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,
70 /* Capabilities supported by this implementation. */
71 #define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
76 /* Actions supported by this implementation. */
77 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
78 | (1 << OFPAT_SET_DL_VLAN) \
79 | (1 << OFPAT_SET_DL_SRC) \
80 | (1 << OFPAT_SET_DL_DST) \
81 | (1 << OFPAT_SET_NW_SRC) \
82 | (1 << OFPAT_SET_NW_DST) \
83 | (1 << OFPAT_SET_TP_SRC) \
84 | (1 << OFPAT_SET_TP_DST) )
87 uint32_t flags; /* BRPF_* flags */
88 uint32_t status; /* BRPS_* flags */
90 struct netdev *netdev;
91 struct list node; /* Element in datapath.ports. */
92 unsigned long long int rx_packets, tx_packets;
93 unsigned long long int rx_bytes, tx_bytes;
94 unsigned long long int tx_dropped;
97 /* The origin of a received OpenFlow message, to enable sending a reply. */
99 struct remote *remote; /* The device that sent the message. */
100 uint32_t xid; /* The OpenFlow transaction ID. */
103 /* A connection to a controller or a management device. */
107 #define TXQ_LIMIT 128 /* Max number of packets to queue for tx. */
108 int n_txq; /* Number of packets queued for tx on rconn. */
110 /* Support for reliable, multi-message replies to requests.
112 * If an incoming request needs to have a reliable reply that might
113 * require multiple messages, it can use remote_start_dump() to set up
114 * a callback that will be called as buffer space for replies. */
115 int (*cb_dump)(struct datapath *, void *aux);
116 void (*cb_done)(void *aux);
121 /* Remote connections. */
122 struct remote *controller; /* Connection to controller. */
123 struct list remotes; /* All connections (including controller). */
124 struct vconn *listen_vconn;
128 /* Unique identifier for this datapath */
131 struct sw_chain *chain; /* Forwarding rules. */
133 /* Configuration set from controller. */
135 uint16_t miss_send_len;
138 struct sw_port ports[OFPP_MAX];
139 struct list port_list; /* List of ports, for flooding. */
142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
144 static struct remote *remote_create(struct datapath *, struct rconn *);
145 static void remote_run(struct datapath *, struct remote *);
146 static void remote_wait(struct remote *);
147 static void remote_destroy(struct remote *);
149 void dp_output_port(struct datapath *, struct buffer *,
150 int in_port, int out_port);
151 void dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm);
152 void dp_output_control(struct datapath *, struct buffer *, int in_port,
153 size_t max_len, int reason);
154 static void send_flow_expired(struct datapath *, struct sw_flow *,
155 enum ofp_flow_expired_reason);
156 static int update_port_status(struct sw_port *p);
157 static void send_port_status(struct sw_port *p, uint8_t status);
158 static void del_switch_port(struct sw_port *p);
159 static void execute_actions(struct datapath *, struct buffer *,
160 int in_port, const struct sw_flow_key *,
161 const struct ofp_action *, int n_actions);
162 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
163 const struct ofp_action *a);
164 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
165 uint8_t nw_proto, const struct ofp_action *a);
166 static void modify_th(struct buffer *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 buffer *, int in_port);
181 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
182 int fwd_control_input(struct datapath *, const struct sender *,
183 const void *, size_t);
185 uint32_t save_buffer(struct buffer *);
186 static struct buffer *retrieve_buffer(uint32_t id);
187 static void discard_buffer(uint32_t id);
189 static int port_no(struct datapath *dp, struct sw_port *p)
191 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
192 return p - dp->ports;
195 /* Generates and returns a random datapath id. */
197 gen_datapath_id(void)
199 uint8_t ea[ETH_ADDR_LEN];
201 return eth_addr_to_uint64(ea);
205 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
209 dp = calloc(1, sizeof *dp);
214 dp->last_timeout = time_now();
215 list_init(&dp->remotes);
216 dp->controller = remote_create(dp, rconn);
217 dp->listen_vconn = NULL;
218 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
219 dp->chain = chain_create();
221 VLOG_ERR("could not create chain");
226 list_init(&dp->port_list);
228 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
234 dp_add_port(struct datapath *dp, const char *name)
