[PATCH net-next 0/3] s390: af_iucv patches

[PATCH net-next 0/3] s390: af_iucv patches

[Ursula Braun]

Hi Dave,

here are improvements for af_iucv relaxing the pressure to allocate
big contiguous kernel buffers.

Thanks, Ursula

Eugene Crosser (3):
  af_iucv: use paged SKBs for big outbound messages
  af_iucv: remove fragment_skb() to use paged SKBs
  af_iucv: use paged SKBs for big inbound messages

 net/iucv/af_iucv.c | 223 +++++++++++++++++++++++++++++------------------------
 1 file changed, 122 insertions(+), 101 deletions(-)

-- 
2.6.6

[PATCH net-next 1/3] af_iucv: use paged SKBs for big outbound messages

[Ursula Braun]

From: Eugene Crosser <Eugene.Crosser@ru.ibm.com>

When an outbound message is bigger than a page, allocate and fill
a paged SKB, and subsequently use IUCV send primitive with IPBUFLST
flag. This relaxes the pressure to allocate big contiguous kernel
buffers.

Signed-off-by: Eugene Crosser <Eugene.Crosser@ru.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
---
 net/iucv/af_iucv.c | 124 +++++++++++++++++++++++++++++++++--------------------
 1 file changed, 77 insertions(+), 47 deletions(-)

diff --git a/net/iucv/af_iucv.c b/net/iucv/af_iucv.c
index fc3598a..38448d1 100644
--- a/net/iucv/af_iucv.c
+++ b/net/iucv/af_iucv.c
@@ -1033,6 +1033,7 @@ static int iucv_sock_sendmsg(struct socket *sock, struct msghdr *msg,
 {
 	struct sock *sk = sock->sk;
 	struct iucv_sock *iucv = iucv_sk(sk);
+	size_t headroom, linear;
 	struct sk_buff *skb;
 	struct iucv_message txmsg = {0};
 	struct cmsghdr *cmsg;
@@ -1110,20 +1111,31 @@ static int iucv_sock_sendmsg(struct socket *sock, struct msghdr *msg,
 	 * this is fine for SOCK_SEQPACKET (unless we want to support
 	 * segmented records using the MSG_EOR flag), but
 	 * for SOCK_STREAM we might want to improve it in future */
-	if (iucv->transport == AF_IUCV_TRANS_HIPER)
-		skb = sock_alloc_send_skb(sk,
-			len + sizeof(struct af_iucv_trans_hdr) + ETH_HLEN,
-			noblock, &err);
-	else
-		skb = sock_alloc_send_skb(sk, len, noblock, &err);
+	headroom = (iucv->transport == AF_IUCV_TRANS_HIPER)
+		   ? sizeof(struct af_iucv_trans_hdr) + ETH_HLEN : 0;
+	if (headroom + len < PAGE_SIZE) {
+		linear = len;
+	} else {
+		/* In nonlinear "classic" iucv skb,
+		 * reserve space for iucv_array
+		 */
+		if (iucv->transport != AF_IUCV_TRANS_HIPER)
+			headroom += sizeof(struct iucv_array) *
+				    (MAX_SKB_FRAGS + 1);
+		linear = PAGE_SIZE - headroom;
+	}
+	skb = sock_alloc_send_pskb(sk, headroom + linear, len - linear,
+				   noblock, &err, 0);
 	if (!skb)
 		goto out;
-	if (iucv->transport == AF_IUCV_TRANS_HIPER)
-		skb_reserve(skb, sizeof(struct af_iucv_trans_hdr) + ETH_HLEN);
-	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
-		err = -EFAULT;
+	if (headroom)
+		skb_reserve(skb, headroom);
+	skb_put(skb, linear);
+	skb->len = len;
+	skb->data_len = len - linear;
+	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
+	if (err)
 		goto fail;
-	}
 
