X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=switch%2Fdatapath.c;h=452d471b38bd0bfaf974cc76d502af02e460a629;hb=be63d0128504bedc9c73ab25d4e462fb188acbbd;hp=1e71411dfaaa2f72c43711ae45a69b9d20e253b4;hpb=95204b0bf0c329fde1ff1ba821f9166abfa18377;p=sliver-openvswitch.git diff --git a/switch/datapath.c b/switch/datapath.c index 1e71411df..452d471b3 100644 --- a/switch/datapath.c +++ b/switch/datapath.c @@ -44,28 +44,21 @@ #include "list.h" #include "netdev.h" #include "ofpbuf.h" -#include "openflow.h" +#include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" #include "rconn.h" +#include "stp.h" #include "switch-flow.h" #include "table.h" #include "timeval.h" #include "vconn.h" #include "xtoxll.h" +#include "dp_act.h" #define THIS_MODULE VLM_datapath #include "vlog.h" -enum br_port_flags { - BRPF_NO_FLOOD = 1 << 0, -}; - -enum br_port_status { - BRPS_PORT_DOWN = 1 << 0, - BRPS_LINK_DOWN = 1 << 1, -}; - extern char mfr_desc; extern char hw_desc; extern char sw_desc; @@ -79,7 +72,9 @@ extern char serial_num; /* Actions supported by this implementation. */ #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \ - | (1 << OFPAT_SET_DL_VLAN) \ + | (1 << OFPAT_SET_VLAN_VID) \ + | (1 << OFPAT_SET_VLAN_PCP) \ + | (1 << OFPAT_STRIP_VLAN) \ | (1 << OFPAT_SET_DL_SRC) \ | (1 << OFPAT_SET_DL_DST) \ | (1 << OFPAT_SET_NW_SRC) \ @@ -88,8 +83,8 @@ extern char serial_num; | (1 << OFPAT_SET_TP_DST) ) struct sw_port { - uint32_t flags; /* BRPF_* flags */ - uint32_t status; /* BRPS_* flags */ + uint32_t config; /* Some subset of OFPPC_* flags. */ + uint32_t state; /* Some subset of OFPPS_* flags. */ struct datapath *dp; struct netdev *netdev; struct list node; /* Element in datapath.ports. */ @@ -121,11 +116,14 @@ struct remote { void *cb_aux; }; +#define DP_MAX_PORTS 255 +BUILD_ASSERT_DECL(DP_MAX_PORTS <= OFPP_MAX); + struct datapath { /* Remote connections. */ struct remote *controller; /* Connection to controller. */ struct list remotes; /* All connections (including controller). */ - struct vconn *listen_vconn; + struct pvconn *listen_pvconn; time_t last_timeout; @@ -139,7 +137,7 @@ struct datapath { uint16_t miss_send_len; /* Switch ports. */ - struct sw_port ports[OFPP_MAX]; + struct sw_port ports[DP_MAX_PORTS]; struct list port_list; /* List of ports, for flooding. */ }; @@ -150,25 +148,12 @@ static void remote_run(struct datapath *, struct remote *); static void remote_wait(struct remote *); static void remote_destroy(struct remote *); -void dp_output_port(struct datapath *, struct ofpbuf *, - int in_port, int out_port); void dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm); -void dp_output_control(struct datapath *, struct ofpbuf *, int in_port, - size_t max_len, int reason); static void send_flow_expired(struct datapath *, struct sw_flow *, enum ofp_flow_expired_reason); static int update_port_status(struct sw_port *p); static void send_port_status(struct sw_port *p, uint8_t status); static void del_switch_port(struct sw_port *p); -static void execute_actions(struct datapath *, struct ofpbuf *, - int in_port, const struct sw_flow_key *, - const struct ofp_action *, int n_actions); -static void modify_vlan(struct ofpbuf *buffer, const struct sw_flow_key *key, - const struct ofp_action *a); -static void modify_nh(struct ofpbuf *buffer, uint16_t eth_proto, - uint8_t nw_proto, const struct ofp_action *a); -static void modify_th(struct ofpbuf *buffer, uint16_t eth_proto, - uint8_t nw_proto, const struct ofp_action *a); /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID * into a buffer number (low bits) and a cookie (high bits). The buffer number @@ -181,8 +166,9 @@ static void modify_th(struct ofpbuf *buffer, uint16_t eth_proto, #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS) -int run_flow_through_tables(struct datapath *, struct ofpbuf *, int in_port); -void fwd_port_input(struct datapath *, struct