1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
35 #include <arpa/inet.h>
49 #include "poll-loop.h"
52 #include "switch-flow.h"
58 #define THIS_MODULE VLM_datapath
64 extern char serial_num;
66 /* Capabilities supported by this implementation. */
67 #define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
72 /* Actions supported by this implementation. */
73 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
74 | (1 << OFPAT_SET_DL_VLAN) \
75 | (1 << OFPAT_SET_DL_SRC) \
76 | (1 << OFPAT_SET_DL_DST) \
77 | (1 << OFPAT_SET_NW_SRC) \
78 | (1 << OFPAT_SET_NW_DST) \
79 | (1 << OFPAT_SET_TP_SRC) \
80 | (1 << OFPAT_SET_TP_DST) )
82 #define PORT_STATUS_BITS (OFPPFL_PORT_DOWN | OFPPFL_LINK_DOWN)
83 #define PORT_FLAG_BITS (~PORT_STATUS_BITS)
86 uint32_t flags; /* Some subset of PORT_FLAG_BITS. */
87 uint32_t status; /* Some subset of PORT_STATUS_BITS. */
89 struct netdev *netdev;
90 struct list node; /* Element in datapath.ports. */
91 unsigned long long int rx_packets, tx_packets;
92 unsigned long long int rx_bytes, tx_bytes;
93 unsigned long long int tx_dropped;
96 /* The origin of a received OpenFlow message, to enable sending a reply. */
98 struct remote *remote; /* The device that sent the message. */
99 uint32_t xid; /* The OpenFlow transaction ID. */
102 /* A connection to a controller or a management device. */
106 #define TXQ_LIMIT 128 /* Max number of packets to queue for tx. */
107 int n_txq; /* Number of packets queued for tx on rconn. */
109 /* Support for reliable, multi-message replies to requests.
111 * If an incoming request needs to have a reliable reply that might
112 * require multiple messages, it can use remote_start_dump() to set up
113 * a callback that will be called as buffer space for replies. */
114 int (*cb_dump)(struct datapath *, void *aux);
115 void (*cb_done)(void *aux);
120 /* Remote connections. */
121 struct remote *controller; /* Connection to controller. */
122 struct list remotes; /* All connections (including controller). */
123 struct pvconn *listen_pvconn;
127 /* Unique identifier for this datapath */
130 struct sw_chain *chain; /* Forwarding rules. */
132 /* Configuration set from controller. */
134 uint16_t miss_send_len;
137 struct sw_port ports[OFPP_MAX];
138 struct list port_list; /* List of ports, for flooding. */
141 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
143 static struct remote *remote_create(struct datapath *, struct rconn *);
144 static void remote_run(struct datapath *, struct remote *);
145 static void remote_wait(struct remote *);
146 static void remote_destroy(struct remote *);
148 void dp_output_port(struct datapath *, struct ofpbuf *,
149 int in_port, int out_port, bool ignore_no_fwd);
150 void dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm);
151 void dp_output_control(struct datapath *, struct ofpbuf *, int in_port,
152 size_t max_len, int reason);
153 static void send_flow_expired(struct datapath *, struct sw_flow *,
154 enum ofp_flow_expired_reason);
155 static int update_port_status(struct sw_port *p);
156 static void send_port_status(struct sw_port *p, uint8_t status);
157 static void del_switch_port(struct sw_port *p);
158 static void execute_actions(struct datapath *, struct ofpbuf *,
159 int in_port, const struct sw_flow_key *,
160 const struct ofp_action *, int n_actions,
162 static void modify_vlan(struct ofpbuf *buffer, const struct sw_flow_key *key,
163 const struct ofp_action *a);
164 static void modify_nh(struct ofpbuf *buffer, uint16_t eth_proto,
165 uint8_t nw_proto, const struct ofp_action *a);
166 static void modify_th(struct ofpbuf *buffer, uint16_t eth_proto,
167 uint8_t nw_proto, const struct ofp_action *a);
169 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
170 * into a buffer number (low bits) and a cookie (high bits). The buffer number
171 * is an index into an array of buffers. The cookie distinguishes between
172 * different packets that have occupied a single buffer. Thus, the more
173 * buffers we have, the lower-quality the cookie... */
174 #define PKT_BUFFER_BITS 8
175 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
176 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
178 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
180 int run_flow_through_tables(struct datapath *, struct ofpbuf *,
182 void fwd_port_input(struct datapath *, struct ofpbuf *, struct sw_port *);
183 int fwd_control_input(struct datapath *, const struct sender *,
184 const void *, size_t);
186 uint32_t save_buffer(struct ofpbuf *);
187 static struct ofpbuf *retrieve_buffer(uint32_t id);
188 static void discard_buffer(uint32_t id);
190 static int port_no(struct datapath *dp, struct sw_port *p)
192 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
193 return p - dp->ports;
196 /* Generates and returns a random datapath id. */
198 gen_datapath_id(void)
200 uint8_t ea[ETH_ADDR_LEN];
202 return eth_addr_to_uint64(ea);
206 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
210 dp = calloc(1, sizeof *dp);
215 dp->last_timeout = time_now();
