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>
47 #include "poll-loop.h"
54 #define THIS_MODULE VLM_datapath
58 BRPF_NO_FLOOD = 1 << 0,
62 BRPS_PORT_DOWN = 1 << 0,
63 BRPS_LINK_DOWN = 1 << 1,
69 extern char serial_num;
71 /* Capabilities supported by this implementation. */
72 #define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
77 /* Actions supported by this implementation. */
78 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
79 | (1 << OFPAT_SET_DL_VLAN) \
80 | (1 << OFPAT_SET_DL_SRC) \
81 | (1 << OFPAT_SET_DL_DST) \
82 | (1 << OFPAT_SET_NW_SRC) \
83 | (1 << OFPAT_SET_NW_DST) \
84 | (1 << OFPAT_SET_TP_SRC) \
85 | (1 << OFPAT_SET_TP_DST) )
88 uint32_t flags; /* BRPF_* flags */
89 uint32_t status; /* BRPS_* flags */
91 struct netdev *netdev;
92 struct list node; /* Element in datapath.ports. */
93 unsigned long long int rx_packets, tx_packets;
94 unsigned long long int rx_bytes, tx_bytes;
95 unsigned long long int tx_dropped;
98 /* The origin of a received OpenFlow message, to enable sending a reply. */
100 struct remote *remote; /* The device that sent the message. */
101 uint32_t xid; /* The OpenFlow transaction ID. */
104 /* A connection to a controller or a management device. */
108 #define TXQ_LIMIT 128 /* Max number of packets to queue for tx. */
109 int n_txq; /* Number of packets queued for tx on rconn. */
111 /* Support for reliable, multi-message replies to requests.
113 * If an incoming request needs to have a reliable reply that might
114 * require multiple messages, it can use remote_start_dump() to set up
115 * a callback that will be called as buffer space for replies. */
116 int (*cb_dump)(struct datapath *, void *aux);
117 void (*cb_done)(void *aux);
122 /* Remote connections. */
123 struct remote *controller; /* Connection to controller. */
124 struct list remotes; /* All connections (including controller). */
125 struct vconn *listen_vconn;
129 /* Unique identifier for this datapath */
132 struct sw_chain *chain; /* Forwarding rules. */
134 /* Configuration set from controller. */
136 uint16_t miss_send_len;
139 struct sw_port ports[OFPP_MAX];
140 struct list port_list; /* List of ports, for flooding. */
143 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
145 static struct remote *remote_create(struct datapath *, struct rconn *);
146 static void remote_run(struct datapath *, struct remote *);
147 static void remote_wait(struct remote *);
148 static void remote_destroy(struct remote *);
150 void dp_output_port(struct datapath *, struct buffer *,
151 int in_port, int out_port);
152 void dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm);
153 void dp_output_control(struct datapath *, struct buffer *, int in_port,
154 size_t max_len, int reason);
155 static void send_flow_expired(struct datapath *, struct sw_flow *,
156 enum ofp_flow_expired_reason);
157 static int update_port_status(struct sw_port *p);
158 static void send_port_status(struct sw_port *p, uint8_t status);
159 static void del_switch_port(struct sw_port *p);
160 static void execute_actions(struct datapath *, struct buffer *,
161 int in_port, const struct sw_flow_key *,
162 const struct ofp_action *, int n_actions);
163 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
164 const struct ofp_action *a);
165 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
166 uint8_t nw_proto, const struct ofp_action *a);
167 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
168 uint8_t nw_proto, const struct ofp_action *a);
170 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
171 * into a buffer number (low bits) and a cookie (high bits). The buffer number
172 * is an index into an array of buffers. The cookie distinguishes between
173 * different packets that have occupied a single buffer. Thus, the more
174 * buffers we have, the lower-quality the cookie... */
175 #define PKT_BUFFER_BITS 8
176 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
177 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
179 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
181 int run_flow_through_tables(struct datapath *, struct buffer *, int in_port);
182 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
183 int fwd_control_input(struct datapath *, const struct sender *,
184 const void *, size_t);
186 uint32_t save_buffer(struct buffer *);
187 static struct buffer *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_vconn = 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_vconn(struct datapath *dp, struct vconn *listen_vconn)
285 assert(!dp->listen_vconn);
286 dp->listen_vconn = listen_vconn;
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 buffer *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 = buffer_new(headroom + hard_header + mtu);
328 buffer->data += headroom;
330 error = netdev_recv(p->netdev, buffer);
333 p->rx_bytes += buffer->size;
334 fwd_port_input(dp, buffer, port_no(dp, 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 buffer_delete(buffer);
343 /* Talk to remotes. */
344 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
347 if (dp->listen_vconn) {
349 struct vconn *new_vconn;
352 retval = vconn_accept(dp->listen_vconn, &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 buffer *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 buffer_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;
446 /* Starts a callback-based, reliable, possibly multi-message reply to a
447 * request made by 'remote'.
