[PATCH v10 0/3] vhost-user,udp: Handle multiple iovec entries per virtqueue element

Laurent Vivier

19 May 2026 19 May '26

5:56 p.m.

Some virtio-net drivers (notably iPXE) provide descriptors where the vnet header and the frame payload are in separate buffers, resulting in two iovec entries per virtqueue element. Currently, the RX (host to guest) path assumes a single iovec per element, which triggers: ASSERTION FAILED in virtqueue_map_desc (virtio.c:403): num_sg < max_num_sg This series reworks the UDP vhost-user receive path to support multiple iovec entries per element, fixing the iPXE crash. This series only addresses the UDP path. TCP vhost-user will be updated to use multi-iov elements in a subsequent series. Based-on: 20260513115218.1662850-1-lvivier@redhat.com v10: - udp_update_hdr4()/udp_update_hdr6(): take separate @uh pointer and @payload - udp_tap_prepare(): extract uh from UDP_IOV_PAYLOAD, @payload is UDP payload only - udp_vu_sock_recv(): extract UDP payload from the function - udp_vu_prepare(): take @payload as parameter, compute checksum conditionally via !*c->pcap instead of hardcoded true - udp_vu_csum(): removed v9: - iov_push_header_(): skip memcpy() when the source pointer already aliases the iov buffer - udp_update_hdr4()//udp_update_hdr6()/usp_csum(): update payload doc - udp_tap_prepare(): use existing (*tap_iov)[UDP_IOV_PAYLOAD] v8: - Add Reviewed-by from David for 1/3 and 2/3 - Remove VLA to use fixed size (VIRTQUEUE_MAX_SIZE) - Rename udp_frame to datagram - Use sizeof(uh) instead of sizeof(struct udphdr) - Push back uh in memory in udp_vu_csum() v7: - In udp_vu_sock_to_tap(), introduce iov_still_needed variable - Fix iov_tail @data doc comments in udp_vu_prepare()/udp_vu_csum() - Don't rewrap the virtio checksum comment in udp_update_hdr6() - Add NULL check for uh in udp_vu_csum() v6: - Rebased on top of [PATCH 00/10] vhost-user: Preparatory series for multiple iovec entries per virtqueue element v5: - This version doesn't change the padding system regarding v4, it's a complex task that will be addressed in another version - reorder patches and add new patches - remove IOV_PUT_HEADER()/with_header() and introduce IOV_PUSH_HEADER() - don't use the iov_tail to provide the headers to the functions - move vu_set_vnethdr() to vu_flush(), extract vu_queue_notify() - move vu_flush() inside loop in tcp_vu_data_from_sock() to flush data by frame and not by full data length v4: - rebase - replace ASSERT() by assert() v3: - include the series "Decouple iovec management from virtqueues elements" - because of this series, drop: "vu_common: Accept explicit iovec counts in vu_set_element()" "vu_common: Accept explicit iovec count per element in vu_init_elem()" "vu_common: Prepare to use multibuffer with guest RX" "vhost-user,udp: Use 2 iovec entries per element" - drop "vu_common: Pass iov_tail to vu_set_vnethdr()" as the specs insures a buffer is big enough to contain vnet header - introduce "with_header()" and merge "udp: Pass iov_tail to udp_update_hdr4()/udp_update_hdr6()" and "udp_vu: Use iov_tail in udp_vu_prepare()" to use it v2: - add iov_truncate(), iov_memset() - remove iov_tail_truncate() and iov_tail_zero_end() - manage 802.3 minimum frame size Laurent Vivier (3): udp_vu: Allow virtqueue elements with multiple iovec entries iov: Introduce IOV_PUSH_HEADER() macro udp: Pass iov_tail to udp_update_hdr4()/udp_update_hdr6() iov.c | 30 ++++++++++++ iov.h | 11 +++++ udp.c | 74 ++++++++++++++-------------- udp_internal.h | 6 ++- udp_vu.c | 129 +++++++++++++++++++------------------------------ 5 files changed, 131 insertions(+), 119 deletions(-) -- 2.54.0

Show replies by date

Laurent Vivier

19 May 19 May

5:56 p.m.

New subject: [PATCH v10 1/3] udp_vu: Allow virtqueue elements with multiple iovec entries