216 list_init(&dp->remotes);
217 dp->controller = remote_create(dp, rconn);
218 dp->listen_pvconn = NULL;
219 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
220 dp->chain = chain_create();
222 VLOG_ERR("could not create chain");
227 list_init(&dp->port_list);
229 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
235 dp_add_port(struct datapath *dp, const char *name)
237 struct netdev *netdev;
243 error = netdev_open(name, NETDEV_ETH_TYPE_ANY, &netdev);
247 error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false);
249 VLOG_ERR("couldn't set promiscuous mode on %s device", name);
250 netdev_close(netdev);
253 if (netdev_get_in4(netdev, &in4)) {
254 VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
256 if (netdev_get_in6(netdev, &in6)) {
257 char in6_name[INET6_ADDRSTRLEN + 1];
258 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
259 VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
262 for (p = dp->ports; ; p++) {
263 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
265 } else if (!p->netdev) {
270 memset(p, '\0', sizeof *p);
274 list_push_back(&dp->port_list, &p->node);
276 /* Notify the ctlpath that this port has been added */
277 send_port_status(p, OFPPR_ADD);
283 dp_add_listen_pvconn(struct datapath *dp, struct pvconn *listen_pvconn)
285 assert(!dp->listen_pvconn);
286 dp->listen_pvconn = listen_pvconn;
290 dp_run(struct datapath *dp)
292 time_t now = time_now();
293 struct sw_port *p, *pn;
294 struct remote *r, *rn;
295 struct ofpbuf *buffer = NULL;
297 if (now != dp->last_timeout) {
298 struct list deleted = LIST_INITIALIZER(&deleted);
299 struct sw_flow *f, *n;
301 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
302 if (update_port_status(p)) {
303 send_port_status(p, OFPPR_MOD);
307 chain_timeout(dp->chain, &deleted);
308 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
309 send_flow_expired(dp, f, f->reason);
310 list_remove(&f->node);
313 dp->last_timeout = now;
315 poll_timer_wait(1000);
317 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
321 /* Allocate buffer with some headroom to add headers in forwarding
322 * to the controller or adding a vlan tag, plus an extra 2 bytes to
323 * allow IP headers to be aligned on a 4-byte boundary. */
324 const int headroom = 128 + 2;
325 const int hard_header = VLAN_ETH_HEADER_LEN;
326 const int mtu = netdev_get_mtu(p->netdev);
327 buffer = ofpbuf_new(headroom + hard_header + mtu);
328 buffer->data = (char*)buffer->data + headroom;
330 error = netdev_recv(p->netdev, buffer);
333 p->rx_bytes += buffer->size;
334 fwd_port_input(dp, buffer, p);
336 } else if (error != EAGAIN) {
337 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
338 netdev_get_name(p->netdev), strerror(error));
341 ofpbuf_delete(buffer);
343 /* Talk to remotes. */
344 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
347 if (dp->listen_pvconn) {
349 struct vconn *new_vconn;
352 retval = pvconn_accept(dp->listen_pvconn, OFP_VERSION, &new_vconn);
354 if (retval != EAGAIN) {
355 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
359 remote_create(dp, rconn_new_from_vconn("passive", new_vconn));
365 remote_run(struct datapath *dp, struct remote *r)
371 /* Do some remote processing, but cap it at a reasonable amount so that
372 * other processing doesn't starve. */
373 for (i = 0; i < 50; i++) {
375 struct ofpbuf *buffer;
376 struct ofp_header *oh;
378 buffer = rconn_recv(r->rconn);
383 if (buffer->size >= sizeof *oh) {
384 struct sender sender;
388 sender.xid = oh->xid;
389 fwd_control_input(dp, &sender, buffer->data, buffer->size);
391 VLOG_WARN_RL(&rl, "received too-short OpenFlow message");
393 ofpbuf_delete(buffer);
395 if (r->n_txq < TXQ_LIMIT) {
396 int error = r->cb_dump(dp, r->cb_aux);
399 VLOG_WARN_RL(&rl, "dump callback error: %s",
402 r->cb_done(r->cb_aux);
411 if (!rconn_is_alive(r->rconn)) {
417 remote_wait(struct remote *r)
419 rconn_run_wait(r->rconn);
420 rconn_recv_wait(r->rconn);
424 remote_destroy(struct remote *r)
427 if (r->cb_dump && r->cb_done) {
428 r->cb_done(r->cb_aux);
430 list_remove(&r->node);
431 rconn_destroy(r->rconn);
436 static struct remote *
437 remote_create(struct datapath *dp, struct rconn *rconn)
439 struct remote *remote = xmalloc(sizeof *remote);
440 list_push_back(&dp->remotes, &remote->node);
441 remote->rconn = rconn;
442 remote->cb_dump = NULL;
447 /* Starts a callback-based, reliable, possibly multi-message reply to a
448 * request made by 'remote'.
450 * 'dump' designates a function that will be called when the 'remote' send
451 * queue has an empty slot. It should compose a message and send it on
452 * 'remote'. On success, it should return 1 if it should be called again when
453 * another send queue slot opens up, 0 if its transmissions are complete, or a
454 * negative errno value on failure.
456 * 'done' designates a function to clean up any resources allocated for the
457 * dump. It must handle being called before the dump is complete (which will
458 * happen if 'remote' is closed unexpectedly).