236 struct netdev *netdev;
242 error = netdev_open(name, NETDEV_ETH_TYPE_ANY, &netdev);
246 error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false);
248 VLOG_ERR("couldn't set promiscuous mode on %s device", name);
249 netdev_close(netdev);
252 if (netdev_get_in4(netdev, &in4)) {
253 VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
255 if (netdev_get_in6(netdev, &in6)) {
256 char in6_name[INET6_ADDRSTRLEN + 1];
257 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
258 VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
261 for (p = dp->ports; ; p++) {
262 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
264 } else if (!p->netdev) {
269 memset(p, '\0', sizeof *p);
273 list_push_back(&dp->port_list, &p->node);
275 /* Notify the ctlpath that this port has been added */
276 send_port_status(p, OFPPR_ADD);
282 dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
284 assert(!dp->listen_vconn);
285 dp->listen_vconn = listen_vconn;
289 dp_run(struct datapath *dp)
291 time_t now = time_now();
292 struct sw_port *p, *pn;
293 struct remote *r, *rn;
294 struct buffer *buffer = NULL;
296 if (now != dp->last_timeout) {
297 struct list deleted = LIST_INITIALIZER(&deleted);
298 struct sw_flow *f, *n;
300 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
301 if (update_port_status(p)) {
302 send_port_status(p, OFPPR_MOD);
306 chain_timeout(dp->chain, &deleted);
307 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
308 send_flow_expired(dp, f, f->reason);
309 list_remove(&f->node);
312 dp->last_timeout = now;
314 poll_timer_wait(1000);
316 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
320 /* Allocate buffer with some headroom to add headers in forwarding
321 * to the controller or adding a vlan tag, plus an extra 2 bytes to
322 * allow IP headers to be aligned on a 4-byte boundary. */
323 const int headroom = 128 + 2;
324 const int hard_header = VLAN_ETH_HEADER_LEN;
325 const int mtu = netdev_get_mtu(p->netdev);
326 buffer = buffer_new(headroom + hard_header + mtu);
327 buffer->data += headroom;
329 error = netdev_recv(p->netdev, buffer);
332 p->rx_bytes += buffer->size;
333 fwd_port_input(dp, buffer, port_no(dp, p));
335 } else if (error != EAGAIN) {
336 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
337 netdev_get_name(p->netdev), strerror(error));
340 buffer_delete(buffer);
342 /* Talk to remotes. */
343 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
346 if (dp->listen_vconn) {
348 struct vconn *new_vconn;
351 retval = vconn_accept(dp->listen_vconn, &new_vconn);
353 if (retval != EAGAIN) {
354 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
358 remote_create(dp, rconn_new_from_vconn("passive", new_vconn));
364 remote_run(struct datapath *dp, struct remote *r)
370 /* Do some remote processing, but cap it at a reasonable amount so that
371 * other processing doesn't starve. */
372 for (i = 0; i < 50; i++) {
374 struct buffer *buffer;
375 struct ofp_header *oh;
377 buffer = rconn_recv(r->rconn);
382 if (buffer->size >= sizeof *oh) {
383 struct sender sender;
387 sender.xid = oh->xid;
388 fwd_control_input(dp, &sender, buffer->data, buffer->size);
390 VLOG_WARN_RL(&rl, "received too-short OpenFlow message");
392 buffer_delete(buffer);
394 if (r->n_txq < TXQ_LIMIT) {
395 int error = r->cb_dump(dp, r->cb_aux);
398 VLOG_WARN_RL(&rl, "dump callback error: %s",
401 r->cb_done(r->cb_aux);
410 if (!rconn_is_alive(r->rconn)) {
416 remote_wait(struct remote *r)
418 rconn_run_wait(r->rconn);
419 rconn_recv_wait(r->rconn);
423 remote_destroy(struct remote *r)
426 if (r->cb_dump && r->cb_done) {
427 r->cb_done(r->cb_aux);
429 list_remove(&r->node);
430 rconn_destroy(r->rconn);
435 static struct remote *
436 remote_create(struct datapath *dp, struct rconn *rconn)
438 struct remote *remote = xmalloc(sizeof *remote);
439 list_push_back(&dp->remotes, &remote->node);
440 remote->rconn = rconn;
441 remote->cb_dump = NULL;
445 /* Starts a callback-based, reliable, possibly multi-message reply to a
446 * request made by 'remote'.
448 * 'dump' designates a function that will be called when the 'remote' send
449 * queue has an empty slot. It should compose a message and send it on
450 * 'remote'. On success, it should return 1 if it should be called again when
451 * another send queue slot opens up, 0 if its transmissions are complete, or a
452 * negative errno value on failure.
