X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=switch%2Fdatapath.c;h=5d4590f1ac57bedb37191de85c070510225c8861;hb=034f125955a786ed35052da688e77d6fc2422def;hp=77b3aaec32e50c4e2ca6ffc4b454547a3fd07851;hpb=9630e4100553c3531605aa44db65e9eaafe07dad;p=sliver-openvswitch.git diff --git a/switch/datapath.c b/switch/datapath.c index 77b3aaec3..5d4590f1a 100644 --- a/switch/datapath.c +++ b/switch/datapath.c @@ -38,27 +38,30 @@ #include #include #include -#include "buffer.h" #include "chain.h" #include "csum.h" #include "flow.h" +#include "list.h" #include "netdev.h" +#include "ofpbuf.h" +#include "openflow.h" #include "packets.h" #include "poll-loop.h" #include "rconn.h" -#include "vconn.h" +#include "stp.h" +#include "switch-flow.h" #include "table.h" #include "timeval.h" +#include "vconn.h" #include "xtoxll.h" #define THIS_MODULE VLM_datapath #include "vlog.h" -#define BRIDGE_PORT_NO_FLOOD 0x00000001 - extern char mfr_desc; extern char hw_desc; extern char sw_desc; +extern char serial_num; /* Capabilities supported by this implementation. */ #define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \ @@ -76,8 +79,12 @@ extern char sw_desc; | (1 << OFPAT_SET_TP_SRC) \ | (1 << OFPAT_SET_TP_DST) ) +#define PORT_STATUS_BITS (OFPPFL_PORT_DOWN | OFPPFL_LINK_DOWN) +#define PORT_FLAG_BITS (~PORT_STATUS_BITS) + struct sw_port { - uint32_t flags; + uint32_t flags; /* Some subset of PORT_FLAG_BITS. */ + uint32_t status; /* Some subset of PORT_STATUS_BITS. */ struct datapath *dp; struct netdev *netdev; struct list node; /* Element in datapath.ports. */ @@ -113,7 +120,7 @@ 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; @@ -131,28 +138,32 @@ struct datapath { struct list port_list; /* List of ports, for flooding. */ }; +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60); + static struct remote *remote_create(struct datapath *, struct rconn *); 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 buffer *, - int in_port, int out_port); -void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp); -void dp_output_control(struct datapath *, struct buffer *, int in_port, +void dp_output_port(struct datapath *, struct ofpbuf *, + int in_port, int out_port, bool ignore_no_fwd); +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 buffer *, +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 buffer *buffer, const struct sw_flow_key *key, + const struct ofp_action *, int n_actions, + bool ignore_no_fwd); +static void modify_vlan(struct ofpbuf *buffer, const struct sw_flow_key *key, const struct ofp_action *a); -static void modify_nh(struct buffer *buffer, uint16_t eth_proto, +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 buffer *buffer, uint16_t eth_proto, +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 @@ -166,13 +177,14 @@ static void modify_th(struct buffer *buffer, uint16_t eth_proto, #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS) -int run_flow_through_tables(struct datapath *, struct buffer *, int in_port); -void fwd_port_input(struct datapath *, struct buffer *, 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); -uint32_t save_buffer(struct buffer *); -static struct buffer *retrieve_buffer(uint32_t id); +uint32_t save_buffer(struct ofpbuf *); +static struct ofpbuf *retrieve_buffer(uint32_t id); static void discard_buffer(uint32_t id); static int port_no(struct datapath *dp, struct sw_port *p) @@ -203,7 +215,7 @@ 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) { @@ -234,7 +246,7 @@ dp_add_port(struct datapath *dp, const char *name) } error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false); if (error) { - VLOG_ERR("Couldn't set promiscuous mode on %s device", name); + VLOG_ERR("couldn't set promiscuous mode on %s device", name); netdev_close(netdev); return error; } @@ -268,10 +280,10 @@ 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 @@ -280,12 +292,18 @@ dp_run(struct datapath *dp) time_t now = time_now(); struct sw_port *p, *pn; struct remote *r, *rn; - struct buffer *buffer = NULL; + struct