460 * 'aux' is passed to 'dump' and 'done'. */
462 remote_start_dump(struct remote *remote,
463 int (*dump)(struct datapath *, void *),
464 void (*done)(void *),
467 assert(!remote->cb_dump);
468 remote->cb_dump = dump;
469 remote->cb_done = done;
470 remote->cb_aux = aux;
474 dp_wait(struct datapath *dp)
479 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
480 netdev_recv_wait(p->netdev);
482 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
485 if (dp->listen_pvconn) {
486 pvconn_wait(dp->listen_pvconn);
490 /* Delete 'p' from switch. */
492 del_switch_port(struct sw_port *p)
494 send_port_status(p, OFPPR_DELETE);
495 netdev_close(p->netdev);
497 list_remove(&p->node);
501 dp_destroy(struct datapath *dp)
503 struct sw_port *p, *n;
509 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
512 chain_destroy(dp->chain);
516 /* Send packets out all the ports except the originating one. If the
517 * "flood" argument is set, don't send out ports with flooding disabled.
520 output_all(struct datapath *dp, struct ofpbuf *buffer, int in_port, int flood)
526 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
527 if (port_no(dp, p) == in_port) {
530 if (flood && p->flags & OFPPFL_NO_FLOOD) {
533 if (prev_port != -1) {
534 dp_output_port(dp, ofpbuf_clone(buffer), in_port, prev_port,
537 prev_port = port_no(dp, p);
540 dp_output_port(dp, buffer, in_port, prev_port, false);
542 ofpbuf_delete(buffer);
548 output_packet(struct datapath *dp, struct ofpbuf *buffer, int out_port)
550 if (out_port >= 0 && out_port < OFPP_MAX) {
551 struct sw_port *p = &dp->ports[out_port];
552 if (p->netdev != NULL && !(p->status & OFPPFL_PORT_DOWN)) {
553 if (!netdev_send(p->netdev, buffer)) {
555 p->tx_bytes += buffer->size;
563 ofpbuf_delete(buffer);
564 VLOG_DBG_RL(&rl, "can't forward to bad port %d\n", out_port);
567 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
570 dp_output_port(struct datapath *dp, struct ofpbuf *buffer,
571 int in_port, int out_port, bool ignore_no_fwd)
575 if (out_port == OFPP_FLOOD) {
576 output_all(dp, buffer, in_port, 1);
577 } else if (out_port == OFPP_ALL) {
578 output_all(dp, buffer, in_port, 0);
579 } else if (out_port == OFPP_CONTROLLER) {
580 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
581 } else if (out_port == OFPP_IN_PORT) {
582 output_packet(dp, buffer, in_port);
583 } else if (out_port == OFPP_TABLE) {
584 struct sw_port *p = in_port < OFPP_MAX ? &dp->ports[in_port] : 0;
585 if (run_flow_through_tables(dp, buffer, p)) {
586 ofpbuf_delete(buffer);
589 if (in_port == out_port) {
590 VLOG_DBG_RL(&rl, "can't directly forward to input port");
593 output_packet(dp, buffer, out_port);
598 make_openflow_reply(size_t openflow_len, uint8_t type,
599 const struct sender *sender, struct ofpbuf **bufferp)
601 return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
606 send_openflow_buffer(struct datapath *dp, struct ofpbuf *buffer,
607 const struct sender *sender)
609 struct remote *remote = sender ? sender->remote : dp->controller;
610 struct rconn *rconn = remote->rconn;
613 update_openflow_length(buffer);
614 retval = rconn_send_with_limit(rconn, buffer, &remote->n_txq, TXQ_LIMIT);
616 VLOG_WARN_RL(&rl, "send to %s failed: %s",
617 rconn_get_name(rconn), strerror(retval));
622 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
623 * packet can be saved in a buffer, then only the first max_len bytes of
624 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
625 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
626 * the caller wants to be sent; a value of 0 indicates the entire packet should
629 dp_output_control(struct datapath *dp, struct ofpbuf *buffer, int in_port,
630 size_t max_len, int reason)
632 struct ofp_packet_in *opi;
636 buffer_id = save_buffer(buffer);
637 total_len = buffer->size;
638 if (buffer_id != UINT32_MAX && max_len && buffer->size > max_len) {
639 buffer->size = max_len;
642 opi = ofpbuf_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
643 opi->header.version = OFP_VERSION;
644 opi->header.type = OFPT_PACKET_IN;
645 opi->header.length = htons(buffer->size);
646 opi->header.xid = htonl(0);
647 opi->buffer_id = htonl(buffer_id);
648 opi->total_len = htons(total_len);
649 opi->in_port = htons(in_port);
650 opi->reason = reason;
652 send_openflow_buffer(dp, buffer, NULL);
655 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
656 struct ofp_phy_port *desc)
658 desc->port_no = htons(port_no(dp, p));
659 strncpy((char *) desc->name, netdev_get_name(p->netdev),
661 desc->name[sizeof desc->name - 1] = '\0';
662 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
664 desc->features = htonl(netdev_get_features(p->netdev));
665 desc->speed = htonl(netdev_get_speed(p->netdev));
666 desc->flags = htonl(p->flags | p->status);
670 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
672 struct ofpbuf *buffer;
673 struct ofp_switch_features *ofr;
676 ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
678 ofr->datapath_id = htonll(dp->id);
679 ofr->n_tables = dp->chain->n_tables;
680 ofr->n_buffers = htonl(N_PKT_BUFFERS);
681 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
682 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
683 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
684 struct ofp_phy_port *opp = ofpbuf_put_uninit(buffer, sizeof *opp);
685 memset(opp, 0, sizeof *opp);
686 fill_port_desc(dp, p, opp);
688 send_openflow_buffer(dp, buffer, sender);
692 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
694 const struct ofp_phy_port *opp = &opm->desc;
695 int port_no = ntohs(opp->port_no);
696 if (port_no < OFPP_MAX) {
697 struct sw_port *p = &dp->ports[port_no];
700 /* Make sure the port id hasn't changed since this was sent */
701 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
702 ETH_ADDR_LEN) != 0) {
707 flag_mask = ntohl(opm->mask) & PORT_FLAG_BITS;
709 p->flags &= ~flag_mask;
710 p->flags |= ntohl(opp->flags) & flag_mask;
713 if (opm->mask & htonl(OFPPFL_PORT_DOWN)) {
714 if ((opp->flags & htonl(OFPPFL_PORT_DOWN))
715 && (p->status & OFPPFL_PORT_DOWN) == 0) {
716 p->status |= OFPPFL_PORT_DOWN;
717 netdev_turn_flags_off(p->netdev, NETDEV_UP, true);
718 } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0
719 && (p->status & OFPPFL_PORT_DOWN)) {
720 p->status &= ~OFPPFL_PORT_DOWN;
721 netdev_turn_flags_on(p->netdev, NETDEV_UP, true);
727 /* Update the port status field of the bridge port. A non-zero return
728 * value indicates some field has changed.