449 * 'dump' designates a function that will be called when the 'remote' send
450 * queue has an empty slot. It should compose a message and send it on
451 * 'remote'. On success, it should return 1 if it should be called again when
452 * another send queue slot opens up, 0 if its transmissions are complete, or a
453 * negative errno value on failure.
455 * 'done' designates a function to clean up any resources allocated for the
456 * dump. It must handle being called before the dump is complete (which will
457 * happen if 'remote' is closed unexpectedly).
459 * 'aux' is passed to 'dump' and 'done'. */
461 remote_start_dump(struct remote *remote,
462 int (*dump)(struct datapath *, void *),
463 void (*done)(void *),
466 assert(!remote->cb_dump);
467 remote->cb_dump = dump;
468 remote->cb_done = done;
469 remote->cb_aux = aux;
473 dp_wait(struct datapath *dp)
478 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
479 netdev_recv_wait(p->netdev);
481 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
484 if (dp->listen_vconn) {
485 vconn_accept_wait(dp->listen_vconn);
489 /* Delete 'p' from switch. */
491 del_switch_port(struct sw_port *p)
493 send_port_status(p, OFPPR_DELETE);
494 netdev_close(p->netdev);
496 list_remove(&p->node);
500 dp_destroy(struct datapath *dp)
502 struct sw_port *p, *n;
508 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
511 chain_destroy(dp->chain);
515 /* Send packets out all the ports except the originating one. If the
516 * "flood" argument is set, don't send out ports with flooding disabled.
519 output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
525 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
526 if (port_no(dp, p) == in_port) {
529 if (flood && p->flags & BRPF_NO_FLOOD) {
532 if (prev_port != -1) {
533 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
535 prev_port = port_no(dp, p);
538 dp_output_port(dp, buffer, in_port, prev_port);
540 buffer_delete(buffer);
546 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
548 if (out_port >= 0 && out_port < OFPP_MAX) {
549 struct sw_port *p = &dp->ports[out_port];
550 if (p->netdev != NULL && !(p->status & BRPS_PORT_DOWN)) {
551 if (!netdev_send(p->netdev, buffer)) {
553 p->tx_bytes += buffer->size;
561 buffer_delete(buffer);
562 VLOG_DBG_RL(&rl, "can't forward to bad port %d\n", out_port);
565 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
568 dp_output_port(struct datapath *dp, struct buffer *buffer,
569 int in_port, int out_port)
573 if (out_port == OFPP_FLOOD) {
574 output_all(dp, buffer, in_port, 1);
575 } else if (out_port == OFPP_ALL) {
576 output_all(dp, buffer, in_port, 0);
577 } else if (out_port == OFPP_CONTROLLER) {
578 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
579 } else if (out_port == OFPP_IN_PORT) {
580 output_packet(dp, buffer, in_port);
581 } else if (out_port == OFPP_TABLE) {
582 if (run_flow_through_tables(dp, buffer, in_port)) {
583 buffer_delete(buffer);
586 if (in_port == out_port) {
587 VLOG_DBG_RL(&rl, "can't directly forward to input port");
590 output_packet(dp, buffer, out_port);
595 make_openflow_reply(size_t openflow_len, uint8_t type,
596 const struct sender *sender, struct buffer **bufferp)
598 return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
603 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
604 const struct sender *sender)
606 struct remote *remote = sender ? sender->remote : dp->controller;
607 struct rconn *rconn = remote->rconn;
610 update_openflow_length(buffer);
611 retval = (remote->n_txq < TXQ_LIMIT
612 ? rconn_send(rconn, buffer, &remote->n_txq)
615 VLOG_WARN_RL(&rl, "send to %s failed: %s",
616 rconn_get_name(rconn), strerror(retval));
617 buffer_delete(buffer);
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 buffer *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 && buffer->size > max_len) {
639 buffer->size = max_len;
642 opi = buffer_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));
667 if (p->flags & BRPF_NO_FLOOD) {
668 desc->flags |= htonl(OFPPFL_NO_FLOOD);
669 } else if (p->status & BRPS_PORT_DOWN) {
670 desc->flags |= htonl(OFPPFL_PORT_DOWN);
671 } else if (p->status & BRPS_LINK_DOWN) {