454 * 'done' designates a function to clean up any resources allocated for the
455 * dump. It must handle being called before the dump is complete (which will
456 * happen if 'remote' is closed unexpectedly).
458 * 'aux' is passed to 'dump' and 'done'. */
460 remote_start_dump(struct remote *remote,
461 int (*dump)(struct datapath *, void *),
462 void (*done)(void *),
465 assert(!remote->cb_dump);
466 remote->cb_dump = dump;
467 remote->cb_done = done;
468 remote->cb_aux = aux;
472 dp_wait(struct datapath *dp)
477 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
478 netdev_recv_wait(p->netdev);
480 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
483 if (dp->listen_vconn) {
484 vconn_accept_wait(dp->listen_vconn);
488 /* Delete 'p' from switch. */
490 del_switch_port(struct sw_port *p)
492 send_port_status(p, OFPPR_DELETE);
493 netdev_close(p->netdev);
495 list_remove(&p->node);
499 dp_destroy(struct datapath *dp)
501 struct sw_port *p, *n;
507 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
510 chain_destroy(dp->chain);
514 /* Send packets out all the ports except the originating one. If the
515 * "flood" argument is set, don't send out ports with flooding disabled.
518 output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
524 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
525 if (port_no(dp, p) == in_port) {
528 if (flood && p->flags & BRPF_NO_FLOOD) {
531 if (prev_port != -1) {
532 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
534 prev_port = port_no(dp, p);
537 dp_output_port(dp, buffer, in_port, prev_port);
539 buffer_delete(buffer);
545 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
547 if (out_port >= 0 && out_port < OFPP_MAX) {
548 struct sw_port *p = &dp->ports[out_port];
549 if (p->netdev != NULL && !(p->status & BRPS_PORT_DOWN)) {
550 if (!netdev_send(p->netdev, buffer)) {
552 p->tx_bytes += buffer->size;
560 buffer_delete(buffer);
561 VLOG_DBG_RL(&rl, "can't forward to bad port %d\n", out_port);
564 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
567 dp_output_port(struct datapath *dp, struct buffer *buffer,
568 int in_port, int out_port)
572 if (out_port == OFPP_FLOOD) {
573 output_all(dp, buffer, in_port, 1);
574 } else if (out_port == OFPP_ALL) {
575 output_all(dp, buffer, in_port, 0);
576 } else if (out_port == OFPP_CONTROLLER) {
577 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
578 } else if (out_port == OFPP_IN_PORT) {
579 output_packet(dp, buffer, in_port);
580 } else if (out_port == OFPP_TABLE) {
581 if (run_flow_through_tables(dp, buffer, in_port)) {
582 buffer_delete(buffer);
585 if (in_port == out_port) {
586 VLOG_DBG_RL(&rl, "can't directly forward to input port");
589 output_packet(dp, buffer, out_port);
594 make_openflow_reply(size_t openflow_len, uint8_t type,
595 const struct sender *sender, struct buffer **bufferp)
597 return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
602 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
603 const struct sender *sender)
605 struct remote *remote = sender ? sender->remote : dp->controller;
606 struct rconn *rconn = remote->rconn;
609 update_openflow_length(buffer);
610 retval = (remote->n_txq < TXQ_LIMIT
611 ? rconn_send(rconn, buffer, &remote->n_txq)
614 VLOG_WARN_RL(&rl, "send to %s failed: %s",
615 rconn_get_name(rconn), strerror(retval));
616 buffer_delete(buffer);
621 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
622 * packet can be saved in a buffer, then only the first max_len bytes of
623 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
624 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
625 * the caller wants to be sent; a value of 0 indicates the entire packet should
628 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
629 size_t max_len, int reason)
631 struct ofp_packet_in *opi;
635 buffer_id = save_buffer(buffer);
636 total_len = buffer->size;
637 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
638 buffer->size = max_len;
641 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
642 opi->header.version = OFP_VERSION;
643 opi->header.type = OFPT_PACKET_IN;
644 opi->header.length = htons(buffer->size);
645 opi->header.xid = htonl(0);
646 opi->buffer_id = htonl(buffer_id);
647 opi->total_len = htons(total_len);
648 opi->in_port = htons(in_port);
649 opi->reason = reason;
651 send_openflow_buffer(dp, buffer, NULL);