 	/* wait if outstanding messages for iucv path has reached */
 	timeo = sock_sndtimeo(sk, noblock);
@@ -1148,49 +1160,67 @@ static int iucv_sock_sendmsg(struct socket *sock, struct msghdr *msg,
 			atomic_dec(&iucv->msg_sent);
 			goto fail;
 		}
-		goto release;
-	}
-	skb_queue_tail(&iucv->send_skb_q, skb);
-
-	if (((iucv->path->flags & IUCV_IPRMDATA) & iucv->flags)
-	      && skb->len <= 7) {
-		err = iucv_send_iprm(iucv->path, &txmsg, skb);
+	} else { /* Classic VM IUCV transport */
+		skb_queue_tail(&iucv->send_skb_q, skb);
+
+		if (((iucv->path->flags & IUCV_IPRMDATA) & iucv->flags) &&
+		    skb->len <= 7) {
+			err = iucv_send_iprm(iucv->path, &txmsg, skb);
+
+			/* on success: there is no message_complete callback */
+			/* for an IPRMDATA msg; remove skb from send queue   */
+			if (err == 0) {
+				skb_unlink(skb, &iucv->send_skb_q);
+				kfree_skb(skb);
+			}
 
-		/* on success: there is no message_complete callback
-		 * for an IPRMDATA msg; remove skb from send queue */
-		if (err == 0) {
-			skb_unlink(skb, &iucv->send_skb_q);
-			kfree_skb(skb);
+			/* this error should never happen since the	*/
+			/* IUCV_IPRMDATA path flag is set... sever path */
+			if (err == 0x15) {
+				pr_iucv->path_sever(iucv->path, NULL);
+				skb_unlink(skb, &iucv->send_skb_q);
+				err = -EPIPE;
+				goto fail;
+			}
+		} else if (skb_is_nonlinear(skb)) {
+			struct iucv_array *iba = (struct iucv_array *)skb->head;
+			int i;
+
+			/* skip iucv_array lying in the headroom */
+			iba[0].address = (u32)(addr_t)skb->data;
+			iba[0].length = (u32)skb_headlen(skb);
+			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+				skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+				iba[i + 1].address =
+					(u32)(addr_t)skb_frag_address(frag);
+				iba[i + 1].length = (u32)skb_frag_size(frag);
+			}
+			err = pr_iucv->message_send(iucv->path, &txmsg,
+						    IUCV_IPBUFLST, 0,
+						    (void *)iba, skb->len);
+		} else { /* non-IPRM Linear skb */
+			err = pr_iucv->message_send(iucv->path, &txmsg,
+					0, 0, (void *)skb->data, skb->len);
 		}
-
-		/* this error should never happen since the
-		 * IUCV_IPRMDATA path flag is set... sever path */
-		if (err == 0x15) {
-			pr_iucv->path_sever(iucv->path, NULL);
+		if (err) {
+			if (err == 3) {
+				user_id[8] = 0;
+				memcpy(user_id, iucv->dst_user_id, 8);
+				appl_id[8] = 0;
+				memcpy(appl_id, iucv->dst_name, 8);
+				pr_err(
+		"Application %s on z/VM guest %s exceeds message limit\n",
+					appl_id, user_id);
+				err = -EAGAIN;
+			} else {
+				err = -EPIPE;
+			}
 			skb_unlink(skb, &iucv->send_skb_q);
-			err = -EPIPE;
 			goto fail;
 		}
-	} else
-		err = pr_iucv->message_send(iucv->path, &txmsg, 0, 0,
-					(void *) skb->data, skb->len);
-	if (err) {
-		if (err == 3) {
-			user_id[8] = 0;
-			memcpy(user_id, iucv->dst_user_id, 8);
-			appl_id[8] = 0;
-			memcpy(appl_id, iucv->dst_name, 8);
-			pr_err("Application %s on z/VM guest %s"
-				" exceeds message limit\n",
-				appl_id, user_id);
-			err = -EAGAIN;
-		} else
-			err = -EPIPE;
-		skb_unlink(skb, &iucv->send_skb_q);
-		goto fail;
 	}
 
-release:
 	release_sock(sk);
 	return len;
 
-- 
2.6.6

[PATCH net-next 2/3] af_iucv: remove fragment_skb() to use paged SKBs

[Ursula Braun]

From: Eugene Crosser <Eugene.Crosser@ru.ibm.com>

Before introducing paged skbs in the receive path, get rid of the
function `iucv_fragment_skb()` that replaces one large linear skb
with several smaller linear skbs.