ofpbuf *, int in_port); +int run_flow_through_tables(struct datapath *, struct ofpbuf *, + struct sw_port *); +void fwd_port_input(struct datapath *, struct ofpbuf *, struct sw_port *); int fwd_control_input(struct datapath *, const struct sender *, const void *, size_t); @@ -218,7 +204,7 @@ dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn) dp->last_timeout = time_now(); list_init(&dp->remotes); dp->controller = remote_create(dp, rconn); - dp->listen_vconn = NULL; + dp->listen_pvconn = NULL; dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id(); dp->chain = chain_create(); if (!dp->chain) { @@ -283,10 +269,10 @@ dp_add_port(struct datapath *dp, const char *name) } void -dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn) +dp_add_listen_pvconn(struct datapath *dp, struct pvconn *listen_pvconn) { - assert(!dp->listen_vconn); - dp->listen_vconn = listen_vconn; + assert(!dp->listen_pvconn); + dp->listen_pvconn = listen_pvconn; } void @@ -303,7 +289,7 @@ dp_run(struct datapath *dp) LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) { if (update_port_status(p)) { - send_port_status(p, OFPPR_MOD); + send_port_status(p, OFPPR_MODIFY); } } @@ -334,7 +320,7 @@ dp_run(struct datapath *dp) if (!error) { p->rx_packets++; p->rx_bytes += buffer->size; - fwd_port_input(dp, buffer, port_no(dp, p)); + fwd_port_input(dp, buffer, p); buffer = NULL; } else if (error != EAGAIN) { VLOG_ERR_RL(&rl, "error receiving data from %s: %s", @@ -347,12 +333,12 @@ dp_run(struct datapath *dp) LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) { remote_run(dp, r); } - if (dp->listen_vconn) { + if (dp->listen_pvconn) { for (;;) { struct vconn *new_vconn; int retval; - retval = vconn_accept(dp->listen_vconn, &new_vconn); + retval = pvconn_accept(dp->listen_pvconn, OFP_VERSION, &new_vconn); if (retval) { if (retval != EAGAIN) { VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); @@ -485,8 +471,8 @@ dp_wait(struct datapath *dp) LIST_FOR_EACH (r, struct remote, node, &dp->remotes) { remote_wait(r); } - if (dp->listen_vconn) { - vconn_accept_wait(dp->listen_vconn); + if (dp->listen_pvconn) { + pvconn_wait(dp->listen_pvconn); } } @@ -530,16 +516,17 @@ output_all(struct datapath *dp, struct ofpbuf *buffer, int in_port, int flood) if (port_no(dp, p) == in_port) { continue; } - if (flood && p->flags & BRPF_NO_FLOOD) { + if (flood && p->config & OFPPC_NO_FLOOD) { continue; } if (prev_port != -1) { - dp_output_port(dp, ofpbuf_clone(buffer), in_port, prev_port); + dp_output_port(dp, ofpbuf_clone(buffer), in_port, prev_port, + false); } prev_port = port_no(dp, p); } if (prev_port != -1) - dp_output_port(dp, buffer, in_port, prev_port); + dp_output_port(dp, buffer, in_port, prev_port, false); else ofpbuf_delete(buffer); @@ -549,9 +536,9 @@ output_all(struct datapath *dp, struct ofpbuf *buffer, int in_port, int flood) void output_packet(struct datapath *dp, struct ofpbuf *buffer, int out_port) { - if (out_port >= 0 && out_port < OFPP_MAX) { + if (out_port >= 0 && out_port < DP_MAX_PORTS) { struct sw_port *p = &dp->ports[out_port]; - if (p->netdev != NULL && !(p->status & BRPS_PORT_DOWN)) { + if (p->netdev != NULL && !(p->config & OFPPC_PORT_DOWN)) { if (!netdev_send(p->netdev, buffer)) { p->tx_packets++; p->tx_bytes += buffer->size; @@ -570,7 +557,7 @@ output_packet(struct datapath *dp, struct ofpbuf *buffer, int out_port) */ void dp_output_port(struct datapath *dp, struct ofpbuf *buffer, - int in_port, int out_port) + int in_port, int out_port, bool ignore_no_fwd) { assert(buffer); @@ -583,7 +570,8 @@ dp_output_port(struct datapath *dp, struct ofpbuf *buffer, } else if (out_port == OFPP_IN_PORT) { output_packet(dp, buffer, in_port); } else if (out_port == OFPP_TABLE) { - if (run_flow_through_tables(dp, buffer, in_port)) { + struct sw_port *p = in_port < DP_MAX_PORTS ? &dp->ports[in_port] : 0; + if (run_flow_through_tables(dp, buffer, p)) { ofpbuf_delete(buffer); } } else { @@ -636,7 +624,7 @@ dp_output_control(struct datapath *dp, struct ofpbuf *buffer, int in_port, buffer_id = save_buffer(buffer); total_len = buffer->size; - if (buffer_id != UINT32_MAX && buffer->size > max_len) { + if (buffer_id != UINT32_MAX && max_len && buffer->size > max_len) { buffer->size = max_len; } @@ -661,17 +649,14 @@ static void fill_port_desc(struct datapath *dp, struct sw_port *p, sizeof desc->name); desc->name[sizeof desc->name - 1] = '\0'; memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN); - desc->flags = 0; - desc->features = htonl(netdev_get_features(p->netdev)); - desc->speed = htonl(netdev_get_speed(p->netdev)); - - if (p->flags & BRPF_NO_FLOOD) { - desc->flags |= htonl(OFPPFL_NO_FLOOD); - } else if (p->status & BRPS_PORT_DOWN) { - desc->flags |= htonl(OFPPFL_PORT_DOWN); - } else if (p->status & BRPS_LINK_DOWN) { - desc->flags |= htonl(OFPPFL_LINK_DOWN); - } + desc->config = htonl(p->config); + desc->state = htonl(p->state); + desc->curr = htonl(netdev_get_features(p->netdev, NETDEV_FEAT_CURRENT)); + desc->supported = htonl(netdev_get_features(p->netdev, + NETDEV_FEAT_SUPPORTED)); + desc->advertised = htonl(netdev_get_features(p->netdev, + NETDEV_FEAT_ADVERTISED)); + desc->peer = htonl(netdev_get_features(p->netdev, NETDEV_FEAT_PEER)); } static void @@ -683,17 +668,13 @@ dp_send_features_reply(struct datapath *dp, const struct sender *sender) ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY, sender, &buffer); - ofr->datapath_id = htonll(dp->id); - ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS); - ofr->n_compression = 0; /* Not supported */ - ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS); - ofr->buffer_mb = htonl(UINT32_MAX); - ofr->n_buffers = htonl(N_PKT_BUFFERS); - ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES); - ofr->actions = htonl(OFP_SUPPORTED_ACTIONS); + ofr->datapath_id = htonll(dp->id); + ofr->n_tables = dp->chain->n_tables; + ofr->n_buffers = htonl(N_PKT_BUFFERS); + ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES); + ofr->actions = htonl(OFP_SUPPORTED_ACTIONS); LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) { - struct ofp_phy_port *opp = ofpbuf_put_uninit(buffer, sizeof *opp); - memset(opp, 0, sizeof *opp); + struct ofp_phy_port *opp = ofpbuf_put_zeros(buffer, sizeof *opp); fill_port_desc(dp, p, opp); } send_openflow_buffer(dp, buffer, sender); @@ -702,33 +683,31 @@ dp_send_features_reply(struct datapath *dp, const struct sender *sender) void dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm) { - const struct ofp_phy_port *opp = &opm->desc; - int port_no = ntohs(opp->port_no); - if (port_no < OFPP_MAX) { + int port_no = ntohs(opm->port_no); + if (port_no < DP_MAX_PORTS) { struct sw_port *p = &dp->ports[port_no]; /* Make sure the port id hasn't changed since this was sent */ - if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev), + if (memcmp(opm->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN) != 0) { return; } - if (opm->mask & htonl(OFPPFL_NO_FLOOD)) { - if (opp->flags & htonl(OFPPFL_NO_FLOOD)) - p->flags |= BRPF_NO_FLOOD; - else - p->flags &= ~BRPF_NO_FLOOD; + if (opm->mask) { + uint32_t config_mask = ntohl(opm->mask); + p->config &= ~config_mask; + p->config |= ntohl(opm->config) & config_mask; } - if (opm->mask & htonl(OFPPFL_PORT_DOWN)) { - if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) - && (p->status & BRPS_PORT_DOWN) == 0) { - p->status |= BRPS_PORT_DOWN; + if (opm->mask & htonl(OFPPC_PORT_DOWN)) { + if ((opm->config & htonl(OFPPC_PORT_DOWN)) + && (p->config & OFPPC_PORT_DOWN) == 0) { + p->config |= OFPPC_PORT_DOWN; netdev_turn_flags_off(p->netdev, NETDEV_UP, true); - } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0 - && (p->status & BRPS_PORT_DOWN)) { - p->status &= ~BRPS_PORT_DOWN; + } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0 + && (p->config & OFPPC_PORT_DOWN)) { + p->config &= ~OFPPC_PORT_DOWN; netdev_turn_flags_on(p->netdev, NETDEV_UP, true); } } @@ -746,7 +725,8 @@ update_port_status(struct sw_port *p) { int retval; enum netdev_flags flags; - uint32_t orig_status = p->status; + uint32_t orig_config = p->config; + uint32_t orig_state = p->state; if (netdev_get_flags(p->netdev, &flags) < 0) { VLOG_WARN_RL(&rl, "could not get netdev flags for %s", @@ -754,9 +734,9 @@ update_port_status(struct sw_port *p) return 0; } else { if (flags & NETDEV_UP) { - p->status &= ~BRPS_PORT_DOWN; + p->config &= ~OFPPC_PORT_DOWN; } else { - p->status |= BRPS_PORT_DOWN; + p->config |= OFPPC_PORT_DOWN; } } @@ -764,12 +744,12 @@ update_port_status(struct sw_port *p) * error. */ retval = netdev_get_link_status(p->netdev); if (retval == 1) { - p->status &= ~BRPS_LINK_DOWN; + p->state &= ~OFPPS_LINK_DOWN; } else if (retval == 0) { - p->status |= BRPS_LINK_DOWN; + p->state |= OFPPS_LINK_DOWN; } - return (orig_status != p->status); + return ((orig_config != p->config) || (orig_state != p->state)); } static void @@ -807,12 +787,11 @@ send_flow_expired(struct datapath *dp, struct sw_flow *flow, void dp_send_error_msg(struct datapath *dp, const struct sender *sender, - uint16_t type, uint16_t code, const uint8_t *data, size_t len) + uint16_t type, uint16_t code, const void *data, size_t len) { struct ofpbuf *buffer; struct ofp_error_msg *oem; - oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG, - sender, &buffer); + oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR, sender, &buffer); oem->type = htons(type); oem->code = htons(code); memcpy(oem->data, data, len); @@ -824,11 +803,10 @@ fill_flow_stats(struct ofpbuf *buffer, struct sw_flow *flow, int table_idx, time_t now) { struct ofp_flow_stats *ofs; - int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions; - ofs = ofpbuf_put_uninit(buffer, length); + int length = sizeof *ofs + flow->sf_acts->actions_len; + ofs = ofpbuf_put_zeros(buffer, length); ofs->length = htons(length); ofs->table_id = table_idx; - ofs->pad = 0; ofs->match.wildcards = htonl(flow->key.wildcards); ofs->match.in_port = flow->key.flow.in_port; memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN); @@ -838,226 +816,62 @@ fill_flow_stats(struct ofpbuf *buffer, struct sw_flow *flow, ofs->match.nw_src = flow->key.flow.nw_src; ofs->match.nw_dst = flow->key.flow.nw_dst; ofs->match.nw_proto = flow->key.flow.nw_proto; - ofs->match.pad = 0; ofs->match.tp_src = flow->key.flow.tp_src; ofs->match.tp_dst = flow->key.flow.tp_dst; ofs->duration = htonl(now - flow->created); ofs->priority = htons(flow->priority); ofs->idle_timeout = htons(flow->idle_timeout); ofs->hard_timeout = htons(flow->hard_timeout); - memset(ofs->pad2, 0, sizeof ofs->pad2); ofs->packet_count = htonll(flow->packet_count); ofs->byte_count = htonll(flow->byte_count); - memcpy(ofs->actions, flow->actions, - sizeof *ofs->actions * flow->n_actions); + memcpy(ofs->actions, flow->sf_acts->actions, flow->sf_acts->actions_len); } -/* 'buffer' was received on 'in_port', a physical switch port between 0 and - * OFPP_MAX. Process it according to 'dp''s flow table. Returns 0 if +/* 'buffer' was received on 'p', which may be a a physical switch port or a + * null pointer. Process it according to 'dp''s flow table. Returns 0 if * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no * matching flow, in which case 'buffer' still belongs to the caller. */ int run_flow_through_tables(struct datapath *dp, struct ofpbuf *buffer, - int in_port) + struct sw_port *p) { struct sw_flow_key key; struct sw_flow *flow; key.wildcards = 0; - if (flow_extract(buffer, in_port, &key.flow) + if (flow_extract(buffer, p ? port_no(dp, p) : OFPP_NONE, &key.flow) && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) { /* Drop fragment. */ ofpbuf_delete(buffer); return 0; } + if (p && p->config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) + && p->config & (!eth_addr_equals(key.flow.dl_dst, stp_eth_addr) + ? OFPPC_NO_RECV : OFPPC_NO_RECV_STP)) { + ofpbuf_delete(buffer); + return 0; + } flow = chain_lookup(dp->chain, &key); if (flow != NULL) { flow_used(flow, buffer); - execute_actions(dp, buffer, in_port, &key, - flow->actions, flow->n_actions); + execute_actions(dp, buffer, &key, flow->sf_acts->actions, + flow->sf_acts->actions_len, false); return 0; } else { return -ESRCH; } } -/* 'buffer' was received on 'in_port', a physical switch port between 0 and - * OFPP_MAX. Process it according to 'dp''s flow table, sending it up to the - * controller if no flow matches. Takes ownership of 'buffer'. */ -void fwd_port_input(struct datapath *dp, struct ofpbuf *buffer, int in_port) -{ - if (run_flow_through_tables(dp, buffer, in_port)) { - dp_output_control(dp, buffer, in_port, dp->miss_send_len, - OFPR_NO_MATCH); - } -} - -static void -do_output(struct datapath *dp, struct ofpbuf *buffer, int in_port, - size_t max_len, int out_port) -{ - if (out_port != OFPP_CONTROLLER) { - dp_output_port(dp, buffer, in_port, out_port); - } else { - dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION); - } -} - -static void -execute_actions(struct datapath *dp, struct ofpbuf *buffer, - int in_port, const struct sw_flow_key *key, - const struct ofp_action *actions, int n_actions) -{ - /* Every output action needs a separate clone of 'buffer', but the common - * case is just a single output action, so that doing a clone and then - * freeing the original buffer is wasteful. So the following code is - * slightly obscure just to avoid that. */ - int prev_port; - size_t max_len=0; /* Initialze to make compiler happy */ - uint16_t eth_proto; - int i; - - prev_port = -1; - eth_proto = ntohs(key->flow.dl_type); - - for (i = 0; i < n_actions; i++) { - const struct ofp_action *a = &actions[i]; - struct eth_header *eh = buffer->l2; - - if (prev_port != -1) { - do_output(dp, ofpbuf_clone(buffer), in_port, max_len, prev_port); - prev_port = -1; - } - - switch (ntohs(a->type)) { - case OFPAT_OUTPUT: - prev_port = ntohs(a->arg.output.port); - max_len = ntohs(a->arg.output.max_len); - break; - - case OFPAT_SET_DL_VLAN: - modify_vlan(buffer, key, a); - break; - - case OFPAT_SET_DL_SRC: - memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src); - break; - - case OFPAT_SET_DL_DST: - memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst); - break; - - case OFPAT_SET_NW_SRC: - case OFPAT_SET_NW_DST: - modify_nh(buffer, eth_proto, key->flow.nw_proto, a); - break; - - case OFPAT_SET_TP_SRC: - case OFPAT_SET_TP_DST: - modify_th(buffer, eth_proto, key->flow.nw_proto, a); - break; - - default: - NOT_REACHED(); - } - } - if (prev_port != -1) - do_output(dp, buffer, in_port, max_len, prev_port); - else - ofpbuf_delete(buffer); -} - -static void modify_nh(struct ofpbuf *buffer, uint16_t eth_proto, - uint8_t nw_proto, const struct ofp_action *a) -{ - if (eth_proto == ETH_TYPE_IP) { - struct ip_header *nh = buffer->l3; - uint32_t new, *field; - - new = a->arg.nw_addr; - field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst; - if (nw_proto == IP_TYPE_TCP) { - struct tcp_header *th = buffer->l4; - th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new); - } else if (nw_proto == IP_TYPE_UDP) { - struct udp_header *th = buffer->l4; - if (th->udp_csum) { - th->udp_csum = recalc_csum32(th->udp_csum, *field, new); - if (!th->udp_csum) { - th->udp_csum = 0xffff; - } - } - } - nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new); - *field = new; - } -} - -static void modify_th(struct ofpbuf *buffer, uint16_t eth_proto, - uint8_t nw_proto, const struct ofp_action *a) -{ - if (eth_proto == ETH_TYPE_IP) { - uint16_t new, *field; - - new = a->arg.tp; - - if (nw_proto == IP_TYPE_TCP) { - struct tcp_header *th = buffer->l4; - field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst; - th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new); - *field = new; - } else if (nw_proto == IP_TYPE_UDP) { - struct udp_header *th = buffer->l4; - field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst; - th->udp_csum = recalc_csum16(th->udp_csum, *field, new); - *field = new; - } - } -} - -static void -modify_vlan(struct ofpbuf *buffer, - const struct sw_flow_key *key, const struct ofp_action *a) +/* 'buffer' was received on 'p', which may be a a physical switch port or a + * null pointer. Process it according to 'dp''s flow table, sending it up to + * the controller if no flow matches. Takes ownership of 'buffer'. */ +void fwd_port_input(struct datapath *dp, struct ofpbuf *buffer, + struct sw_port *p) { - uint16_t new_id = a->arg.vlan_id; - struct vlan_eth_header *veh; - - if (new_id != htons(OFP_VLAN_NONE)) { - if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) { - /* Modify vlan id, but maintain other TCI values */ - veh = buffer->l2; - veh->veth_tci &= ~htons(VLAN_VID); - veh->veth_tci |= new_id; - } else { - /* Insert new vlan id. */ - struct eth_header *eh = buffer->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 = new_id; - tmp.veth_next_type = eh->eth_type; - - veh = ofpbuf_push_uninit(buffer, VLAN_HEADER_LEN); - memcpy(veh, &tmp, sizeof tmp); - buffer->l2 = (char*)buffer->l2 - VLAN_HEADER_LEN; - } - } else { - /* Remove an existing vlan header if it exists */ - veh = buffer->l2; - if (veh->veth_type == htons(ETH_TYPE_VLAN)) { - struct eth_header tmp; - - memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN); - memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN); - tmp.eth_type = veh->veth_next_type; - - buffer->size -= VLAN_HEADER_LEN; - buffer->data = (char*)buffer->data + VLAN_HEADER_LEN; - buffer->l2 = (char*)buffer->l2 + VLAN_HEADER_LEN; - memcpy(buffer->data, &tmp, sizeof tmp); - } + if (run_flow_through_tables(dp, buffer, p)) { + dp_output_control(dp, buffer, port_no(dp, p), + dp->miss_send_len, OFPR_NO_MATCH); } } @@ -1103,25 +917,25 @@ recv_set_config(struct datapath *dp, const struct sender *sender UNUSED, } static int -recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED, +recv_packet_out(struct datapath *dp, const struct sender *sender, const void *msg) { const struct ofp_packet_out *opo = msg; struct sw_flow_key key; + uint16_t v_code; struct ofpbuf *buffer; - int n_actions = ntohs(opo->n_actions); - int act_len = n_actions * sizeof opo->actions[0]; + size_t actions_len = ntohs(opo->actions_len); - if (act_len > (ntohs(opo->header.length) - sizeof *opo)) { + if (actions_len > (ntohs(opo->header.length) - sizeof *opo)) { VLOG_DBG_RL(&rl, "message too short for number of actions"); return -EINVAL; } if (ntohl(opo->buffer_id) == (uint32_t) -1) { /* FIXME: can we avoid copying data here? */ - int data_len = ntohs(opo->header.length) - sizeof *opo - act_len; + int data_len = ntohs(opo->header.length) - sizeof *opo - actions_len; buffer = ofpbuf_new(data_len); - ofpbuf_put(buffer, &opo->actions[n_actions], data_len); + ofpbuf_put(buffer, (uint8_t *)opo->actions + actions_len, data_len); } else { buffer = retrieve_buffer(ntohl(opo->buffer_id)); if (!buffer) { @@ -1130,10 +944,21 @@ recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED, } flow_extract(buffer, ntohs(opo->in_port), &key.flow); - execute_actions(dp, buffer, ntohs(opo->in_port), - &key, opo->actions, n_actions); - return 0; + v_code = validate_actions(dp, &key, opo->actions, actions_len); + if (v_code != ACT_VALIDATION_OK) { + dp_send_error_msg(dp, sender, OFPET_BAD_ACTION, v_code, + msg, ntohs(opo->header.length)); + goto error; + } + + execute_actions(dp, buffer, &key, opo->actions, actions_len, true); + + return 0; + +error: + ofpbuf_delete(buffer); + return -EINVAL; } static int @@ -1148,50 +973,45 @@ recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED, } static