ofpbuf *buffer = NULL; if (now != dp->last_timeout) { struct list deleted = LIST_INITIALIZER(&deleted); struct sw_flow *f, *n; + LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) { + if (update_port_status(p)) { + send_port_status(p, OFPPR_MOD); + } + } + chain_timeout(dp->chain, &deleted); LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) { send_flow_expired(dp, f, f->reason); @@ -306,36 +324,35 @@ dp_run(struct datapath *dp) const int headroom = 128 + 2; const int hard_header = VLAN_ETH_HEADER_LEN; const int mtu = netdev_get_mtu(p->netdev); - buffer = buffer_new(headroom + hard_header + mtu); - buffer->data += headroom; + buffer = ofpbuf_new(headroom + hard_header + mtu); + buffer->data = (char*)buffer->data + headroom; } error = netdev_recv(p->netdev, buffer); 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("Error receiving data from %s: %s", - netdev_get_name(p->netdev), strerror(error)); - del_switch_port(p); + VLOG_ERR_RL(&rl, "error receiving data from %s: %s", + netdev_get_name(p->netdev), strerror(error)); } } - buffer_delete(buffer); + ofpbuf_delete(buffer); /* Talk to remotes. */ 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("accept failed (%s)", strerror(retval)); + VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); } break; } @@ -355,7 +372,7 @@ remote_run(struct datapath *dp, struct remote *r) * other processing doesn't starve. */ for (i = 0; i < 50; i++) { if (!r->cb_dump) { - struct buffer *buffer; + struct ofpbuf *buffer; struct ofp_header *oh; buffer = rconn_recv(r->rconn); @@ -371,15 +388,16 @@ remote_run(struct datapath *dp, struct remote *r) sender.xid = oh->xid; fwd_control_input(dp, &sender, buffer->data, buffer->size); } else { - VLOG_WARN("received too-short OpenFlow message"); + VLOG_WARN_RL(&rl, "received too-short OpenFlow message"); } - buffer_delete(buffer); + ofpbuf_delete(buffer); } else { if (r->n_txq < TXQ_LIMIT) { int error = r->cb_dump(dp, r->cb_aux); if (error <= 0) { if (error) { - VLOG_WARN("dump callback error: %s", strerror(-error)); + VLOG_WARN_RL(&rl, "dump callback error: %s", + strerror(-error)); } r->cb_done(r->cb_aux); r->cb_dump = NULL; @@ -422,6 +440,7 @@ remote_create(struct datapath *dp, struct rconn *rconn) list_push_back(&dp->remotes, &remote->node); remote->rconn = rconn; remote->cb_dump = NULL; + remote->n_txq = 0; return remote; } @@ -463,8 +482,8 @@ 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); } } @@ -498,7 +517,7 @@ dp_destroy(struct datapath *dp) * "flood" argument is set, don't send out ports with flooding disabled. */ static int -output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood) +output_all(struct datapath *dp, struct ofpbuf *buffer, int in_port, int flood) { struct sw_port *p; int prev_port; @@ -508,28 +527,29 @@ output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood) if (port_no(dp, p) == in_port) { continue; } - if (flood && p->flags & BRIDGE_PORT_NO_FLOOD) { + if (flood && p->flags & OFPPFL_NO_FLOOD) { continue; } if (prev_port != -1) { - dp_output_port(dp, buffer_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 - buffer_delete(buffer); + ofpbuf_delete(buffer); return 0; } void -output_packet(struct datapath *dp, struct buffer *buffer, int out_port) +output_packet(struct datapath *dp, struct ofpbuf *buffer, int out_port) { if (out_port >= 0 && out_port < OFPP_MAX) { struct sw_port *p = &dp->ports[out_port]; - if (p->netdev != NULL) { + if (p->netdev != NULL && !