public inbox for passt-dev@passt.top
 help / color / mirror / code / Atom feed
From: Laurent Vivier <lvivier@redhat.com>
To: passt-dev@passt.top
Cc: Laurent Vivier <lvivier@redhat.com>
Subject: [PATCH v2 02/13] vhost-user: Centralise 802.3 frame padding in vu_collect() and vu_flush()
Date: Mon,  9 Mar 2026 10:47:33 +0100	[thread overview]
Message-ID: <20260309094744.1907754-3-lvivier@redhat.com> (raw)
In-Reply-To: <20260309094744.1907754-1-lvivier@redhat.com>

The per-protocol padding done by vu_pad() in tcp_vu.c and udp_vu.c was
only correct for single-buffer frames, and assumed the padding area always
fell within the first iov.  It also relied on each caller computing the
right MAX(..., ETH_ZLEN + VNET_HLEN) size for vu_collect() and calling
vu_pad() at the right point.

Centralise padding logic into the two shared vhost-user helpers instead:

 - vu_collect() now ensures at least ETH_ZLEN + VNET_HLEN bytes of buffer
   space are collected, so there is always room for a minimum-sized frame.

 - vu_flush() computes the actual frame length (accounting for
   VIRTIO_NET_F_MRG_RXBUF multi-buffer frames) and passes the padded
   length to vu_queue_fill().

A new iov_memset() helper in iov.c zero-fills the padding area in each
buffer before iov_truncate() sets the logical frame size.  The callers in
tcp_vu.c, udp_vu.c and vu_send_single() use iov_memset() directly,
replacing the now-removed vu_pad() helper and the MAX(..., ETH_ZLEN +
VNET_HLEN) size calculations passed to vu_collect().

Centralising padding here will also ease the move to multi-iovec per
element support, since there will be a single place to update.

In vu_send_single(), fix padding, truncation and data copy to use the
requested frame size rather than the total available buffer space from
vu_collect(), which could be larger.  Also add matching padding, truncation
and explicit size to vu_collect() for the DUP_ACK path in
tcp_vu_send_flag().

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
---
 iov.c       | 24 +++++++++++++++++++
 iov.h       |  2 ++
 tcp_vu.c    | 35 +++++++++++++++++----------
 udp_vu.c    | 12 ++++++----
 vu_common.c | 69 +++++++++++++++++++++++++++++++----------------------
 vu_common.h |  1 -
 6 files changed, 96 insertions(+), 47 deletions(-)

diff --git a/iov.c b/iov.c
index 31a3f5bc29e5..cd48667226f3 100644
--- a/iov.c
+++ b/iov.c
@@ -169,6 +169,30 @@ size_t iov_truncate(struct iovec *iov, size_t iov_cnt, size_t size)
 	return i;
 }
 
+/**
+ * iov_memset() - Set bytes of an IO vector to a given value
+ * @iov:	IO vector
+ * @iov_cnt:	Number of elements in @iov
+ * @offset:	Byte offset in the iovec at which to start
+ * @c:		Byte value to fill with
+ * @length:	Number of bytes to set
+ */
+void iov_memset(const struct iovec *iov, size_t iov_cnt, size_t offset, int c,
+		size_t length)
+{
+	size_t i;
+
+	i = iov_skip_bytes(iov, iov_cnt, offset, &offset);
+
+	for ( ; i < iov_cnt; i++) {
+		size_t n = MIN(iov[i].iov_len - offset, length);
+
+		memset((char *)iov[i].iov_base + offset, c, n);
+		offset = 0;
+		length -= n;
+	}
+}
+
 /**
  * iov_tail_prune() - Remove any unneeded buffers from an IOV tail
  * @tail:	IO vector tail (modified)
diff --git a/iov.h b/iov.h
index b4e50b0fca5a..d295d05b3bab 100644
--- a/iov.h
+++ b/iov.h
@@ -30,6 +30,8 @@ size_t iov_to_buf(const struct iovec *iov, size_t iov_cnt,
 		  size_t offset, void *buf, size_t bytes);
 size_t iov_size(const struct iovec *iov, size_t iov_cnt);
 size_t iov_truncate(struct iovec *iov, size_t iov_cnt, size_t size);
+void iov_memset(const struct iovec *iov, size_t iov_cnt, size_t offset, int c,
+		size_t length);
 