672 desc->flags |= htonl(OFPPFL_LINK_DOWN);
677 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
679 struct buffer *buffer;
680 struct ofp_switch_features *ofr;
683 ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
685 ofr->datapath_id = htonll(dp->id);
686 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
687 ofr->n_compression = 0; /* Not supported */
688 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
689 ofr->buffer_mb = htonl(UINT32_MAX);
690 ofr->n_buffers = htonl(N_PKT_BUFFERS);
691 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
692 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
693 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
694 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
695 memset(opp, 0, sizeof *opp);
696 fill_port_desc(dp, p, opp);
698 send_openflow_buffer(dp, buffer, sender);
702 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
704 const struct ofp_phy_port *opp = &opm->desc;
705 int port_no = ntohs(opp->port_no);
706 if (port_no < OFPP_MAX) {
707 struct sw_port *p = &dp->ports[port_no];
709 /* Make sure the port id hasn't changed since this was sent */
710 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
711 ETH_ADDR_LEN) != 0) {
716 if (opm->mask & htonl(OFPPFL_NO_FLOOD)) {
717 if (opp->flags & htonl(OFPPFL_NO_FLOOD))
718 p->flags |= BRPF_NO_FLOOD;
720 p->flags &= ~BRPF_NO_FLOOD;
723 if (opm->mask & htonl(OFPPFL_PORT_DOWN)) {
724 if ((opp->flags & htonl(OFPPFL_PORT_DOWN))
725 && (p->status & BRPS_PORT_DOWN) == 0) {
726 p->status |= BRPS_PORT_DOWN;
727 netdev_turn_flags_off(p->netdev, NETDEV_UP, true);
728 } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0
729 && (p->status & BRPS_PORT_DOWN)) {
730 p->status &= ~BRPS_PORT_DOWN;
731 netdev_turn_flags_on(p->netdev, NETDEV_UP, true);
737 /* Update the port status field of the bridge port. A non-zero return
738 * value indicates some field has changed.
740 * NB: Callers of this function may hold the RCU read lock, so any
741 * additional checks must not sleep.
744 update_port_status(struct sw_port *p)
747 enum netdev_flags flags;
748 uint32_t orig_status = p->status;
750 if (netdev_get_flags(p->netdev, &flags) < 0) {
751 VLOG_WARN_RL(&rl, "could not get netdev flags for %s",
752 netdev_get_name(p->netdev));
755 if (flags & NETDEV_UP) {
756 p->status &= ~BRPS_PORT_DOWN;
758 p->status |= BRPS_PORT_DOWN;
762 /* Not all cards support this getting link status, so don't warn on
764 retval = netdev_get_link_status(p->netdev);
766 p->status &= ~BRPS_LINK_DOWN;
767 } else if (retval == 0) {
768 p->status |= BRPS_LINK_DOWN;
771 return (orig_status != p->status);
775 send_port_status(struct sw_port *p, uint8_t status)
777 struct buffer *buffer;
778 struct ofp_port_status *ops;
779 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
780 ops->reason = status;
781 memset(ops->pad, 0, sizeof ops->pad);
782 fill_port_desc(p->dp, p, &ops->desc);
784 send_openflow_buffer(p->dp, buffer, NULL);
788 send_flow_expired(struct datapath *dp, struct sw_flow *flow,
789 enum ofp_flow_expired_reason reason)
791 struct buffer *buffer;
792 struct ofp_flow_expired *ofe;
793 ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
794 flow_fill_match(&ofe->match, &flow->key);
796 ofe->priority = htons(flow->priority);
797 ofe->reason = reason;
798 memset(ofe->pad, 0, sizeof ofe->pad);
800 ofe->duration = htonl(time_now() - flow->created);
801 memset(ofe->pad2, 0, sizeof ofe->pad2);
802 ofe->packet_count = htonll(flow->packet_count);
803 ofe->byte_count = htonll(flow->byte_count);
804 send_openflow_buffer(dp, buffer, NULL);
808 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
809 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
811 struct buffer *buffer;
812 struct ofp_error_msg *oem;
813 oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG,
815 oem->type = htons(type);
816 oem->code = htons(code);
817 memcpy(oem->data, data, len);
818 send_openflow_buffer(dp, buffer, sender);
822 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
823 int table_idx, time_t now)
825 struct ofp_flow_stats *ofs;
826 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
827 ofs = buffer_put_uninit(buffer, length);
828 ofs->length = htons(length);