654 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
655 struct ofp_phy_port *desc)
657 desc->port_no = htons(port_no(dp, p));
658 strncpy((char *) desc->name, netdev_get_name(p->netdev),
660 desc->name[sizeof desc->name - 1] = '\0';
661 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
663 desc->features = htonl(netdev_get_features(p->netdev));
664 desc->speed = htonl(netdev_get_speed(p->netdev));
666 if (p->flags & BRPF_NO_FLOOD) {
667 desc->flags |= htonl(OFPPFL_NO_FLOOD);
668 } else if (p->status & BRPS_PORT_DOWN) {
669 desc->flags |= htonl(OFPPFL_PORT_DOWN);
670 } else if (p->status & BRPS_LINK_DOWN) {
671 desc->flags |= htonl(OFPPFL_LINK_DOWN);
676 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
678 struct buffer *buffer;
679 struct ofp_switch_features *ofr;
682 ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
684 ofr->datapath_id = htonll(dp->id);
685 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
686 ofr->n_compression = 0; /* Not supported */
687 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
688 ofr->buffer_mb = htonl(UINT32_MAX);
689 ofr->n_buffers = htonl(N_PKT_BUFFERS);
690 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
691 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
692 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
693 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
694 memset(opp, 0, sizeof *opp);
695 fill_port_desc(dp, p, opp);
697 send_openflow_buffer(dp, buffer, sender);
701 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
703 const struct ofp_phy_port *opp = &opm->desc;
704 int port_no = ntohs(opp->port_no);
705 if (port_no < OFPP_MAX) {
706 struct sw_port *p = &dp->ports[port_no];
708 /* Make sure the port id hasn't changed since this was sent */
709 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
710 ETH_ADDR_LEN) != 0) {
715 if (opm->mask & htonl(OFPPFL_NO_FLOOD)) {
716 if (opp->flags & htonl(OFPPFL_NO_FLOOD))
717 p->flags |= BRPF_NO_FLOOD;
719 p->flags &= ~BRPF_NO_FLOOD;
722 if (opm->mask & htonl(OFPPFL_PORT_DOWN)) {
723 if ((opp->flags & htonl(OFPPFL_PORT_DOWN))
724 && (p->status & BRPS_PORT_DOWN) == 0) {
725 p->status |= BRPS_PORT_DOWN;
726 netdev_turn_flags_off(p->netdev, NETDEV_UP, true);
727 } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0
728 && (p->status & BRPS_PORT_DOWN)) {
729 p->status &= ~BRPS_PORT_DOWN;
730 netdev_turn_flags_on(p->netdev, NETDEV_UP, true);
736 /* Update the port status field of the bridge port. A non-zero return
737 * value indicates some field has changed.
739 * NB: Callers of this function may hold the RCU read lock, so any
740 * additional checks must not sleep.
743 update_port_status(struct sw_port *p)
746 enum netdev_flags flags;
747 uint32_t orig_status = p->status;
749 if (netdev_get_flags(p->netdev, &flags) < 0) {
750 VLOG_WARN_RL(&rl, "could not get netdev flags for %s",
751 netdev_get_name(p->netdev));
754 if (flags & NETDEV_UP) {
755 p->status &= ~BRPS_PORT_DOWN;
757 p->status |= BRPS_PORT_DOWN;
761 /* Not all cards support this getting link status, so don't warn on
763 retval = netdev_get_link_status(p->netdev);
765 p->status &= ~BRPS_LINK_DOWN;
766 } else if (retval == 0) {
767 p->status |= BRPS_LINK_DOWN;
770 return (orig_status != p->status);
774 send_port_status(struct sw_port *p, uint8_t status)
776 struct buffer *buffer;
777 struct ofp_port_status *ops;
778 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
779 ops->reason = status;
780 memset(ops->pad, 0, sizeof ops->pad);
781 fill_port_desc(p->dp, p, &ops->desc);
783 send_openflow_buffer(p->dp, buffer, NULL);
787 send_flow_expired(struct datapath *dp, struct sw_flow *flow,
788 enum ofp_flow_expired_reason reason)
790 struct buffer *buffer;
791 struct ofp_flow_expired *ofe;
792 ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
793 flow_fill_match(&ofe->match, &flow->key);
795 ofe->priority = htons(flow->priority);
796 ofe->reason = reason;
797 memset(ofe->pad, 0, sizeof ofe->pad);
799 ofe->duration = htonl(time_now() - flow->created);
800 memset(ofe->pad2, 0, sizeof ofe->pad2);
801 ofe->packet_count = htonll(flow->packet_count);
802 ofe->byte_count = htonll(flow->byte_count);
803 send_openflow_buffer(dp, buffer, NULL);
807 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
808 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
810 struct buffer *buffer;
811 struct ofp_error_msg *oem;
812 oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG,