(p->status & OFPPFL_PORT_DOWN)) { if (!netdev_send(p->netdev, buffer)) { p->tx_packets++; p->tx_bytes += buffer->size; @@ -540,16 +560,15 @@ output_packet(struct datapath *dp, struct buffer *buffer, int out_port) } } - buffer_delete(buffer); - /* FIXME: ratelimit */ - VLOG_DBG("can't forward to bad port %d\n", out_port); + ofpbuf_delete(buffer); + VLOG_DBG_RL(&rl, "can't forward to bad port %d\n", out_port); } /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'. */ void -dp_output_port(struct datapath *dp, struct buffer *buffer, - int in_port, int out_port) +dp_output_port(struct datapath *dp, struct ofpbuf *buffer, + int in_port, int out_port, bool ignore_no_fwd) { assert(buffer); @@ -562,13 +581,13 @@ dp_output_port(struct datapath *dp, struct buffer *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)) { - buffer_delete(buffer); + struct sw_port *p = in_port < OFPP_MAX ? &dp->ports[in_port] : 0; + if (run_flow_through_tables(dp, buffer, p)) { + ofpbuf_delete(buffer); } } else { if (in_port == out_port) { - /* FIXME: ratelimit */ - VLOG_DBG("can't directly forward to input port"); + VLOG_DBG_RL(&rl, "can't directly forward to input port"); return; } output_packet(dp, buffer, out_port); @@ -577,14 +596,14 @@ dp_output_port(struct datapath *dp, struct buffer *buffer, static void * make_openflow_reply(size_t openflow_len, uint8_t type, - const struct sender *sender, struct buffer **bufferp) + const struct sender *sender, struct ofpbuf **bufferp) { return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0, bufferp); } static int -send_openflow_buffer(struct datapath *dp, struct buffer *buffer, +send_openflow_buffer(struct datapath *dp, struct ofpbuf *buffer, const struct sender *sender) { struct remote *remote = sender ? sender->remote : dp->controller; @@ -592,13 +611,10 @@ send_openflow_buffer(struct datapath *dp, struct buffer *buffer, int retval; update_openflow_length(buffer); - retval = (remote->n_txq < TXQ_LIMIT - ? rconn_send(rconn, buffer, &remote->n_txq) - : EAGAIN); + retval = rconn_send_with_limit(rconn, buffer, &remote->n_txq, TXQ_LIMIT); if (retval) { - VLOG_WARN("send to %s failed: %s", - rconn_get_name(rconn), strerror(retval)); - buffer_delete(buffer); + VLOG_WARN_RL(&rl, "send to %s failed: %s", + rconn_get_name(rconn), strerror(retval)); } return retval; } @@ -610,7 +626,7 @@ send_openflow_buffer(struct datapath *dp, struct buffer *buffer, * the caller wants to be sent; a value of 0 indicates the entire packet should * be sent. */ void -dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port, +dp_output_control(struct datapath *dp, struct ofpbuf *buffer, int in_port, size_t max_len, int reason) { struct ofp_packet_in *opi; @@ -619,11 +635,11 @@ dp_output_control(struct datapath *dp, struct buffer *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; } - opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data)); + opi = ofpbuf_push_uninit(buffer, offsetof(struct ofp_packet_in, data)); opi->header.version = OFP_VERSION; opi->header.type = OFPT_PACKET_IN; opi->header.length = htons(buffer->size); @@ -644,30 +660,28 @@ 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 = htonl(p->flags); + desc->flags = 0; desc->features = htonl(netdev_get_features(p->netdev)); desc->speed = htonl(netdev_get_speed(p->netdev)); + desc->flags = htonl(p->flags | p->status); } static void dp_send_features_reply(struct datapath *dp, const struct sender *sender) { - struct buffer *buffer; + struct ofpbuf *buffer; struct ofp_switch_features *ofr; struct sw_port *p; 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 = buffer_put_uninit(buffer, sizeof *opp); + struct ofp_phy_port *opp = ofpbuf_put_uninit(buffer, sizeof *opp); memset(opp, 0, sizeof *opp); fill_port_desc(dp, p, opp); } @@ -675,25 +689,82 @@ dp_send_features_reply(struct datapath *dp, const struct sender *sender) } void -dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp) +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) { struct sw_port *p = &dp->ports[port_no]; + uint32_t flag_mask; /* Make sure the port id hasn't changed since this was sent */ if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN) != 0) { return; } - p->flags = htonl(opp->flags); + + + flag_mask = ntohl(opm->mask) & PORT_FLAG_BITS; + if (flag_mask) { + p->flags &= ~flag_mask; + p->flags |= ntohl(opp->flags) & flag_mask; + } + + if (opm->mask & htonl(OFPPFL_PORT_DOWN)) { + if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) + && (p->status & OFPPFL_PORT_DOWN) == 0) { + p->status |= OFPPFL_PORT_DOWN; + netdev_turn_flags_off(p->netdev, NETDEV_UP, true); + } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0 + && (p->status & OFPPFL_PORT_DOWN)) { + p->status &= ~OFPPFL_PORT_DOWN; + netdev_turn_flags_on(p->netdev, NETDEV_UP, true); + } + } + } +} + +/* Update the port status field of the bridge port. A non-zero return + * value indicates some field has changed. + * + * NB: Callers of this function may hold the RCU read lock, so any + * additional checks must not sleep. + */ +static int +update_port_status(struct sw_port *p) +{ + int retval; + enum netdev_flags flags; + uint32_t orig_status = p->status; + + if (netdev_get_flags(p->netdev, &flags) < 0) { + VLOG_WARN_RL(&rl, "could not get netdev flags for %s", + netdev_get_name(p->netdev)); + return 0; + } else { + if (flags & NETDEV_UP) { + p->status &= ~OFPPFL_PORT_DOWN; + } else { + p->status |= OFPPFL_PORT_DOWN; + } } + + /* Not all cards support this getting link status, so don't warn on + * error. */ + retval = netdev_get_link_status(p->netdev); + if (retval == 1) { + p->status &= ~OFPPFL_LINK_DOWN; + } else if (retval == 0) { + p->status |= OFPPFL_LINK_DOWN; + } + + return (orig_status != p->status); } static void send_port_status(struct sw_port *p, uint8_t status) { - struct buffer *buffer; + struct ofpbuf *buffer; struct ofp_port_status *ops; ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer); ops->reason = status; @@ -707,7 +778,7 @@ void send_flow_expired(struct datapath *dp, struct sw_flow *flow, enum ofp_flow_expired_reason reason) { - struct buffer *buffer; + struct ofpbuf *buffer; struct ofp_flow_expired *ofe; ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer); flow_fill_match(&ofe->match, &flow->key); @@ -725,12 +796,11 @@ 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 buffer *buffer; + 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); @@ -738,12 +808,12 @@ dp_send_error_msg(struct datapath *dp, const struct sender *sender, } static void -fill_flow_stats(struct buffer *buffer, struct sw_flow *flow, +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 = buffer_put_uninit(buffer, length); + int length = sizeof *ofs + sizeof *ofs->actions * flow->sf_acts->n_actions; + ofs = ofpbuf_put_uninit(buffer, length); ofs->length = htons(length); ofs->table_id = table_idx; ofs->pad = 0; @@ -766,66 +836,75 @@ fill_flow_stats(struct buffer *buffer, struct sw_flow *flow, 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, + sizeof *ofs->actions * flow->sf_acts->n_actions); } -/* '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 buffer *buffer, - int in_port) +int run_flow_through_tables(struct datapath *dp, struct ofpbuf *buffer, + 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. */ - buffer_delete(buffer); + ofpbuf_delete(buffer); return 0; } + if (p && p->flags & (OFPPFL_NO_RECV | OFPPFL_NO_RECV_STP) + && p->flags & (!eth_addr_equals(key.flow.dl_dst, stp_eth_addr) + ? OFPPFL_NO_RECV : OFPPFL_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, port_no(dp, p), + &key, flow->sf_acts->actions, + flow->sf_acts->n_actions, 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 buffer *buffer, int in_port) +/* '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) { - if (run_flow_through_tables(dp, buffer, in_port)) { - dp_output_control(dp, buffer, in_port, dp->miss_send_len, - OFPR_NO_MATCH); + if (run_flow_through_tables(dp, buffer, p)) { + dp_output_control(dp, buffer, port_no(dp, p), + dp->miss_send_len, OFPR_NO_MATCH); } } static void -do_output(struct datapath *dp, struct buffer *buffer, int in_port, - size_t max_len, int out_port) +do_output(struct datapath *dp, struct ofpbuf *buffer, int in_port, + size_t max_len, int out_port, bool ignore_no_fwd) { if (out_port != OFPP_CONTROLLER) { - dp_output_port(dp, buffer, in_port, out_port); + dp_output_port(dp, buffer, in_port, out_port, ignore_no_fwd); } else { dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION); } } static void -execute_actions(struct datapath *dp, struct buffer *buffer, +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) + const