730 * NB: Callers of this function may hold the RCU read lock, so any
731 * additional checks must not sleep.
734 update_port_status(struct sw_port *p)
737 enum netdev_flags flags;
738 uint32_t orig_status = p->status;
740 if (netdev_get_flags(p->netdev, &flags) < 0) {
741 VLOG_WARN_RL(&rl, "could not get netdev flags for %s",
742 netdev_get_name(p->netdev));
745 if (flags & NETDEV_UP) {
746 p->status &= ~OFPPFL_PORT_DOWN;
748 p->status |= OFPPFL_PORT_DOWN;
752 /* Not all cards support this getting link status, so don't warn on
754 retval = netdev_get_link_status(p->netdev);
756 p->status &= ~OFPPFL_LINK_DOWN;
757 } else if (retval == 0) {
758 p->status |= OFPPFL_LINK_DOWN;
761 return (orig_status != p->status);
765 send_port_status(struct sw_port *p, uint8_t status)
767 struct ofpbuf *buffer;
768 struct ofp_port_status *ops;
769 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
770 ops->reason = status;
771 memset(ops->pad, 0, sizeof ops->pad);
772 fill_port_desc(p->dp, p, &ops->desc);
774 send_openflow_buffer(p->dp, buffer, NULL);
778 send_flow_expired(struct datapath *dp, struct sw_flow *flow,
779 enum ofp_flow_expired_reason reason)
781 struct ofpbuf *buffer;
782 struct ofp_flow_expired *ofe;
783 ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
784 flow_fill_match(&ofe->match, &flow->key);
786 ofe->priority = htons(flow->priority);
787 ofe->reason = reason;
788 memset(ofe->pad, 0, sizeof ofe->pad);
790 ofe->duration = htonl(time_now() - flow->created);
791 memset(ofe->pad2, 0, sizeof ofe->pad2);
792 ofe->packet_count = htonll(flow->packet_count);
793 ofe->byte_count = htonll(flow->byte_count);
794 send_openflow_buffer(dp, buffer, NULL);
798 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
799 uint16_t type, uint16_t code, const void *data, size_t len)
801 struct ofpbuf *buffer;
802 struct ofp_error_msg *oem;
803 oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR, sender, &buffer);
804 oem->type = htons(type);
805 oem->code = htons(code);
806 memcpy(oem->data, data, len);
807 send_openflow_buffer(dp, buffer, sender);
811 fill_flow_stats(struct ofpbuf *buffer, struct sw_flow *flow,
812 int table_idx, time_t now)
814 struct ofp_flow_stats *ofs;
815 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
816 ofs = ofpbuf_put_uninit(buffer, length);
817 ofs->length = htons(length);
818 ofs->table_id = table_idx;
820 ofs->match.wildcards = htonl(flow->key.wildcards);
821 ofs->match.in_port = flow->key.flow.in_port;
822 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
823 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
824 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
825 ofs->match.dl_type = flow->key.flow.dl_type;
826 ofs->match.nw_src = flow->key.flow.nw_src;
827 ofs->match.nw_dst = flow->key.flow.nw_dst;
828 ofs->match.nw_proto = flow->key.flow.nw_proto;
830 ofs->match.tp_src = flow->key.flow.tp_src;
831 ofs->match.tp_dst = flow->key.flow.tp_dst;
832 ofs->duration = htonl(now - flow->created);
833 ofs->priority = htons(flow->priority);
834 ofs->idle_timeout = htons(flow->idle_timeout);
835 ofs->hard_timeout = htons(flow->hard_timeout);
836 memset(ofs->pad2, 0, sizeof ofs->pad2);
837 ofs->packet_count = htonll(flow->packet_count);
838 ofs->byte_count = htonll(flow->byte_count);
839 memcpy(ofs->actions, flow->actions,
840 sizeof *ofs->actions * flow->n_actions);
844 /* 'buffer' was received on 'p', which may be a a physical switch port or a
845 * null pointer. Process it according to 'dp''s flow table. Returns 0 if
846 * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no
847 * matching flow, in which case 'buffer' still belongs to the caller. */
848 int run_flow_through_tables(struct datapath *dp, struct ofpbuf *buffer,
851 struct sw_flow_key key;
852 struct sw_flow *flow;
855 if (flow_extract(buffer, p ? port_no(dp, p) : OFPP_NONE, &key.flow)
856 && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
858 ofpbuf_delete(buffer);
861 if (p && p->flags & (OFPPFL_NO_RECV | OFPPFL_NO_RECV_STP)
862 && p->flags & (!eth_addr_equals(key.flow.dl_dst, stp_eth_addr)
863 ? OFPPFL_NO_RECV : OFPPFL_NO_RECV_STP)) {
864 ofpbuf_delete(buffer);
868 flow = chain_lookup(dp->chain, &key);
870 flow_used(flow, buffer);
871 execute_actions(dp, buffer, port_no(dp, p),
872 &key, flow->actions, flow->n_actions, false);
879 /* 'buffer' was received on 'p', which may be a a physical switch port or a
880 * null pointer. Process it according to 'dp''s flow table, sending it up to
881 * the controller if no flow matches. Takes ownership of 'buffer'. */
882 void fwd_port_input(struct datapath *dp, struct ofpbuf *buffer,
885 if (run_flow_through_tables(dp, buffer, p)) {
