[qemu.git] / net / eth.c

/*
 * QEMU network structures definitions and helper functions
 *
 * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
 *
 * Developed by Daynix Computing LTD (http://www.daynix.com)
 *
 * Authors:
 * Dmitry Fleytman <[email protected]>
 * Tamir Shomer <[email protected]>
 * Yan Vugenfirer <[email protected]>
 *
 * This work is licensed under the terms of the GNU GPL, version 2 or later.
 * See the COPYING file in the top-level directory.
 *
 */

#include "qemu/osdep.h"
#include "net/eth.h"
#include "net/checksum.h"
#include "net/tap.h"

void eth_setup_vlan_headers_ex(struct eth_header *ehdr, uint16_t vlan_tag,
    uint16_t vlan_ethtype, bool *is_new)
{
    struct vlan_header *vhdr = PKT_GET_VLAN_HDR(ehdr);

    switch (be16_to_cpu(ehdr->h_proto)) {
    case ETH_P_VLAN:
    case ETH_P_DVLAN:
        /* vlan hdr exists */
        *is_new = false;
        break;

    default:
        /* No VLAN header, put a new one */
        vhdr->h_proto = ehdr->h_proto;
        ehdr->h_proto = cpu_to_be16(vlan_ethtype);
        *is_new = true;
        break;
    }
    vhdr->h_tci = cpu_to_be16(vlan_tag);
}

uint8_t
eth_get_gso_type(uint16_t l3_proto, uint8_t *l3_hdr, uint8_t l4proto)
{
    uint8_t ecn_state = 0;

    if (l3_proto == ETH_P_IP) {
        struct ip_header *iphdr = (struct ip_header *) l3_hdr;

        if (IP_HEADER_VERSION(iphdr) == IP_HEADER_VERSION_4) {
            if (IPTOS_ECN(iphdr->ip_tos) == IPTOS_ECN_CE) {
                ecn_state = VIRTIO_NET_HDR_GSO_ECN;
            }
            if (l4proto == IP_PROTO_TCP) {
                return VIRTIO_NET_HDR_GSO_TCPV4 | ecn_state;
            } else if (l4proto == IP_PROTO_UDP) {
                return VIRTIO_NET_HDR_GSO_UDP | ecn_state;
            }
        }
    } else if (l3_proto == ETH_P_IPV6) {
        struct ip6_header *ip6hdr = (struct ip6_header *) l3_hdr;

        if (IP6_ECN(ip6hdr->ip6_ecn_acc) == IP6_ECN_CE) {
            ecn_state = VIRTIO_NET_HDR_GSO_ECN;
        }

        if (l4proto == IP_PROTO_TCP) {
            return VIRTIO_NET_HDR_GSO_TCPV6 | ecn_state;
        }
    }

    /* Unsupported offload */
    g_assert_not_reached();

    return VIRTIO_NET_HDR_GSO_NONE | ecn_state;
}

uint16_t
eth_get_l3_proto(const struct iovec *l2hdr_iov, int iovcnt, size_t l2hdr_len)
{
    uint16_t proto;
    size_t copied;
    size_t size = iov_size(l2hdr_iov, iovcnt);
    size_t proto_offset = l2hdr_len - sizeof(proto);

    if (size < proto_offset) {
        return ETH_P_UNKNOWN;
    }

    copied = iov_to_buf(l2hdr_iov, iovcnt, proto_offset,
                        &proto, sizeof(proto));

    return (copied == sizeof(proto)) ? be16_to_cpu(proto) : ETH_P_UNKNOWN;
}

static bool
_eth_copy_chunk(size_t input_size,
                const struct iovec *iov, int iovcnt,
                size_t offset, size_t length,
                void *buffer)
{
    size_t copied;

    if (input_size < offset) {
        return false;
    }

    copied = iov_to_buf(iov, iovcnt, offset, buffer, length);

    if (copied < length) {
        return false;
    }

    return true;
}

static bool
_eth_tcp_has_data(bool is_ip4,
                  const struct ip_header  *ip4_hdr,
                  const struct ip6_header *ip6_hdr,
                  size_t full_ip6hdr_len,
                  const struct tcp_header *tcp)
{
    uint32_t l4len;

    if (is_ip4) {
        l4len = be16_to_cpu(ip4_hdr->ip_len) - IP_HDR_GET_LEN(ip4_hdr);
    } else {
        size_t opts_len = full_ip6hdr_len - sizeof(struct ip6_header);
        l4len = be16_to_cpu(ip6_hdr->ip6_ctlun.ip6_un1.ip6_un1_plen) - opts_len;
    }

    return l4len > TCP_HEADER_DATA_OFFSET(tcp);
}

void eth_get_protocols(const struct iovec *iov, int iovcnt,
                       bool *isip4, bool *isip6,
                       bool *isudp, bool *istcp,
                       size_t *l3hdr_off,
                       size_t *l4hdr_off,
                       size_t *l5hdr_off,
                       eth_ip6_hdr_info *ip6hdr_info,
                       eth_ip4_hdr_info *ip4hdr_info,
                       eth_l4_hdr_info  *l4hdr_info)
{
    int proto;
    bool fragment = false;
    size_t l2hdr_len = eth_get_l2_hdr_length_iov(iov, iovcnt);
    size_t input_size = iov_size(iov, iovcnt);
    size_t copied;

    *isip4 = *isip6 = *isudp = *istcp = false;

    proto = eth_get_l3_proto(iov, iovcnt, l2hdr_len);