829 ofs->table_id = table_idx;
831 ofs->match.wildcards = htonl(flow->key.wildcards);
832 ofs->match.in_port = flow->key.flow.in_port;
833 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
834 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
835 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
836 ofs->match.dl_type = flow->key.flow.dl_type;
837 ofs->match.nw_src = flow->key.flow.nw_src;
838 ofs->match.nw_dst = flow->key.flow.nw_dst;
839 ofs->match.nw_proto = flow->key.flow.nw_proto;
841 ofs->match.tp_src = flow->key.flow.tp_src;
842 ofs->match.tp_dst = flow->key.flow.tp_dst;
843 ofs->duration = htonl(now - flow->created);
844 ofs->priority = htons(flow->priority);
845 ofs->idle_timeout = htons(flow->idle_timeout);
846 ofs->hard_timeout = htons(flow->hard_timeout);
847 memset(ofs->pad2, 0, sizeof ofs->pad2);
848 ofs->packet_count = htonll(flow->packet_count);
849 ofs->byte_count = htonll(flow->byte_count);
850 memcpy(ofs->actions, flow->actions,
851 sizeof *ofs->actions * flow->n_actions);
855 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
856 * OFPP_MAX. Process it according to 'dp''s flow table. Returns 0 if
857 * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no
858 * matching flow, in which case 'buffer' still belongs to the caller. */
859 int run_flow_through_tables(struct datapath *dp, struct buffer *buffer,
862 struct sw_flow_key key;
863 struct sw_flow *flow;
866 if (flow_extract(buffer, in_port, &key.flow)
867 && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
869 buffer_delete(buffer);
873 flow = chain_lookup(dp->chain, &key);
875 flow_used(flow, buffer);
876 execute_actions(dp, buffer, in_port, &key,
877 flow->actions, flow->n_actions);
884 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
885 * OFPP_MAX. Process it according to 'dp''s flow table, sending it up to the
886 * controller if no flow matches. Takes ownership of 'buffer'. */
887 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
889 if (run_flow_through_tables(dp, buffer, in_port)) {
890 dp_output_control(dp, buffer, in_port, dp->miss_send_len,
896 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
897 size_t max_len, int out_port)
899 if (out_port != OFPP_CONTROLLER) {
900 dp_output_port(dp, buffer, in_port, out_port);
902 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
907 execute_actions(struct datapath *dp, struct buffer *buffer,
908 int in_port, const struct sw_flow_key *key,
909 const struct ofp_action *actions, int n_actions)
911 /* Every output action needs a separate clone of 'buffer', but the common
912 * case is just a single output action, so that doing a clone and then
913 * freeing the original buffer is wasteful. So the following code is
914 * slightly obscure just to avoid that. */
916 size_t max_len=0; /* Initialze to make compiler happy */
921 eth_proto = ntohs(key->flow.dl_type);
923 for (i = 0; i < n_actions; i++) {
924 const struct ofp_action *a = &actions[i];
925 struct eth_header *eh = buffer->l2;
927 if (prev_port != -1) {
928 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
932 switch (ntohs(a->type)) {
934 prev_port = ntohs(a->arg.output.port);
935 max_len = ntohs(a->arg.output.max_len);
938 case OFPAT_SET_DL_VLAN:
939 modify_vlan(buffer, key, a);
942 case OFPAT_SET_DL_SRC:
943 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
946 case OFPAT_SET_DL_DST:
947 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
950 case OFPAT_SET_NW_SRC:
951 case OFPAT_SET_NW_DST:
952 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
955 case OFPAT_SET_TP_SRC:
956 case OFPAT_SET_TP_DST:
957 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
965 do_output(dp, buffer, in_port, max_len, prev_port);
967 buffer_delete(buffer);
970 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
971 uint8_t nw_proto, const struct ofp_action *a)
973 if (eth_proto == ETH_TYPE_IP) {
974 struct ip_header *nh = buffer->l3;
975 uint32_t new, *field;
977 new = a->arg.nw_addr;
978 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
979 if (nw_proto == IP_TYPE_TCP) {
980 struct tcp_header *th = buffer->l4;
981 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