814 oem->type = htons(type);
815 oem->code = htons(code);
816 memcpy(oem->data, data, len);
817 send_openflow_buffer(dp, buffer, sender);
821 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
822 int table_idx, time_t now)
824 struct ofp_flow_stats *ofs;
825 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
826 ofs = buffer_put_uninit(buffer, length);
827 ofs->length = htons(length);
828 ofs->table_id = table_idx;
830 ofs->match.wildcards = htonl(flow->key.wildcards);
831 ofs->match.in_port = flow->key.flow.in_port;
832 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
833 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
834 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
835 ofs->match.dl_type = flow->key.flow.dl_type;
836 ofs->match.nw_src = flow->key.flow.nw_src;
837 ofs->match.nw_dst = flow->key.flow.nw_dst;
838 ofs->match.nw_proto = flow->key.flow.nw_proto;
840 ofs->match.tp_src = flow->key.flow.tp_src;
841 ofs->match.tp_dst = flow->key.flow.tp_dst;
842 ofs->duration = htonl(now - flow->created);
843 ofs->priority = htons(flow->priority);
844 ofs->idle_timeout = htons(flow->idle_timeout);
845 ofs->hard_timeout = htons(flow->hard_timeout);
846 memset(ofs->pad2, 0, sizeof ofs->pad2);
847 ofs->packet_count = htonll(flow->packet_count);
848 ofs->byte_count = htonll(flow->byte_count);
849 memcpy(ofs->actions, flow->actions,
850 sizeof *ofs->actions * flow->n_actions);
854 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
855 * OFPP_MAX. Process it according to 'dp''s flow table. Returns 0 if
856 * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no
857 * matching flow, in which case 'buffer' still belongs to the caller. */
858 int run_flow_through_tables(struct datapath *dp, struct buffer *buffer,
861 struct sw_flow_key key;
862 struct sw_flow *flow;
865 if (flow_extract(buffer, in_port, &key.flow)
866 && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
868 buffer_delete(buffer);
872 flow = chain_lookup(dp->chain, &key);
874 flow_used(flow, buffer);
875 execute_actions(dp, buffer, in_port, &key,
876 flow->actions, flow->n_actions);
883 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
884 * OFPP_MAX. Process it according to 'dp''s flow table, sending it up to the
885 * controller if no flow matches. Takes ownership of 'buffer'. */
886 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
888 if (run_flow_through_tables(dp, buffer, in_port)) {
889 dp_output_control(dp, buffer, in_port, dp->miss_send_len,
895 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
896 size_t max_len, int out_port)
898 if (out_port != OFPP_CONTROLLER) {
899 dp_output_port(dp, buffer, in_port, out_port);
901 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
906 execute_actions(struct datapath *dp, struct buffer *buffer,
907 int in_port, const struct sw_flow_key *key,
908 const struct ofp_action *actions, int n_actions)
910 /* Every output action needs a separate clone of 'buffer', but the common
911 * case is just a single output action, so that doing a clone and then
912 * freeing the original buffer is wasteful. So the following code is
913 * slightly obscure just to avoid that. */
915 size_t max_len=0; /* Initialze to make compiler happy */
920 eth_proto = ntohs(key->flow.dl_type);
922 for (i = 0; i < n_actions; i++) {
923 const struct ofp_action *a = &actions[i];
924 struct eth_header *eh = buffer->l2;
926 if (prev_port != -1) {
927 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
931 switch (ntohs(a->type)) {
933 prev_port = ntohs(a->arg.output.port);
934 max_len = ntohs(a->arg.output.max_len);
937 case OFPAT_SET_DL_VLAN:
938 modify_vlan(buffer, key, a);
941 case OFPAT_SET_DL_SRC:
942 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
945 case OFPAT_SET_DL_DST:
946 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
949 case OFPAT_SET_NW_SRC:
950 case OFPAT_SET_NW_DST:
951 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
954 case OFPAT_SET_TP_SRC:
955 case OFPAT_SET_TP_DST:
956 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
964 do_output(dp, buffer, in_port, max_len, prev_port);
966 buffer_delete(buffer);
969 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
970 uint8_t nw_proto, const struct ofp_action *a)
972 if (eth_proto == ETH_TYPE_IP) {
973 struct ip_header *nh = buffer->l3;
974 uint32_t new, *field;
976 new = a->arg.nw_addr;