int -add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm) +add_flow(struct datapath *dp, const struct sender *sender, + const struct ofp_flow_mod *ofm) { int error = -ENOMEM; - int n_actions; - int i; - struct sw_flow *flow; - - - /* To prevent loops, make sure there's no action to send to the - * OFP_TABLE virtual port. - */ - n_actions = (ntohs(ofm->header.length) - sizeof *ofm) - / sizeof *ofm->actions; - for (i=0; iactions[i]; - - if (a->type == htons(OFPAT_OUTPUT) - && (a->arg.output.port == htons(OFPP_TABLE) - || a->arg.output.port == htons(OFPP_NONE) - || a->arg.output.port == ofm->match.in_port)) { - /* xxx Send fancy new error message? */ - goto error; - } - } + uint16_t v_code; + struct sw_flow *flow; + size_t actions_len = ntohs(ofm->header.length) - sizeof *ofm; /* Allocate memory. */ - flow = flow_alloc(n_actions); + flow = flow_alloc(actions_len); if (flow == NULL) goto error; - /* Fill out flow. */ flow_extract_match(&flow->key, &ofm->match); + + v_code = validate_actions(dp, &flow->key, ofm->actions, actions_len); + if (v_code != ACT_VALIDATION_OK) { + dp_send_error_msg(dp, sender, OFPET_BAD_ACTION, v_code, + ofm, ntohs(ofm->header.length)); + goto error_free_flow; + } + + /* Fill out flow. */ flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1; flow->idle_timeout = ntohs(ofm->idle_timeout); flow->hard_timeout = ntohs(ofm->hard_timeout); flow->used = flow->created = time_now(); - flow->n_actions = n_actions; + flow->sf_acts->actions_len = actions_len; flow->byte_count = 0; flow->packet_count = 0; - memcpy(flow->actions, ofm->actions, n_actions * sizeof *flow->actions); + memcpy(flow->sf_acts->actions, ofm->actions, actions_len); /* Act. */ error = chain_insert(dp->chain, flow); - if (error) { + if (error == -ENOBUFS) { + dp_send_error_msg(dp, sender, OFPET_FLOW_MOD_FAILED, + OFPFMFC_ALL_TABLES_FULL, ofm, ntohs(ofm->header.length)); + goto error_free_flow; + } else if (error) { goto error_free_flow; } error = 0; @@ -1202,7 +1022,8 @@ add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm) uint16_t in_port = ntohs(ofm->match.in_port); flow_used(flow, buffer); flow_extract(buffer, in_port, &key.flow); - execute_actions(dp, buffer, in_port, &key, ofm->actions, n_actions); + execute_actions(dp, buffer, &key, + ofm->actions, actions_len, false); } else { error = -ESRCH; } @@ -1218,24 +1039,73 @@ error: } static int -recv_flow(struct datapath *dp, const struct sender *sender UNUSED, +mod_flow(struct datapath *dp, const struct sender *sender, + const struct ofp_flow_mod *ofm) +{ + int error = -ENOMEM; + uint16_t v_code; + size_t actions_len; + struct sw_flow_key key; + uint16_t priority; + int strict; + + flow_extract_match(&key, &ofm->match); + + actions_len = ntohs(ofm->header.length) - sizeof *ofm; + + v_code = validate_actions(dp, &key, ofm->actions, actions_len); + if (v_code != ACT_VALIDATION_OK) { + dp_send_error_msg(dp, sender, OFPET_BAD_ACTION, v_code, + ofm, ntohs(ofm->header.length)); + goto error; + } + + priority = key.wildcards ? ntohs(ofm->priority) : -1; + strict = (ofm->command == htons(OFPFC_MODIFY_STRICT)) ? 1 : 0; + chain_modify(dp->chain, &key, priority, strict, ofm->actions, actions_len); + + if (ntohl(ofm->buffer_id) != UINT32_MAX) { + struct ofpbuf *buffer = retrieve_buffer(ntohl(ofm->buffer_id)); + if (buffer) { + struct sw_flow_key skb_key; + uint16_t in_port = ntohs(ofm->match.in_port); + flow_extract(buffer, in_port, &skb_key.flow); + execute_actions(dp, buffer, &skb_key, + ofm->actions, actions_len, false); + } else { + error = -ESRCH; + } + } + return error; + +error: + if (ntohl(ofm->buffer_id) != (uint32_t) -1) + discard_buffer(ntohl(ofm->buffer_id)); + return error; +} + +static int +recv_flow(struct datapath *dp, const struct sender *sender, const void *msg) { const struct ofp_flow_mod *ofm = msg; uint16_t command = ntohs(ofm->command); if (command == OFPFC_ADD) { - return add_flow(dp, ofm); + return add_flow(dp, sender, ofm); + } else if ((command == OFPFC_MODIFY) || (command == OFPFC_MODIFY_STRICT)) { + return mod_flow(dp, sender, ofm); } else if (command == OFPFC_DELETE) { struct sw_flow_key key; flow_extract_match(&key, &ofm->match); - return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH; + return chain_delete(dp->chain, &key, ofm->out_port, 0, 0) ? 