 /*
  * DOC: Theory of Operation, struct iov_tail
diff --git a/tcp_vu.c b/tcp_vu.c
index fd734e857b3b..3adead5f33fa 100644
--- a/tcp_vu.c
+++ b/tcp_vu.c
@@ -72,12 +72,12 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
 	struct vu_dev *vdev = c->vdev;
 	struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
 	struct vu_virtq_element flags_elem[2];
-	size_t optlen, hdrlen, l2len;
 	struct ipv6hdr *ip6h = NULL;
 	struct iphdr *ip4h = NULL;
 	struct iovec flags_iov[2];
 	struct tcp_syn_opts *opts;
 	struct iov_tail payload;
+	size_t optlen, hdrlen;
 	struct tcphdr *th;
 	struct ethhdr *eh;
 	uint32_t seq;
@@ -90,7 +90,7 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
 	vu_set_element(&flags_elem[0], NULL, &flags_iov[0]);
 
 	elem_cnt = vu_collect(vdev, vq, &flags_elem[0], 1,
-			      MAX(hdrlen + sizeof(*opts), ETH_ZLEN + VNET_HLEN), NULL);
+			      hdrlen + sizeof(*opts), NULL);
 	if (elem_cnt != 1)
 		return -1;
 
@@ -131,6 +131,11 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
 		return ret;
 	}
 
+	/* Pad short frames to ETH_ZLEN */
+	if (ETH_ZLEN + VNET_HLEN > hdrlen + optlen) {
+		iov_memset(&flags_iov[0], 1, hdrlen + optlen, 0,
+			   ETH_ZLEN + VNET_HLEN - (hdrlen + optlen));
+	}
 	iov_truncate(&flags_iov[0], 1, hdrlen + optlen);
 	payload = IOV_TAIL(flags_elem[0].in_sg, 1, hdrlen);
 
@@ -140,9 +145,6 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
 	tcp_fill_headers(c, conn, eh, ip4h, ip6h, th, &payload,
 			 NULL, seq, !*c->pcap);
 
-	l2len = optlen + hdrlen - VNET_HLEN;
-	vu_pad(&flags_elem[0].in_sg[0], l2len);
-
 	if (*c->pcap)
 		pcap_iov(&flags_elem[0].in_sg[0], 1, VNET_HLEN);
 	nb_ack = 1;
@@ -151,10 +153,17 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
 		vu_set_element(&flags_elem[1], NULL, &flags_iov[1]);
 
 		elem_cnt = vu_collect(vdev, vq, &flags_elem[1], 1,
-				      flags_elem[0].in_sg[0].iov_len, NULL);
+				      hdrlen + optlen, NULL);
 		if (elem_cnt == 1 &&
 		    flags_elem[1].in_sg[0].iov_len >=
 		    flags_elem[0].in_sg[0].iov_len) {
+			/* Pad short frames to ETH_ZLEN */
+			if (ETH_ZLEN + VNET_HLEN > hdrlen + optlen) {
+				iov_memset(&flags_iov[1], 1, hdrlen + optlen, 0,
+					   ETH_ZLEN + VNET_HLEN -
+					   (hdrlen + optlen));
+			}
+			iov_truncate(&flags_iov[1], 1, hdrlen + optlen);
 			memcpy(flags_elem[1].in_sg[0].iov_base,
 			       flags_elem[0].in_sg[0].iov_base,
 			       flags_elem[0].in_sg[0].iov_len);
@@ -212,8 +221,7 @@ static ssize_t tcp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq,
 
 		cnt = vu_collect(vdev, vq, &elem[elem_cnt],
 				 VIRTQUEUE_MAX_SIZE - elem_cnt,
-				 MAX(MIN(mss, fillsize) + hdrlen, ETH_ZLEN + VNET_HLEN),
-				 &frame_size);
+				 MIN(mss, fillsize) + hdrlen, &frame_size);
 		if (cnt == 0)
 			break;
 