The previous code assumed a 1:1 mapping between virtqueue elements and iovec entries (enforced by an assert). Drop that assumption to allow elements that span multiple iovecs: track elem_used separately by walking the element list against the iov count returned after padding. This also fixes vu_queue_rewind() and vu_flush() to use the element count rather than the iov count. Use iov_tail_clone() in udp_vu_sock_recv() to handle header offset, replacing the manual base/len adjustment and restore pattern. Signed-off-by: Laurent Vivier Reviewed-by: David Gibson Reviewed-by: Jon Maloy --- udp_vu.c | 31 ++++++++++++++++--------------- 1 file changed, 16 insertions(+), 15 deletions(-) diff --git a/udp_vu.c b/udp_vu.c index dfff7bb1cd53..74bf79d57969 100644 --- a/udp_vu.c +++ b/udp_vu.c @@ -66,30 +66,25 @@ static size_t udp_vu_hdrlen(bool v6) */ static ssize_t udp_vu_sock_recv(struct iovec *iov, size_t *cnt, int s, bool v6) { + struct iovec msg_iov[VIRTQUEUE_MAX_SIZE]; struct msghdr msg = { 0 }; + struct iov_tail payload; size_t hdrlen, iov_used; ssize_t dlen; /* compute L2 header length */ hdrlen = udp_vu_hdrlen(v6); - /* reserve space for the headers */ - assert(iov[0].iov_len >= MAX(hdrlen, ETH_ZLEN + VNET_HLEN)); - iov[0].iov_base = (char *)iov[0].iov_base + hdrlen; - iov[0].iov_len -= hdrlen; + payload = IOV_TAIL(iov, *cnt, hdrlen); - /* read data from the socket */ - msg.msg_iov = iov; - msg.msg_iovlen = *cnt; + msg.msg_iov = msg_iov; + msg.msg_iovlen = iov_tail_clone(msg.msg_iov, payload.cnt, &payload); + /* read data from the socket */ dlen = recvmsg(s, &msg, 0); if (dlen < 0) return -1; - /* restore the pointer to the headers address */ - iov[0].iov_base = (char *)iov[0].iov_base - hdrlen; - iov[0].iov_len += hdrlen; - iov_used = iov_skip_bytes(iov, *cnt, MAX(dlen + hdrlen, VNET_HLEN + ETH_ZLEN), NULL); @@ -202,7 +197,7 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) } for (i = 0; i < n; i++) { - unsigned elem_cnt, elem_used; + unsigned elem_cnt, elem_used, j, k; size_t iov_cnt; ssize_t dlen; @@ -212,15 +207,21 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) if (elem_cnt == 0) break; - assert((size_t)elem_cnt == iov_cnt); /* one iovec per element */ - dlen = udp_vu_sock_recv(iov_vu, &iov_cnt, s, v6); if (dlen < 0) { vu_queue_rewind(vq, elem_cnt); break; } - elem_used = iov_cnt; /* one iovec per element */ + elem_used = 0; + for (j = 0, k = 0; k < iov_cnt && j < elem_cnt; j++) { + size_t iov_still_needed = iov_cnt - k; + + if (elem[j].in_num > iov_still_needed) + elem[j].in_num = iov_still_needed; + k += elem[j].in_num; + elem_used++; + } /* release unused buffers */ vu_queue_rewind(vq, elem_cnt - elem_used); -- 2.54.0

Laurent Vivier

5:56 p.m.

New subject: [PATCH v10 2/3] iov: Introduce IOV_PUSH_HEADER() macro

Add iov_push_header_() and its typed wrapper IOV_PUSH_HEADER() to write a header into an iov_tail at the current offset and advance past it. This is the write counterpart to IOV_PEEK_HEADER() / IOV_REMOVE_HEADER(), using iov_from_buf() to copy the header data across iovec boundaries. Signed-off-by: Laurent Vivier Reviewed-by: David Gibson --- iov.c | 31 +++++++++++++++++++++++++++++++ iov.h | 11 +++++++++++ 2 files changed, 42 insertions(+) diff --git a/iov.c b/iov.c index 7c18941c0395..6a5d7d35b67f 100644 --- a/iov.c +++ b/iov.c @@ -359,6 +359,37 @@ void *iov_peek_header_(struct iov_tail *tail, void *v, size_t len, size_t align) return v; } +/** + * iov_push_header_() - Write a new header to an IOV tail + * @tail: IOV tail to write header to + * @v: Pointer to header data to write + * @len: Length of header to write, in bytes + * + * Return: number of bytes written + */ +/* cppcheck-suppress unusedFunction */ +size_t iov_push_header_(struct iov_tail *tail, const void *v, size_t len) +{ + size_t l; + + if (!iov_tail_prune(tail)) + return 0; /* No space */ + + if ((char *)tail->iov[0].iov_base + tail->off == v) /* already in place */ + l = len; + else + l = iov_from_buf(tail->iov, tail->cnt, tail->off, v, len); + + if (l != len) { + warn("iov_push_header_(): Not enough space to store header" + " (%zu != %zu)", l, len); + } + + tail->off = tail->off + l; + + return l; +} + /** * iov_remove_header_() - Remove a header from an IOV tail * @tail: IOV tail to remove header from (modified) diff --git a/iov.h b/iov.h index 3c63308e554f..4fdf14a85b19 100644 --- a/iov.h +++ b/iov.h @@ -93,6 +93,7 @@ bool iov_tail_prune(struct iov_tail *tail); size_t iov_tail_size(struct iov_tail *tail); bool iov_drop_header(struct iov_tail *tail, size_t len); void *iov_peek_header_(struct iov_tail *tail, void *v, size_t len, size_t align); +size_t iov_push_header_(struct iov_tail *tail, const void *v, size_t len); void *iov_remove_header_(struct iov_tail *tail, void *v, size_t len, size_t align); ssize_t iov_tail_clone(struct iovec *dst_iov, size_t dst_iov_cnt, struct iov_tail *tail); @@ -115,6 +116,16 @@ ssize_t iov_tail_clone(struct iovec *dst_iov, size_t dst_iov_cnt, sizeof(var_), \ __alignof__(var_)))) +/** + * IOV_PUSH_HEADER() - Write a new header to an IOV tail + * @tail_: IOV tail to write header to + * @var_: A variable containing the header data to write + * + * Return: number of bytes written + */ +#define IOV_PUSH_HEADER(tail_, var_) \ + (iov_push_header_((tail_), &(var_), sizeof(var_))) + /** * IOV_REMOVE_HEADER() - Remove and return typed header from an IOV tail * @tail_: IOV tail to remove header from (modified) -- 2.54.0

Laurent Vivier

5:56 p.m.