Signed-off-by: Eugene Crosser <Eugene.Crosser@ru.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
---
 net/iucv/af_iucv.c | 59 +++---------------------------------------------------
 1 file changed, 3 insertions(+), 56 deletions(-)

diff --git a/net/iucv/af_iucv.c b/net/iucv/af_iucv.c
index 38448d1..9ed2adf 100644
--- a/net/iucv/af_iucv.c
+++ b/net/iucv/af_iucv.c
@@ -1231,44 +1231,6 @@ out:
 	return err;
 }
 
-/* iucv_fragment_skb() - Fragment a single IUCV message into multiple skb's
- *
- * Locking: must be called with message_q.lock held
- */
-static int iucv_fragment_skb(struct sock *sk, struct sk_buff *skb, int len)
-{
-	int dataleft, size, copied = 0;
-	struct sk_buff *nskb;
-
-	dataleft = len;
-	while (dataleft) {
-		if (dataleft >= sk->sk_rcvbuf / 4)
-			size = sk->sk_rcvbuf / 4;
-		else
-			size = dataleft;
-
-		nskb = alloc_skb(size, GFP_ATOMIC | GFP_DMA);
-		if (!nskb)
-			return -ENOMEM;
-
-		/* copy target class to control buffer of new skb */
-		IUCV_SKB_CB(nskb)->class = IUCV_SKB_CB(skb)->class;
-
-		/* copy data fragment */
-		memcpy(nskb->data, skb->data + copied, size);
-		copied += size;
-		dataleft -= size;
-
-		skb_reset_transport_header(nskb);
-		skb_reset_network_header(nskb);
-		nskb->len = size;
-
-		skb_queue_tail(&iucv_sk(sk)->backlog_skb_q, nskb);
-	}
-
-	return 0;
-}
-
 /* iucv_process_message() - Receive a single outstanding IUCV message
  *
  * Locking: must be called with message_q.lock held
@@ -1300,24 +1262,9 @@ static void iucv_process_message(struct sock *sk, struct sk_buff *skb,
 			kfree_skb(skb);
 			return;
 		}
-		/* we need to fragment iucv messages for SOCK_STREAM only;
-		 * for SOCK_SEQPACKET, it is only relevant if we support
-		 * record segmentation using MSG_EOR (see also recvmsg()) */
-		if (sk->sk_type == SOCK_STREAM &&
-		    skb->truesize >= sk->sk_rcvbuf / 4) {
-			rc = iucv_fragment_skb(sk, skb, len);
-			kfree_skb(skb);
-			skb = NULL;
-			if (rc) {
-				pr_iucv->path_sever(path, NULL);
-				return;
-			}
-			skb = skb_dequeue(&iucv_sk(sk)->backlog_skb_q);
-		} else {
-			skb_reset_transport_header(skb);
-			skb_reset_network_header(skb);
-			skb->len = len;
-		}
+		skb_reset_transport_header(skb);
+		skb_reset_network_header(skb);
+		skb->len = len;
 	}
 
 	IUCV_SKB_CB(skb)->offset = 0;
-- 
2.6.6

[PATCH net-next 3/3] af_iucv: use paged SKBs for big inbound messages

[Ursula Braun]

From: Eugene Crosser <Eugene.Crosser@ru.ibm.com>

When an inbound message is bigger than a page, allocate a paged SKB,
and subsequently use IUCV receive primitive with IPBUFLST flag.
This relaxes the pressure to allocate big contiguous kernel buffers.