977 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
978 if (nw_proto == IP_TYPE_TCP) {
979 struct tcp_header *th = buffer->l4;
980 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
981 } else if (nw_proto == IP_TYPE_UDP) {
982 struct udp_header *th = buffer->l4;
984 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
986 th->udp_csum = 0xffff;
990 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
995 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
996 uint8_t nw_proto, const struct ofp_action *a)
998 if (eth_proto == ETH_TYPE_IP) {
999 uint16_t new, *field;
1003 if (nw_proto == IP_TYPE_TCP) {
1004 struct tcp_header *th = buffer->l4;
1005 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
1006 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
1008 } else if (nw_proto == IP_TYPE_UDP) {
1009 struct udp_header *th = buffer->l4;
1010 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
1011 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
1018 modify_vlan(struct buffer *buffer,
1019 const struct sw_flow_key *key, const struct ofp_action *a)
1021 uint16_t new_id = a->arg.vlan_id;
1022 struct vlan_eth_header *veh;
1024 if (new_id != htons(OFP_VLAN_NONE)) {
1025 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
1026 /* Modify vlan id, but maintain other TCI values */
1028 veh->veth_tci &= ~htons(VLAN_VID);
1029 veh->veth_tci |= new_id;
1031 /* Insert new vlan id. */
1032 struct eth_header *eh = buffer->l2;
1033 struct vlan_eth_header tmp;
1034 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
1035 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
1036 tmp.veth_type = htons(ETH_TYPE_VLAN);
1037 tmp.veth_tci = new_id;
1038 tmp.veth_next_type = eh->eth_type;
1040 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
1041 memcpy(veh, &tmp, sizeof tmp);
1042 buffer->l2 -= VLAN_HEADER_LEN;
1045 /* Remove an existing vlan header if it exists */
1047 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
1048 struct eth_header tmp;
1050 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1051 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1052 tmp.eth_type = veh->veth_next_type;
1054 buffer->size -= VLAN_HEADER_LEN;
1055 buffer->data += VLAN_HEADER_LEN;
1056 buffer->l2 += VLAN_HEADER_LEN;
1057 memcpy(buffer->data, &tmp, sizeof tmp);
1063 recv_features_request(struct datapath *dp, const struct sender *sender,
1066 dp_send_features_reply(dp, sender);
1071 recv_get_config_request(struct datapath *dp, const struct sender *sender,
1074 struct buffer *buffer;
1075 struct ofp_switch_config *osc;
1077 osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
1080 osc->flags = htons(dp->flags);
1081 osc->miss_send_len = htons(dp->miss_send_len);
1083 return send_openflow_buffer(dp, buffer, sender);
1087 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
1090 const struct ofp_switch_config *osc = msg;
1093 flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
1094 if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
1095 && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
1096 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
1099 dp->miss_send_len = ntohs(osc->miss_send_len);
1104 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1107 const struct ofp_packet_out *opo = msg;
1108 struct sw_flow_key key;
1109 struct buffer *buffer;
1110 int n_actions = ntohs(opo->n_actions);
1111 int act_len = n_actions * sizeof opo->actions[0];
1113 if (act_len > (ntohs(opo->header.length) - sizeof *opo)) {
1114 VLOG_DBG_RL(&rl, "message too short for number of actions");
1118 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1119 /* FIXME: can we avoid copying data here? */
1120 int data_len = ntohs(opo->header.length) - sizeof *opo - act_len;
1121 buffer = buffer_new(data_len);
1122 buffer_put(buffer, &opo->actions[n_actions], data_len);
1124 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1130 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1131 execute_actions(dp, buffer, ntohs(opo->in_port),
1132 &key, opo->actions, n_actions);
1138 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1141 const struct ofp_port_mod *opm = msg;
1143 dp_update_port_flags(dp, opm);
1149 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1151 int error = -ENOMEM;
1154 struct sw_flow *flow;
1157 /* To prevent loops, make sure there's no action to send to the
1158 * OFP_TABLE virtual port.