982 } else if (nw_proto == IP_TYPE_UDP) {
983 struct udp_header *th = buffer->l4;
985 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
987 th->udp_csum = 0xffff;
991 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
996 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
997 uint8_t nw_proto, const struct ofp_action *a)
999 if (eth_proto == ETH_TYPE_IP) {
1000 uint16_t new, *field;
1004 if (nw_proto == IP_TYPE_TCP) {
1005 struct tcp_header *th = buffer->l4;
1006 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
1007 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
1009 } else if (nw_proto == IP_TYPE_UDP) {
1010 struct udp_header *th = buffer->l4;
1011 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
1012 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
1019 modify_vlan(struct buffer *buffer,
1020 const struct sw_flow_key *key, const struct ofp_action *a)
1022 uint16_t new_id = a->arg.vlan_id;
1023 struct vlan_eth_header *veh;
1025 if (new_id != htons(OFP_VLAN_NONE)) {
1026 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
1027 /* Modify vlan id, but maintain other TCI values */
1029 veh->veth_tci &= ~htons(VLAN_VID);
1030 veh->veth_tci |= new_id;
1032 /* Insert new vlan id. */
1033 struct eth_header *eh = buffer->l2;
1034 struct vlan_eth_header tmp;
1035 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
1036 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
1037 tmp.veth_type = htons(ETH_TYPE_VLAN);
1038 tmp.veth_tci = new_id;
1039 tmp.veth_next_type = eh->eth_type;
1041 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
1042 memcpy(veh, &tmp, sizeof tmp);
1043 buffer->l2 -= VLAN_HEADER_LEN;
1046 /* Remove an existing vlan header if it exists */
1048 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
1049 struct eth_header tmp;
1051 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1052 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1053 tmp.eth_type = veh->veth_next_type;
1055 buffer->size -= VLAN_HEADER_LEN;
1056 buffer->data += VLAN_HEADER_LEN;
1057 buffer->l2 += VLAN_HEADER_LEN;
1058 memcpy(buffer->data, &tmp, sizeof tmp);
1064 recv_features_request(struct datapath *dp, const struct sender *sender,
1067 dp_send_features_reply(dp, sender);
1072 recv_get_config_request(struct datapath *dp, const struct sender *sender,
1075 struct buffer *buffer;
1076 struct ofp_switch_config *osc;
1078 osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
1081 osc->flags = htons(dp->flags);
1082 osc->miss_send_len = htons(dp->miss_send_len);
1084 return send_openflow_buffer(dp, buffer, sender);
1088 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
1091 const struct ofp_switch_config *osc = msg;
1094 flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
1095 if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
1096 && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
1097 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
1100 dp->miss_send_len = ntohs(osc->miss_send_len);
1105 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1108 const struct ofp_packet_out *opo = msg;
1109 struct sw_flow_key key;
1110 struct buffer *buffer;
1111 int n_actions = ntohs(opo->n_actions);
1112 int act_len = n_actions * sizeof opo->actions[0];
1114 if (act_len > (ntohs(opo->header.length) - sizeof *opo)) {
1115 VLOG_DBG_RL(&rl, "message too short for number of actions");
1119 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1120 /* FIXME: can we avoid copying data here? */
1121 int data_len = ntohs(opo->header.length) - sizeof *opo - act_len;
1122 buffer = buffer_new(data_len);
1123 buffer_put(buffer, &opo->actions[n_actions], data_len);
1125 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1131 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1132 execute_actions(dp, buffer, ntohs(opo->in_port),
1133 &key, opo->actions, n_actions);
1139 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1142 const struct ofp_port_mod *opm = msg;
1144 dp_update_port_flags(dp, opm);
1150 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1152 int error = -ENOMEM;
1155 struct sw_flow *flow;
1158 /* To prevent loops, make sure there's no action to send to the
1159 * OFP_TABLE virtual port.