    *l3hdr_off = l2hdr_len;

    if (proto == ETH_P_IP) {
        struct ip_header *iphdr = &ip4hdr_info->ip4_hdr;

        if (input_size < l2hdr_len) {
            return;
        }

        copied = iov_to_buf(iov, iovcnt, l2hdr_len, iphdr, sizeof(*iphdr));

        *isip4 = true;

        if (copied < sizeof(*iphdr)) {
            return;
        }

        if (IP_HEADER_VERSION(iphdr) == IP_HEADER_VERSION_4) {
            if (iphdr->ip_p == IP_PROTO_TCP) {
                *istcp = true;
            } else if (iphdr->ip_p == IP_PROTO_UDP) {
                *isudp = true;
            }
        }

        ip4hdr_info->fragment = IP4_IS_FRAGMENT(iphdr);
        *l4hdr_off = l2hdr_len + IP_HDR_GET_LEN(iphdr);

        fragment = ip4hdr_info->fragment;
    } else if (proto == ETH_P_IPV6) {

        *isip6 = true;
        if (eth_parse_ipv6_hdr(iov, iovcnt, l2hdr_len,
                               ip6hdr_info)) {
            if (ip6hdr_info->l4proto == IP_PROTO_TCP) {
                *istcp = true;
            } else if (ip6hdr_info->l4proto == IP_PROTO_UDP) {
                *isudp = true;
            }
        } else {
            return;
        }

        *l4hdr_off = l2hdr_len + ip6hdr_info->full_hdr_len;
        fragment = ip6hdr_info->fragment;
    }

    if (!fragment) {
        if (*istcp) {
            *istcp = _eth_copy_chunk(input_size,
                                     iov, iovcnt,
                                     *l4hdr_off, sizeof(l4hdr_info->hdr.tcp),
                                     &l4hdr_info->hdr.tcp);

            if (*istcp) {
                *l5hdr_off = *l4hdr_off +
                    TCP_HEADER_DATA_OFFSET(&l4hdr_info->hdr.tcp);

                l4hdr_info->has_tcp_data =
                    _eth_tcp_has_data(proto == ETH_P_IP,
                                      &ip4hdr_info->ip4_hdr,
                                      &ip6hdr_info->ip6_hdr,
                                      *l4hdr_off - *l3hdr_off,
                                      &l4hdr_info->hdr.tcp);
            }
        } else if (*isudp) {
            *isudp = _eth_copy_chunk(input_size,
                                     iov, iovcnt,
                                     *l4hdr_off, sizeof(l4hdr_info->hdr.udp),
                                     &l4hdr_info->hdr.udp);
            *l5hdr_off = *l4hdr_off + sizeof(l4hdr_info->hdr.udp);
        }
    }
}

size_t
eth_strip_vlan(const struct iovec *iov, int iovcnt, size_t iovoff,
               uint8_t *new_ehdr_buf,
               uint16_t *payload_offset, uint16_t *tci)
{
    struct vlan_header vlan_hdr;
    struct eth_header *new_ehdr = (struct eth_header *) new_ehdr_buf;

    size_t copied = iov_to_buf(iov, iovcnt, iovoff,
                               new_ehdr, sizeof(*new_ehdr));

    if (copied < sizeof(*new_ehdr)) {
        return 0;
    }

    switch (be16_to_cpu(new_ehdr->h_proto)) {
    case ETH_P_VLAN:
    case ETH_P_DVLAN:
        copied = iov_to_buf(iov, iovcnt, iovoff + sizeof(*new_ehdr),
                            &vlan_hdr, sizeof(vlan_hdr));

        if (copied < sizeof(vlan_hdr)) {
            return 0;
        }

        new_ehdr->h_proto = vlan_hdr.h_proto;

        *tci = be16_to_cpu(vlan_hdr.h_tci);
        *payload_offset = iovoff + sizeof(*new_ehdr) + sizeof(vlan_hdr);

        if (be16_to_cpu(new_ehdr->h_proto) == ETH_P_VLAN) {

            copied = iov_to_buf(iov, iovcnt, *payload_offset,
                                PKT_GET_VLAN_HDR(new_ehdr), sizeof(vlan_hdr));

            if (copied < sizeof(vlan_hdr)) {
                return 0;
            }

            *payload_offset += sizeof(vlan_hdr);

            return sizeof(struct eth_header) + sizeof(struct vlan_header);
        } else {
            return sizeof(struct eth_header);
        }
    default:
        return 0;
    }
}

size_t
eth_strip_vlan_ex(const struct iovec *iov, int iovcnt, size_t iovoff,
                  uint16_t vet, uint8_t *new_ehdr_buf,
                  uint16_t *payload_offset, uint16_t *tci)
{
    struct vlan_header vlan_hdr;
    struct eth_header *new_ehdr = (struct eth_header *) new_ehdr_buf;

    size_t copied = iov_to_buf(iov, iovcnt, iovoff,
                               new_ehdr, sizeof(*new_ehdr));

    if (copied < sizeof(*new_ehdr)) {
        return 0;
    }

    if (be16_to_cpu(new_ehdr->h_proto) == vet) {
        copied = iov_to_buf(iov, iovcnt, iovoff + sizeof(*new_ehdr),
                            &vlan_hdr, sizeof(vlan_hdr));

        if (copied < sizeof(vlan_hdr)) {
            return 0;
        }

        new_ehdr->h_proto = vlan_hdr.h_proto;