1160 n_actions = (ntohs(ofm->header.length) - sizeof *ofm)
1161 / sizeof *ofm->actions;
1162 for (i=0; i<n_actions; i++) {
1163 const struct ofp_action *a = &ofm->actions[i];
1165 if (a->type == htons(OFPAT_OUTPUT)
1166 && (a->arg.output.port == htons(OFPP_TABLE)
1167 || a->arg.output.port == htons(OFPP_NONE)
1168 || a->arg.output.port == ofm->match.in_port)) {
1169 /* xxx Send fancy new error message? */
1174 /* Allocate memory. */
1175 flow = flow_alloc(n_actions);
1179 /* Fill out flow. */
1180 flow_extract_match(&flow->key, &ofm->match);
1181 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1182 flow->idle_timeout = ntohs(ofm->idle_timeout);
1183 flow->hard_timeout = ntohs(ofm->hard_timeout);
1184 flow->used = flow->created = time_now();
1185 flow->n_actions = n_actions;
1186 flow->byte_count = 0;
1187 flow->packet_count = 0;
1188 memcpy(flow->actions, ofm->actions, n_actions * sizeof *flow->actions);
1191 error = chain_insert(dp->chain, flow);
1193 goto error_free_flow;
1196 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1197 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1199 struct sw_flow_key key;
1200 uint16_t in_port = ntohs(ofm->match.in_port);
1201 flow_used(flow, buffer);
1202 flow_extract(buffer, in_port, &key.flow);
1203 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_actions);
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 version_stats_dump(struct datapath *dp, void *state,
1243 struct buffer *buffer)
1245 struct ofp_version_stats *ovs = buffer_put_uninit(buffer, sizeof *ovs);
1247 strncpy(ovs->mfr_desc, &mfr_desc, sizeof ovs->mfr_desc);
1248 strncpy(ovs->hw_desc, &hw_desc, sizeof ovs->hw_desc);
1249 strncpy(ovs->sw_desc, &sw_desc, sizeof ovs->sw_desc);
1254 struct flow_stats_state {
1256 struct sw_table_position position;
1257 struct ofp_flow_stats_request rq;
1260 struct buffer *buffer;
1263 #define MAX_FLOW_STATS_BYTES 4096
1265 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1268 const struct ofp_flow_stats_request *fsr = body;
1269 struct flow_stats_state *s = xmalloc(sizeof *s);
1270 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1271 memset(&s->position, 0, sizeof s->position);
1277 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1279 struct flow_stats_state *s = private;
1280 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1281 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1284 static int flow_stats_dump(struct datapath *dp, void *state,
1285 struct buffer *buffer)
1287 struct flow_stats_state *s = state;
1288 struct sw_flow_key match_key;
1290 flow_extract_match(&match_key, &s->rq.match);
1292 s->now = time_now();
1293 while (s->table_idx < dp->chain->n_tables
1294 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1296 struct sw_table *table = dp->chain->tables[s->table_idx];
1298 if (table->iterate(table, &match_key, &s->position,
1299 flow_stats_dump_callback, s))
1303 memset(&s->position, 0, sizeof s->position);
1305 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1308 static void flow_stats_done(void *state)
1313 struct aggregate_stats_state {
1314 struct ofp_aggregate_stats_request rq;
1317 static int aggregate_stats_init(struct datapath *dp,
1318 const void *body, int body_len,
1321 const struct ofp_aggregate_stats_request *rq = body;
1322 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1328 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1330 struct ofp_aggregate_stats_reply *rpy = private;
1331 rpy->packet_count += flow->packet_count;
1332 rpy->byte_count += flow->byte_count;
1337 static int aggregate_stats_dump(struct datapath *dp, void *state,
1338 struct buffer *buffer)
1340 struct aggregate_stats_state *s = state;
1341 struct ofp_aggregate_stats_request *rq = &s->rq;
1342 struct ofp_aggregate_stats_reply *rpy;
1343 struct sw_table_position position;
1344 struct sw_flow_key match_key;
1347 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1348 memset(rpy, 0, sizeof *rpy);
1350 flow_extract_match(&match_key, &rq->match);
1351 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1352 memset(&position, 0, sizeof position);
1353 while (table_idx < dp->chain->n_tables
1354 && (rq->table_id == 0xff || rq->table_id == table_idx))
1356 struct sw_table *table = dp->chain->tables[table_idx];
1359 error = table->iterate(table, &match_key, &position,