0 : -ESRCH; } else if (command == OFPFC_DELETE_STRICT) { struct sw_flow_key key; uint16_t priority; flow_extract_match(&key, &ofm->match); priority = key.wildcards ? ntohs(ofm->priority) : -1; - return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH; + return chain_delete(dp->chain, &key, ofm->out_port, + priority, 1) ? 0 : -ESRCH; } else { return -ENODEV; } @@ -1244,7 +1114,7 @@ recv_flow(struct datapath *dp, const struct sender *sender UNUSED, static int desc_stats_dump(struct datapath *dp, void *state, struct ofpbuf *buffer) { - struct ofp_desc_stats *ods = ofpbuf_put_uninit(buffer, sizeof *ods); + struct ofp_desc_stats *ods = ofpbuf_put_zeros(buffer, sizeof *ods); strncpy(ods->mfr_desc, &mfr_desc, sizeof ods->mfr_desc); strncpy(ods->hw_desc, &hw_desc, sizeof ods->hw_desc); @@ -1298,8 +1168,8 @@ static int flow_stats_dump(struct datapath *dp, void *state, { struct sw_table *table = dp->chain->tables[s->table_idx]; - if (table->iterate(table, &match_key, &s->position, - flow_stats_dump_callback, s)) + if (table->iterate(table, &match_key, s->rq.out_port, + &s->position, flow_stats_dump_callback, s)) break; s->table_idx++; @@ -1347,8 +1217,7 @@ static int aggregate_stats_dump(struct datapath *dp, void *state, struct sw_flow_key match_key; int table_idx; - rpy = ofpbuf_put_uninit(buffer, sizeof *rpy); - memset(rpy, 0, sizeof *rpy); + rpy = ofpbuf_put_zeros(buffer, sizeof *rpy); flow_extract_match(&match_key, &rq->match); table_idx = rq->table_id == 0xff ? 0 : rq->table_id; @@ -1359,7 +1228,7 @@ static int aggregate_stats_dump(struct datapath *dp, void *state, struct sw_table *table = dp->chain->tables[table_idx]; int error; - error = table->iterate(table, &match_key, &position, + error = table->iterate(table, &match_key, rq->out_port, &position, aggregate_stats_dump_callback, rpy); if (error) return error; @@ -1384,14 +1253,15 @@ static int table_stats_dump(struct datapath *dp, void *state, { int i; for (i = 0; i < dp->chain->n_tables; i++) { - struct ofp_table_stats *ots = ofpbuf_put_uninit(buffer, sizeof *ots); + struct ofp_table_stats *ots = ofpbuf_put_zeros(buffer, sizeof *ots); struct sw_table_stats stats; dp->chain->tables[i]->stats(dp->chain->tables[i], &stats); strncpy(ots->name, stats.name, sizeof ots->name); ots->table_id = i; - memset(ots->pad, 0, sizeof ots->pad); + ots->wildcards = htonl(stats.wildcards); ots->max_entries = htonl(stats.max_flows); ots->active_count = htonl(stats.n_flows); + ots->lookup_count = htonll(stats.n_lookup); ots->matched_count = htonll(stats.n_matched); } return 0; @@ -1416,15 +1286,14 @@ static int port_stats_dump(struct datapath *dp, void *state, struct port_stats_state *s = state; int i; - for (i = s->port; i < OFPP_MAX; i++) { + for (i = s->port; i < DP_MAX_PORTS; i++) { struct sw_port *p = &dp->ports[i]; struct ofp_port_stats *ops; if (!p->netdev) { continue; } - ops = ofpbuf_put_uninit(buffer, sizeof *ops); + ops = ofpbuf_put_zeros(buffer, sizeof *ops); ops->port_no = htons(port_no(dp, p)); - memset(ops->pad, 0, sizeof ops->pad); ops->rx_packets = htonll(p->rx_packets); ops->tx_packets = htonll(p->tx_packets); ops->rx_bytes = htonll(p->rx_bytes); @@ -1700,7 +1569,9 @@ fwd_control_input(struct datapath *dp, const struct sender *sender, handler = recv_echo_reply; break; default: - return -ENOSYS; + dp_send_error_msg(dp, sender, OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE, + msg, length); + return -EINVAL; } /* Handle it. */