        *tci = be16_to_cpu(vlan_hdr.h_tci);
        *payload_offset = iovoff + sizeof(*new_ehdr) + sizeof(vlan_hdr);
        return sizeof(struct eth_header);
    }

    return 0;
}

void
eth_setup_ip4_fragmentation(const void *l2hdr, size_t l2hdr_len,
                            void *l3hdr, size_t l3hdr_len,
                            size_t l3payload_len,
                            size_t frag_offset, bool more_frags)
{
    const struct iovec l2vec = {
        .iov_base = (void *) l2hdr,
        .iov_len = l2hdr_len
    };

    if (eth_get_l3_proto(&l2vec, 1, l2hdr_len) == ETH_P_IP) {
        uint16_t orig_flags;
        struct ip_header *iphdr = (struct ip_header *) l3hdr;
        uint16_t frag_off_units = frag_offset / IP_FRAG_UNIT_SIZE;
        uint16_t new_ip_off;

        assert(frag_offset % IP_FRAG_UNIT_SIZE == 0);
        assert((frag_off_units & ~IP_OFFMASK) == 0);

        orig_flags = be16_to_cpu(iphdr->ip_off) & ~(IP_OFFMASK|IP_MF);
        new_ip_off = frag_off_units | orig_flags  | (more_frags ? IP_MF : 0);
        iphdr->ip_off = cpu_to_be16(new_ip_off);
        iphdr->ip_len = cpu_to_be16(l3payload_len + l3hdr_len);
    }
}

void
eth_fix_ip4_checksum(void *l3hdr, size_t l3hdr_len)
{
    struct ip_header *iphdr = (struct ip_header *) l3hdr;
    iphdr->ip_sum = 0;
    iphdr->ip_sum = cpu_to_be16(net_raw_checksum(l3hdr, l3hdr_len));
}

uint32_t
eth_calc_ip4_pseudo_hdr_csum(struct ip_header *iphdr,
                             uint16_t csl,
                             uint32_t *cso)
{
    struct ip_pseudo_header ipph;
    ipph.ip_src = iphdr->ip_src;
    ipph.ip_dst = iphdr->ip_dst;
    ipph.ip_payload = cpu_to_be16(csl);
    ipph.ip_proto = iphdr->ip_p;
    ipph.zeros = 0;
    *cso = sizeof(ipph);
    return net_checksum_add(*cso, (uint8_t *) &ipph);
}

uint32_t
eth_calc_ip6_pseudo_hdr_csum(struct ip6_header *iphdr,
                             uint16_t csl,
                             uint8_t l4_proto,
                             uint32_t *cso)
{
    struct ip6_pseudo_header ipph;
    ipph.ip6_src = iphdr->ip6_src;
    ipph.ip6_dst = iphdr->ip6_dst;
    ipph.len = cpu_to_be16(csl);
    ipph.zero[0] = 0;
    ipph.zero[1] = 0;
    ipph.zero[2] = 0;
    ipph.next_hdr = l4_proto;
    *cso = sizeof(ipph);
    return net_checksum_add(*cso, (uint8_t *)&ipph);
}

static bool
eth_is_ip6_extension_header_type(uint8_t hdr_type)
{
    switch (hdr_type) {
    case IP6_HOP_BY_HOP:
    case IP6_ROUTING:
    case IP6_FRAGMENT:
    case IP6_ESP:
    case IP6_AUTHENTICATION:
    case IP6_DESTINATON:
    case IP6_MOBILITY:
        return true;
    default:
        return false;
    }
}

static bool
_eth_get_rss_ex_dst_addr(const struct iovec *pkt, int pkt_frags,
                        size_t rthdr_offset,
                        struct ip6_ext_hdr *ext_hdr,
                        struct in6_address *dst_addr)
{
    struct ip6_ext_hdr_routing *rthdr = (struct ip6_ext_hdr_routing *) ext_hdr;

    if ((rthdr->rtype == 2) &&
        (rthdr->len == sizeof(struct in6_address) / 8) &&
        (rthdr->segleft == 1)) {

        size_t input_size = iov_size(pkt, pkt_frags);
        size_t bytes_read;

        if (input_size < rthdr_offset + sizeof(*ext_hdr)) {
            return false;
        }

        bytes_read = iov_to_buf(pkt, pkt_frags,
                                rthdr_offset + sizeof(*ext_hdr),
                                dst_addr, sizeof(*dst_addr));

        return bytes_read == sizeof(*dst_addr);
    }

    return false;
}

static bool
_eth_get_rss_ex_src_addr(const struct iovec *pkt, int pkt_frags,
                        size_t dsthdr_offset,
                        struct ip6_ext_hdr *ext_hdr,
                        struct in6_address *src_addr)
{
    size_t bytes_left = (ext_hdr->ip6r_len + 1) * 8 - sizeof(*ext_hdr);
    struct ip6_option_hdr opthdr;
    size_t opt_offset = dsthdr_offset + sizeof(*ext_hdr);

    while (bytes_left > sizeof(opthdr)) {
        size_t input_size = iov_size(pkt, pkt_frags);
        size_t bytes_read, optlen;

        if (input_size < opt_offset) {
            return false;
        }

        bytes_read = iov_to_buf(pkt, pkt_frags, opt_offset,
                                &opthdr, sizeof(opthdr));

        if (bytes_read != sizeof(opthdr)) {
            return false;
        }

        optlen = (opthdr.type == IP6_OPT_PAD1) ? 1
                                               : (opthdr.len + sizeof(opthdr));