struct ofp_action *actions, int n_actions, + bool ignore_no_fwd) { /* 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 @@ -844,7 +923,8 @@ execute_actions(struct datapath *dp, struct buffer *buffer, struct eth_header *eh = buffer->l2; if (prev_port != -1) { - do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port); + do_output(dp, ofpbuf_clone(buffer), in_port, max_len, prev_port, + ignore_no_fwd); prev_port = -1; } @@ -881,12 +961,12 @@ execute_actions(struct datapath *dp, struct buffer *buffer, } } if (prev_port != -1) - do_output(dp, buffer, in_port, max_len, prev_port); + do_output(dp, buffer, in_port, max_len, prev_port, ignore_no_fwd); else - buffer_delete(buffer); + ofpbuf_delete(buffer); } -static void modify_nh(struct buffer *buffer, uint16_t eth_proto, +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) { @@ -912,7 +992,7 @@ static void modify_nh(struct buffer *buffer, uint16_t eth_proto, } } -static void modify_th(struct buffer *buffer, uint16_t eth_proto, +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) { @@ -935,7 +1015,7 @@ static void modify_th(struct buffer *buffer, uint16_t eth_proto, } static void -modify_vlan(struct buffer *buffer, +modify_vlan(struct ofpbuf *buffer, const struct sw_flow_key *key, const struct ofp_action *a) { uint16_t new_id = a->arg.vlan_id; @@ -957,9 +1037,9 @@ modify_vlan(struct buffer *buffer, tmp.veth_tci = new_id; tmp.veth_next_type = eh->eth_type; - veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN); + veh = ofpbuf_push_uninit(buffer, VLAN_HEADER_LEN); memcpy(veh, &tmp, sizeof tmp); - buffer->l2 -= VLAN_HEADER_LEN; + buffer->l2 = (char*)buffer->l2 - VLAN_HEADER_LEN; } } else { /* Remove an existing vlan header if it exists */ @@ -972,8 +1052,8 @@ modify_vlan(struct buffer *buffer, tmp.eth_type = veh->veth_next_type; buffer->size -= VLAN_HEADER_LEN; - buffer->data += VLAN_HEADER_LEN; - buffer->l2 += 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); } } @@ -991,7 +1071,7 @@ static int recv_get_config_request(struct datapath *dp, const struct sender *sender, const void *msg) { - struct buffer *buffer; + struct ofpbuf *buffer; struct ofp_switch_config *osc; osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY, @@ -1026,20 +1106,20 @@ recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED, { const struct ofp_packet_out *opo = msg; struct sw_flow_key key; - struct buffer *buffer; + struct ofpbuf *buffer; int n_actions = ntohs(opo->n_actions); int act_len = n_actions * sizeof opo->actions[0]; if (act_len > (ntohs(opo->header.length) - sizeof *opo)) { - VLOG_DBG("message too short for number of actions"); + 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; - buffer = buffer_new(data_len); - buffer_put(buffer, &opo->actions[n_actions], data_len); + buffer = ofpbuf_new(data_len); + ofpbuf_put(buffer, &opo->actions[n_actions], data_len); } else { buffer = retrieve_buffer(ntohl(opo->buffer_id)); if (!buffer) { @@ -1049,7 +1129,7 @@ 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); + &key, opo->actions, n_actions, true); return 0; } @@ -1060,7 +1140,7 @@ recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED, { const struct ofp_port_mod *opm = msg; - dp_update_port_flags(dp, &opm->desc); + dp_update_port_flags(dp, opm); return 0; } @@ -1102,10 +1182,11 @@ add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm) 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->n_actions = n_actions; 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, + n_actions * sizeof *flow->sf_acts->actions); /* Act. */ error = chain_insert(dp->chain, flow); @@ -1114,13 +1195,14 @@ add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm) } error = 0; if (ntohl(ofm->buffer_id) != UINT32_MAX) { - struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id)); + struct ofpbuf *buffer = retrieve_buffer(ntohl(ofm->buffer_id)); if (buffer) { struct sw_flow_key key; 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, in_port, &key, + ofm->actions, n_actions, false); } else { error = -ESRCH; } @@ -1135,6 +1217,59 @@ error: return error; } +static int +mod_flow(struct datapath *dp, const struct ofp_flow_mod *ofm) +{ + int error = -ENOMEM; + int n_actions; + int i; + struct sw_flow_key key; + uint16_t priority; + int strict; + + + /* 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; + } + } + + flow_extract_match(&key, &ofm->match); + 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, n_actions); + + 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, in_port, &skb_key, + ofm->actions, n_actions, 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 UNUSED, const void *msg) @@ -1144,6 +1279,8 @@ recv_flow(struct datapath *dp, const struct sender *sender UNUSED, if (command == OFPFC_ADD) { return add_flow(dp, ofm); + } else if ((command == OFPFC_MODIFY) || (command == OFPFC_MODIFY_STRICT)) { + return mod_flow(dp, ofm); } else if (command == OFPFC_DELETE) { struct sw_flow_key key; flow_extract_match(&key, &ofm->match); @@ -1159,14 +1296,15 @@ recv_flow(struct datapath *dp, const struct sender *sender UNUSED, } } -static int version_stats_dump(struct datapath *dp, void *state, - struct buffer *buffer) +static int desc_stats_dump(struct datapath *dp, void *state, + struct ofpbuf *buffer) { - struct ofp_version_stats *ovs = buffer_put_uninit(buffer, sizeof *ovs); + struct ofp_desc_stats *ods = ofpbuf_put_uninit(buffer, sizeof *ods); - strncpy(ovs->mfr_desc, &mfr_desc, sizeof ovs->mfr_desc); - strncpy(ovs->hw_desc, &hw_desc, sizeof ovs->hw_desc); - strncpy(ovs->sw_desc, &sw_desc, sizeof ovs->sw_desc); + strncpy(ods->mfr_desc, &mfr_desc, sizeof ods->mfr_desc); + strncpy(ods->hw_desc, &hw_desc, sizeof ods->hw_desc); + strncpy(ods->sw_desc, &sw_desc, sizeof ods->sw_desc); + strncpy(ods->serial_num, &serial_num, sizeof ods->serial_num); return 0; } @@ -1177,7 +1315,7 @@ struct flow_stats_state { struct ofp_flow_stats_request rq; time_t now; - struct buffer *buffer; + struct ofpbuf *buffer; }; #define MAX_FLOW_STATS_BYTES 4096 @@ -1202,7 +1340,7 @@ static int flow_stats_dump_callback(struct sw_flow *flow, void *private) } static int flow_stats_dump(struct datapath *dp, void *state, - struct buffer *buffer) + struct ofpbuf *buffer) { struct flow_stats_state *s = state; struct sw_flow_key match_key; @@ -1255,7 +1393,7 @@ static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private) } static int aggregate_stats_dump(struct datapath *dp, void *state, - struct buffer *buffer) + struct ofpbuf *buffer) { struct aggregate_stats_state *s = state; struct ofp_aggregate_stats_request *rq = &s->rq; @@ -1264,7 +1402,7 @@ static int aggregate_stats_dump(struct datapath *dp, void *state, struct sw_flow_key match_key; int table_idx; - rpy = buffer_put_uninit(buffer, sizeof *rpy); + rpy = ofpbuf_put_uninit(buffer, sizeof *rpy); memset(rpy, 0, sizeof *rpy); flow_extract_match(&match_key, &rq->match); @@ -1297,18 +1435,20 @@ static void aggregate_stats_done(void *state) } static int table_stats_dump(struct datapath *dp, void *state, - struct buffer *buffer) + struct ofpbuf *buffer) { int i; for (i = 0; i < dp->chain->n_tables; i++) { - struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots); + struct ofp_table_stats *ots = ofpbuf_put_uninit(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; + ots->wildcards = htonl(stats.wildcards); memset(ots->pad, 0, sizeof ots->pad); 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; @@ -1328,7 +1468,7 @@ static int port_stats_init(struct datapath *dp, const void *body, int body_len, } static int port_stats_dump(struct datapath *dp, void *state, - struct buffer *buffer) + struct ofpbuf *buffer) { struct port_stats_state *s = state; int i; @@ -1339,7 +1479,7 @@ static int port_stats_dump(struct datapath *dp, void *state, if (!p->netdev) { continue; } - ops = buffer_put_uninit(buffer, sizeof *ops); + ops = ofpbuf_put_uninit(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); @@ -1366,6 +1506,9 @@ static void port_stats_done(void *state) } struct stats_type { + /* Value for 'type' member of struct ofp_stats_request. */ + int type; + /* Minimum and maximum acceptable number of bytes in body member of * struct ofp_stats_request. */ size_t min_body, max_body; @@ -1382,7 +1525,7 @@ struct stats_type { * struct ofp_stats_reply. On success, it should return 1 if it should be * called again later with another buffer, 0 if it is done, or a negative * errno value on failure. */ - int (*dump)(struct datapath *dp, void *state, struct buffer *buffer); + int (*dump)(struct datapath *dp, void *state, struct ofpbuf *buffer); /* Cleans any state created by the init or dump functions. May be null * if no cleanup is required. */ @@ -1390,35 +1533,40 @@ struct stats_type { }; static const struct stats_type stats[] = { - [OFPST_VERSION] = { + { + OFPST_DESC, 0, 0, NULL, - version_stats_dump, + desc_stats_dump, NULL }, - [OFPST_FLOW] = { + { + OFPST_FLOW, sizeof(struct ofp_flow_stats_request), sizeof(struct ofp_flow_stats_request), flow_stats_init, flow_stats_dump, flow_stats_done }, - [OFPST_AGGREGATE] = { + { + OFPST_AGGREGATE, sizeof(struct ofp_aggregate_stats_request), sizeof(struct ofp_aggregate_stats_request), aggregate_stats_init, aggregate_stats_dump, aggregate_stats_done }, - [OFPST_TABLE] = { + { + OFPST_TABLE, 0, 0, NULL, table_stats_dump, NULL }, - [OFPST_PORT] = { + { + OFPST_PORT, 0, 0, port_stats_init, @@ -1440,7 +1588,7 @@ stats_dump(struct datapath *dp, void *cb_) { struct stats_dump_cb *cb = cb_; struct ofp_stats_reply *osr; - struct buffer *buffer; + struct ofpbuf *buffer; int err; if (cb->done) { @@ -1449,7 +1597,7 @@ stats_dump(struct datapath *dp, void *cb_) osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender, &buffer); - osr->type = htons(cb->s - stats); + osr->type = htons(cb->s->type); osr->flags = 0; err = cb->s->dump(dp, cb->state, buffer); @@ -1459,7 +1607,7 @@ stats_dump(struct datapath *dp, void *cb_) cb->done = true; } else { /* Buffer might have been reallocated, so find our data again. */ - osr = buffer_at_assert(buffer, 0, sizeof *osr); + osr = ofpbuf_at_assert(buffer, 0, sizeof *osr); osr->flags = ntohs(OFPSF_REPLY_MORE); } err2 = send_openflow_buffer(dp, buffer, &cb->sender); @@ -1489,27 +1637,33 @@ recv_stats_request(struct datapath *dp, const struct sender *sender, { const struct ofp_stats_request *rq = oh; size_t rq_len = ntohs(rq->header.length); + const struct stats_type *st; struct stats_dump_cb *cb; int type, body_len; int err; type = ntohs(rq->type); - if (type >= ARRAY_SIZE(stats) || !stats[type].dump) { - VLOG_WARN("received stats request of unknown type %d", type); - return -EINVAL; + for (st = stats; ; st++) { + if (st >= &stats[ARRAY_SIZE(stats)]) { + VLOG_WARN_RL(&rl, "received stats request of unknown type %d", + type); + return -EINVAL; + } else if (type == st->type) { + break; + } } cb = xmalloc(sizeof *cb); cb->done = false; cb->rq = xmemdup(rq, rq_len); cb->sender = *sender; - cb->s = &stats[type]; + cb->s = st; cb->state = NULL; body_len = rq_len - offsetof(struct ofp_stats_request, body); if (body_len < cb->s->min_body || body_len > cb->s->max_body) { - VLOG_WARN("stats request type %d with bad body length %d", - type, body_len); + VLOG_WARN_RL(&rl, "stats request type %d with bad body length %d", + type, body_len); err = -EINVAL; goto error; } @@ -1517,8 +1671,9 @@ recv_stats_request(struct datapath *dp, const struct sender *sender, if (cb->s->init) { err = cb->s->init(dp, rq->body, body_len, &cb->state); if (err) { - VLOG_WARN("failed initialization of stats request type %d: %s", - type, strerror(-err)); + VLOG_WARN_RL(&rl, + "failed initialization of stats request type %d: %s", + type, strerror(-err)); goto error; } } @@ -1552,65 +1707,65 @@ int fwd_control_input(struct datapath *dp, const struct sender *sender, const void *msg, size_t length) { - struct openflow_packet { - size_t min_size; - int (*handler)(struct datapath *, const struct sender *, const void *); - }; - - static const struct openflow_packet packets[] = { - [OFPT_FEATURES_REQUEST] = { - sizeof (struct ofp_header), - recv_features_request, - }, - [OFPT_GET_CONFIG_REQUEST] = { - sizeof (struct ofp_header), - recv_get_config_request, - }, - [OFPT_SET_CONFIG] = { - sizeof (struct ofp_switch_config), - recv_set_config, - }, - [OFPT_PACKET_OUT] = { - sizeof (struct ofp_packet_out), - recv_packet_out, - }, - [OFPT_FLOW_MOD] = { - sizeof (struct ofp_flow_mod), - recv_flow, - }, - [OFPT_PORT_MOD] = { - sizeof (struct ofp_port_mod), - recv_port_mod, - }, - [OFPT_STATS_REQUEST] = { - sizeof (struct ofp_stats_request), - recv_stats_request, - }, - [OFPT_ECHO_REQUEST] = { - sizeof (struct ofp_header), - recv_echo_request, - }, - [OFPT_ECHO_REPLY] = { - sizeof (struct ofp_header), - recv_echo_reply, - }, - }; - - const struct openflow_packet *pkt; + int (*handler)(struct datapath *, const struct sender *, const void *); struct ofp_header *oh; + size_t min_size; + /* Check encapsulated length. */ oh = (struct ofp_header *) msg; + if (ntohs(oh->length) > length) { + return -EINVAL; + } assert(oh->version == OFP_VERSION); - if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length) + + /* Figure out how to handle it. */ + switch (oh->type) { + case OFPT_FEATURES_REQUEST: + min_size = sizeof(struct ofp_header); + handler = recv_features_request; + break; + case OFPT_GET_CONFIG_REQUEST: + min_size = sizeof(struct ofp_header); + handler = recv_get_config_request; + break; + case OFPT_SET_CONFIG: + min_size = sizeof(struct ofp_switch_config); + handler = recv_set_config; + break; + case OFPT_PACKET_OUT: + min_size = sizeof(struct ofp_packet_out); + handler = recv_packet_out; + break; + case OFPT_FLOW_MOD: + min_size = sizeof(struct ofp_flow_mod); + handler = recv_flow; + break; + case OFPT_PORT_MOD: + min_size = sizeof(struct ofp_port_mod); + handler = recv_port_mod; + break; + case OFPT_STATS_REQUEST: + min_size = sizeof(struct ofp_stats_request); + handler = recv_stats_request; + break; + case OFPT_ECHO_REQUEST: + min_size = sizeof(struct ofp_header); + handler = recv_echo_request; + break; + case OFPT_ECHO_REPLY: + min_size = sizeof(struct ofp_header); + handler = recv_echo_reply; + break; + default: + dp_send_error_msg(dp, sender, OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE, + msg, length); return -EINVAL; + } - pkt = &packets[oh->type]; - if (!pkt->handler) - return -ENOSYS; - if (length < pkt->min_size) + /* Handle it. */ + if (length < min_size) return -EFAULT; - - return pkt->handler(dp, sender, msg); + return handler(dp, sender, msg); } /* Packet buffering. */ @@ -1618,7 +1773,7 @@ fwd_control_input(struct datapath *dp, const struct sender *sender, #define OVERWRITE_SECS 1 struct packet_buffer { - struct buffer *buffer; + struct ofpbuf *buffer; uint32_t cookie; time_t timeout; }; @@ -1626,7 +1781,7 @@ struct packet_buffer { static struct packet_buffer buffers[N_PKT_BUFFERS]; static unsigned int buffer_idx; -uint32_t save_buffer(struct buffer *buffer) +uint32_t save_buffer(struct ofpbuf *buffer) { struct packet_buffer *p; uint32_t id; @@ -1639,23 +1794,23 @@ uint32_t save_buffer(struct buffer *buffer) if (time_now() < p->timeout) { /* FIXME */ return -1; } else { - buffer_delete(p->buffer); + ofpbuf_delete(p->buffer); } } /* Don't use maximum cookie value since the all-bits-1 id is * special. */ if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1) p->cookie = 0; - p->buffer = buffer_clone(buffer); /* FIXME */ + p->buffer = ofpbuf_clone(buffer); /* FIXME */ p->timeout = time_now() + OVERWRITE_SECS; /* FIXME */ id = buffer_idx | (p->cookie << PKT_BUFFER_BITS); return id; } -static struct buffer *retrieve_buffer(uint32_t id) +static struct ofpbuf *retrieve_buffer(uint32_t id) { - struct buffer *buffer = NULL; + struct ofpbuf *buffer = NULL; struct packet_buffer *p; p = &buffers[id & PKT_BUFFER_MASK]; @@ -1676,7 +1831,7 @@ static void discard_buffer(uint32_t id) p = &buffers[id & PKT_BUFFER_MASK]; if (p->cookie == id >> PKT_BUFFER_BITS) { - buffer_delete(p->buffer); + ofpbuf_delete(p->buffer); p->buffer = NULL; } }