@@ -222,6 +230,7 @@ static ssize_t tcp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq,
 		/* reserve space for headers in iov */
 		iov = &elem[elem_cnt].in_sg[0];
 		ASSERT(iov->iov_len >= hdrlen);
+
 		iov->iov_base = (char *)iov->iov_base + hdrlen;
 		iov->iov_len -= hdrlen;
 		head[(*head_cnt)++] = elem_cnt;
@@ -246,6 +255,11 @@ static ssize_t tcp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq,
 	if (!peek_offset_cap)
 		ret -= already_sent;
 
+	/* Pad short frames to ETH_ZLEN */
+	if (ETH_ZLEN + VNET_HLEN > (size_t)ret + hdrlen) {
+		iov_memset(&iov_vu[DISCARD_IOV_NUM], elem_cnt, ret, 0,
+			   ETH_ZLEN + VNET_HLEN - (ret + hdrlen));
+	}
 	/* adjust iov number and length of the last iov */
 	i = iov_truncate(&iov_vu[DISCARD_IOV_NUM], elem_cnt, ret);
 
@@ -443,7 +457,6 @@ int tcp_vu_data_from_sock(const struct ctx *c, struct tcp_tap_conn *conn)
 		size_t frame_size = iov_size(iov, buf_cnt);
 		bool push = i == head_cnt - 1;
 		ssize_t dlen;
-		size_t l2len;
 
 		ASSERT(frame_size >= hdrlen);
 
@@ -457,10 +470,6 @@ int tcp_vu_data_from_sock(const struct ctx *c, struct tcp_tap_conn *conn)
 
 		tcp_vu_prepare(c, conn, iov, buf_cnt, &check, !*c->pcap, push);
 
-		/* Pad first/single buffer only, it's at least ETH_ZLEN long */
-		l2len = dlen + hdrlen - VNET_HLEN;
-		vu_pad(iov, l2len);
-
 		if (*c->pcap)
 			pcap_iov(iov, buf_cnt, VNET_HLEN);
 
diff --git a/udp_vu.c b/udp_vu.c
index 5effca777e0a..ef9d26118eaf 100644
--- a/udp_vu.c
+++ b/udp_vu.c
@@ -73,7 +73,7 @@ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s,
 	const struct vu_dev *vdev = c->vdev;
 	struct msghdr msg  = { 0 };
 	int iov_cnt, iov_used;
-	size_t hdrlen, l2len;
+	size_t hdrlen;
 
 	ASSERT(!c->no_udp);
 
@@ -98,6 +98,7 @@ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s,
 
 	/* reserve space for the headers */
 	ASSERT(iov_vu[0].iov_len >= MAX(hdrlen, ETH_ZLEN + VNET_HLEN));
+
 	iov_vu[0].iov_base = (char *)iov_vu[0].iov_base + hdrlen;
 	iov_vu[0].iov_len -= hdrlen;
 
@@ -115,12 +116,13 @@ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s,
 	iov_vu[0].iov_base = (char *)iov_vu[0].iov_base - hdrlen;
 	iov_vu[0].iov_len += hdrlen;
 
+	/* Pad short frames to ETH_ZLEN */
+	if (ETH_ZLEN + VNET_HLEN > *dlen + hdrlen) {
+		iov_memset(iov_vu, iov_cnt, *dlen + hdrlen, 0,
+			   ETH_ZLEN + VNET_HLEN - (*dlen + hdrlen));
+	}
 	iov_used = iov_truncate(iov_vu, iov_cnt, *dlen + hdrlen);
 