        if (optlen > bytes_left) {
            return false;
        }

        if (opthdr.type == IP6_OPT_HOME) {
            size_t input_size = iov_size(pkt, pkt_frags);

            if (input_size < opt_offset + sizeof(opthdr)) {
                return false;
            }

            bytes_read = iov_to_buf(pkt, pkt_frags,
                                    opt_offset + sizeof(opthdr),
                                    src_addr, sizeof(*src_addr));

            return bytes_read == sizeof(*src_addr);
        }

        opt_offset += optlen;
        bytes_left -= optlen;
    }

    return false;
}

bool eth_parse_ipv6_hdr(const struct iovec *pkt, int pkt_frags,
                        size_t ip6hdr_off, eth_ip6_hdr_info *info)
{
    struct ip6_ext_hdr ext_hdr;
    size_t bytes_read;
    uint8_t curr_ext_hdr_type;
    size_t input_size = iov_size(pkt, pkt_frags);

    info->rss_ex_dst_valid = false;
    info->rss_ex_src_valid = false;
    info->fragment = false;

    if (input_size < ip6hdr_off) {
        return false;
    }

    bytes_read = iov_to_buf(pkt, pkt_frags, ip6hdr_off,
                            &info->ip6_hdr, sizeof(info->ip6_hdr));
    if (bytes_read < sizeof(info->ip6_hdr)) {
        return false;
    }

    info->full_hdr_len = sizeof(struct ip6_header);

    curr_ext_hdr_type = info->ip6_hdr.ip6_nxt;

    if (!eth_is_ip6_extension_header_type(curr_ext_hdr_type)) {
        info->l4proto = info->ip6_hdr.ip6_nxt;
        info->has_ext_hdrs = false;
        return true;
    }

    info->has_ext_hdrs = true;

    do {
        if (input_size < ip6hdr_off + info->full_hdr_len) {
            return false;
        }

        bytes_read = iov_to_buf(pkt, pkt_frags, ip6hdr_off + info->full_hdr_len,
                                &ext_hdr, sizeof(ext_hdr));

        if (bytes_read < sizeof(ext_hdr)) {
            return false;
        }

        if (curr_ext_hdr_type == IP6_ROUTING) {
            info->rss_ex_dst_valid =
                _eth_get_rss_ex_dst_addr(pkt, pkt_frags,
                                         ip6hdr_off + info->full_hdr_len,
                                         &ext_hdr, &info->rss_ex_dst);
        } else if (curr_ext_hdr_type == IP6_DESTINATON) {
            info->rss_ex_src_valid =
                _eth_get_rss_ex_src_addr(pkt, pkt_frags,
                                         ip6hdr_off + info->full_hdr_len,
                                         &ext_hdr, &info->rss_ex_src);
        } else if (curr_ext_hdr_type == IP6_FRAGMENT) {
            info->fragment = true;
        }

        info->full_hdr_len += (ext_hdr.ip6r_len + 1) * IP6_EXT_GRANULARITY;
        curr_ext_hdr_type = ext_hdr.ip6r_nxt;
    } while (eth_is_ip6_extension_header_type(curr_ext_hdr_type));