New subject: [PATCH v10 3/3] udp: Pass iov_tail to udp_update_hdr4()/udp_update_hdr6()

Change udp_update_hdr4() and udp_update_hdr6() to take an iov_tail pointing at the UDP frame instead of a contiguous udp_payload_t buffer and explicit data length. This lets vhost-user pass scatter-gather virtqueue buffers directly without an intermediate copy. The UDP header is built into a local struct udphdr and written back with IOV_PUSH_HEADER(). On the tap side, udp_tap_prepare() wraps the existing udp_payload_t in a two-element iov to match the new interface. Signed-off-by: Laurent Vivier --- iov.c | 1 - udp.c | 74 +++++++++++++++++----------------- udp_internal.h | 6 ++- udp_vu.c | 106 ++++++++++++++++++------------------------------- 4 files changed, 78 insertions(+), 109 deletions(-) diff --git a/iov.c b/iov.c index 6a5d7d35b67f..9248ba95a9f2 100644 --- a/iov.c +++ b/iov.c @@ -367,7 +367,6 @@ void *iov_peek_header_(struct iov_tail *tail, void *v, size_t len, size_t align) * * Return: number of bytes written */ -/* cppcheck-suppress unusedFunction */ size_t iov_push_header_(struct iov_tail *tail, const void *v, size_t len) { size_t l; diff --git a/udp.c b/udp.c index 66dc7766868c..cfc7b2439881 100644 --- a/udp.c +++ b/udp.c @@ -255,20 +255,22 @@ static void udp_iov_init(const struct ctx *c) /** * udp_update_hdr4() - Update headers for one IPv4 datagram * @ip4h: Pre-filled IPv4 header (except for tot_len and saddr) - * @bp: Pointer to udp_payload_t to update + * @uh: UDP header to fill + * @payload: UDP payload * @toside: Flowside for destination side * @dlen: Length of UDP payload * @no_udp_csum: Do not set UDP checksum * - * Return: size of IPv4 payload (UDP header + data) + * Return: size of datagram (UDP header + data) */ -size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, +size_t udp_update_hdr4(struct iphdr *ip4h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum) { const struct in_addr *src = inany_v4(&toside->oaddr); const struct in_addr *dst = inany_v4(&toside->eaddr); - size_t l4len = dlen + sizeof(bp->uh); + size_t l4len = dlen + sizeof(*uh); size_t l3len = l4len + sizeof(*ip4h); assert(src && dst); @@ -278,19 +280,13 @@ size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, ip4h->saddr = src->s_addr; ip4h->check = csum_ip4_header(l3len, IPPROTO_UDP, *src, *dst); - bp->uh.source = htons(toside->oport); - bp->uh.dest = htons(toside->eport); - bp->uh.len = htons(l4len); - if (no_udp_csum) { - bp->uh.check = 0; - } else { - const struct iovec iov = { - .iov_base = bp->data, - .iov_len = dlen - }; - struct iov_tail data = IOV_TAIL(&iov, 1, 0); - csum_udp4(&bp->uh, *src, *dst, &data, dlen); - } + uh->source = htons(toside->oport); + uh->dest = htons(toside->eport); + uh->len = htons(l4len); + if (no_udp_csum) + uh->check = 0; + else + csum_udp4(uh, *src, *dst, payload, dlen); return l4len; } @@ -299,18 +295,20 @@ size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, * udp_update_hdr6() - Update headers for one IPv6 datagram * @ip6h: Pre-filled IPv6 header (except for payload_len and * addresses) - * @bp: Pointer to udp_payload_t to update + * @uh: UDP header to fill + * @payload: UDP payload * @toside: Flowside for destination side * @dlen: Length of UDP payload * @no_udp_csum: Do not set UDP checksum * - * Return: size of IPv6 payload (UDP header + data) + * Return: size of datagram (UDP header + data) */ -size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, +size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum) { - uint16_t l4len = dlen + sizeof(bp->uh); + uint16_t l4len = dlen + sizeof(*uh); ip6h->payload_len = htons(l4len); ip6h->daddr = toside->eaddr.a6; @@ -319,23 +317,19 @@ size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, ip6h->nexthdr = IPPROTO_UDP; ip6h->hop_limit = 255; - bp->uh.source = htons(toside->oport); - bp->uh.dest = htons(toside->eport); - bp->uh.len = ip6h->payload_len; + uh->source = htons(toside->oport); + uh->dest = htons(toside->eport); + uh->len = htons(l4len); + if (no_udp_csum) { /* 0 is an invalid checksum for UDP IPv6 and dropped by * the kernel stack, even if the