X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fdpif-netdev.c;h=c7179ec19deb90fc748caaa5a80d5117450bd87f;hb=66642cb40b12594c62f3d3037f1e4efa528416b7;hp=6c813299f7aea5302e7c41bceb17093f0912a84d;hpb=401eeb92d32ac0fa07f34f5b803d67b8032b6403;p=openvswitch diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 6c813299..c7179ec1 100644 --- a/lib/dpif-netdev.c +++ b/lib/dpif-netdev.c @@ -47,12 +47,11 @@ #include "util.h" #include "vlog.h" -VLOG_DEFINE_THIS_MODULE(dpif_netdev) +VLOG_DEFINE_THIS_MODULE(dpif_netdev); /* Configuration parameters. */ enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */ enum { MAX_QUEUE_LEN = 100 }; /* Maximum number of packets per queue. */ -enum { N_GROUPS = 16 }; /* Number of port groups. */ enum { MAX_PORTS = 256 }; /* Maximum number of ports. */ enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ @@ -70,7 +69,6 @@ struct dp_netdev { bool drop_frags; /* Drop all IP fragments, if true. */ struct ovs_queue queues[N_QUEUES]; /* Messages queued for dpif_recv(). */ struct hmap flow_table; /* Flow table. */ - struct odp_port_group groups[N_GROUPS]; /* Statistics. */ long long int n_frags; /* Number of dropped IP fragments. */ @@ -96,13 +94,13 @@ struct dp_netdev_port { /* A flow in dp_netdev's 'flow_table'. */ struct dp_netdev_flow { struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */ - flow_t key; + struct flow key; /* Statistics. */ - struct timespec used; /* Last used time. */ - long long int packet_count; /* Number of packets matched. */ - long long int byte_count; /* Number of bytes matched. */ - uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */ + struct timespec used; /* Last used time. */ + long long int packet_count; /* Number of packets matched. */ + long long int byte_count; /* Number of bytes matched. */ + uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */ /* Actions. */ union odp_action *actions; @@ -137,7 +135,7 @@ static int do_del_port(struct dp_netdev *, uint16_t port_no); static int dp_netdev_output_control(struct dp_netdev *, const struct ofpbuf *, int queue_no, int port_no, uint32_t arg); static int dp_netdev_execute_actions(struct dp_netdev *, - struct ofpbuf *, const flow_t *, + struct ofpbuf *, struct flow *, const union odp_action *, int n); static struct dpif_netdev * @@ -228,11 +226,6 @@ create_dp_netdev(const char *name, int dp_idx, struct dpif **dpifp) queue_init(&dp->queues[i]); } hmap_init(&dp->flow_table); - for (i = 0; i < N_GROUPS; i++) { - dp->groups[i].ports = NULL; - dp->groups[i].n_ports = 0; - dp->groups[i].group = i; - } list_init(&dp->port_list); error = do_add_port(dp, name, ODP_PORT_INTERNAL, ODPP_LOCAL); if (error) { @@ -245,8 +238,8 @@ create_dp_netdev(const char *name, int dp_idx, struct dpif **dpifp) } static int -dpif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create, - struct dpif **dpifp) +dpif_netdev_open(const struct dpif_class *class OVS_UNUSED, const char *name, + bool create, struct dpif **dpifp) { if (create) { if (find_dp_netdev(name)) { @@ -294,9 +287,6 @@ dp_netdev_free(struct dp_netdev *dp) queue_destroy(&dp->queues[i]); } hmap_destroy(&dp->flow_table); - for (i = 0; i < N_GROUPS; i++) { - free(dp->groups[i].ports); - } dp_netdevs[dp->dp_idx] = NULL; list_remove(&dp->node); free(dp); @@ -331,7 +321,6 @@ dpif_netdev_get_stats(const struct dpif *dpif, struct odp_stats *stats) stats->max_capacity = MAX_FLOWS; stats->n_ports = dp->n_ports; stats->max_ports = MAX_PORTS; - stats->max_groups = N_GROUPS; stats->n_frags = dp->n_frags; stats->n_hit = dp->n_hit; stats->n_missed = dp->n_missed; @@ -457,7 +446,7 @@ get_port_by_name(struct dp_netdev *dp, { struct dp_netdev_port *port; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { + LIST_FOR_EACH (port, node, &dp->port_list) { if (!strcmp(netdev_get_name(port->netdev), devname)) { *portp = port; return 0; @@ -545,8 +534,7 @@ dp_netdev_flow_flush(struct dp_netdev *dp) { struct dp_netdev_flow *flow, *next; - HMAP_FOR_EACH_SAFE (flow, next, struct dp_netdev_flow, node, - &dp->flow_table) { + HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) { dp_netdev_free_flow(dp, flow); } } @@ -567,7 +555,7 @@ dpif_netdev_port_list(const struct dpif *dpif, struct odp_port *ports, int n) int i; i = 0; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { + LIST_FOR_EACH (port, node, &dp->port_list) { struct odp_port *odp_port = &ports[i]; if (i >= n) { break; @@ -599,70 +587,12 @@ dpif_netdev_port_poll_wait(const struct dpif *dpif_) } } -static int -get_port_group(const struct dpif *dpif, int group_no, - struct odp_port_group **groupp) -{ - struct dp_netdev *dp = get_dp_netdev(dpif); - - if (group_no >= 0 && group_no < N_GROUPS) { - *groupp = &dp->groups[group_no]; - return 0; - } else { - *groupp = NULL; - return EINVAL; - } -} - -static int -dpif_netdev_port_group_get(const struct dpif *dpif, int group_no, - uint16_t ports[], int n) -{ - struct odp_port_group *group; - int error; - - if (n < 0) { - return -EINVAL; - } - - error = get_port_group(dpif, group_no, &group); - if (!error) { - memcpy(ports, group->ports, MIN(n, group->n_ports) * sizeof *ports); - return group->n_ports; - } else { - return -error; - } -} - -static int -dpif_netdev_port_group_set(struct dpif *dpif, int group_no, - const uint16_t ports[], int n) -{ - struct odp_port_group *group; - int error; - - if (n < 0 || n > MAX_PORTS) { - return EINVAL; - } - - error = get_port_group(dpif, group_no, &group); - if (!error) { - free(group->ports); - group->ports = xmemdup(ports, n * sizeof *group->ports); - group->n_ports = n; - group->group = group_no; - } - return error; -} - static struct dp_netdev_flow * -dp_netdev_lookup_flow(const struct dp_netdev *dp, const flow_t *key) +dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key) { struct dp_netdev_flow *flow; - assert(!key->reserved[0] && !key->reserved[1] && !key->reserved[2]); - HMAP_FOR_EACH_WITH_HASH (flow, struct dp_netdev_flow, node, - flow_hash(key, 0), &dp->flow_table) { + HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) { if (flow_equal(&flow->key, key)) { return flow; } @@ -670,12 +600,12 @@ dp_netdev_lookup_flow(const struct dp_netdev *dp, const flow_t *key) return NULL; } +/* The caller must fill in odp_flow->key itself. */ static void answer_flow_query(struct dp_netdev_flow *flow, uint32_t query_flags, struct odp_flow *odp_flow) { if (flow) { - odp_flow->key = flow->key; odp_flow->stats.n_packets = flow->packet_count; odp_flow->stats.n_bytes = flow->byte_count; odp_flow->stats.used_sec = flow->used.tv_sec; @@ -707,7 +637,10 @@ dpif_netdev_flow_get(const struct dpif *dpif, struct odp_flow flows[], int n) for (i = 0; i < n; i++) { struct odp_flow *odp_flow = &flows[i]; - answer_flow_query(dp_netdev_lookup_flow(dp, &odp_flow->key), + struct flow key; + + odp_flow_key_to_flow(&odp_flow->key, &key); + answer_flow_query(dp_netdev_lookup_flow(dp, &key), odp_flow->flags, odp_flow); } return 0; @@ -729,26 +662,12 @@ dpif_netdev_validate_actions(const union odp_action *actions, int n_actions, } break; - case ODPAT_OUTPUT_GROUP: - *mutates = true; - if (a->output_group.group >= N_GROUPS) { - return EINVAL; - } - break; - case ODPAT_CONTROLLER: break; - case ODPAT_SET_VLAN_VID: + case ODPAT_SET_DL_TCI: *mutates = true; - if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) { - return EINVAL; - } - break; - - case ODPAT_SET_VLAN_PCP: - *mutates = true; - if (a->vlan_pcp.vlan_pcp & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) { + if (a->dl_tci.tci & htons(VLAN_CFI)) { return EINVAL; } break; @@ -808,8 +727,7 @@ add_flow(struct dpif *dpif, struct odp_flow *odp_flow) int error; flow = xzalloc(sizeof *flow); - flow->key = odp_flow->key; - memset(flow->key.reserved, 0, sizeof flow->key.reserved); + odp_flow_key_to_flow(&odp_flow->key, &flow->key); error = set_flow_actions(flow, odp_flow); if (error) { @@ -836,8 +754,10 @@ dpif_netdev_flow_put(struct dpif *dpif, struct odp_flow_put *put) { struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_flow *flow; + struct flow key; - flow = dp_netdev_lookup_flow(dp, &put->flow.key); + odp_flow_key_to_flow(&put->flow.key, &key); + flow = dp_netdev_lookup_flow(dp, &key); if (!flow) { if (put->flags & ODPPF_CREATE) { if (hmap_count(&dp->flow_table) < MAX_FLOWS) { @@ -867,8 +787,10 @@ dpif_netdev_flow_del(struct dpif *dpif, struct odp_flow *odp_flow) { struct dp_netdev *dp = get_dp_netdev(dpif); struct dp_netdev_flow *flow; + struct flow key; - flow = dp_netdev_lookup_flow(dp, &odp_flow->key); + odp_flow_key_to_flow(&odp_flow->key, &key); + flow = dp_netdev_lookup_flow(dp, &key); if (flow) { answer_flow_query(flow, 0, odp_flow); dp_netdev_free_flow(dp, flow); @@ -886,24 +808,27 @@ dpif_netdev_flow_list(const struct dpif *dpif, struct odp_flow flows[], int n) int i; i = 0; - HMAP_FOR_EACH (flow, struct dp_netdev_flow, node, &dp->flow_table) { + HMAP_FOR_EACH (flow, node, &dp->flow_table) { if (i >= n) { break; } - answer_flow_query(flow, 0, &flows[i++]); + + odp_flow_key_from_flow(&flows[i].key, &flow->key); + answer_flow_query(flow, 0, &flows[i]); + i++; } return hmap_count(&dp->flow_table); } static int -dpif_netdev_execute(struct dpif *dpif, uint16_t in_port, +dpif_netdev_execute(struct dpif *dpif, const union odp_action actions[], int n_actions, const struct ofpbuf *packet) { struct dp_netdev *dp = get_dp_netdev(dpif); struct ofpbuf copy; bool mutates; - flow_t flow; + struct flow key; int error; if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) { @@ -928,8 +853,8 @@ dpif_netdev_execute(struct dpif *dpif, uint16_t in_port, * if we don't. */ copy = *packet; } - flow_extract(©, 0, in_port, &flow); - error = dp_netdev_execute_actions(dp, ©, &flow, actions, n_actions); + flow_extract(©, 0, -1, &key); + error = dp_netdev_execute_actions(dp, ©, &key, actions, n_actions); if (mutates) { ofpbuf_uninit(©); } @@ -998,7 +923,7 @@ dpif_netdev_recv_wait(struct dpif *dpif) } static void -dp_netdev_flow_used(struct dp_netdev_flow *flow, const flow_t *key, +dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key, const struct ofpbuf *packet) { time_timespec(&flow->used); @@ -1015,7 +940,7 @@ dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port, struct ofpbuf *packet) { struct dp_netdev_flow *flow; - flow_t key; + struct flow key; if (packet->size < ETH_HEADER_LEN) { return; @@ -1044,10 +969,10 @@ dp_netdev_run(void) struct dp_netdev *dp; ofpbuf_init(&packet, DP_NETDEV_HEADROOM + max_mtu); - LIST_FOR_EACH (dp, struct dp_netdev, node, &dp_netdev_list) { + LIST_FOR_EACH (dp, node, &dp_netdev_list) { struct dp_netdev_port *port; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { + LIST_FOR_EACH (port, node, &dp->port_list) { int error; /* Reset packet contents. */ @@ -1072,25 +997,21 @@ dp_netdev_wait(void) { struct dp_netdev *dp; - LIST_FOR_EACH (dp, struct dp_netdev, node, &dp_netdev_list) { + LIST_FOR_EACH (dp, node, &dp_netdev_list) { struct dp_netdev_port *port; - LIST_FOR_EACH (port, struct dp_netdev_port, node, &dp->port_list) { + LIST_FOR_EACH (port, node, &dp->port_list) { netdev_recv_wait(port->netdev); } } } -/* Modify the TCI field of 'packet'. If a VLAN tag is not present, one - * is added with the TCI field set to 'tci'. If a VLAN tag is present, - * then 'mask' bits are cleared before 'tci' is logically OR'd into the - * TCI field. - * - * Note that the function does not ensure that 'tci' does not affect - * bits outside