diff options
Diffstat (limited to 'hw')
-rw-r--r-- | hw/net/net_tx_pkt.c | 248 |
1 files changed, 206 insertions, 42 deletions
diff --git a/hw/net/net_tx_pkt.c b/hw/net/net_tx_pkt.c index 6afd3f6..4a35e84 100644 --- a/hw/net/net_tx_pkt.c +++ b/hw/net/net_tx_pkt.c @@ -326,7 +326,8 @@ bool net_tx_pkt_build_vheader(struct NetTxPkt *pkt, bool tso_enable, case VIRTIO_NET_HDR_GSO_TCPV6: bytes_read = iov_to_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags, 0, &l4hdr, sizeof(l4hdr)); - if (bytes_read < sizeof(l4hdr)) { + if (bytes_read < sizeof(l4hdr) || + l4hdr.th_off * sizeof(uint32_t) < sizeof(l4hdr)) { return false; } @@ -466,15 +467,14 @@ void net_tx_pkt_reset(struct NetTxPkt *pkt) pkt->l4proto = 0; } -static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt) +static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt, + struct iovec *iov, uint32_t iov_len, + uint16_t csl) { - struct iovec *iov = &pkt->vec[NET_TX_PKT_L2HDR_FRAG]; uint32_t csum_cntr; uint16_t csum = 0; uint32_t cso; /* num of iovec without vhdr */ - uint32_t iov_len = pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1; - uint16_t csl; size_t csum_offset = pkt->virt_hdr.csum_start + pkt->virt_hdr.csum_offset; uint16_t l3_proto = eth_get_l3_proto(iov, 1, iov->iov_len); @@ -482,8 +482,6 @@ static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt) iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum); /* Calculate L4 TCP/UDP checksum */ - csl = pkt->payload_len; - csum_cntr = 0; cso = 0; /* add pseudo header to csum */ @@ -509,14 +507,13 @@ static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt) #define NET_MAX_FRAG_SG_LIST (64) static size_t net_tx_pkt_fetch_fragment(struct NetTxPkt *pkt, - int *src_idx, size_t *src_offset, struct iovec *dst, int *dst_idx) + int *src_idx, size_t *src_offset, size_t src_len, + struct iovec *dst, int *dst_idx) { size_t fetched = 0; struct iovec *src = pkt->vec; - *dst_idx = NET_TX_PKT_PL_START_FRAG; - - while (fetched < IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size)) { + while (fetched < src_len) { /* no more place in fragment iov */ if (*dst_idx == NET_MAX_FRAG_SG_LIST) { @@ -531,7 +528,7 @@ static size_t net_tx_pkt_fetch_fragment(struct NetTxPkt *pkt, dst[*dst_idx].iov_base = src[*src_idx].iov_base + *src_offset; dst[*dst_idx].iov_len = MIN(src[*src_idx].iov_len - *src_offset, - IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size) - fetched); + src_len - fetched); *src_offset += dst[*dst_idx].iov_len; fetched += dst[*dst_idx].iov_len; @@ -560,58 +557,223 @@ static void net_tx_pkt_sendv( } } +static bool net_tx_pkt_tcp_fragment_init(struct NetTxPkt *pkt, + struct iovec *fragment, + int *pl_idx, + size_t *l4hdr_len, + int *src_idx, + size_t *src_offset, + size_t *src_len) +{ + struct iovec *l4 = fragment + NET_TX_PKT_PL_START_FRAG; + size_t bytes_read = 0; + struct tcp_hdr *th; + + if (!pkt->payload_frags) { + return false; + } + + l4->iov_len = pkt->virt_hdr.hdr_len - pkt->hdr_len; + l4->iov_base = g_malloc(l4->iov_len); + + *src_idx = NET_TX_PKT_PL_START_FRAG; + while (pkt->vec[*src_idx].iov_len < l4->iov_len - bytes_read) { + memcpy((char *)l4->iov_base + bytes_read, pkt->vec[*src_idx].iov_base, + pkt->vec[*src_idx].iov_len); + + bytes_read += pkt->vec[*src_idx].iov_len; + + (*src_idx)++; + if (*src_idx >= pkt->payload_frags + NET_TX_PKT_PL_START_FRAG) { + g_free(l4->iov_base); + return false; + } + } + + *src_offset = l4->iov_len - bytes_read; + memcpy((char *)l4->iov_base + bytes_read, pkt->vec[*src_idx].iov_base, + *src_offset); + + th = l4->iov_base; + th->th_flags &= ~(TH_FIN | TH_PUSH); + + *pl_idx = NET_TX_PKT_PL_START_FRAG + 1; + *l4hdr_len = l4->iov_len; + *src_len = pkt->virt_hdr.gso_size; + + return true; +} + +static void net_tx_pkt_tcp_fragment_deinit(struct iovec *fragment) +{ + g_free(fragment[NET_TX_PKT_PL_START_FRAG].iov_base); +} + +static void net_tx_pkt_tcp_fragment_fix(struct NetTxPkt *pkt, + struct iovec *fragment, + size_t fragment_len, + uint8_t gso_type) +{ + struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG; + struct iovec *l4hdr = fragment + NET_TX_PKT_PL_START_FRAG; + struct ip_header *ip = l3hdr->iov_base; + struct ip6_header *ip6 = l3hdr->iov_base; + size_t len = l3hdr->iov_len + l4hdr->iov_len + fragment_len; + + switch (gso_type) { + case VIRTIO_NET_HDR_GSO_TCPV4: + ip->ip_len = cpu_to_be16(len); + eth_fix_ip4_checksum(l3hdr->iov_base, l3hdr->iov_len); + break; + + case VIRTIO_NET_HDR_GSO_TCPV6: + len -= sizeof(struct ip6_header); + ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = cpu_to_be16(len); + break; + } +} + +static void net_tx_pkt_tcp_fragment_advance(struct NetTxPkt *pkt, + struct iovec *fragment, + size_t fragment_len, + uint8_t gso_type) +{ + struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG; + struct iovec *l4hdr = fragment + NET_TX_PKT_PL_START_FRAG; + struct ip_header *ip = l3hdr->iov_base; + struct tcp_hdr *th = l4hdr->iov_base; + + if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4) { + ip->ip_id = cpu_to_be16(be16_to_cpu(ip->ip_id) + 1); + } + + th->th_seq = cpu_to_be32(be32_to_cpu(th->th_seq) + fragment_len); + th->th_flags &= ~TH_CWR; +} + +static void net_tx_pkt_udp_fragment_init(struct NetTxPkt *pkt, + int *pl_idx, + size_t *l4hdr_len, + int *src_idx, size_t *src_offset, + size_t *src_len) +{ + *pl_idx = NET_TX_PKT_PL_START_FRAG; + *l4hdr_len = 0; + *src_idx = NET_TX_PKT_PL_START_FRAG; + *src_offset = 0; + *src_len = IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size); +} + +static void net_tx_pkt_udp_fragment_fix(struct NetTxPkt *pkt, + struct iovec *fragment, + size_t fragment_offset, + size_t fragment_len) +{ + bool more_frags = fragment_offset + fragment_len < pkt->payload_len; + uint16_t orig_flags; + struct iovec *l3hdr = fragment + NET_TX_PKT_L3HDR_FRAG; + struct ip_header *ip = l3hdr->iov_base; + uint16_t frag_off_units = fragment_offset / IP_FRAG_UNIT_SIZE; + uint16_t new_ip_off; + + assert(fragment_offset % IP_FRAG_UNIT_SIZE == 0); + assert((frag_off_units & ~IP_OFFMASK) == 0); + + orig_flags = be16_to_cpu(ip->ip_off) & ~(IP_OFFMASK | IP_MF); + new_ip_off = frag_off_units | orig_flags | (more_frags ? IP_MF : 0); + ip->ip_off = cpu_to_be16(new_ip_off); + ip->ip_len = cpu_to_be16(l3hdr->iov_len + fragment_len); + + eth_fix_ip4_checksum(l3hdr->iov_base, l3hdr->iov_len); +} + static bool net_tx_pkt_do_sw_fragmentation(struct NetTxPkt *pkt, NetTxPktCallback callback, void *context) { + uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN; + struct iovec fragment[NET_MAX_FRAG_SG_LIST]; - size_t fragment_len = 0; - bool more_frags = false; - - /* some pointers for shorter code */ - void *l2_iov_base, *l3_iov_base; - size_t l2_iov_len, l3_iov_len; - int src_idx = NET_TX_PKT_PL_START_FRAG, dst_idx; - size_t src_offset = 0; + size_t fragment_len; + size_t l4hdr_len; + size_t src_len; + + int src_idx, dst_idx, pl_idx; + size_t src_offset; size_t fragment_offset = 0; struct virtio_net_hdr virt_hdr = { .flags = pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM ? VIRTIO_NET_HDR_F_DATA_VALID : 0 }; - l2_iov_base = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base; - l2_iov_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len; - l3_iov_base = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base; - l3_iov_len = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len; - /* Copy headers */ fragment[NET_TX_PKT_VHDR_FRAG].iov_base = &virt_hdr; fragment[NET_TX_PKT_VHDR_FRAG].iov_len = sizeof(virt_hdr); - fragment[NET_TX_PKT_L2HDR_FRAG].iov_base = l2_iov_base; - fragment[NET_TX_PKT_L2HDR_FRAG].iov_len = l2_iov_len; - fragment[NET_TX_PKT_L3HDR_FRAG].iov_base = l3_iov_base; - fragment[NET_TX_PKT_L3HDR_FRAG].iov_len = l3_iov_len; + fragment[NET_TX_PKT_L2HDR_FRAG] = pkt->vec[NET_TX_PKT_L2HDR_FRAG]; + fragment[NET_TX_PKT_L3HDR_FRAG] = pkt->vec[NET_TX_PKT_L3HDR_FRAG]; + switch (gso_type) { + case VIRTIO_NET_HDR_GSO_TCPV4: + case VIRTIO_NET_HDR_GSO_TCPV6: + if (!net_tx_pkt_tcp_fragment_init(pkt, fragment, &pl_idx, &l4hdr_len, + &src_idx, &src_offset, &src_len)) { + return false; + } + break; - /* Put as much data as possible and send */ - do { - fragment_len = net_tx_pkt_fetch_fragment(pkt, &src_idx, &src_offset, - fragment, &dst_idx); + case VIRTIO_NET_HDR_GSO_UDP: + net_tx_pkt_do_sw_csum(pkt, &pkt->vec[NET_TX_PKT_L2HDR_FRAG], + pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1, + pkt->payload_len); + net_tx_pkt_udp_fragment_init(pkt, &pl_idx, &l4hdr_len, + &src_idx, &src_offset, &src_len); + break; - more_frags = (fragment_offset + fragment_len < pkt->payload_len); + default: + abort(); + } - eth_setup_ip4_fragmentation(l2_iov_base, l2_iov_len, l3_iov_base, - l3_iov_len, fragment_len, fragment_offset, more_frags); + /* Put as much data as possible and send */ + while (true) { + dst_idx = pl_idx; + fragment_len = net_tx_pkt_fetch_fragment(pkt, + &src_idx, &src_offset, src_len, fragment, &dst_idx); + if (!fragment_len) { + break; + } - eth_fix_ip4_checksum(l3_iov_base, l3_iov_len); + switch (gso_type) { + case VIRTIO_NET_HDR_GSO_TCPV4: + case VIRTIO_NET_HDR_GSO_TCPV6: + net_tx_pkt_tcp_fragment_fix(pkt, fragment, fragment_len, gso_type); + net_tx_pkt_do_sw_csum(pkt, fragment + NET_TX_PKT_L2HDR_FRAG, + dst_idx - NET_TX_PKT_L2HDR_FRAG, + l4hdr_len + fragment_len); + break; + + case VIRTIO_NET_HDR_GSO_UDP: + net_tx_pkt_udp_fragment_fix(pkt, fragment, fragment_offset, + fragment_len); + break; + } callback(context, fragment + NET_TX_PKT_L2HDR_FRAG, dst_idx - NET_TX_PKT_L2HDR_FRAG, fragment + NET_TX_PKT_VHDR_FRAG, dst_idx - NET_TX_PKT_VHDR_FRAG); + if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 || + gso_type == VIRTIO_NET_HDR_GSO_TCPV6) { + net_tx_pkt_tcp_fragment_advance(pkt, fragment, fragment_len, + gso_type); + } + fragment_offset += fragment_len; + } - } while (fragment_len && more_frags); + if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 || + gso_type == VIRTIO_NET_HDR_GSO_TCPV6) { + net_tx_pkt_tcp_fragment_deinit(fragment); + } return true; } @@ -627,10 +789,6 @@ bool net_tx_pkt_send_custom(struct NetTxPkt *pkt, bool offload, { assert(pkt); - if (!offload && pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { - net_tx_pkt_do_sw_csum(pkt); - } - /* * Since underlying infrastructure does not support IP datagrams longer * than 64K we should drop such packets and don't even try to send @@ -644,6 +802,12 @@ bool net_tx_pkt_send_custom(struct NetTxPkt *pkt, bool offload, } if (offload || pkt->virt_hdr.gso_type == VIRTIO_NET_HDR_GSO_NONE) { + if (!offload && pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { + net_tx_pkt_do_sw_csum(pkt, &pkt->vec[NET_TX_PKT_L2HDR_FRAG], + pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1, + pkt->payload_len); + } + net_tx_pkt_fix_ip6_payload_len(pkt); callback(context, pkt->vec + NET_TX_PKT_L2HDR_FRAG, pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - NET_TX_PKT_L2HDR_FRAG, |