checksum is disabled by virtio * flags. We need to put any non-zero value here. */ - bp->uh.check = 0xffff; + uh->check = 0xffff; } else { - const struct iovec iov = { - .iov_base = bp->data, - .iov_len = dlen - }; - struct iov_tail data = IOV_TAIL(&iov, 1, 0); - csum_udp6(&bp->uh, &toside->oaddr.a6, &toside->eaddr.a6, &data, - dlen); + csum_udp6(uh, &toside->oaddr.a6, &toside->eaddr.a6, + payload, dlen); } return l4len; @@ -372,15 +366,20 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, bool no_udp_csum) { struct iovec (*tap_iov)[UDP_NUM_IOVS] = &udp_l2_iov[idx]; + struct udphdr *uh = (*tap_iov)[UDP_IOV_PAYLOAD].iov_base; + struct iov_tail payload = IOV_TAIL(&(*tap_iov)[UDP_IOV_PAYLOAD], 1, + sizeof(*uh)); struct ethhdr *eh = (*tap_iov)[UDP_IOV_ETH].iov_base; - struct udp_payload_t *bp = &udp_payload[idx]; struct udp_meta_t *bm = &udp_meta[idx]; size_t l4len, l2len; + l4len = sizeof(struct udphdr) + mmh[idx].msg_len; + (*tap_iov)[UDP_IOV_PAYLOAD].iov_len = l4len; + eth_update_mac(eh, NULL, tap_omac); if (!inany_v4(&toside->eaddr) || !inany_v4(&toside->oaddr)) { - l4len = udp_update_hdr6(&bm->ip6h, bp, toside, - mmh[idx].msg_len, no_udp_csum); + udp_update_hdr6(&bm->ip6h, uh, &payload, toside, + mmh[idx].msg_len, no_udp_csum); l2len = MAX(l4len + sizeof(bm->ip6h) + ETH_HLEN, ETH_ZLEN); tap_hdr_update(&bm->taph, l2len); @@ -388,8 +387,8 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, eh->h_proto = htons_constant(ETH_P_IPV6); (*tap_iov)[UDP_IOV_IP] = IOV_OF_LVALUE(bm->ip6h); } else { - l4len = udp_update_hdr4(&bm->ip4h, bp, toside, - mmh[idx].msg_len, no_udp_csum); + udp_update_hdr4(&bm->ip4h, uh, &payload, toside, + mmh[idx].msg_len, no_udp_csum); l2len = MAX(l4len + sizeof(bm->ip4h) + ETH_HLEN, ETH_ZLEN); tap_hdr_update(&bm->taph, l2len); @@ -397,7 +396,6 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, eh->h_proto = htons_constant(ETH_P_IP); (*tap_iov)[UDP_IOV_IP] = IOV_OF_LVALUE(bm->ip4h); } - (*tap_iov)[UDP_IOV_PAYLOAD].iov_len = l4len; udp_tap_pad(*tap_iov); } diff --git a/udp_internal.h b/udp_internal.h index 64e457748324..361cc7495a01 100644 --- a/udp_internal.h +++ b/udp_internal.h @@ -25,10 +25,12 @@ struct udp_payload_t { } __attribute__ ((packed, aligned(__alignof__(unsigned int)))); #endif -size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, +size_t udp_update_hdr4(struct iphdr *ip4h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum); -size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, +size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum); void udp_sock_fwd(const struct ctx *c, int s, int rule_hint, diff --git a/udp_vu.c b/udp_vu.c index 74bf79d57969..888367ef4e16 100644 --- a/udp_vu.c +++ b/udp_vu.c @@ -55,40 +55,33 @@ static size_t udp_vu_hdrlen(bool v6) /** * udp_vu_sock_recv() - Receive datagrams from socket into vhost-user buffers - * @iov: IO vector for the frame (in/out) - * @cnt: Number of available entries in @iov (input) - * Number of used entries in @iov to store the datagram (output) + * @payload: UDP payload + * @cnt: Number of used entries in @payload to store the datagram (output) * Unchanged on failure * @s: Socket to receive from - * @v6: Set for IPv6 connections * * Return: size of received data, -1 on error */ -static ssize_t udp_vu_sock_recv(struct iovec *iov, size_t *cnt, int s, bool v6) +static ssize_t udp_vu_sock_recv(struct iov_tail *payload, size_t *cnt, int s) { struct iovec msg_iov[VIRTQUEUE_MAX_SIZE]; struct msghdr msg = { 0 }; - struct iov_tail payload; - size_t hdrlen, iov_used; + size_t iov_used; ssize_t dlen; - /* compute L2 header length */ - hdrlen = udp_vu_hdrlen(v6); - - payload = IOV_TAIL(iov, *cnt, hdrlen); - msg.msg_iov = msg_iov; - msg.msg_iovlen = iov_tail_clone(msg.msg_iov, payload.cnt, &payload); + msg.msg_iovlen = iov_tail_clone(msg.msg_iov, ARRAY_SIZE(msg_iov), + payload); /* read data from the socket */ dlen = recvmsg(s, &msg, 0); if (dlen < 0) return -1; - iov_used = iov_skip_bytes(iov, *cnt, - MAX(dlen + hdrlen, VNET_HLEN + ETH_ZLEN), - NULL); - if (iov_used < *cnt) + iov_used = iov_skip_bytes(payload->iov, payload->cnt, + MAX(dlen + payload->off, + VNET_HLEN + ETH_ZLEN), NULL); + if (iov_used < payload->cnt) iov_used++; *cnt = iov_used; /* one iovec per element */ @@ -98,69 +91,44 @@ static ssize_t udp_vu_sock_recv(struct iovec *iov, size_t *cnt, int s, bool v6) /** * udp_vu_prepare() - Prepare the packet header * @c: Execution context - * @iov: IO vector for the frame (including vnet header) + * @data: IO vector tail for the L2 frame, on return points to the L4 header + * @payload: UDP payload * @toside: Address information for one side of the flow * @dlen: Packet data length */ -static void udp_vu_prepare(const struct ctx *c, const struct iovec *iov, - const struct flowside *toside, ssize_t dlen) +static void udp_vu_prepare(const struct ctx *c, struct iov_tail *data, + struct iov_tail *payload, + const struct flowside *toside, size_t dlen) { - struct ethhdr *eh; + bool ipv4 = inany_v4(&toside->eaddr) && inany_v4(&toside->oaddr); + struct ethhdr eh; + struct udphdr uh; /* ethernet header */ - eh = vu_eth(iov[0].iov_base); + memcpy(eh.h_dest, c->guest_mac, sizeof(eh.h_dest)); + memcpy(eh.h_source, c->our_tap_mac, sizeof(eh.h_source)); - memcpy(eh->h_dest, c->guest_mac, sizeof(eh->h_dest)); - memcpy(eh->h_source, c->our_tap_mac, sizeof(eh->h_source)); + if (ipv4) + eh.h_proto = htons(ETH_P_IP); + else + eh.h_proto = htons(ETH_P_IPV6); + IOV_PUSH_HEADER(data, eh); /* initialize header */ - if (inany_v4(&toside->eaddr) && inany_v4(&toside->oaddr)) { - struct iphdr *iph = vu_ip(iov[0].iov_base); - struct udp_payload_t *bp = vu_payloadv4(iov[0].iov_base); - - eh->h_proto = htons(ETH_P_IP); + if (ipv4) { + struct iphdr iph = (struct iphdr)L2_BUF_IP4_INIT(IPPROTO_UDP); - *iph = (struct iphdr)L2_BUF_IP4_INIT(IPPROTO_UDP); + udp_update_hdr4(&iph, &uh, payload, toside, dlen, !*c->pcap); - udp_update_hdr4(iph, bp, toside, dlen, true); + IOV_PUSH_HEADER(data, iph); } else { - struct ipv6hdr *ip6h = vu_ip(iov[0].iov_base); - struct udp_payload_t *bp = vu_payloadv6(iov[0].iov_base); + struct ipv6hdr ip6h = (struct ipv6hdr)L2_BUF_IP6_INIT(IPPROTO_UDP); - eh->h_proto = htons(ETH_P_IPV6); + udp_update_hdr6(&ip6h, &uh, payload, toside, dlen, !*c->pcap); - *ip6h = (struct ipv6hdr)L2_BUF_IP6_INIT(IPPROTO_UDP); - - udp_update_hdr6(ip6h, bp, toside, dlen, true); - } -} - -/** - * udp_vu_csum() - Calculate and set checksum for a UDP packet - * @toside: Address information for one side of the flow - * @iov: IO vector for the frame - * @cnt: Number of IO vector entries - * @dlen: Data length - */ -static void udp_vu_csum(const struct flowside *toside, const struct iovec *iov, - size_t cnt, size_t dlen) -{ - const struct in_addr *src4 = inany_v4(&toside->oaddr); - const struct in_addr *dst4 = inany_v4(&toside->eaddr); - char *base = iov[0].iov_base; - struct udp_payload_t *bp; - struct iov_tail data; - - if (src4 && dst4) { - bp = vu_payloadv4(base); - data = IOV_TAIL(iov, cnt, (char *)&bp->data - base); - csum_udp4(&bp->uh, *src4, *dst4, &data, dlen); - } else { - bp = vu_payloadv6(base); - data = IOV_TAIL(iov, cnt, (char *)&bp->data - base); - csum_udp6(&bp->uh, &toside->oaddr.a6, &toside->eaddr.a6, &data, - dlen); + IOV_PUSH_HEADER(data, ip6h); } + IOV_PUSH_HEADER(data, uh); } /** @@ -198,6 +166,7 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) for (i = 0; i < n; i++) { unsigned elem_cnt, elem_used, j, k; + struct iov_tail payload; size_t iov_cnt; ssize_t dlen; @@ -207,7 +176,8 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) if (elem_cnt == 0) break; - dlen = udp_vu_sock_recv(iov_vu, &iov_cnt, s, v6); + payload = IOV_TAIL(iov_vu, iov_cnt, hdrlen); + dlen = udp_vu_sock_recv(&payload, &iov_cnt, s); if (dlen < 0) { vu_queue_rewind(vq, elem_cnt); break; @@ -227,9 +197,9 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) vu_queue_rewind(vq, elem_cnt - elem_used); if (iov_cnt > 0) { - udp_vu_prepare(c, iov_vu, toside, dlen); + struct iov_tail data = IOV_TAIL(iov_vu, iov_cnt, VNET_HLEN); + udp_vu_prepare(c, &data, &payload, toside, dlen); if (*c->pcap) { - udp_vu_csum(toside, iov_vu, iov_cnt, dlen); pcap_iov(iov_vu, iov_cnt, VNET_HLEN, hdrlen + dlen - VNET_HLEN); } -- 2.54.0