886 dp_output_control(dp, buffer, port_no(dp, p),
887 dp->miss_send_len, OFPR_NO_MATCH);
892 do_output(struct datapath *dp, struct ofpbuf *buffer, int in_port,
893 size_t max_len, int out_port, bool ignore_no_fwd)
895 if (out_port != OFPP_CONTROLLER) {
896 dp_output_port(dp, buffer, in_port, out_port, ignore_no_fwd);
898 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
903 execute_actions(struct datapath *dp, struct ofpbuf *buffer,
904 int in_port, const struct sw_flow_key *key,
905 const struct ofp_action *actions, int n_actions,
908 /* Every output action needs a separate clone of 'buffer', but the common
909 * case is just a single output action, so that doing a clone and then
910 * freeing the original buffer is wasteful. So the following code is
911 * slightly obscure just to avoid that. */
913 size_t max_len=0; /* Initialze to make compiler happy */
918 eth_proto = ntohs(key->flow.dl_type);
920 for (i = 0; i < n_actions; i++) {
921 const struct ofp_action *a = &actions[i];
922 struct eth_header *eh = buffer->l2;
924 if (prev_port != -1) {
925 do_output(dp, ofpbuf_clone(buffer), in_port, max_len, prev_port,
930 switch (ntohs(a->type)) {
932 prev_port = ntohs(a->arg.output.port);
933 max_len = ntohs(a->arg.output.max_len);
936 case OFPAT_SET_DL_VLAN:
937 modify_vlan(buffer, key, a);
940 case OFPAT_SET_DL_SRC:
941 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
944 case OFPAT_SET_DL_DST:
945 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
948 case OFPAT_SET_NW_SRC:
949 case OFPAT_SET_NW_DST:
950 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
953 case OFPAT_SET_TP_SRC:
954 case OFPAT_SET_TP_DST:
955 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
963 do_output(dp, buffer, in_port, max_len, prev_port, ignore_no_fwd);
965 ofpbuf_delete(buffer);
968 static void modify_nh(struct ofpbuf *buffer, uint16_t eth_proto,
969 uint8_t nw_proto, const struct ofp_action *a)
971 if (eth_proto == ETH_TYPE_IP) {
972 struct ip_header *nh = buffer->l3;
973 uint32_t new, *field;
975 new = a->arg.nw_addr;
976 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
977 if (nw_proto == IP_TYPE_TCP) {
978 struct tcp_header *th = buffer->l4;
979 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
980 } else if (nw_proto == IP_TYPE_UDP) {
981 struct udp_header *th = buffer->l4;
983 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
985 th->udp_csum = 0xffff;
989 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
994 static void modify_th(struct ofpbuf *buffer, uint16_t eth_proto,
995 uint8_t nw_proto, const struct ofp_action *a)
997 if (eth_proto == ETH_TYPE_IP) {
998 uint16_t new, *field;
1002 if (nw_proto == IP_TYPE_TCP) {
1003 struct tcp_header *th = buffer->l4;
1004 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
1005 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
1007 } else if (nw_proto == IP_TYPE_UDP) {
1008 struct udp_header *th = buffer->l4;
1009 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
1010 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
1017 modify_vlan(struct ofpbuf *buffer,
1018 const struct sw_flow_key *key, const struct ofp_action *a)
1020 uint16_t new_id = a->arg.vlan_id;
1021 struct vlan_eth_header *veh;
1023 if (new_id != htons(OFP_VLAN_NONE)) {
1024 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
1025 /* Modify vlan id, but maintain other TCI values */
1027 veh->veth_tci &= ~htons(VLAN_VID);
1028 veh->veth_tci |= new_id;
1030 /* Insert new vlan id. */
1031 struct eth_header *eh = buffer->l2;
1032 struct vlan_eth_header tmp;
1033 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
1034 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
1035 tmp.veth_type = htons(ETH_TYPE_VLAN);
1036 tmp.veth_tci = new_id;
1037 tmp.veth_next_type = eh->eth_type;
1039 veh = ofpbuf_push_uninit(buffer, VLAN_HEADER_LEN);
1040 memcpy(veh, &tmp, sizeof tmp);
1041 buffer->l2 = (char*)buffer->l2 - VLAN_HEADER_LEN;
1044 /* Remove an existing vlan header if it exists */
1046 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
1047 struct eth_header tmp;
1049 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1050 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1051 tmp.eth_type = veh->veth_next_type;
1053 buffer->size -= VLAN_HEADER_LEN;
1054 buffer->data = (char*)buffer->data + VLAN_HEADER_LEN;
1055 buffer->l2 = (char*)buffer->l2 + VLAN_HEADER_LEN;
1056 memcpy(buffer->data, &tmp, sizeof tmp);
1062 recv_features_request(struct datapath *dp, const struct sender *sender,
1065 dp_send_features_reply(dp, sender);
1070 recv_get_config_request(struct datapath *dp, const struct sender *sender,
1073 struct ofpbuf *buffer;