1161 n_actions = (ntohs(ofm->header.length) - sizeof *ofm)
1162 / sizeof *ofm->actions;
1163 for (i=0; i<n_actions; i++) {
1164 const struct ofp_action *a = &ofm->actions[i];
1166 if (a->type == htons(OFPAT_OUTPUT)
1167 && (a->arg.output.port == htons(OFPP_TABLE)
1168 || a->arg.output.port == htons(OFPP_NONE)
1169 || a->arg.output.port == ofm->match.in_port)) {
1170 /* xxx Send fancy new error message? */
1175 /* Allocate memory. */
1176 flow = flow_alloc(n_actions);
1180 /* Fill out flow. */
1181 flow_extract_match(&flow->key, &ofm->match);
1182 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1183 flow->idle_timeout = ntohs(ofm->idle_timeout);
1184 flow->hard_timeout = ntohs(ofm->hard_timeout);
1185 flow->used = flow->created = time_now();
1186 flow->n_actions = n_actions;
1187 flow->byte_count = 0;
1188 flow->packet_count = 0;
1189 memcpy(flow->actions, ofm->actions, n_actions * sizeof *flow->actions);
1192 error = chain_insert(dp->chain, flow);
1194 goto error_free_flow;
1197 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1198 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1200 struct sw_flow_key key;
1201 uint16_t in_port = ntohs(ofm->match.in_port);
1202 flow_used(flow, buffer);
1203 flow_extract(buffer, in_port, &key.flow);
1204 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_actions);
1214 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1215 discard_buffer(ntohl(ofm->buffer_id));
1220 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1223 const struct ofp_flow_mod *ofm = msg;
1224 uint16_t command = ntohs(ofm->command);
1226 if (command == OFPFC_ADD) {
1227 return add_flow(dp, ofm);
1228 } else if (command == OFPFC_DELETE) {
1229 struct sw_flow_key key;
1230 flow_extract_match(&key, &ofm->match);
1231 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1232 } else if (command == OFPFC_DELETE_STRICT) {
1233 struct sw_flow_key key;
1235 flow_extract_match(&key, &ofm->match);
1236 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1237 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1243 static int desc_stats_dump(struct datapath *dp, void *state,
1244 struct buffer *buffer)
1246 struct ofp_desc_stats *ods = buffer_put_uninit(buffer, sizeof *ods);
1248 strncpy(ods->mfr_desc, &mfr_desc, sizeof ods->mfr_desc);
1249 strncpy(ods->hw_desc, &hw_desc, sizeof ods->hw_desc);
1250 strncpy(ods->sw_desc, &sw_desc, sizeof ods->sw_desc);
1251 strncpy(ods->serial_num, &serial_num, sizeof ods->serial_num);
1256 struct flow_stats_state {
1258 struct sw_table_position position;
1259 struct ofp_flow_stats_request rq;
1262 struct buffer *buffer;
1265 #define MAX_FLOW_STATS_BYTES 4096
1267 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1270 const struct ofp_flow_stats_request *fsr = body;
1271 struct flow_stats_state *s = xmalloc(sizeof *s);
1272 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1273 memset(&s->position, 0, sizeof s->position);
1279 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1281 struct flow_stats_state *s = private;
1282 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1283 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1286 static int flow_stats_dump(struct datapath *dp, void *state,
1287 struct buffer *buffer)
1289 struct flow_stats_state *s = state;
1290 struct sw_flow_key match_key;
1292 flow_extract_match(&match_key, &s->rq.match);
1294 s->now = time_now();
1295 while (s->table_idx < dp->chain->n_tables
1296 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1298 struct sw_table *table = dp->chain->tables[s->table_idx];
1300 if (table->iterate(table, &match_key, &s->position,
1301 flow_stats_dump_callback, s))
1305 memset(&s->position, 0, sizeof s->position);
1307 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1310 static void flow_stats_done(void *state)
1315 struct aggregate_stats_state {
1316 struct ofp_aggregate_stats_request rq;
1319 static int aggregate_stats_init(struct datapath *dp,
1320 const void *body, int body_len,
1323 const struct ofp_aggregate_stats_request *rq = body;
1324 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1330 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1332 struct ofp_aggregate_stats_reply *rpy = private;
1333 rpy->packet_count += flow->packet_count;
1334 rpy->byte_count += flow->byte_count;
1339 static int aggregate_stats_dump(struct datapath *dp, void *state,
1340 struct buffer *buffer)
1342 struct aggregate_stats_state *s = state;
1343 struct ofp_aggregate_stats_request *rq = &s->rq;
1344 struct ofp_aggregate_stats_reply *rpy;