Signed-off-by: Eugene Crosser <Eugene.Crosser@ru.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
---
 net/iucv/af_iucv.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 50 insertions(+), 6 deletions(-)

diff --git a/net/iucv/af_iucv.c b/net/iucv/af_iucv.c
index 9ed2adf..37d674e 100644
--- a/net/iucv/af_iucv.c
+++ b/net/iucv/af_iucv.c
@@ -1231,6 +1231,34 @@ out:
 	return err;
 }
 
+static struct sk_buff *alloc_iucv_recv_skb(unsigned long len)
+{
+	size_t headroom, linear;
+	struct sk_buff *skb;
+	int err;
+
+	if (len < PAGE_SIZE) {
+		headroom = 0;
+		linear = len;
+	} else {
+		headroom = sizeof(struct iucv_array) * (MAX_SKB_FRAGS + 1);
+		linear = PAGE_SIZE - headroom;
+	}
+	skb = alloc_skb_with_frags(headroom + linear, len - linear,
+				   0, &err, GFP_ATOMIC | GFP_DMA);
+	WARN_ONCE(!skb,
+		  "alloc of recv iucv skb len=%lu failed with errcode=%d\n",
+		  len, err);
+	if (skb) {
+		if (headroom)
+			skb_reserve(skb, headroom);
+		skb_put(skb, linear);
+		skb->len = len;
+		skb->data_len = len - linear;
+	}
+	return skb;
+}
+
 /* iucv_process_message() - Receive a single outstanding IUCV message
  *
  * Locking: must be called with message_q.lock held
@@ -1255,16 +1283,32 @@ static void iucv_process_message(struct sock *sk, struct sk_buff *skb,
 			skb->len = 0;
 		}
 	} else {
-		rc = pr_iucv->message_receive(path, msg,
+		if (skb_is_nonlinear(skb)) {
+			struct iucv_array *iba = (struct iucv_array *)skb->head;
+			int i;
+
+			iba[0].address = (u32)(addr_t)skb->data;
+			iba[0].length = (u32)skb_headlen(skb);
+			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+				skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+				iba[i + 1].address =
+					(u32)(addr_t)skb_frag_address(frag);
+				iba[i + 1].length = (u32)skb_frag_size(frag);
+			}
+			rc = pr_iucv->message_receive(path, msg,
+					      IUCV_IPBUFLST,
+					      (void *)iba, len, NULL);
+		} else {
+			rc = pr_iucv->message_receive(path, msg,
 					      msg->flags & IUCV_IPRMDATA,
 					      skb->data, len, NULL);
+		}
 		if (rc) {
 			kfree_skb(skb);
 			return;
 		}
-		skb_reset_transport_header(skb);
-		skb_reset_network_header(skb);
-		skb->len = len;
+		WARN_ON_ONCE(skb->len != len);
 	}
 
 	IUCV_SKB_CB(skb)->offset = 0;
@@ -1283,7 +1327,7 @@ static void iucv_process_message_q(struct sock *sk)
 	struct sock_msg_q *p, *n;
 
 	list_for_each_entry_safe(p, n, &iucv->message_q.list, list) {
-		skb = alloc_skb(iucv_msg_length(&p->msg), GFP_ATOMIC | GFP_DMA);
+		skb = alloc_iucv_recv_skb(iucv_msg_length(&p->msg));
 		if (!skb)
 			break;
 		iucv_process_message(sk, skb, p->path, &p->msg);
@@ -1778,7 +1822,7 @@ static void iucv_callback_rx(struct iucv_path *path, struct iucv_message *msg)
 	if (len > sk->sk_rcvbuf)
 		goto save_message;
 
-	skb = alloc_skb(iucv_msg_length(msg), GFP_ATOMIC | GFP_DMA);
+	skb = alloc_iucv_recv_skb(iucv_msg_length(msg));
 	if (!skb)
 		goto save_message;
 
-- 
2.6.6

Re: [PATCH net-next 0/3] s390: af_iucv patches

[David Miller]

From: Ursula Braun <ubraun@linux.vnet.ibm.com>
Date: Mon, 13 Jun 2016 18:46:13 +0200

> here are improvements for af_iucv relaxing the pressure to allocate
> big contiguous kernel buffers.

Series applied.