    info->l4proto = ext_hdr.ip6r_nxt;
    return true;
}
Commit	Line	Data
75020a70 DF	1	/*
	2	* QEMU network structures definitions and helper functions
	3	*
	4	* Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
	5	*
	6	* Developed by Daynix Computing LTD (http://www.daynix.com)
	7	*
	8	* Authors:
	9	* Dmitry Fleytman <[email protected]>
	10	* Tamir Shomer <[email protected]>
	11	* Yan Vugenfirer <[email protected]>
	12	*
	13	* This work is licensed under the terms of the GNU GPL, version 2 or later.
	14	* See the COPYING file in the top-level directory.
	15	*
	16	*/
	17
2744d920	18	#include "qemu/osdep.h"
75020a70 DF	19	#include "net/eth.h"
75020a70 DF	20	#include "net/checksum.h"
75020a70 DF	21	#include "net/tap.h"
75020a70 DF	22
eb700029 DF	23	void eth_setup_vlan_headers_ex(struct eth_header *ehdr, uint16_t vlan_tag,
eb700029 DF	24	uint16_t vlan_ethtype, bool *is_new)
75020a70 DF	25	{
	26	struct vlan_header *vhdr = PKT_GET_VLAN_HDR(ehdr);
	27
	28	switch (be16_to_cpu(ehdr->h_proto)) {
	29	case ETH_P_VLAN:
	30	case ETH_P_DVLAN:
	31	/* vlan hdr exists */
	32	*is_new = false;
	33	break;
	34
	35	default:
	36	/* No VLAN header, put a new one */
	37	vhdr->h_proto = ehdr->h_proto;
eb700029	38	ehdr->h_proto = cpu_to_be16(vlan_ethtype);
75020a70 DF	39	*is_new = true;
	40	break;
	41	}
	42	vhdr->h_tci = cpu_to_be16(vlan_tag);
	43	}
	44
	45	uint8_t
	46	eth_get_gso_type(uint16_t l3_proto, uint8_t *l3_hdr, uint8_t l4proto)
	47	{
	48	uint8_t ecn_state = 0;
	49
	50	if (l3_proto == ETH_P_IP) {
	51	struct ip_header iphdr = (struct ip_header ) l3_hdr;
	52
	53	if (IP_HEADER_VERSION(iphdr) == IP_HEADER_VERSION_4) {
	54	if (IPTOS_ECN(iphdr->ip_tos) == IPTOS_ECN_CE) {
	55	ecn_state = VIRTIO_NET_HDR_GSO_ECN;
	56	}
	57	if (l4proto == IP_PROTO_TCP) {
	58	return VIRTIO_NET_HDR_GSO_TCPV4 \| ecn_state;
	59	} else if (l4proto == IP_PROTO_UDP) {
	60	return VIRTIO_NET_HDR_GSO_UDP \| ecn_state;
	61	}
	62	}
	63	} else if (l3_proto == ETH_P_IPV6) {
	64	struct ip6_header ip6hdr = (struct ip6_header ) l3_hdr;
	65
	66	if (IP6_ECN(ip6hdr->ip6_ecn_acc) == IP6_ECN_CE) {
	67	ecn_state = VIRTIO_NET_HDR_GSO_ECN;
	68	}
	69
	70	if (l4proto == IP_PROTO_TCP) {
	71	return VIRTIO_NET_HDR_GSO_TCPV6 \| ecn_state;
	72	}
	73	}
	74
	75	/* Unsupported offload */
dfc6f865	76	g_assert_not_reached();
75020a70 DF	77
	78	return VIRTIO_NET_HDR_GSO_NONE \| ecn_state;
	79	}
	80
eb700029 DF	81	uint16_t
	82	eth_get_l3_proto(const struct iovec *l2hdr_iov, int iovcnt, size_t l2hdr_len)
	83	{
	84	uint16_t proto;
	85	size_t copied;
	86	size_t size = iov_size(l2hdr_iov, iovcnt);
	87	size_t proto_offset = l2hdr_len - sizeof(proto);
	88
	89	if (size < proto_offset) {
	90	return ETH_P_UNKNOWN;
	91	}
	92
	93	copied = iov_to_buf(l2hdr_iov, iovcnt, proto_offset,
	94	&proto, sizeof(proto));
	95
	96	return (copied == sizeof(proto)) ? be16_to_cpu(proto) : ETH_P_UNKNOWN;
	97	}
	98
	99	static bool
	100	_eth_copy_chunk(size_t input_size,
	101	const struct iovec *iov, int iovcnt,
	102	size_t offset, size_t length,
	103	void *buffer)
	104	{
	105	size_t copied;
	106
	107	if (input_size < offset) {
	108	return false;
	109	}
	110
	111	copied = iov_to_buf(iov, iovcnt, offset, buffer, length);
	112
	113	if (copied < length) {
	114	return false;
	115	}
	116
	117	return true;
	118	}
	119
	120	static bool
	121	_eth_tcp_has_data(bool is_ip4,
	122	const struct ip_header *ip4_hdr,
	123	const struct ip6_header *ip6_hdr,
	124	size_t full_ip6hdr_len,
	125	const struct tcp_header *tcp)
	126	{
	127	uint32_t l4len;
	128
	129	if (is_ip4) {
	130	l4len = be16_to_cpu(ip4_hdr->ip_len) - IP_HDR_GET_LEN(ip4_hdr);
	131	} else {
	132	size_t opts_len = full_ip6hdr_len - sizeof(struct ip6_header);
	133	l4len = be16_to_cpu(ip6_hdr->ip6_ctlun.ip6_un1.ip6_un1_plen) - opts_len;
	134	}
	135
	136	return l4len > TCP_HEADER_DATA_OFFSET(tcp);
	137	}
	138
	139	void eth_get_protocols(const struct iovec *iov, int iovcnt,
75020a70	140	bool isip4, bool isip6,
eb700029 DF	141	bool isudp, bool istcp,