1345 struct sw_table_position position;
1346 struct sw_flow_key match_key;
1349 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1350 memset(rpy, 0, sizeof *rpy);
1352 flow_extract_match(&match_key, &rq->match);
1353 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1354 memset(&position, 0, sizeof position);
1355 while (table_idx < dp->chain->n_tables
1356 && (rq->table_id == 0xff || rq->table_id == table_idx))
1358 struct sw_table *table = dp->chain->tables[table_idx];
1361 error = table->iterate(table, &match_key, &position,
1362 aggregate_stats_dump_callback, rpy);
1367 memset(&position, 0, sizeof position);
1370 rpy->packet_count = htonll(rpy->packet_count);
1371 rpy->byte_count = htonll(rpy->byte_count);
1372 rpy->flow_count = htonl(rpy->flow_count);
1376 static void aggregate_stats_done(void *state)
1381 static int table_stats_dump(struct datapath *dp, void *state,
1382 struct buffer *buffer)
1385 for (i = 0; i < dp->chain->n_tables; i++) {
1386 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1387 struct sw_table_stats stats;
1388 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1389 strncpy(ots->name, stats.name, sizeof ots->name);
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 buffer *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 = buffer_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 /* Minimum and maximum acceptable number of bytes in body member of
1452 * struct ofp_stats_request. */
1453 size_t min_body, max_body;
1455 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1456 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1457 * Returns zero if successful, otherwise a negative error code.
1458 * May initialize '*state' to state information. May be null if no
1459 * initialization is required.*/
1460 int (*init)(struct datapath *dp, const void *body, int body_len,
1463 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1464 * struct ofp_stats_reply. On success, it should return 1 if it should be
1465 * called again later with another buffer, 0 if it is done, or a negative
1466 * errno value on failure. */
1467 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1469 /* Cleans any state created by the init or dump functions. May be null
1470 * if no cleanup is required. */
1471 void (*done)(void *state);
1474 static const struct stats_type stats[] = {
1483 sizeof(struct ofp_flow_stats_request),
1484 sizeof(struct ofp_flow_stats_request),
1489 [OFPST_AGGREGATE] = {
1490 sizeof(struct ofp_aggregate_stats_request),
1491 sizeof(struct ofp_aggregate_stats_request),
1492 aggregate_stats_init,
1493 aggregate_stats_dump,
1494 aggregate_stats_done
1512 struct stats_dump_cb {
1514 struct ofp_stats_request *rq;
1515 struct sender sender;
1516 const struct stats_type *s;
1521 stats_dump(struct datapath *dp, void *cb_)
1523 struct stats_dump_cb *cb = cb_;
1524 struct ofp_stats_reply *osr;
1525 struct buffer *buffer;
1532 osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1534 osr->type = htons(cb->s - stats);
1537 err = cb->s->dump(dp, cb->state, buffer);
1543 /* Buffer might have been reallocated, so find our data again. */
1544 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1545 osr->flags = ntohs(OFPSF_REPLY_MORE);
1547 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1557 stats_done(void *cb_)
1559 struct stats_dump_cb *cb = cb_;
1562 cb->s->done(cb->state);
1569 recv_stats_request(struct datapath *dp, const struct sender *sender,
1572 const struct ofp_stats_request *rq = oh;
1573 size_t rq_len = ntohs(rq->header.length);
1574 struct stats_dump_cb *cb;
1578 type = ntohs(rq->type);
1579 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1580 VLOG_WARN_RL(&rl, "received stats request of unknown type %d", type);
1584 cb = xmalloc(sizeof *cb);
1586 cb->rq = xmemdup(rq, rq_len);
1587 cb->sender = *sender;
1588 cb->s = &stats[type];
1591 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1592 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1593 VLOG_WARN_RL(&rl, "stats request type %d with bad body length %d",
1600 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1603 "failed initialization of stats request type %d: %s",
1604 type, strerror(-err));
1609 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1619 recv_echo_request(struct datapath *dp, const struct sender *sender,
1622 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1626 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1627 const void *oh UNUSED)
1632 /* 'msg', which is 'length' bytes long, was received from the control path.
