X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fflow.c;h=460771ccb1fb31e3aa579a4e5185a89a7db1a7e7;hb=b53055f4da4f8de17c591ae07ddd782829d21cd8;hp=6928f74eca1c29c718ba198366a4be45f26375a4;hpb=46aef12ee5fcb028d20bac654c8a08cf993cfac8;p=sliver-openvswitch.git diff --git a/lib/flow.c b/lib/flow.c index 6928f74ec..460771ccb 100644 --- a/lib/flow.c +++ b/lib/flow.c @@ -16,6 +16,7 @@ #include #include #include "flow.h" +#include #include #include #include @@ -134,19 +135,16 @@ parse_ethertype(struct ofpbuf *b) static int parse_ipv6(struct ofpbuf *packet, struct flow *flow) { - struct ip6_hdr *nh; - int nh_len = sizeof(struct ip6_hdr); - int payload_len; + const struct ip6_hdr *nh; ovs_be32 tc_flow; int nexthdr; - if (packet->size < sizeof *nh) { - return -EINVAL; + nh = ofpbuf_try_pull(packet, sizeof *nh); + if (!nh) { + return EINVAL; } - nh = packet->data; nexthdr = nh->ip6_nxt; - payload_len = ntohs(nh->ip6_plen); flow->ipv6_src = nh->ip6_src; flow->ipv6_dst = nh->ip6_dst; @@ -155,15 +153,6 @@ parse_ipv6(struct ofpbuf *packet, struct flow *flow) flow->nw_tos = (ntohl(tc_flow) >> 4) & IP_DSCP_MASK; flow->nw_proto = IPPROTO_NONE; - /* We don't process jumbograms. */ - if (!payload_len) { - return -EINVAL; - } - - if (packet->size < sizeof *nh + payload_len) { - return -EINVAL; - } - while (1) { if ((nexthdr != IPPROTO_HOPOPTS) && (nexthdr != IPPROTO_ROUTING) @@ -179,9 +168,10 @@ parse_ipv6(struct ofpbuf *packet, struct flow *flow) /* We only verify that at least 8 bytes of the next header are * available, but many of these headers are longer. Ensure that * accesses within the extension header are within those first 8 + * bytes. All extension headers are required to be at least 8 * bytes. */ - if (packet->size < nh_len + 8) { - return -EINVAL; + if (packet->size < 8) { + return EINVAL; } if ((nexthdr == IPPROTO_HOPOPTS) @@ -189,28 +179,28 @@ parse_ipv6(struct ofpbuf *packet, struct flow *flow) || (nexthdr == IPPROTO_DSTOPTS)) { /* These headers, while different, have the fields we care about * in the same location and with the same interpretation. */ - struct ip6_ext *ext_hdr; - - ext_hdr = (struct ip6_ext *)((char *)packet->data + nh_len); + const struct ip6_ext *ext_hdr = (struct ip6_ext *)packet->data; nexthdr = ext_hdr->ip6e_nxt; - nh_len += (ext_hdr->ip6e_len + 1) * 8; + if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 1) * 8)) { + return EINVAL; + } } else if (nexthdr == IPPROTO_AH) { /* A standard AH definition isn't available, but the fields * we care about are in the same location as the generic * option header--only the header length is calculated * differently. */ - struct ip6_ext *ext_hdr; - - ext_hdr = (struct ip6_ext *)((char *)packet->data + nh_len); + const struct ip6_ext *ext_hdr = (struct ip6_ext *)packet->data; nexthdr = ext_hdr->ip6e_nxt; - nh_len += (ext_hdr->ip6e_len + 2) * 4; + if (!ofpbuf_try_pull(packet, (ext_hdr->ip6e_len + 2) * 4)) { + return EINVAL; + } } else if (nexthdr == IPPROTO_FRAGMENT) { - struct ip6_frag *frag_hdr; - - frag_hdr = (struct ip6_frag *)((char *)packet->data + nh_len); + const struct ip6_frag *frag_hdr = (struct ip6_frag *)packet->data; nexthdr = frag_hdr->ip6f_nxt; - nh_len += sizeof *frag_hdr; + if (!ofpbuf_try_pull(packet, sizeof *frag_hdr)) { + return