-	/* pad frame to 60 bytes: first buffer is at least ETH_ZLEN long */
-	l2len = *dlen + hdrlen - VNET_HLEN;
-	vu_pad(&iov_vu[0], l2len);
-
 	vu_set_vnethdr(iov_vu[0].iov_base, iov_used);
 
 	/* release unused buffers */
diff --git a/vu_common.c b/vu_common.c
index 5f2ce18e5b71..8ea05dd30890 100644
--- a/vu_common.c
+++ b/vu_common.c
@@ -87,8 +87,8 @@ int vu_collect(const struct vu_dev *vdev, struct vu_virtq *vq,
 	size_t current_size = 0;
 	int elem_cnt = 0;
 
+	size = MAX(size, ETH_ZLEN + VNET_HLEN); /* 802.3 minimum size */
 	while (current_size < size && elem_cnt < max_elem) {
-		struct iovec *iov;
 		int ret;
 
 		ret = vu_queue_pop(vdev, vq, &elem[elem_cnt]);
@@ -101,12 +101,12 @@ int vu_collect(const struct vu_dev *vdev, struct vu_virtq *vq,
 			break;
 		}
 
-		iov = &elem[elem_cnt].in_sg[0];
+		elem[elem_cnt].in_num = iov_truncate(elem[elem_cnt].in_sg,
+						     elem[elem_cnt].in_num,
+						     size - current_size);
 
-		if (iov->iov_len > size - current_size)
-			iov->iov_len = size - current_size;
-
-		current_size += iov->iov_len;
+		current_size += iov_size(elem[elem_cnt].in_sg,
+					 elem[elem_cnt].in_num);
 		elem_cnt++;
 
 		if (!vu_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
@@ -143,10 +143,30 @@ void vu_set_vnethdr(struct virtio_net_hdr_mrg_rxbuf *vnethdr, int num_buffers)
 void vu_flush(const struct vu_dev *vdev, struct vu_virtq *vq,
 	      struct vu_virtq_element *elem, int elem_cnt)
 {
-	int i;
-
-	for (i = 0; i < elem_cnt; i++)
-		vu_queue_fill(vdev, vq, &elem[i], elem[i].in_sg[0].iov_len, i);
+	int i, j, num_buffers;
+
+	for (i = 0; i < elem_cnt; i += num_buffers) {
+		const struct virtio_net_hdr_mrg_rxbuf *vnethdr;
+		size_t len, padding, elem_size;
+
+		vnethdr = elem[i].in_sg[0].iov_base;
+		num_buffers = le16toh(vnethdr->num_buffers);
+
+		len = 0;
+		for (j = 0; j < num_buffers - 1; j++) {
+			elem_size = iov_size(elem[i + j].in_sg,
+					     elem[i + j].in_num);
+			vu_queue_fill(vdev, vq, &elem[i + j],
+				      elem_size, i + j);
+			len += elem_size;
+		}
+		/* pad the last element to have an 802.3 minimum frame size */
+		elem_size = iov_size(elem[i + j].in_sg, elem[i + j].in_num);
+		padding = MAX(0, (ssize_t)(ETH_ZLEN + VNET_HLEN) -
+			         (ssize_t)(len + elem_size));
+		vu_queue_fill(vdev, vq, &elem[i + j], elem_size + padding,
+			      i + j);
+	}
 
 	vu_queue_flush(vdev, vq, elem_cnt);
 	vu_queue_notify(vdev, vq);
@@ -268,38 +288,31 @@ int vu_send_single(const struct ctx *c, const void *buf, size_t size)
 		goto err;
 	}
 
+	/* Pad short frames to ETH_ZLEN */
+	if (size < ETH_ZLEN + VNET_HLEN) {
+		iov_memset(in_sg, elem_cnt, size, 0,
+			   ETH_ZLEN + VNET_HLEN - size);
+	}
+	elem_cnt = iov_truncate(in_sg, elem_cnt, size);
+
 	vu_set_vnethdr(in_sg[0].iov_base, elem_cnt);
 