David Gibson

20 May 20 May

3:30 a.m.

New subject: [PATCH v10 3/3] udp: Pass iov_tail to udp_update_hdr4()/udp_update_hdr6()

On Tue, May 19, 2026 at 05:56:13PM +0200, Laurent Vivier wrote:

...

Change udp_update_hdr4() and udp_update_hdr6() to take an iov_tail pointing at the UDP frame instead of a contiguous udp_payload_t buffer and explicit data length. This lets vhost-user pass scatter-gather virtqueue buffers directly without an intermediate copy.

The UDP header is built into a local struct udphdr and written back with IOV_PUSH_HEADER(). On the tap side, udp_tap_prepare() wraps the existing udp_payload_t in a two-element iov to match the new interface.

Signed-off-by: Laurent Vivier --- iov.c | 1 - udp.c | 74 +++++++++++++++++----------------- udp_internal.h | 6 ++- udp_vu.c | 106 ++++++++++++++++++------------------------------- 4 files changed, 78 insertions(+), 109 deletions(-)

Lovely :). Reviewed-by: David Gibson Couple of cosmetic nits below, follow up at your discretion.

...

diff --git a/iov.c b/iov.c index 6a5d7d35b67f..9248ba95a9f2 100644 --- a/iov.c +++ b/iov.c @@ -367,7 +367,6 @@ void *iov_peek_header_(struct iov_tail *tail, void *v, size_t len, size_t align) * * Return: number of bytes written */ -/* cppcheck-suppress unusedFunction */ size_t iov_push_header_(struct iov_tail *tail, const void *v, size_t len) { size_t l; diff --git a/udp.c b/udp.c index 66dc7766868c..cfc7b2439881 100644 --- a/udp.c +++ b/udp.c @@ -255,20 +255,22 @@ static void udp_iov_init(const struct ctx *c) /** * udp_update_hdr4() - Update headers for one IPv4 datagram * @ip4h: Pre-filled IPv4 header (except for tot_len and saddr) - * @bp: Pointer to udp_payload_t to update + * @uh: UDP header to fill + * @payload: UDP payload * @toside: Flowside for destination side * @dlen: Length of UDP payload * @no_udp_csum: Do not set UDP checksum * - * Return: size of IPv4 payload (UDP header + data) + * Return: size of datagram (UDP header + data) */ -size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, +size_t udp_update_hdr4(struct iphdr *ip4h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum) { const struct in_addr *src = inany_v4(&toside->oaddr); const struct in_addr *dst = inany_v4(&toside->eaddr); - size_t l4len = dlen + sizeof(bp->uh); + size_t l4len = dlen + sizeof(*uh); size_t l3len = l4len + sizeof(*ip4h);

assert(src && dst); @@ -278,19 +280,13 @@ size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, ip4h->saddr = src->s_addr; ip4h->check = csum_ip4_header(l3len, IPPROTO_UDP, *src, *dst);

- bp->uh.source = htons(toside->oport); - bp->uh.dest = htons(toside->eport); - bp->uh.len = htons(l4len); - if (no_udp_csum) { - bp->uh.check = 0; - } else { - const struct iovec iov = { - .iov_base = bp->data, - .iov_len = dlen - }; - struct iov_tail data = IOV_TAIL(&iov, 1, 0); - csum_udp4(&bp->uh, *src, *dst, &data, dlen); - } + uh->source = htons(toside->oport); + uh->dest = htons(toside->eport); + uh->len = htons(l4len); + if (no_udp_csum) + uh->check = 0; + else + csum_udp4(uh, *src, *dst, payload, dlen);

return l4len; } @@ -299,18 +295,20 @@ size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, * udp_update_hdr6() - Update headers for one IPv6 datagram * @ip6h: Pre-filled IPv6 header (except for payload_len and * addresses) - * @bp: Pointer to udp_payload_t to update + * @uh: UDP header to fill + * @payload: UDP payload * @toside: Flowside for destination side * @dlen: Length of UDP payload * @no_udp_csum: Do not set UDP checksum * - * Return: size of IPv6 payload (UDP header + data) + * Return: size of datagram (UDP header + data) */ -size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, +size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum) { - uint16_t l4len = dlen + sizeof(bp->uh); + uint16_t l4len = dlen + sizeof(*uh);

ip6h->payload_len = htons(l4len); ip6h->daddr = toside->eaddr.a6; @@ -319,23 +317,19 @@ size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, ip6h->nexthdr = IPPROTO_UDP; ip6h->hop_limit = 255;

- bp->uh.source = htons(toside->oport); - bp->uh.dest = htons(toside->eport); - bp->uh.len = ip6h->payload_len; + uh->source = htons(toside->oport); + uh->dest = htons(toside->eport); + uh->len = htons(l4len); + if (no_udp_csum) { /* 0 is an invalid checksum for UDP IPv6 and dropped by * the kernel stack, even if the checksum is disabled by virtio * flags. We need to put any non-zero value here. */ - bp->uh.check = 0xffff; + uh->check = 0xffff; } else { - const struct iovec iov = { - .iov_base = bp->data, - .iov_len = dlen - }; - struct iov_tail data = IOV_TAIL(&iov, 1, 0); - csum_udp6(&bp->uh, &toside->oaddr.a6, &toside->eaddr.a6, &data, - dlen); + csum_udp6(uh, &toside->oaddr.a6, &toside->eaddr.a6, + payload, dlen); }

return l4len; @@ -372,15 +366,20 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, bool no_udp_csum) { struct iovec (*tap_iov)[UDP_NUM_IOVS] = &udp_l2_iov[idx]; + struct udphdr *uh = (*tap_iov)[UDP_IOV_PAYLOAD].iov_base; + struct iov_tail payload = IOV_TAIL(&(*tap_iov)[UDP_IOV_PAYLOAD], 1, + sizeof(*uh)); struct ethhdr *eh = (*tap_iov)[UDP_IOV_ETH].iov_base; - struct udp_payload_t *bp = &udp_payload[idx]; struct udp_meta_t *bm = &udp_meta[idx]; size_t l4len, l2len;

+ l4len = sizeof(struct udphdr) + mmh[idx].msg_len;

Nit: sizeof(*uh).