1074 struct ofp_switch_config *osc;
1076 osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
1079 osc->flags = htons(dp->flags);
1080 osc->miss_send_len = htons(dp->miss_send_len);
1082 return send_openflow_buffer(dp, buffer, sender);
1086 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
1089 const struct ofp_switch_config *osc = msg;
1092 flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
1093 if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
1094 && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
1095 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
1098 dp->miss_send_len = ntohs(osc->miss_send_len);
1103 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1106 const struct ofp_packet_out *opo = msg;
1107 struct sw_flow_key key;
1108 struct ofpbuf *buffer;
1109 int n_actions = ntohs(opo->n_actions);
1110 int act_len = n_actions * sizeof opo->actions[0];
1112 if (act_len > (ntohs(opo->header.length) - sizeof *opo)) {
1113 VLOG_DBG_RL(&rl, "message too short for number of actions");
1117 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1118 /* FIXME: can we avoid copying data here? */
1119 int data_len = ntohs(opo->header.length) - sizeof *opo - act_len;
1120 buffer = ofpbuf_new(data_len);
1121 ofpbuf_put(buffer, &opo->actions[n_actions], data_len);
1123 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1129 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1130 execute_actions(dp, buffer, ntohs(opo->in_port),
1131 &key, opo->actions, n_actions, true);
1137 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1140 const struct ofp_port_mod *opm = msg;
1142 dp_update_port_flags(dp, opm);
1148 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1150 int error = -ENOMEM;
1153 struct sw_flow *flow;
1156 /* To prevent loops, make sure there's no action to send to the
1157 * OFP_TABLE virtual port.
1159 n_actions = (ntohs(ofm->header.length) - sizeof *ofm)
1160 / sizeof *ofm->actions;
1161 for (i=0; i<n_actions; i++) {
1162 const struct ofp_action *a = &ofm->actions[i];
1164 if (a->type == htons(OFPAT_OUTPUT)
1165 && (a->arg.output.port == htons(OFPP_TABLE)
1166 || a->arg.output.port == htons(OFPP_NONE)
1167 || a->arg.output.port == ofm->match.in_port)) {
1168 /* xxx Send fancy new error message? */
1173 /* Allocate memory. */
1174 flow = flow_alloc(n_actions);
1178 /* Fill out flow. */
1179 flow_extract_match(&flow->key, &ofm->match);
1180 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1181 flow->idle_timeout = ntohs(ofm->idle_timeout);
1182 flow->hard_timeout = ntohs(ofm->hard_timeout);
1183 flow->used = flow->created = time_now();
1184 flow->n_actions = n_actions;
1185 flow->byte_count = 0;
1186 flow->packet_count = 0;
1187 memcpy(flow->actions, ofm->actions, n_actions * sizeof *flow->actions);
1190 error = chain_insert(dp->chain, flow);
1192 goto error_free_flow;
1195 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1196 struct ofpbuf *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1198 struct sw_flow_key key;
1199 uint16_t in_port = ntohs(ofm->match.in_port);
1200 flow_used(flow, buffer);
1201 flow_extract(buffer, in_port, &key.flow);
1202 execute_actions(dp, buffer, in_port, &key,
1203 ofm->actions, n_actions, false);
1213 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1214 discard_buffer(ntohl(ofm->buffer_id));
1219 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1222 const struct ofp_flow_mod *ofm = msg;
1223 uint16_t command = ntohs(ofm->command);
1225 if (command == OFPFC_ADD) {
1226 return add_flow(dp, ofm);
1227 } else if (command == OFPFC_DELETE) {
1228 struct sw_flow_key key;
1229 flow_extract_match(&key, &ofm->match);
1230 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1231 } else if (command == OFPFC_DELETE_STRICT) {
1232 struct sw_flow_key key;
1234 flow_extract_match(&key, &ofm->match);
1235 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1236 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1242 static int desc_stats_dump(struct datapath *dp, void *state,
1243 struct ofpbuf *buffer)
1245 struct ofp_desc_stats *ods = ofpbuf_put_uninit(buffer, sizeof *ods);
1247 strncpy(ods->mfr_desc, &mfr_desc, sizeof ods->mfr_desc);
1248 strncpy(ods->hw_desc, &hw_desc, sizeof ods->hw_desc);
1249 strncpy(ods->sw_desc, &sw_desc, sizeof ods->sw_desc);
1250 strncpy(ods->serial_num, &serial_num, sizeof ods->serial_num);
1255 struct flow_stats_state {
1257 struct sw_table_position position;
1258 struct ofp_flow_stats_request rq;
1261 struct ofpbuf *buffer;
1264 #define MAX_FLOW_STATS_BYTES 4096
1266 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1269 const struct ofp_flow_stats_request *fsr = body;