EINVAL; + } /* We only process the first fragment. */ if ((frag_hdr->ip6f_offlg & IP6F_OFF_MASK) != htons(0)) { @@ -220,24 +210,34 @@ parse_ipv6(struct ofpbuf *packet, struct flow *flow) } } - /* The payload length claims to be smaller than the size of the - * headers we've already processed. */ - if (payload_len < nh_len - sizeof *nh) { - return -EINVAL; - } - flow->nw_proto = nexthdr; - return nh_len; + return 0; } -/* Neighbor Discovery Solicitation and Advertisement messages are - * identical in structure, so we'll just use one of them. To be safe, - * we'll assert that they're still identical. */ -BUILD_ASSERT_DECL(sizeof(struct nd_neighbor_solicit) - == sizeof(struct nd_neighbor_advert)); +static void +parse_tcp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow) +{ + const struct tcp_header *tcp = pull_tcp(b); + if (tcp) { + flow->tp_src = tcp->tcp_src; + flow->tp_dst = tcp->tcp_dst; + packet->l7 = b->data; + } +} + +static void +parse_udp(struct ofpbuf *packet, struct ofpbuf *b, struct flow *flow) +{ + const struct udp_header *udp = pull_udp(b); + if (udp) { + flow->tp_src = udp->udp_src; + flow->tp_dst = udp->udp_dst; + packet->l7 = b->data; + } +} static bool -parse_icmpv6(struct ofpbuf *b, struct flow *flow, int icmp_len) +parse_icmpv6(struct ofpbuf *b, struct flow *flow) { const struct icmp6_hdr *icmp = pull_icmpv6(b); @@ -250,34 +250,26 @@ parse_icmpv6(struct ofpbuf *b, struct flow *flow, int icmp_len) flow->icmp_type = htons(icmp->icmp6_type); flow->icmp_code = htons(icmp->icmp6_code); - if (!icmp->icmp6_code - && ((icmp->icmp6_type == ND_NEIGHBOR_SOLICIT) - || (icmp->icmp6_type == ND_NEIGHBOR_ADVERT))) { - struct nd_neighbor_solicit *nd_ns; /* Identical to ND advert */ + if (icmp->icmp6_code == 0 && + (icmp->icmp6_type == ND_NEIGHBOR_SOLICIT || + icmp->icmp6_type == ND_NEIGHBOR_ADVERT)) { + const struct in6_addr *nd_target; - /* In order to process neighbor discovery options, we need the - * entire packet. */ - if ((icmp_len < sizeof *nd_ns) - || (!ofpbuf_try_pull(b, sizeof *nd_ns - sizeof *icmp))) { + nd_target = ofpbuf_try_pull(b, sizeof *nd_target); + if (!nd_target) { return false; } - nd_ns = (struct nd_neighbor_solicit *)icmp; - flow->nd_target = nd_ns->nd_ns_target; - - icmp_len -= sizeof(*nd_ns); - while (icmp_len >= 8) { - struct nd_opt_hdr *nd_opt; - int opt_len; - const uint8_t *data; + flow->nd_target = *nd_target; + while (b->size >= 8) { /* The minimum size of an option is 8 bytes, which also is * the size of Ethernet link-layer options. */ - nd_opt = ofpbuf_pull(b, 8); - if (!nd_opt->nd_opt_len || nd_opt->nd_opt_len * 8 > icmp_len) { + const struct nd_opt_hdr *nd_opt = b->data; + int opt_len = nd_opt->nd_opt_len * 8; + + if (!opt_len || opt_len > b->size) { goto invalid; } - opt_len = nd_opt->nd_opt_len * 8; - data = (const uint8_t *)(nd_opt + 1); /* Store the link layer address if the appropriate option is * provided. It is considered an error if the same link @@ -285,40 +277,37 @@ parse_icmpv6(struct ofpbuf *b, struct flow *flow, int icmp_len) if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR && opt_len == 8) { if (eth_addr_is_zero(flow->arp_sha)) { - memcpy(flow->arp_sha, data, ETH_ADDR_LEN); + memcpy(flow->arp_sha, nd_opt + 1, ETH_ADDR_LEN); } else { goto invalid; } } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR && opt_len == 8) { if (eth_addr_is_zero(flow->arp_tha)) { - memcpy(flow->arp_tha, data, ETH_ADDR_LEN); + memcpy(flow->arp_tha, nd_opt + 1, ETH_ADDR_LEN); } else { goto invalid; } } - /* Pull the rest of this option. */ - if (!ofpbuf_try_pull(b, opt_len - 8)) { + if (!ofpbuf_try_pull(b, opt_len)) { goto invalid; } - - icmp_len -= opt_len; } } return true; invalid: - memset(&flow->nd_target, '\0', sizeof(flow->nd_target)); - memset(flow->arp_sha, '\0', sizeof(flow->arp_sha)); - memset(flow->arp_tha, '\0', sizeof(flow->arp_tha)); + memset(&flow->nd_target, 0, sizeof(flow->nd_target)); + memset(flow->arp_sha, 0, sizeof(flow->arp_sha)); + memset(flow->arp_tha, 0, sizeof(flow->arp_tha)); return false; } -/* Initializes 'flow' members from 'packet', 'tun_id', and 'in_port. +/* Initializes 'flow' members from 'packet', 'tun_id', and 'ofp_in_port'. * Initializes 'packet' header pointers as follows: * * - packet->l2 to the start of the Ethernet header. @@ -334,7 +323,7 @@ invalid: * present and has a correct length, and otherwise NULL. */ int -flow_extract(struct ofpbuf *packet, ovs_be64 tun_id, uint16_t in_port, +flow_extract(struct ofpbuf *packet, ovs_be64 tun_id, uint16_t ofp_in_port, struct flow *flow) { struct ofpbuf b = *packet; @@ -345,7 +334,7 @@ flow_extract(struct ofpbuf *packet, ovs_be64 tun_id, uint16_t in_port, memset(flow, 0, sizeof *flow); flow->tun_id = tun_id; - flow->in_port = in_port; + flow->in_port = ofp_in_port; packet->l2 = b.data; packet->l3 = NULL; @@ -380,19 +369,9 @@ flow_extract(struct ofpbuf *packet, ovs_be64 tun_id, uint16_t in_port, packet->l4 = b.data; if (!IP_IS_FRAGMENT(nh->ip_frag_off)) { if (flow->nw_proto == IPPROTO_TCP) { - const struct tcp_header *tcp = pull_tcp(&b); - if (tcp) { - flow->tp_src = tcp->tcp_src; - flow->tp_dst = tcp->tcp_dst; - packet->l7 = b.data; - } + parse_tcp(packet, &b, flow); } else if (flow->nw_proto == IPPROTO_UDP) { - const struct udp_header *udp = pull_udp(&b); - if (udp) { - flow->tp_src = udp->udp_src; - flow->tp_dst = udp->udp_dst; - packet->l7 = b.data; - } + parse_udp(packet, &b, flow); } else if (flow->nw_proto == IPPROTO_ICMP) { const struct icmp_header *icmp = pull_icmp(&b); if (icmp) { @@ -406,36 +385,20 @@ flow_extract(struct ofpbuf *packet, ovs_be64 tun_id, uint16_t in_port, } } } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) { - int nh_len; - const struct ip6_hdr *nh; - nh_len = parse_ipv6(&b, flow); - if (nh_len < 0) { + retval = parse_ipv6(&b, flow); + if (retval) { return 0; } - nh = ofpbuf_try_pull(&b, nh_len); - if (nh) { - packet->l4 = b.data; - if (flow->nw_proto == IPPROTO_TCP) { - const struct tcp_header *tcp = pull_tcp(&b); - if (tcp) { - flow->tp_src = tcp->tcp_src; - flow->tp_dst = tcp->tcp_dst; - packet->l7 = b.data; - } - } else if (flow->nw_proto == IPPROTO_UDP) { - const struct udp_header *udp = pull_udp(&b); - if (udp) { - flow->tp_src = udp->udp_src; - flow->tp_dst = udp->udp_dst; - packet->l7 = b.data; - } - } else if (flow->nw_proto == IPPROTO_ICMPV6) { - int icmp_len = ntohs(nh->ip6_plen) + sizeof *nh - nh_len; - if (parse_icmpv6(&b, flow, icmp_len)) { - packet->l7 = b.data; - } + packet->l4 = b.data; + if (flow->nw_proto == IPPROTO_TCP) { + parse_tcp(packet, &b, flow); + } else if (flow->nw_proto == IPPROTO_UDP) { + parse_udp(packet, &b, flow); + } else if (flow->nw_proto == IPPROTO_ICMPV6) { + if (parse_icmpv6(&b, flow)) { + packet->l7 = b.data; } } } else if (flow->dl_type == htons(ETH_TYPE_ARP)) { @@ -494,7 +457,7 @@ void flow_format(struct ds *ds, const struct flow *flow) { ds_put_format(ds, "tunnel%#"PRIx64":in_port%04"PRIx16":tci(", - flow->tun_id, flow->in_port); + ntohll(flow->tun_id), flow->in_port); if (flow->vlan_tci) { ds_put_format(ds, "vlan%"PRIu16",pcp%d", vlan_tci_to_vid(flow->vlan_tci), @@ -514,7 +477,7 @@ flow_format(struct ds *ds, const struct flow *flow) print_ipv6_addr(ds, &flow->ipv6_src); ds_put_cstr(ds, "->"); print_ipv6_addr(ds, &flow->ipv6_dst); - + } else { ds_put_format(ds, " proto%"PRIu8 " tos%"PRIu8 @@ -594,7 +557,7 @@ flow_wildcards_is_exact(const struct flow_wildcards *wc) } for (i = 0; i < FLOW_N_REGS; i++) { - if (wc->reg_masks[i] != htonl(UINT32_MAX)) { + if (wc->reg_masks[i] != UINT32_MAX) { return false; } } @@ -628,13 +591,13 @@ flow_wildcards_combine(struct flow_wildcards *dst, /* Returns a hash of the wildcards in 'wc'. */ uint32_t -flow_wildcards_hash(const struct flow_wildcards *wc) +flow_wildcards_hash(const struct flow_wildcards *wc, uint32_t basis) { /* If you change struct flow_wildcards and thereby trigger this * assertion, please check that the new struct flow_wildcards has no holes * in it before you update the assertion. */ BUILD_ASSERT_DECL(sizeof *wc == 56 + FLOW_N_REGS * 4); - return hash_bytes(wc, sizeof *wc, 0); + return hash_bytes(wc, sizeof *wc, basis); } /* Returns true if 'a' and 'b' represent the same wildcards, false if they are @@ -649,7 +612,7 @@ flow_wildcards_equal(const struct flow_wildcards *a, || a->tun_id_mask != b->tun_id_mask || a->nw_src_mask != b->nw_src_mask || a->nw_dst_mask != b->nw_dst_mask - || a->vlan_tci_mask != b->vlan_tci_mask + || a->vlan_tci_mask != b->vlan_tci_mask || !ipv6_addr_equals(&a->ipv6_src_mask, &b->ipv6_src_mask) || !ipv6_addr_equals(&a->ipv6_dst_mask, &b->ipv6_dst_mask)) { return false; @@ -764,6 +727,83 @@ flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask) wc->reg_masks[idx] = mask; } +/* Returns the wildcard bitmask for the Ethernet destination address + * that 'wc' specifies. The bitmask has a 0 in each bit that is wildcarded + * and a 1 in each bit that must match. */ +const uint8_t * +flow_wildcards_to_dl_dst_mask(flow_wildcards_t wc) +{ + static const uint8_t no_wild[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; + static const uint8_t addr_wild[] = {0x01, 0x00, 0x00, 0x00, 0x00, 0x00}; + static const uint8_t mcast_wild[] = {0xfe, 0xff, 0xff, 0xff, 0xff, 0xff}; + static const uint8_t all_wild[] = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00}; + + switch (wc & (FWW_DL_DST | FWW_ETH_MCAST)) { + case 0: return no_wild; + case FWW_DL_DST: return addr_wild; + case FWW_ETH_MCAST: return mcast_wild; + case FWW_DL_DST | FWW_ETH_MCAST: return all_wild; + } + NOT_REACHED(); +} + +/* Returns true if 'mask' is a valid wildcard bitmask for the Ethernet + * destination address. Valid bitmasks are either all-bits-0 or all-bits-1, + * except that the multicast bit may differ from the rest of the bits. So, + * there