	142	size_t *l3hdr_off,
	143	size_t *l4hdr_off,
	144	size_t *l5hdr_off,
	145	eth_ip6_hdr_info *ip6hdr_info,
	146	eth_ip4_hdr_info *ip4hdr_info,
	147	eth_l4_hdr_info *l4hdr_info)
75020a70 DF	148	{
75020a70 DF	149	int proto;
eb700029 DF	150	bool fragment = false;
	151	size_t l2hdr_len = eth_get_l2_hdr_length_iov(iov, iovcnt);
	152	size_t input_size = iov_size(iov, iovcnt);
	153	size_t copied;
	154
75020a70 DF	155	isip4 = isip6 = isudp = istcp = false;
75020a70 DF	156
eb700029 DF	157	proto = eth_get_l3_proto(iov, iovcnt, l2hdr_len);
	158
	159	*l3hdr_off = l2hdr_len;
	160
75020a70	161	if (proto == ETH_P_IP) {
eb700029	162	struct ip_header *iphdr = &ip4hdr_info->ip4_hdr;
75020a70	163
eb700029 DF	164	if (input_size < l2hdr_len) {
	165	return;
	166	}
	167
	168	copied = iov_to_buf(iov, iovcnt, l2hdr_len, iphdr, sizeof(*iphdr));
75020a70	169
eb700029	170	*isip4 = true;
75020a70	171
eb700029 DF	172	if (copied < sizeof(*iphdr)) {
	173	return;
	174	}
75020a70 DF	175
	176	if (IP_HEADER_VERSION(iphdr) == IP_HEADER_VERSION_4) {
	177	if (iphdr->ip_p == IP_PROTO_TCP) {
	178	*istcp = true;
	179	} else if (iphdr->ip_p == IP_PROTO_UDP) {
	180	*isudp = true;
	181	}
	182	}
75020a70	183
eb700029 DF	184	ip4hdr_info->fragment = IP4_IS_FRAGMENT(iphdr);
	185	*l4hdr_off = l2hdr_len + IP_HDR_GET_LEN(iphdr);
	186
	187	fragment = ip4hdr_info->fragment;
	188	} else if (proto == ETH_P_IPV6) {
75020a70 DF	189
75020a70 DF	190	*isip6 = true;
eb700029 DF	191	if (eth_parse_ipv6_hdr(iov, iovcnt, l2hdr_len,
	192	ip6hdr_info)) {
	193	if (ip6hdr_info->l4proto == IP_PROTO_TCP) {
75020a70	194	*istcp = true;
eb700029	195	} else if (ip6hdr_info->l4proto == IP_PROTO_UDP) {
75020a70 DF	196	*isudp = true;
75020a70 DF	197	}
eb700029 DF	198	} else {
	199	return;
	200	}
	201
	202	*l4hdr_off = l2hdr_len + ip6hdr_info->full_hdr_len;
	203	fragment = ip6hdr_info->fragment;
	204	}
	205
	206	if (!fragment) {
	207	if (*istcp) {
	208	*istcp = _eth_copy_chunk(input_size,
	209	iov, iovcnt,
	210	*l4hdr_off, sizeof(l4hdr_info->hdr.tcp),
	211	&l4hdr_info->hdr.tcp);
	212
2c5e564f	213	if (*istcp) {
eb700029 DF	214	l5hdr_off = l4hdr_off +
	215	TCP_HEADER_DATA_OFFSET(&l4hdr_info->hdr.tcp);
	216
	217	l4hdr_info->has_tcp_data =
	218	_eth_tcp_has_data(proto == ETH_P_IP,
	219	&ip4hdr_info->ip4_hdr,
	220	&ip6hdr_info->ip6_hdr,
	221	l4hdr_off - l3hdr_off,
	222	&l4hdr_info->hdr.tcp);
	223	}
	224	} else if (*isudp) {
	225	*isudp = _eth_copy_chunk(input_size,
	226	iov, iovcnt,
	227	*l4hdr_off, sizeof(l4hdr_info->hdr.udp),
	228	&l4hdr_info->hdr.udp);
	229	l5hdr_off = l4hdr_off + sizeof(l4hdr_info->hdr.udp);
	230	}
	231	}
	232	}
	233
566342c3	234	size_t
eb700029 DF	235	eth_strip_vlan(const struct iovec *iov, int iovcnt, size_t iovoff,
	236	uint8_t *new_ehdr_buf,
	237	uint16_t payload_offset, uint16_t tci)
	238	{
	239	struct vlan_header vlan_hdr;
	240	struct eth_header new_ehdr = (struct eth_header ) new_ehdr_buf;
	241
	242	size_t copied = iov_to_buf(iov, iovcnt, iovoff,
	243	new_ehdr, sizeof(*new_ehdr));
	244
	245	if (copied < sizeof(*new_ehdr)) {
566342c3	246	return 0;
eb700029 DF	247	}
	248
	249	switch (be16_to_cpu(new_ehdr->h_proto)) {
	250	case ETH_P_VLAN:
	251	case ETH_P_DVLAN:
	252	copied = iov_to_buf(iov, iovcnt, iovoff + sizeof(*new_ehdr),
	253	&vlan_hdr, sizeof(vlan_hdr));
	254
	255	if (copied < sizeof(vlan_hdr)) {
566342c3	256	return 0;
eb700029 DF	257	}
	258
	259	new_ehdr->h_proto = vlan_hdr.h_proto;
	260
	261	*tci = be16_to_cpu(vlan_hdr.h_tci);
	262	payload_offset = iovoff + sizeof(new_ehdr) + sizeof(vlan_hdr);
	263
	264	if (be16_to_cpu(new_ehdr->h_proto) == ETH_P_VLAN) {
	265
	266	copied = iov_to_buf(iov, iovcnt, *payload_offset,
	267	PKT_GET_VLAN_HDR(new_ehdr), sizeof(vlan_hdr));
	268
	269	if (copied < sizeof(vlan_hdr)) {
566342c3	270	return 0;
eb700029 DF	271	}
	272
	273	*payload_offset += sizeof(vlan_hdr);
566342c3 DF	274
	275	return sizeof(struct eth_header) + sizeof(struct vlan_header);
	276	} else {
	277	return sizeof(struct eth_header);
eb700029	278	}
