1 files changed, 216 insertions, 111 deletions

diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c
index 794ba4ca0994..0b6dad7580a1 100644
--- a/net/sunrpc/xprtrdma/frwr_ops.c
+++ b/net/sunrpc/xprtrdma/frwr_ops.c
@@ -144,6 +144,26 @@ frwr_mr_recycle_worker(struct work_struct *work)
 	frwr_release_mr(mr);
 }
 
+/* frwr_reset - Place MRs back on the free list
+ * @req: request to reset
+ *
+ * Used after a failed marshal. For FRWR, this means the MRs
+ * don't have to be fully released and recreated.
+ *
+ * NB: This is safe only as long as none of @req's MRs are
+ * involved with an ongoing asynchronous FAST_REG or LOCAL_INV
+ * Work Request.
+ */
+void frwr_reset(struct rpcrdma_req *req)
+{
+	while (!list_empty(&req->rl_registered)) {
+		struct rpcrdma_mr *mr;
+
+		mr = rpcrdma_mr_pop(&req->rl_registered);
+		rpcrdma_mr_unmap_and_put(mr);
+	}
+}
+
 /**
  * frwr_init_mr - Initialize one MR
  * @ia: interface adapter
@@ -168,7 +188,6 @@ int frwr_init_mr(struct rpcrdma_ia *ia, struct rpcrdma_mr *mr)
 		goto out_list_err;
 
 	mr->frwr.fr_mr = frmr;
-	mr->frwr.fr_state = FRWR_IS_INVALID;
 	mr->mr_dir = DMA_NONE;
 	INIT_LIST_HEAD(&mr->mr_list);
 	INIT_WORK(&mr->mr_recycle, frwr_mr_recycle_worker);
@@ -298,65 +317,6 @@ size_t frwr_maxpages(struct rpcrdma_xprt *r_xprt)
 }
 
 /**
- * frwr_wc_fastreg - Invoked by RDMA provider for a flushed FastReg WC
- * @cq:	completion queue (ignored)
- * @wc:	completed WR
- *
- */
-static void
-frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc)
-{
-	struct ib_cqe *cqe = wc->wr_cqe;
-	struct rpcrdma_frwr *frwr =
-			container_of(cqe, struct rpcrdma_frwr, fr_cqe);
-
-	/* WARNING: Only wr_cqe and status are reliable at this point */
-	if (wc->status != IB_WC_SUCCESS)
-		frwr->fr_state = FRWR_FLUSHED_FR;
-	trace_xprtrdma_wc_fastreg(wc, frwr);
-}
-
-/**
- * frwr_wc_localinv - Invoked by RDMA provider for a flushed LocalInv WC
- * @cq:	completion queue (ignored)
- * @wc:	completed WR
- *
- */
-static void
-frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc)
-{
-	struct ib_cqe *cqe = wc->wr_cqe;
-	struct rpcrdma_frwr *frwr = container_of(cqe, struct rpcrdma_frwr,
-						 fr_cqe);
-
-	/* WARNING: Only wr_cqe and status are reliable at this point */
-	if (wc->status != IB_WC_SUCCESS)
-		frwr->fr_state = FRWR_FLUSHED_LI;
-	trace_xprtrdma_wc_li(wc, frwr);
-}
-
-/**
- * frwr_wc_localinv_wake - Invoked by RDMA provider for a signaled LocalInv WC
- * @cq:	completion queue (ignored)
- * @wc:	completed WR
- *
- * Awaken anyone waiting for an MR to finish being fenced.
- */
-static void
-frwr_wc_localinv_wake(struct ib_cq *cq, struct ib_wc *wc)
-{
-	struct ib_cqe *cqe = wc->wr_cqe;
-	struct rpcrdma_frwr *frwr = container_of(cqe, struct rpcrdma_frwr,
-						 fr_cqe);
-
-	/* WARNING: Only wr_cqe and status are reliable at this point */
-	if (wc->status != IB_WC_SUCCESS)
-		frwr->fr_state = FRWR_FLUSHED_LI;
-	trace_xprtrdma_wc_li_wake(wc, frwr);
-	complete(&frwr->fr_linv_done);
-}
-
-/**
  * frwr_map - Register a memory region
  * @r_xprt: controlling transport
  * @seg: memory region co-ordinates
@@ -378,23 +338,15 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt,
 {
 	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
 	bool holes_ok = ia->ri_mrtype == IB_MR_TYPE_SG_GAPS;
-	struct rpcrdma_frwr *frwr;
 	struct rpcrdma_mr *mr;
 	struct ib_mr *ibmr;
 	struct ib_reg_wr *reg_wr;
 	int i, n;
 	u8 key;
 