...

+ (*tap_iov)[UDP_IOV_PAYLOAD].iov_len = l4len; + eth_update_mac(eh, NULL, tap_omac); if (!inany_v4(&toside->eaddr) || !inany_v4(&toside->oaddr)) { - l4len = udp_update_hdr6(&bm->ip6h, bp, toside, - mmh[idx].msg_len, no_udp_csum); + udp_update_hdr6(&bm->ip6h, uh, &payload, toside, + mmh[idx].msg_len, no_udp_csum);

l2len = MAX(l4len + sizeof(bm->ip6h) + ETH_HLEN, ETH_ZLEN); tap_hdr_update(&bm->taph, l2len); @@ -388,8 +387,8 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, eh->h_proto = htons_constant(ETH_P_IPV6); (*tap_iov)[UDP_IOV_IP] = IOV_OF_LVALUE(bm->ip6h); } else { - l4len = udp_update_hdr4(&bm->ip4h, bp, toside, - mmh[idx].msg_len, no_udp_csum); + udp_update_hdr4(&bm->ip4h, uh, &payload, toside, + mmh[idx].msg_len, no_udp_csum);

l2len = MAX(l4len + sizeof(bm->ip4h) + ETH_HLEN, ETH_ZLEN); tap_hdr_update(&bm->taph, l2len); @@ -397,7 +396,6 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, eh->h_proto = htons_constant(ETH_P_IP); (*tap_iov)[UDP_IOV_IP] = IOV_OF_LVALUE(bm->ip4h); } - (*tap_iov)[UDP_IOV_PAYLOAD].iov_len = l4len;

udp_tap_pad(*tap_iov); } diff --git a/udp_internal.h b/udp_internal.h index 64e457748324..361cc7495a01 100644 --- a/udp_internal.h +++ b/udp_internal.h @@ -25,10 +25,12 @@ struct udp_payload_t { } __attribute__ ((packed, aligned(__alignof__(unsigned int)))); #endif

-size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, +size_t udp_update_hdr4(struct iphdr *ip4h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum); -size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, +size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udphdr *uh, + struct iov_tail *payload, const struct flowside *toside, size_t dlen, bool no_udp_csum); void udp_sock_fwd(const struct ctx *c, int s, int rule_hint, diff --git a/udp_vu.c b/udp_vu.c index 74bf79d57969..888367ef4e16 100644 --- a/udp_vu.c +++ b/udp_vu.c @@ -55,40 +55,33 @@ static size_t udp_vu_hdrlen(bool v6)

/** * udp_vu_sock_recv() - Receive datagrams from socket into vhost-user buffers - * @iov: IO vector for the frame (in/out) - * @cnt: Number of available entries in @iov (input) - * Number of used entries in @iov to store the datagram (output) + * @payload: UDP payload

Nit: "Buffer(s) for UDP payload" maybe?

...

+ * @cnt: Number of used entries in @payload to store the datagram (output) * Unchanged on failure * @s: Socket to receive from - * @v6: Set for IPv6 connections * * Return: size of received data, -1 on error */ -static ssize_t udp_vu_sock_recv(struct iovec *iov, size_t *cnt, int s, bool v6) +static ssize_t udp_vu_sock_recv(struct iov_tail *payload, size_t *cnt, int s) { struct iovec msg_iov[VIRTQUEUE_MAX_SIZE]; struct msghdr msg = { 0 }; - struct iov_tail payload; - size_t hdrlen, iov_used; + size_t iov_used; ssize_t dlen;

- /* compute L2 header length */ - hdrlen = udp_vu_hdrlen(v6); - - payload = IOV_TAIL(iov, *cnt, hdrlen); - msg.msg_iov = msg_iov; - msg.msg_iovlen = iov_tail_clone(msg.msg_iov, payload.cnt, &payload); + msg.msg_iovlen = iov_tail_clone(msg.msg_iov, ARRAY_SIZE(msg_iov), + payload);

/* read data from the socket */ dlen = recvmsg(s, &msg, 0); if (dlen < 0) return -1;

- iov_used = iov_skip_bytes(iov, *cnt, - MAX(dlen + hdrlen, VNET_HLEN + ETH_ZLEN), - NULL); - if (iov_used < *cnt) + iov_used = iov_skip_bytes(payload->iov, payload->cnt, + MAX(dlen + payload->off, + VNET_HLEN + ETH_ZLEN), NULL); + if (iov_used < payload->cnt) iov_used++; *cnt = iov_used; /* one iovec per element */

