rxe_mr_copy(mr, res->read.va, payload_addr(&ack_pkt),
 			  payload, RXE_FROM_MR_OBJ);
 	if (err)
 		pr_err("Failed copying memory\n");
+	if (mr)
+		rxe_drop_ref(mr);
 
 	if (bth_pad(&ack_pkt)) {
 		u8 *pad = payload_addr(&ack_pkt) + payload;