-	mr = NULL;
-	do {
-		if (mr)
-			rpcrdma_mr_recycle(mr);
-		mr = rpcrdma_mr_get(r_xprt);
-		if (!mr)
-			return ERR_PTR(-EAGAIN);
-	} while (mr->frwr.fr_state != FRWR_IS_INVALID);
-	frwr = &mr->frwr;
-	frwr->fr_state = FRWR_IS_VALID;
+	mr = rpcrdma_mr_get(r_xprt);
+	if (!mr)
+		goto out_getmr_err;
 
 	if (nsegs > ia->ri_max_frwr_depth)
 		nsegs = ia->ri_max_frwr_depth;
@@ -423,7 +375,7 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt,
 	if (!mr->mr_nents)
 		goto out_dmamap_err;
 
-	ibmr = frwr->fr_mr;
+	ibmr = mr->frwr.fr_mr;
 	n = ib_map_mr_sg(ibmr, mr->mr_sg, mr->mr_nents, NULL, PAGE_SIZE);
 	if (unlikely(n != mr->mr_nents))
 		goto out_mapmr_err;
@@ -433,7 +385,7 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt,
 	key = (u8)(ibmr->rkey & 0x000000FF);
 	ib_update_fast_reg_key(ibmr, ++key);
 
-	reg_wr = &frwr->fr_regwr;
+	reg_wr = &mr->frwr.fr_regwr;
 	reg_wr->mr = ibmr;
 	reg_wr->key = ibmr->rkey;
 	reg_wr->access = writing ?
@@ -448,6 +400,10 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt,
 	*out = mr;
 	return seg;
 
+out_getmr_err:
+	xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
+	return ERR_PTR(-EAGAIN);
+
 out_dmamap_err:
 	mr->mr_dir = DMA_NONE;
 	trace_xprtrdma_frwr_sgerr(mr, i);
@@ -461,6 +417,23 @@ out_mapmr_err:
 }
 
 /**
+ * frwr_wc_fastreg - Invoked by RDMA provider for a flushed FastReg WC
+ * @cq:	completion queue (ignored)
+ * @wc:	completed WR
+ *
+ */
+static void frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc)
+{
+	struct ib_cqe *cqe = wc->wr_cqe;
+	struct rpcrdma_frwr *frwr =
+		container_of(cqe, struct rpcrdma_frwr, fr_cqe);
+
+	/* WARNING: Only wr_cqe and status are reliable at this point */
+	trace_xprtrdma_wc_fastreg(wc, frwr);
+	/* The MR will get recycled when the associated req is retransmitted */
+}
+
+/**
  * frwr_send - post Send WR containing the RPC Call message
  * @ia: interface adapter
  * @req: Prepared RPC Call
@@ -512,31 +485,75 @@ void frwr_reminv(struct rpcrdma_rep *rep, struct list_head *mrs)
 		if (mr->mr_handle == rep->rr_inv_rkey) {
 			list_del_init(&mr->mr_list);
 			trace_xprtrdma_mr_remoteinv(mr);
-			mr->frwr.fr_state = FRWR_IS_INVALID;
 			rpcrdma_mr_unmap_and_put(mr);
 			break;	/* only one invalidated MR per RPC */
 		}
 }
 