1270 struct flow_stats_state *s = xmalloc(sizeof *s);
1271 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1272 memset(&s->position, 0, sizeof s->position);
1278 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1280 struct flow_stats_state *s = private;
1281 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1282 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1285 static int flow_stats_dump(struct datapath *dp, void *state,
1286 struct ofpbuf *buffer)
1288 struct flow_stats_state *s = state;
1289 struct sw_flow_key match_key;
1291 flow_extract_match(&match_key, &s->rq.match);
1293 s->now = time_now();
1294 while (s->table_idx < dp->chain->n_tables
1295 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1297 struct sw_table *table = dp->chain->tables[s->table_idx];
1299 if (table->iterate(table, &match_key, &s->position,
1300 flow_stats_dump_callback, s))
1304 memset(&s->position, 0, sizeof s->position);
1306 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1309 static void flow_stats_done(void *state)
1314 struct aggregate_stats_state {
1315 struct ofp_aggregate_stats_request rq;
1318 static int aggregate_stats_init(struct datapath *dp,
1319 const void *body, int body_len,
1322 const struct ofp_aggregate_stats_request *rq = body;
1323 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1329 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1331 struct ofp_aggregate_stats_reply *rpy = private;
1332 rpy->packet_count += flow->packet_count;
1333 rpy->byte_count += flow->byte_count;
1338 static int aggregate_stats_dump(struct datapath *dp, void *state,
1339 struct ofpbuf *buffer)
1341 struct aggregate_stats_state *s = state;
1342 struct ofp_aggregate_stats_request *rq = &s->rq;
1343 struct ofp_aggregate_stats_reply *rpy;
1344 struct sw_table_position position;
1345 struct sw_flow_key match_key;
1348 rpy = ofpbuf_put_uninit(buffer, sizeof *rpy);
1349 memset(rpy, 0, sizeof *rpy);
1351 flow_extract_match(&match_key, &rq->match);
1352 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1353 memset(&position, 0, sizeof position);
1354 while (table_idx < dp->chain->n_tables
1355 && (rq->table_id == 0xff || rq->table_id == table_idx))
1357 struct sw_table *table = dp->chain->tables[table_idx];
1360 error = table->iterate(table, &match_key, &position,
1361 aggregate_stats_dump_callback, rpy);
1366 memset(&position, 0, sizeof position);
1369 rpy->packet_count = htonll(rpy->packet_count);
1370 rpy->byte_count = htonll(rpy->byte_count);
1371 rpy->flow_count = htonl(rpy->flow_count);
1375 static void aggregate_stats_done(void *state)
1380 static int table_stats_dump(struct datapath *dp, void *state,
1381 struct ofpbuf *buffer)
1384 for (i = 0; i < dp->chain->n_tables; i++) {
1385 struct ofp_table_stats *ots = ofpbuf_put_uninit(buffer, sizeof *ots);
1386 struct sw_table_stats stats;
1387 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1388 strncpy(ots->name, stats.name, sizeof ots->name);
1390 ots->wildcards = htonl(stats.wildcards);
1391 memset(ots->pad, 0, sizeof ots->pad);
1392 ots->max_entries = htonl(stats.max_flows);
1393 ots->active_count = htonl(stats.n_flows);
1394 ots->matched_count = htonll(stats.n_matched);
1399 struct port_stats_state {
1403 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1406 struct port_stats_state *s = xmalloc(sizeof *s);
1412 static int port_stats_dump(struct datapath *dp, void *state,
1413 struct ofpbuf *buffer)
1415 struct port_stats_state *s = state;
1418 for (i = s->port; i < OFPP_MAX; i++) {
1419 struct sw_port *p = &dp->ports[i];
1420 struct ofp_port_stats *ops;
1424 ops = ofpbuf_put_uninit(buffer, sizeof *ops);
1425 ops->port_no = htons(port_no(dp, p));
1426 memset(ops->pad, 0, sizeof ops->pad);
1427 ops->rx_packets = htonll(p->rx_packets);
1428 ops->tx_packets = htonll(p->tx_packets);
1429 ops->rx_bytes = htonll(p->rx_bytes);
1430 ops->tx_bytes = htonll(p->tx_bytes);
1431 ops->rx_dropped = htonll(-1);
1432 ops->tx_dropped = htonll(p->tx_dropped);
1433 ops->rx_errors = htonll(-1);
1434 ops->tx_errors = htonll(-1);
1435 ops->rx_frame_err = htonll(-1);
1436 ops->rx_over_err = htonll(-1);
1437 ops->rx_crc_err = htonll(-1);
1438 ops->collisions = htonll(-1);
1445 static void port_stats_done(void *state)
1451 /* Value for 'type' member of struct ofp_stats_request. */
1454 /* Minimum and maximum acceptable number of bytes in body member of
1455 * struct ofp_stats_request. */
1456 size_t min_body, max_body;
1458 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1459 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1460 * Returns zero if successful, otherwise a negative error code.