1633 * Apply it to 'chain'. */
1635 fwd_control_input(struct datapath *dp, const struct sender *sender,
1636 const void *msg, size_t length)
1638 struct openflow_packet {
1640 int (*handler)(struct datapath *, const struct sender *, const void *);
1643 static const struct openflow_packet packets[] = {
1644 [OFPT_FEATURES_REQUEST] = {
1645 sizeof (struct ofp_header),
1646 recv_features_request,
1648 [OFPT_GET_CONFIG_REQUEST] = {
1649 sizeof (struct ofp_header),
1650 recv_get_config_request,
1652 [OFPT_SET_CONFIG] = {
1653 sizeof (struct ofp_switch_config),
1656 [OFPT_PACKET_OUT] = {
1657 sizeof (struct ofp_packet_out),
1661 sizeof (struct ofp_flow_mod),
1665 sizeof (struct ofp_port_mod),
1668 [OFPT_STATS_REQUEST] = {
1669 sizeof (struct ofp_stats_request),
1672 [OFPT_ECHO_REQUEST] = {
1673 sizeof (struct ofp_header),
1676 [OFPT_ECHO_REPLY] = {
1677 sizeof (struct ofp_header),
1682 const struct openflow_packet *pkt;
1683 struct ofp_header *oh;
1685 oh = (struct ofp_header *) msg;
1686 assert(oh->version == OFP_VERSION);
1687 if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length)
1690 pkt = &packets[oh->type];
1693 if (length < pkt->min_size)
1696 return pkt->handler(dp, sender, msg);
1699 /* Packet buffering. */
1701 #define OVERWRITE_SECS 1
1703 struct packet_buffer {
1704 struct buffer *buffer;
1709 static struct packet_buffer buffers[N_PKT_BUFFERS];
1710 static unsigned int buffer_idx;
1712 uint32_t save_buffer(struct buffer *buffer)
1714 struct packet_buffer *p;
1717 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1718 p = &buffers[buffer_idx];
1720 /* Don't buffer packet if existing entry is less than
1721 * OVERWRITE_SECS old. */
1722 if (time_now() < p->timeout) { /* FIXME */
1725 buffer_delete(p->buffer);
1728 /* Don't use maximum cookie value since the all-bits-1 id is
1730 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1732 p->buffer = buffer_clone(buffer); /* FIXME */
1733 p->timeout = time_now() + OVERWRITE_SECS; /* FIXME */
1734 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1739 static struct buffer *retrieve_buffer(uint32_t id)
1741 struct buffer *buffer = NULL;
1742 struct packet_buffer *p;
1744 p = &buffers[id & PKT_BUFFER_MASK];
1745 if (p->cookie == id >> PKT_BUFFER_BITS) {
1749 printf("cookie mismatch: %x != %x\n",
1750 id >> PKT_BUFFER_BITS, p->cookie);
1756 static void discard_buffer(uint32_t id)
1758 struct packet_buffer *p;
1760 p = &buffers[id & PKT_BUFFER_MASK];
1761 if (p->cookie == id >> PKT_BUFFER_BITS) {
1762 buffer_delete(p->buffer);