+static void __frwr_release_mr(struct ib_wc *wc, struct rpcrdma_mr *mr)
+{
+	if (wc->status != IB_WC_SUCCESS)
+		rpcrdma_mr_recycle(mr);
+	else
+		rpcrdma_mr_unmap_and_put(mr);
+}
+
 /**
- * frwr_unmap_sync - invalidate memory regions that were registered for @req
- * @r_xprt: controlling transport
- * @mrs: list of MRs to process
+ * frwr_wc_localinv - Invoked by RDMA provider for a LOCAL_INV WC
+ * @cq:	completion queue (ignored)
+ * @wc:	completed WR
+ *
+ */
+static void frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc)
+{
+	struct ib_cqe *cqe = wc->wr_cqe;
+	struct rpcrdma_frwr *frwr =
+		container_of(cqe, struct rpcrdma_frwr, fr_cqe);
+	struct rpcrdma_mr *mr = container_of(frwr, struct rpcrdma_mr, frwr);
+
+	/* WARNING: Only wr_cqe and status are reliable at this point */
+	trace_xprtrdma_wc_li(wc, frwr);
+	__frwr_release_mr(wc, mr);
+}
+
+/**
+ * frwr_wc_localinv_wake - Invoked by RDMA provider for a LOCAL_INV WC
+ * @cq:	completion queue (ignored)
+ * @wc:	completed WR
  *
- * Sleeps until it is safe for the host CPU to access the
- * previously mapped memory regions.
+ * Awaken anyone waiting for an MR to finish being fenced.
+ */
+static void frwr_wc_localinv_wake(struct ib_cq *cq, struct ib_wc *wc)
+{
+	struct ib_cqe *cqe = wc->wr_cqe;
+	struct rpcrdma_frwr *frwr =
+		container_of(cqe, struct rpcrdma_frwr, fr_cqe);
+	struct rpcrdma_mr *mr = container_of(frwr, struct rpcrdma_mr, frwr);
+
+	/* WARNING: Only wr_cqe and status are reliable at this point */
+	trace_xprtrdma_wc_li_wake(wc, frwr);
+	complete(&frwr->fr_linv_done);
+	__frwr_release_mr(wc, mr);
+}
+
+/**
+ * frwr_unmap_sync - invalidate memory regions that were registered for @req
+ * @r_xprt: controlling transport instance
+ * @req: rpcrdma_req with a non-empty list of MRs to process
  *
- * Caller ensures that @mrs is not empty before the call. This
- * function empties the list.
+ * Sleeps until it is safe for the host CPU to access the previously mapped
+ * memory regions. This guarantees that registered MRs are properly fenced
+ * from the server before the RPC consumer accesses the data in them. It
+ * also ensures proper Send flow control: waking the next RPC waits until
+ * this RPC has relinquished all its Send Queue entries.
  */
-void frwr_unmap_sync(struct rpcrdma_xprt *r_xprt, struct list_head *mrs)
+void frwr_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
 {
 	struct ib_send_wr *first, **prev, *last;
 	const struct ib_send_wr *bad_wr;
-	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
 	struct rpcrdma_frwr *frwr;
 	struct rpcrdma_mr *mr;
-	int count, rc;
+	int rc;
 
 	/* ORDER: Invalidate all of the MRs first
 	 *
@@ -544,33 +561,32 @@ void frwr_unmap_sync(struct rpcrdma_xprt *r_xprt, struct list_head *mrs)
 	 * a single ib_post_send() call.
 	 */
 	frwr = NULL;
-	count = 0;
 	prev = &first;
-	list_for_each_entry(mr, mrs, mr_list) {
-		mr->frwr.fr_state = FRWR_IS_INVALID;
+	while (!list_empty(&req->rl_registered)) {
+		mr = rpcrdma_mr_pop(&req->rl_registered);
 
-		frwr = &mr->frwr;
 		trace_xprtrdma_mr_localinv(mr);
+		r_xprt->rx_stats.local_inv_needed++;
 
+		frwr = &mr->frwr;
 		frwr->fr_cqe.done = frwr_wc_localinv;
 		last = &frwr->fr_invwr;
-		memset(last, 0, sizeof(*last));
+		last->next = NULL;
 		last->wr_cqe = &frwr->fr_cqe;
+		last->sg_list = NULL;
+		last->num_sge = 0;
 		last->opcode = IB_WR_LOCAL_INV;
+		last->send_flags = IB_SEND_SIGNALED;
 		last->ex.invalidate_rkey = mr->mr_handle;
-		count++;
 
 		*prev = last;
 		prev = &last->next;
 	}
-	if (!frwr)
-		goto unmap;
 
 	/* Strong send queue ordering guarantees that when the
 	 * last WR in the chain completes, all WRs in the chain
 	 * are complete.
 	 */
-	last->send_flags = IB_SEND_SIGNALED;
 	frwr->fr_cqe.done = frwr_wc_localinv_wake;
 	reinit_completion(&frwr->fr_linv_done);
 
@@ -578,37 +594,126 @@ void frwr_unmap_sync(struct rpcrdma_xprt *r_xprt, struct list_head *mrs)
 	 * replaces the QP. The RPC reply handler won't call us
 	 * unless ri_id->qp is a valid pointer.
 	 */
-	r_xprt->rx_stats.local_inv_needed++;
 	bad_wr = NULL;
-	rc = ib_post_send(ia->ri_id->qp, first, &bad_wr);
+	rc = ib_post_send(r_xprt->rx_ia.ri_id->qp, first, &bad_wr);
+	trace_xprtrdma_post_send(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 don't wait in that case.
+	 */
 	if (bad_wr != first)
 		wait_for_completion(&frwr->fr_linv_done);
-	if (rc)
-		goto out_release;
+	if (!rc)
+		return;
 