1461 * May initialize '*state' to state information. May be null if no
1462 * initialization is required.*/
1463 int (*init)(struct datapath *dp, const void *body, int body_len,
1466 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1467 * struct ofp_stats_reply. On success, it should return 1 if it should be
1468 * called again later with another buffer, 0 if it is done, or a negative
1469 * errno value on failure. */
1470 int (*dump)(struct datapath *dp, void *state, struct ofpbuf *buffer);
1472 /* Cleans any state created by the init or dump functions. May be null
1473 * if no cleanup is required. */
1474 void (*done)(void *state);
1477 static const struct stats_type stats[] = {
1488 sizeof(struct ofp_flow_stats_request),
1489 sizeof(struct ofp_flow_stats_request),
1496 sizeof(struct ofp_aggregate_stats_request),
1497 sizeof(struct ofp_aggregate_stats_request),
1498 aggregate_stats_init,
1499 aggregate_stats_dump,
1500 aggregate_stats_done
1520 struct stats_dump_cb {
1522 struct ofp_stats_request *rq;
1523 struct sender sender;
1524 const struct stats_type *s;
1529 stats_dump(struct datapath *dp, void *cb_)
1531 struct stats_dump_cb *cb = cb_;
1532 struct ofp_stats_reply *osr;
1533 struct ofpbuf *buffer;
1540 osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1542 osr->type = htons(cb->s->type);
1545 err = cb->s->dump(dp, cb->state, buffer);
1551 /* Buffer might have been reallocated, so find our data again. */
1552 osr = ofpbuf_at_assert(buffer, 0, sizeof *osr);
1553 osr->flags = ntohs(OFPSF_REPLY_MORE);
1555 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1565 stats_done(void *cb_)
1567 struct stats_dump_cb *cb = cb_;
1570 cb->s->done(cb->state);
1577 recv_stats_request(struct datapath *dp, const struct sender *sender,
1580 const struct ofp_stats_request *rq = oh;
1581 size_t rq_len = ntohs(rq->header.length);
1582 const struct stats_type *st;
1583 struct stats_dump_cb *cb;
1587 type = ntohs(rq->type);
1588 for (st = stats; ; st++) {
1589 if (st >= &stats[ARRAY_SIZE(stats)]) {
1590 VLOG_WARN_RL(&rl, "received stats request of unknown type %d",
1593 } else if (type == st->type) {
1598 cb = xmalloc(sizeof *cb);
1600 cb->rq = xmemdup(rq, rq_len);
1601 cb->sender = *sender;
1605 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1606 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1607 VLOG_WARN_RL(&rl, "stats request type %d with bad body length %d",
1614 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1617 "failed initialization of stats request type %d: %s",
1618 type, strerror(-err));
1623 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1633 recv_echo_request(struct datapath *dp, const struct sender *sender,
1636 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1640 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1641 const void *oh UNUSED)
1646 /* 'msg', which is 'length' bytes long, was received from the control path.
1647 * Apply it to 'chain'. */
1649 fwd_control_input(struct datapath *dp, const struct sender *sender,
1650 const void *msg, size_t length)
1652 int (*handler)(struct datapath *, const struct sender *, const void *);
1653 struct ofp_header *oh;
1656 /* Check encapsulated length. */
1657 oh = (struct ofp_header *) msg;
1658 if (ntohs(oh->length) > length) {
1661 assert(oh->version == OFP_VERSION);
1663 /* Figure out how to handle it. */
1665 case OFPT_FEATURES_REQUEST:
1666 min_size = sizeof(struct ofp_header);
1667 handler = recv_features_request;
1669 case OFPT_GET_CONFIG_REQUEST:
1670 min_size = sizeof(struct ofp_header);
1671 handler = recv_get_config_request;
1673 case OFPT_SET_CONFIG:
1674 min_size = sizeof(struct ofp_switch_config);
1675 handler = recv_set_config;
1677 case OFPT_PACKET_OUT:
1678 min_size = sizeof(struct ofp_packet_out);
1679 handler = recv_packet_out;
1682 min_size = sizeof(struct ofp_flow_mod);
1683 handler = recv_flow;
1686 min_size = sizeof(struct ofp_port_mod);
1687 handler = recv_port_mod;
1689 case OFPT_STATS_REQUEST:
1690 min_size = sizeof(struct ofp_stats_request);
1691 handler = recv_stats_request;
1693 case OFPT_ECHO_REQUEST:
1694 min_size = sizeof(struct ofp_header);
1695 handler = recv_echo_request;
1697 case OFPT_ECHO_REPLY:
1698 min_size = sizeof(struct ofp_header);
1699 handler = recv_echo_reply;
1702 dp_send_error_msg(dp, sender, OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE,
1708 if (length < min_size)
1710 return handler(dp, sender, msg);
1713 /* Packet buffering. */
1715 #define OVERWRITE_SECS 1
1717 struct packet_buffer {
1718 struct ofpbuf *buffer;
1723 static struct packet_buffer buffers[N_PKT_BUFFERS];
1724 static unsigned int buffer_idx;
1726 uint32_t save_buffer(struct ofpbuf *buffer)
1728 struct packet_buffer *p;
1731 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1732 p = &buffers[buffer_idx];
1734 /* Don't buffer packet if existing entry is less than
1735 * OVERWRITE_SECS old. */
1736 if (time_now() < p->timeout) { /* FIXME */
1739 ofpbuf_delete(p->buffer);
1742 /* Don't use maximum cookie value since the all-bits-1 id is
1744 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1746 p->buffer = ofpbuf_clone(buffer); /* FIXME */
1747 p->timeout = time_now() + OVERWRITE_SECS; /* FIXME */
1748 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1753 static struct ofpbuf *retrieve_buffer(uint32_t id)
1755 struct ofpbuf *buffer = NULL;
1756 struct packet_buffer *p;
1758 p = &buffers[id & PKT_BUFFER_MASK];
1759 if (p->cookie == id >> PKT_BUFFER_BITS) {
1763 printf("cookie mismatch: %x != %x\n",
1764 id >> PKT_BUFFER_BITS, p->cookie);
1770 static void discard_buffer(uint32_t id)
1772 struct packet_buffer *p;
1774 p = &buffers[id & PKT_BUFFER_MASK];
1775 if (p->cookie == id >> PKT_BUFFER_BITS) {
1776 ofpbuf_delete(p->buffer);