@@ -98,69 +91,44 @@ static ssize_t udp_vu_sock_recv(struct iovec *iov, size_t *cnt, int s, bool v6) /** * udp_vu_prepare() - Prepare the packet header * @c: Execution context - * @iov: IO vector for the frame (including vnet header) + * @data: IO vector tail for the L2 frame, on return points to the L4 header + * @payload: UDP payload * @toside: Address information for one side of the flow * @dlen: Packet data length */ -static void udp_vu_prepare(const struct ctx *c, const struct iovec *iov, - const struct flowside *toside, ssize_t dlen) +static void udp_vu_prepare(const struct ctx *c, struct iov_tail *data, + struct iov_tail *payload, + const struct flowside *toside, size_t dlen) { - struct ethhdr *eh; + bool ipv4 = inany_v4(&toside->eaddr) && inany_v4(&toside->oaddr); + struct ethhdr eh; + struct udphdr uh;

/* ethernet header */ - eh = vu_eth(iov[0].iov_base); + memcpy(eh.h_dest, c->guest_mac, sizeof(eh.h_dest)); + memcpy(eh.h_source, c->our_tap_mac, sizeof(eh.h_source));

- memcpy(eh->h_dest, c->guest_mac, sizeof(eh->h_dest)); - memcpy(eh->h_source, c->our_tap_mac, sizeof(eh->h_source)); + if (ipv4) + eh.h_proto = htons(ETH_P_IP); + else + eh.h_proto = htons(ETH_P_IPV6); + IOV_PUSH_HEADER(data, eh);

/* initialize header */ - if (inany_v4(&toside->eaddr) && inany_v4(&toside->oaddr)) { - struct iphdr *iph = vu_ip(iov[0].iov_base); - struct udp_payload_t *bp = vu_payloadv4(iov[0].iov_base); - - eh->h_proto = htons(ETH_P_IP); + if (ipv4) { + struct iphdr iph = (struct iphdr)L2_BUF_IP4_INIT(IPPROTO_UDP);

- *iph = (struct iphdr)L2_BUF_IP4_INIT(IPPROTO_UDP); + udp_update_hdr4(&iph, &uh, payload, toside, dlen, !*c->pcap);

- udp_update_hdr4(iph, bp, toside, dlen, true); + IOV_PUSH_HEADER(data, iph); } else { - struct ipv6hdr *ip6h = vu_ip(iov[0].iov_base); - struct udp_payload_t *bp = vu_payloadv6(iov[0].iov_base); + struct ipv6hdr ip6h = (struct ipv6hdr)L2_BUF_IP6_INIT(IPPROTO_UDP);

- eh->h_proto = htons(ETH_P_IPV6); + udp_update_hdr6(&ip6h, &uh, payload, toside, dlen, !*c->pcap);

- *ip6h = (struct ipv6hdr)L2_BUF_IP6_INIT(IPPROTO_UDP); - - udp_update_hdr6(ip6h, bp, toside, dlen, true); - } -} - -/** - * udp_vu_csum() - Calculate and set checksum for a UDP packet - * @toside: Address information for one side of the flow - * @iov: IO vector for the frame - * @cnt: Number of IO vector entries - * @dlen: Data length - */ -static void udp_vu_csum(const struct flowside *toside, const struct iovec *iov, - size_t cnt, size_t dlen) -{ - const struct in_addr *src4 = inany_v4(&toside->oaddr); - const struct in_addr *dst4 = inany_v4(&toside->eaddr); - char *base = iov[0].iov_base; - struct udp_payload_t *bp; - struct iov_tail data; - - if (src4 && dst4) { - bp = vu_payloadv4(base); - data = IOV_TAIL(iov, cnt, (char *)&bp->data - base); - csum_udp4(&bp->uh, *src4, *dst4, &data, dlen); - } else { - bp = vu_payloadv6(base); - data = IOV_TAIL(iov, cnt, (char *)&bp->data - base); - csum_udp6(&bp->uh, &toside->oaddr.a6, &toside->eaddr.a6, &data, - dlen); + IOV_PUSH_HEADER(data, ip6h); } + IOV_PUSH_HEADER(data, uh); }

/** @@ -198,6 +166,7 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx)

for (i = 0; i < n; i++) { unsigned elem_cnt, elem_used, j, k; + struct iov_tail payload; size_t iov_cnt; ssize_t dlen;

@@ -207,7 +176,8 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) if (elem_cnt == 0) break;

- dlen = udp_vu_sock_recv(iov_vu, &iov_cnt, s, v6); + payload = IOV_TAIL(iov_vu, iov_cnt, hdrlen); + dlen = udp_vu_sock_recv(&payload, &iov_cnt, s); if (dlen < 0) { vu_queue_rewind(vq, elem_cnt); break; @@ -227,9 +197,9 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) vu_queue_rewind(vq, elem_cnt - elem_used);

if (iov_cnt > 0) { - udp_vu_prepare(c, iov_vu, toside, dlen); + struct iov_tail data = IOV_TAIL(iov_vu, iov_cnt, VNET_HLEN); + udp_vu_prepare(c, &data, &payload, toside, dlen); if (*c->pcap) { - udp_vu_csum(toside, iov_vu, iov_cnt, dlen); pcap_iov(iov_vu, iov_cnt, VNET_HLEN, hdrlen + dlen - VNET_HLEN); } -- 2.54.0

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

Age (days ago)

Last active (days ago)

List overview

Download

4 comments

2 participants

participants (2)

David Gibson
Laurent Vivier