-	/* ORDER: Now DMA unmap all of the MRs, and return
-	 * them to the free MR list.
+	/* Recycle MRs in the LOCAL_INV chain that did not get posted.
 	 */
-unmap:
-	while (!list_empty(mrs)) {
-		mr = rpcrdma_mr_pop(mrs);
-		rpcrdma_mr_unmap_and_put(mr);
+	while (bad_wr) {
+		frwr = container_of(bad_wr, struct rpcrdma_frwr,
+				    fr_invwr);
+		mr = container_of(frwr, struct rpcrdma_mr, frwr);
+		bad_wr = bad_wr->next;
+
+		list_del_init(&mr->mr_list);
+		rpcrdma_mr_recycle(mr);
 	}
-	return;
+}
 
-out_release:
-	pr_err("rpcrdma: FRWR invalidate ib_post_send returned %i\n", rc);
+/**
+ * frwr_wc_localinv_done - Invoked by RDMA provider for a signaled LOCAL_INV WC
+ * @cq:	completion queue (ignored)
+ * @wc:	completed WR
+ *
+ */
+static void frwr_wc_localinv_done(struct ib_cq *cq, struct ib_wc *wc)
+{
+	struct ib_cqe *cqe = wc->wr_cqe;
+	struct rpcrdma_frwr *frwr =
+		container_of(cqe, struct rpcrdma_frwr, fr_cqe);
+	struct rpcrdma_mr *mr = container_of(frwr, struct rpcrdma_mr, frwr);
 
-	/* Unmap and release the MRs in the LOCAL_INV WRs that did not
-	 * get posted.
+	/* WARNING: Only wr_cqe and status are reliable at this point */
+	trace_xprtrdma_wc_li_done(wc, frwr);
+	rpcrdma_complete_rqst(frwr->fr_req->rl_reply);
+	__frwr_release_mr(wc, mr);
+}
+
+/**
+ * frwr_unmap_async - invalidate memory regions that were registered for @req
+ * @r_xprt: controlling transport instance
+ * @req: rpcrdma_req with a non-empty list of MRs to process
+ *
+ * This guarantees that registered MRs are properly fenced from the
+ * server before the RPC consumer accesses the data in them. It also
+ * ensures proper Send flow control: waking the next RPC waits until
+ * this RPC has relinquished all its Send Queue entries.
+ */
+void frwr_unmap_async(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
+{
+	struct ib_send_wr *first, *last, **prev;
+	const struct ib_send_wr *bad_wr;
+	struct rpcrdma_frwr *frwr;
+	struct rpcrdma_mr *mr;
+	int rc;
+
+	/* Chain the LOCAL_INV Work Requests and post them with
+	 * a single ib_post_send() call.
+	 */
+	frwr = NULL;
+	prev = &first;
+	while (!list_empty(&req->rl_registered)) {
+		mr = rpcrdma_mr_pop(&req->rl_registered);
+
+		trace_xprtrdma_mr_localinv(mr);
+		r_xprt->rx_stats.local_inv_needed++;
+
+		frwr = &mr->frwr;
+		frwr->fr_cqe.done = frwr_wc_localinv;
+		frwr->fr_req = req;
+		last = &frwr->fr_invwr;
+		last->next = NULL;
+		last->wr_cqe = &frwr->fr_cqe;
+		last->sg_list = NULL;
+		last->num_sge = 0;
+		last->opcode = IB_WR_LOCAL_INV;
+		last->send_flags = IB_SEND_SIGNALED;
+		last->ex.invalidate_rkey = mr->mr_handle;
+
+		*prev = last;
+		prev = &last->next;
+	}
+
+	/* Strong send queue ordering guarantees that when the
+	 * last WR in the chain completes, all WRs in the chain
+	 * are complete. The last completion will wake up the
+	 * RPC waiter.
+	 */
+	frwr->fr_cqe.done = frwr_wc_localinv_done;
+
+	/* Transport disconnect drains the receive CQ before it
+	 * replaces the QP. The RPC reply handler won't call us
+	 * unless ri_id->qp is a valid pointer.
+	 */
+	bad_wr = NULL;
+	rc = ib_post_send(r_xprt->rx_ia.ri_id->qp, first, &bad_wr);
+	trace_xprtrdma_post_send(req, rc);
+	if (!rc)
+		return;
+
+	/* Recycle MRs in the LOCAL_INV chain that did not get posted.
 	 */
 	while (bad_wr) {
-		frwr = container_of(bad_wr, struct rpcrdma_frwr,
-				    fr_invwr);
+		frwr = container_of(bad_wr, struct rpcrdma_frwr, fr_invwr);
 		mr = container_of(frwr, struct rpcrdma_mr, frwr);
 		bad_wr = bad_wr->next;
 
-		list_del_init(&mr->mr_list);
 		rpcrdma_mr_recycle(mr);
 	}
+
+	/* 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.
+	 */
+	rpcrdma_complete_rqst(req->rl_reply);
 }