of 'mask'. +/* Modify the TCI field of 'packet'. If a VLAN tag is present, its TCI field + * is replaced by 'tci'. If a VLAN tag is not present, one is added with the + * TCI field set to 'tci'. */ static void -dp_netdev_modify_vlan_tci(struct ofpbuf *packet, uint16_t tci, uint16_t mask) +dp_netdev_set_dl_tci(struct ofpbuf *packet, uint16_t tci) { struct vlan_eth_header *veh; struct eth_header *eh; @@ -1098,18 +1019,15 @@ dp_netdev_modify_vlan_tci(struct ofpbuf *packet, uint16_t tci, uint16_t mask) eh = packet->l2; if (packet->size >= sizeof(struct vlan_eth_header) && eh->eth_type == htons(ETH_TYPE_VLAN)) { - /* Clear 'mask' bits, but maintain other TCI bits. */ veh = packet->l2; - veh->veth_tci &= ~htons(mask); - veh->veth_tci |= htons(tci); + veh->veth_tci = tci; } else { /* Insert new 802.1Q header. */ - struct eth_header *eh = packet->l2; struct vlan_eth_header tmp; memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN); memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN); tmp.veth_type = htons(ETH_TYPE_VLAN); - tmp.veth_tci = htons(tci); + tmp.veth_tci = tci; tmp.veth_next_type = eh->eth_type; veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN); @@ -1151,19 +1069,25 @@ dp_netdev_set_dl_dst(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN]) memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst); } +static bool +is_ip(const struct ofpbuf *packet, const struct flow *key) +{ + return key->dl_type == htons(ETH_TYPE_IP) && packet->l4; +} + static void -dp_netdev_set_nw_addr(struct ofpbuf *packet, const flow_t *key, +dp_netdev_set_nw_addr(struct ofpbuf *packet, struct flow *key, const struct odp_action_nw_addr *a) { - if (key->dl_type == htons(ETH_TYPE_IP)) { + if (is_ip(packet, key)) { struct ip_header *nh = packet->l3; uint32_t *field; field = a->type == ODPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst; - if (key->nw_proto == IP_TYPE_TCP) { + if (key->nw_proto == IP_TYPE_TCP && packet->l7) { struct tcp_header *th = packet->l4; th->tcp_csum = recalc_csum32(th->tcp_csum, *field, a->nw_addr); - } else if (key->nw_proto == IP_TYPE_UDP) { + } else if (key->nw_proto == IP_TYPE_UDP && packet->l7) { struct udp_header *uh = packet->l4; if (uh->udp_csum) { uh->udp_csum = recalc_csum32(uh->udp_csum, *field, a->nw_addr); @@ -1178,10 +1102,10 @@ dp_netdev_set_nw_addr(struct ofpbuf *packet, const flow_t *key, } static void -dp_netdev_set_nw_tos(struct ofpbuf *packet, const flow_t *key, +dp_netdev_set_nw_tos(struct ofpbuf *packet, struct flow *key, const struct odp_action_nw_tos *a) { - if (key->dl_type == htons(ETH_TYPE_IP)) { + if (is_ip(packet, key)) { struct ip_header *nh = packet->l3; uint8_t *field = &nh->ip_tos; @@ -1195,17 +1119,17 @@ dp_netdev_set_nw_tos(struct ofpbuf *packet, const flow_t *key, } static void -dp_netdev_set_tp_port(struct ofpbuf *packet, const flow_t *key, +dp_netdev_set_tp_port(struct ofpbuf *packet, struct flow *key, const struct odp_action_tp_port *a) { - if (key->dl_type == htons(ETH_TYPE_IP)) { + if (is_ip(packet, key)) { uint16_t *field; - if (key->nw_proto == IPPROTO_TCP) { + if (key->nw_proto == IPPROTO_TCP && packet->l7) { struct tcp_header *th = packet->l4; field = a->type == ODPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst; th->tcp_csum = recalc_csum16(th->tcp_csum, *field, a->tp_port); *field = a->tp_port; - } else if (key->nw_proto == IPPROTO_UDP) { + } else if (key->nw_proto == IPPROTO_UDP && packet->l7) { struct udp_header *uh = packet->l4; field = a->type == ODPAT_SET_TP_SRC ? &uh->udp_src : &uh->udp_dst; uh->udp_csum = recalc_csum16(uh->udp_csum, *field, a->tp_port); @@ -1220,27 +1144,12 @@ static void dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet, uint16_t out_port) { - struct dp_netdev_port *p = dp->ports[out_port]; + struct dp_netdev_port *p = dp->ports[out_port]; if (p) { netdev_send(p->netdev, packet); } } -static void -dp_netdev_output_group(struct dp_netdev *dp, uint16_t group, uint16_t in_port, - struct ofpbuf *packet) -{ - struct odp_port_group *g = &dp->groups[group]; - int i; - - for (i = 0; i < g->n_ports; i++) { - uint16_t out_port = g->ports[i]; - if (out_port != in_port) { - dp_netdev_output_port(dp, packet, out_port); - } - } -} - static int dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet, int queue_no, int port_no, uint32_t arg) @@ -1256,8 +1165,7 @@ dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet, } msg_size = sizeof *header + packet->size; - msg = ofpbuf_new(msg_size + DPIF_RECV_MSG_PADDING); - ofpbuf_reserve(msg, DPIF_RECV_MSG_PADDING); + msg = ofpbuf_new_with_headroom(msg_size, DPIF_RECV_MSG_PADDING); header = ofpbuf_put_uninit(msg, sizeof *header); header->type = queue_no; header->length = msg_size; @@ -1273,7 +1181,7 @@ dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet, * screwy or truncated header fields or one whose inner and outer Ethernet * address differ. */ static bool -dp_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct odp_flow_key *key) +dp_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct flow *key) { struct arp_eth_header *arp; struct eth_header *eth; @@ -1299,71 +1207,59 @@ dp_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct odp_flow_key *key) static int dp_netdev_execute_actions(struct dp_netdev *dp, - struct ofpbuf *packet, const flow_t *key, + struct ofpbuf *packet, struct flow *key, const union odp_action *actions, int n_actions) { int i; for (i = 0; i < n_actions; i++) { const union odp_action *a = &actions[i]; - switch (a->type) { - case ODPAT_OUTPUT: + switch (a->type) { + case ODPAT_OUTPUT: dp_netdev_output_port(dp, packet, a->output.port); - break; - - case ODPAT_OUTPUT_GROUP: - dp_netdev_output_group(dp, a->output_group.group, key->in_port, - packet); - break; + break; - case ODPAT_CONTROLLER: + case ODPAT_CONTROLLER: dp_netdev_output_control(dp, packet, _ODPL_ACTION_NR, key->in_port, a->controller.arg); - break; - - case ODPAT_SET_VLAN_VID: - dp_netdev_modify_vlan_tci(packet, ntohs(a->vlan_vid.vlan_vid), - VLAN_VID_MASK); break; - case ODPAT_SET_VLAN_PCP: - dp_netdev_modify_vlan_tci(packet, - a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT, - VLAN_PCP_MASK); + case ODPAT_SET_DL_TCI: + dp_netdev_set_dl_tci(packet, a->dl_tci.tci); break; - case ODPAT_STRIP_VLAN: - dp_netdev_strip_vlan(packet); - break; + case ODPAT_STRIP_VLAN: + dp_netdev_strip_vlan(packet); + break; - case ODPAT_SET_DL_SRC: + case ODPAT_SET_DL_SRC: dp_netdev_set_dl_src(packet, a->dl_addr.dl_addr); - break; + break; - case ODPAT_SET_DL_DST: + case ODPAT_SET_DL_DST: dp_netdev_set_dl_dst(packet, a->dl_addr.dl_addr); - break; + break; - case ODPAT_SET_NW_SRC: - case ODPAT_SET_NW_DST: - dp_netdev_set_nw_addr(packet, key, &a->nw_addr); - break; + case ODPAT_SET_NW_SRC: + case ODPAT_SET_NW_DST: + dp_netdev_set_nw_addr(packet, key, &a->nw_addr); + break; - case ODPAT_SET_NW_TOS: - dp_netdev_set_nw_tos(packet, key, &a->nw_tos); - break; + case ODPAT_SET_NW_TOS: + dp_netdev_set_nw_tos(packet, key, &a->nw_tos); + break; - case ODPAT_SET_TP_SRC: - case ODPAT_SET_TP_DST: - dp_netdev_set_tp_port(packet, key, &a->tp_port); - break; + case ODPAT_SET_TP_SRC: + case ODPAT_SET_TP_DST: + dp_netdev_set_tp_port(packet, key, &a->tp_port); + break; case ODPAT_DROP_SPOOFED_ARP: if (dp_netdev_is_spoofed_arp(packet, key)) { return 0; } - } - } + } + } return 0; } @@ -1386,8 +1282,6 @@ const struct dpif_class dpif_netdev_class = { dpif_netdev_port_list, dpif_netdev_port_poll, dpif_netdev_port_poll_wait, - dpif_netdev_port_group_get, - dpif_netdev_port_group_set, dpif_netdev_flow_get, dpif_netdev_flow_put, dpif_netdev_flow_del,