-	total -= VNET_HLEN;
+	size -= VNET_HLEN;
 
 	/* copy data from the buffer to the iovec */
-	iov_from_buf(in_sg, elem_cnt, VNET_HLEN, buf, total);
+	iov_from_buf(in_sg, elem_cnt, VNET_HLEN, buf, size);
 
 	if (*c->pcap)
 		pcap_iov(in_sg, elem_cnt, VNET_HLEN);
 
 	vu_flush(vdev, vq, elem, elem_cnt);
 
-	trace("vhost-user sent %zu", total);
+	trace("vhost-user sent %zu", size);
 
-	return total;
+	return size;
 err:
 	for (i = 0; i < elem_cnt; i++)
 		vu_queue_detach_element(vq);
 
 	return -1;
 }
-
-/**
- * vu_pad() - Pad 802.3 frame to minimum length (60 bytes) if needed
- * @iov:	Buffer in iovec array where end of 802.3 frame is stored
- * @l2len:	Layer-2 length already filled in frame
- */
-void vu_pad(struct iovec *iov, size_t l2len)
-{
-	if (l2len >= ETH_ZLEN)
-		return;
-
-	memset((char *)iov->iov_base + iov->iov_len, 0, ETH_ZLEN - l2len);
-	iov->iov_len += ETH_ZLEN - l2len;
-}
diff --git a/vu_common.h b/vu_common.h
index 865d9771fa89..5de0c987b936 100644
--- a/vu_common.h
+++ b/vu_common.h
@@ -61,6 +61,5 @@ void vu_flush(const struct vu_dev *vdev, struct vu_virtq *vq,
 void vu_kick_cb(struct vu_dev *vdev, union epoll_ref ref,
 		const struct timespec *now);
 int vu_send_single(const struct ctx *c, const void *buf, size_t size);
-void vu_pad(struct iovec *iov, size_t l2len);
 
 #endif /* VU_COMMON_H */
-- 
2.53.0


  parent reply	other threads:[~2026-03-09  9:47 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-03-09  9:47 [PATCH v2 00/13] vhost-user,udp: Handle multiple iovec entries per virtqueue element Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 01/13] iov: Add iov_truncate() helper and use it in vu handlers Laurent Vivier
2026-03-09  9:47 ` Laurent Vivier [this message]
2026-03-09  9:47 ` [PATCH v2 03/13] vhost-user: Use ARRAY_SIZE(elem) instead of VIRTQUEUE_MAX_SIZE Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 04/13] udp_vu: Use iov_tail to manage virtqueue buffers Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 05/13] udp_vu: Move virtqueue management from udp_vu_sock_recv() to its caller Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 06/13] iov: Add IOV_PUT_HEADER() to write header data back to iov_tail Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 07/13] udp: Pass iov_tail to udp_update_hdr4()/udp_update_hdr6() Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 08/13] udp_vu: Use iov_tail in udp_vu_prepare() Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 09/13] vu_common: Pass iov_tail to vu_set_vnethdr() Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 10/13] vu_common: Accept explicit iovec counts in vu_set_element() Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 11/13] vu_common: Accept explicit iovec count per element in vu_init_elem() Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 12/13] vu_common: Prepare to use multibuffer with guest RX Laurent Vivier
2026-03-09  9:47 ` [PATCH v2 13/13] vhost-user,udp: Use 2 iovec entries per element Laurent Vivier

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=20260309094744.1907754-3-lvivier@redhat.com \
    --to=lvivier@redhat.com \
    --cc=passt-dev@passt.top \
    /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
Be sure your reply has a Subject: header at the top and a blank line before the message body.
Code repositories for project(s) associated with this public inbox

	https://passt.top/passt

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for IMAP folder(s).