eb700029	279	default:
566342c3	280	return 0;
eb700029 DF	281	}
	282	}
	283
566342c3	284	size_t
eb700029 DF	285	eth_strip_vlan_ex(const struct iovec *iov, int iovcnt, size_t iovoff,
	286	uint16_t vet, uint8_t *new_ehdr_buf,
	287	uint16_t payload_offset, uint16_t tci)
	288	{
	289	struct vlan_header vlan_hdr;
	290	struct eth_header new_ehdr = (struct eth_header ) new_ehdr_buf;
	291
	292	size_t copied = iov_to_buf(iov, iovcnt, iovoff,
	293	new_ehdr, sizeof(*new_ehdr));
	294
	295	if (copied < sizeof(*new_ehdr)) {
566342c3	296	return 0;
eb700029 DF	297	}
	298
	299	if (be16_to_cpu(new_ehdr->h_proto) == vet) {
	300	copied = iov_to_buf(iov, iovcnt, iovoff + sizeof(*new_ehdr),
	301	&vlan_hdr, sizeof(vlan_hdr));
	302
	303	if (copied < sizeof(vlan_hdr)) {
566342c3	304	return 0;
75020a70	305	}
eb700029 DF	306
	307	new_ehdr->h_proto = vlan_hdr.h_proto;
	308
	309	*tci = be16_to_cpu(vlan_hdr.h_tci);
	310	payload_offset = iovoff + sizeof(new_ehdr) + sizeof(vlan_hdr);
566342c3	311	return sizeof(struct eth_header);
75020a70	312	}
eb700029	313
566342c3	314	return 0;
75020a70 DF	315	}
	316
	317	void
	318	eth_setup_ip4_fragmentation(const void *l2hdr, size_t l2hdr_len,
	319	void *l3hdr, size_t l3hdr_len,
	320	size_t l3payload_len,
	321	size_t frag_offset, bool more_frags)
	322	{
eb700029 DF	323	const struct iovec l2vec = {
	324	.iov_base = (void *) l2hdr,
	325	.iov_len = l2hdr_len
	326	};
	327
	328	if (eth_get_l3_proto(&l2vec, 1, l2hdr_len) == ETH_P_IP) {
75020a70 DF	329	uint16_t orig_flags;
	330	struct ip_header iphdr = (struct ip_header ) l3hdr;
	331	uint16_t frag_off_units = frag_offset / IP_FRAG_UNIT_SIZE;
	332	uint16_t new_ip_off;
	333
	334	assert(frag_offset % IP_FRAG_UNIT_SIZE == 0);
	335	assert((frag_off_units & ~IP_OFFMASK) == 0);
	336
	337	orig_flags = be16_to_cpu(iphdr->ip_off) & ~(IP_OFFMASK\|IP_MF);
	338	new_ip_off = frag_off_units \| orig_flags \| (more_frags ? IP_MF : 0);
	339	iphdr->ip_off = cpu_to_be16(new_ip_off);
	340	iphdr->ip_len = cpu_to_be16(l3payload_len + l3hdr_len);
	341	}
	342	}
	343
	344	void
	345	eth_fix_ip4_checksum(void *l3hdr, size_t l3hdr_len)
	346	{
	347	struct ip_header iphdr = (struct ip_header ) l3hdr;
	348	iphdr->ip_sum = 0;
	349	iphdr->ip_sum = cpu_to_be16(net_raw_checksum(l3hdr, l3hdr_len));
	350	}
	351
	352	uint32_t
eb700029 DF	353	eth_calc_ip4_pseudo_hdr_csum(struct ip_header *iphdr,
	354	uint16_t csl,
	355	uint32_t *cso)
75020a70 DF	356	{
	357	struct ip_pseudo_header ipph;
	358	ipph.ip_src = iphdr->ip_src;
	359	ipph.ip_dst = iphdr->ip_dst;
	360	ipph.ip_payload = cpu_to_be16(csl);
	361	ipph.ip_proto = iphdr->ip_p;
	362	ipph.zeros = 0;
eb700029 DF	363	*cso = sizeof(ipph);
	364	return net_checksum_add(cso, (uint8_t ) &ipph);
	365	}
	366
	367	uint32_t
	368	eth_calc_ip6_pseudo_hdr_csum(struct ip6_header *iphdr,
	369	uint16_t csl,
	370	uint8_t l4_proto,
	371	uint32_t *cso)
	372	{
	373	struct ip6_pseudo_header ipph;
	374	ipph.ip6_src = iphdr->ip6_src;
	375	ipph.ip6_dst = iphdr->ip6_dst;
	376	ipph.len = cpu_to_be16(csl);
	377	ipph.zero[0] = 0;
	378	ipph.zero[1] = 0;
	379	ipph.zero[2] = 0;
	380	ipph.next_hdr = l4_proto;
	381	*cso = sizeof(ipph);
	382	return net_checksum_add(cso, (uint8_t )&ipph);
75020a70 DF	383	}
	384
	385	static bool
	386	eth_is_ip6_extension_header_type(uint8_t hdr_type)
	387	{
	388	switch (hdr_type) {
	389	case IP6_HOP_BY_HOP:
	390	case IP6_ROUTING:
	391	case IP6_FRAGMENT:
	392	case IP6_ESP:
	393	case IP6_AUTHENTICATION:
	394	case IP6_DESTINATON:
	395	case IP6_MOBILITY:
	396	return true;
	397	default:
	398	return false;
	399	}
	400	}
	401
eb700029 DF	402	static bool
	403	_eth_get_rss_ex_dst_addr(const struct iovec *pkt, int pkt_frags,
	404	size_t rthdr_offset,
	405	struct ip6_ext_hdr *ext_hdr,
	406	struct in6_address *dst_addr)
	407	{
	408	struct ip6_ext_hdr_routing rthdr = (struct ip6_ext_hdr_routing ) ext_hdr;
	409
	410	if ((rthdr->rtype == 2) &&
	411	(rthdr->len == sizeof(struct in6_address) / 8) &&
	412	(rthdr->segleft == 1)) {
	413