are four possible valid bitmasks: + * + * - 00:00:00:00:00:00 + * - 01:00:00:00:00:00 + * - fe:ff:ff:ff:ff:ff + * - ff:ff:ff:ff:ff:ff + * + * All other bitmasks are invalid. */ +bool +flow_wildcards_is_dl_dst_mask_valid(const uint8_t mask[ETH_ADDR_LEN]) +{ + switch (mask[0]) { + case 0x00: + case 0x01: + return (mask[1] | mask[2] | mask[3] | mask[4] | mask[5]) == 0x00; + + case 0xfe: + case 0xff: + return (mask[1] & mask[2] & mask[3] & mask[4] & mask[5]) == 0xff; + + default: + return false; + } +} + +/* Returns 'wc' with the FWW_DL_DST and FWW_ETH_MCAST bits modified + * appropriately to match 'mask'. + * + * This function will assert-fail if 'mask' is invalid. Only 'mask' values + * accepted by flow_wildcards_is_dl_dst_mask_valid() are allowed. */ +flow_wildcards_t +flow_wildcards_set_dl_dst_mask(flow_wildcards_t wc, + const uint8_t mask[ETH_ADDR_LEN]) +{ + assert(flow_wildcards_is_dl_dst_mask_valid(mask)); + + switch (mask[0]) { + case 0x00: + return wc | FWW_DL_DST | FWW_ETH_MCAST; + + case 0x01: + return (wc | FWW_DL_DST) & ~FWW_ETH_MCAST; + + case 0xfe: + return (wc & ~FWW_DL_DST) | FWW_ETH_MCAST; + + case 0xff: + return wc & ~(FWW_DL_DST | FWW_ETH_MCAST); + + default: + NOT_REACHED(); + } +} + /* Hashes 'flow' based on its L2 through L4 protocol information. */ uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis) @@ -788,10 +828,13 @@ flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis) } fields.vlan_tci = flow->vlan_tci & htons(VLAN_VID_MASK); fields.eth_type = flow->dl_type; + + /* UDP source and destination port are not taken into account because they + * will not necessarily be symmetric in a bidirectional flow. */ if (fields.eth_type == htons(ETH_TYPE_IP)) { fields.ipv4_addr = flow->nw_src ^ flow->nw_dst; fields.ip_proto = flow->nw_proto; - if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_UDP) { + if (fields.ip_proto == IPPROTO_TCP) { fields.tp_addr = flow->tp_src ^ flow->tp_dst; } } else if (fields.eth_type == htons(ETH_TYPE_IPV6)) { @@ -803,9 +846,45 @@ flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis) ipv6_addr[i] = a[i] ^ b[i]; } fields.ip_proto = flow->nw_proto; - if (fields.ip_proto == IPPROTO_TCP || fields.ip_proto == IPPROTO_UDP) { + if (fields.ip_proto == IPPROTO_TCP) { fields.tp_addr = flow->tp_src ^ flow->tp_dst; } } return hash_bytes(&fields, sizeof fields, basis); } + +/* Hashes the portions of 'flow' designated by 'fields'. */ +uint32_t +flow_hash_fields(const struct flow *flow, enum nx_hash_fields fields, + uint16_t basis) +{ + switch (fields) { + + case NX_HASH_FIELDS_ETH_SRC: + return hash_bytes(flow->dl_src, sizeof flow->dl_src, basis); + + case NX_HASH_FIELDS_SYMMETRIC_L4: + return flow_hash_symmetric_l4(flow, basis); + } + + NOT_REACHED(); +} + +/* Returns a string representation of 'fields'. */ +const char * +flow_hash_fields_to_str(enum nx_hash_fields fields) +{ + switch (fields) { + case NX_HASH_FIELDS_ETH_SRC: return "eth_src"; + case NX_HASH_FIELDS_SYMMETRIC_L4: return "symmetric_l4"; + default: return ""; + } +} + +/* Returns true if the value of 'fields' is supported. Otherwise false. */ +bool +flow_hash_fields_valid(enum nx_hash_fields fields) +{ + return fields == NX_HASH_FIELDS_ETH_SRC + || fields == NX_HASH_FIELDS_SYMMETRIC_L4; +}