	414	size_t input_size = iov_size(pkt, pkt_frags);
	415	size_t bytes_read;
	416
	417	if (input_size < rthdr_offset + sizeof(*ext_hdr)) {
	418	return false;
	419	}
	420
	421	bytes_read = iov_to_buf(pkt, pkt_frags,
	422	rthdr_offset + sizeof(*ext_hdr),
4555ca68	423	dst_addr, sizeof(*dst_addr));
eb700029	424
b2caa3b8	425	return bytes_read == sizeof(*dst_addr);
eb700029 DF	426	}
	427
	428	return false;
	429	}
	430
	431	static bool
	432	_eth_get_rss_ex_src_addr(const struct iovec *pkt, int pkt_frags,
	433	size_t dsthdr_offset,
	434	struct ip6_ext_hdr *ext_hdr,
	435	struct in6_address *src_addr)
	436	{
	437	size_t bytes_left = (ext_hdr->ip6r_len + 1) * 8 - sizeof(*ext_hdr);
	438	struct ip6_option_hdr opthdr;
	439	size_t opt_offset = dsthdr_offset + sizeof(*ext_hdr);
	440
	441	while (bytes_left > sizeof(opthdr)) {
	442	size_t input_size = iov_size(pkt, pkt_frags);
	443	size_t bytes_read, optlen;
	444
	445	if (input_size < opt_offset) {
	446	return false;
	447	}
	448
	449	bytes_read = iov_to_buf(pkt, pkt_frags, opt_offset,
	450	&opthdr, sizeof(opthdr));
	451
	452	if (bytes_read != sizeof(opthdr)) {
	453	return false;
	454	}
	455
	456	optlen = (opthdr.type == IP6_OPT_PAD1) ? 1
	457	: (opthdr.len + sizeof(opthdr));
	458
	459	if (optlen > bytes_left) {
	460	return false;
	461	}
	462
	463	if (opthdr.type == IP6_OPT_HOME) {
	464	size_t input_size = iov_size(pkt, pkt_frags);
	465
	466	if (input_size < opt_offset + sizeof(opthdr)) {
	467	return false;
	468	}
	469
	470	bytes_read = iov_to_buf(pkt, pkt_frags,
	471	opt_offset + sizeof(opthdr),
4555ca68	472	src_addr, sizeof(*src_addr));
eb700029	473
b2caa3b8	474	return bytes_read == sizeof(*src_addr);
eb700029 DF	475	}
	476
	477	opt_offset += optlen;
	478	bytes_left -= optlen;
	479	}
	480
	481	return false;
	482	}
	483
	484	bool eth_parse_ipv6_hdr(const struct iovec *pkt, int pkt_frags,
	485	size_t ip6hdr_off, eth_ip6_hdr_info *info)
75020a70	486	{
75020a70 DF	487	struct ip6_ext_hdr ext_hdr;
75020a70 DF	488	size_t bytes_read;
eb700029 DF	489	uint8_t curr_ext_hdr_type;
	490	size_t input_size = iov_size(pkt, pkt_frags);
	491
	492	info->rss_ex_dst_valid = false;
	493	info->rss_ex_src_valid = false;
	494	info->fragment = false;
	495
	496	if (input_size < ip6hdr_off) {
	497	return false;
	498	}
75020a70 DF	499
75020a70 DF	500	bytes_read = iov_to_buf(pkt, pkt_frags, ip6hdr_off,
eb700029 DF	501	&info->ip6_hdr, sizeof(info->ip6_hdr));
eb700029 DF	502	if (bytes_read < sizeof(info->ip6_hdr)) {
75020a70 DF	503	return false;
	504	}
	505
eb700029 DF	506	info->full_hdr_len = sizeof(struct ip6_header);
	507
	508	curr_ext_hdr_type = info->ip6_hdr.ip6_nxt;
75020a70	509
eb700029 DF	510	if (!eth_is_ip6_extension_header_type(curr_ext_hdr_type)) {
	511	info->l4proto = info->ip6_hdr.ip6_nxt;
	512	info->has_ext_hdrs = false;
75020a70 DF	513	return true;
	514	}
	515
eb700029 DF	516	info->has_ext_hdrs = true;
eb700029 DF	517
75020a70	518	do {
eb700029 DF	519	if (input_size < ip6hdr_off + info->full_hdr_len) {
	520	return false;
	521	}
	522
	523	bytes_read = iov_to_buf(pkt, pkt_frags, ip6hdr_off + info->full_hdr_len,
75020a70	524	&ext_hdr, sizeof(ext_hdr));
75020a70	525
eb700029 DF	526	if (bytes_read < sizeof(ext_hdr)) {
	527	return false;
	528	}
	529
	530	if (curr_ext_hdr_type == IP6_ROUTING) {
	531	info->rss_ex_dst_valid =
	532	_eth_get_rss_ex_dst_addr(pkt, pkt_frags,
	533	ip6hdr_off + info->full_hdr_len,
	534	&ext_hdr, &info->rss_ex_dst);
	535	} else if (curr_ext_hdr_type == IP6_DESTINATON) {
	536	info->rss_ex_src_valid =
	537	_eth_get_rss_ex_src_addr(pkt, pkt_frags,
	538	ip6hdr_off + info->full_hdr_len,
	539	&ext_hdr, &info->rss_ex_src);
	540	} else if (curr_ext_hdr_type == IP6_FRAGMENT) {
	541	info->fragment = true;
	542	}
	543
	544	info->full_hdr_len += (ext_hdr.ip6r_len + 1) * IP6_EXT_GRANULARITY;
	545	curr_ext_hdr_type = ext_hdr.ip6r_nxt;
	546	} while (eth_is_ip6_extension_header_type(curr_ext_hdr_type));
	547
	548	info->l4proto = ext_hdr.ip6r_nxt;
75020a70 DF	549	return true;
75020a70 DF	550	}