1360 aggregate_stats_dump_callback, rpy);
1365 memset(&position, 0, sizeof position);
1368 rpy->packet_count = htonll(rpy->packet_count);
1369 rpy->byte_count = htonll(rpy->byte_count);
1370 rpy->flow_count = htonl(rpy->flow_count);
1374 static void aggregate_stats_done(void *state)
1379 static int table_stats_dump(struct datapath *dp, void *state,
1380 struct buffer *buffer)
1383 for (i = 0; i < dp->chain->n_tables; i++) {
1384 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1385 struct sw_table_stats stats;
1386 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1387 strncpy(ots->name, stats.name, sizeof ots->name);
1389 memset(ots->pad, 0, sizeof ots->pad);
1390 ots->max_entries = htonl(stats.max_flows);
1391 ots->active_count = htonl(stats.n_flows);
1392 ots->matched_count = htonll(stats.n_matched);
1397 struct port_stats_state {
1401 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1404 struct port_stats_state *s = xmalloc(sizeof *s);
1410 static int port_stats_dump(struct datapath *dp, void *state,
1411 struct buffer *buffer)
1413 struct port_stats_state *s = state;
1416 for (i = s->port; i < OFPP_MAX; i++) {
1417 struct sw_port *p = &dp->ports[i];
1418 struct ofp_port_stats *ops;
1422 ops = buffer_put_uninit(buffer, sizeof *ops);
1423 ops->port_no = htons(port_no(dp, p));
1424 memset(ops->pad, 0, sizeof ops->pad);
1425 ops->rx_packets = htonll(p->rx_packets);
1426 ops->tx_packets = htonll(p->tx_packets);
1427 ops->rx_bytes = htonll(p->rx_bytes);
1428 ops->tx_bytes = htonll(p->tx_bytes);
1429 ops->rx_dropped = htonll(-1);
1430 ops->tx_dropped = htonll(p->tx_dropped);
1431 ops->rx_errors = htonll(-1);
1432 ops->tx_errors = htonll(-1);
1433 ops->rx_frame_err = htonll(-1);
1434 ops->rx_over_err = htonll(-1);
1435 ops->rx_crc_err = htonll(-1);
1436 ops->collisions = htonll(-1);
1443 static void port_stats_done(void *state)
1449 /* Minimum and maximum acceptable number of bytes in body member of
1450 * struct ofp_stats_request. */
1451 size_t min_body, max_body;
1453 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1454 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1455 * Returns zero if successful, otherwise a negative error code.
1456 * May initialize '*state' to state information. May be null if no
1457 * initialization is required.*/
1458 int (*init)(struct datapath *dp, const void *body, int body_len,
1461 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1462 * struct ofp_stats_reply. On success, it should return 1 if it should be
1463 * called again later with another buffer, 0 if it is done, or a negative
1464 * errno value on failure. */
1465 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1467 /* Cleans any state created by the init or dump functions. May be null
1468 * if no cleanup is required. */
1469 void (*done)(void *state);
1472 static const struct stats_type stats[] = {
1481 sizeof(struct ofp_flow_stats_request),
1482 sizeof(struct ofp_flow_stats_request),
1487 [OFPST_AGGREGATE] = {
1488 sizeof(struct ofp_aggregate_stats_request),
1489 sizeof(struct ofp_aggregate_stats_request),
1490 aggregate_stats_init,
1491 aggregate_stats_dump,
1492 aggregate_stats_done
1510 struct stats_dump_cb {
1512 struct ofp_stats_request *rq;
1513 struct sender sender;
1514 const struct stats_type *s;
1519 stats_dump(struct datapath *dp, void *cb_)
1521 struct stats_dump_cb *cb = cb_;
1522 struct ofp_stats_reply *osr;
1523 struct buffer *buffer;
1530 osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1532 osr->type = htons(cb->s - stats);
1535 err = cb->s->dump(dp, cb->state, buffer);
1541 /* Buffer might have been reallocated, so find our data again. */
1542 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1543 osr->flags = ntohs(OFPSF_REPLY_MORE);
1545 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1555 stats_done(void *cb_)
1557 struct stats_dump_cb *cb = cb_;
1560 cb->s->done(cb->state);
1567 recv_stats_request(struct datapath *dp, const struct sender *sender,
1570 const struct ofp_stats_request *rq = oh;
1571 size_t rq_len = ntohs(rq->header.length);
1572 struct stats_dump_cb *cb;
1576 type = ntohs(rq->type);
1577 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1578 VLOG_WARN_RL(&rl, "received stats request of unknown type %d", type);
1582 cb = xmalloc(sizeof *cb);
1584 cb->rq = xmemdup(rq, rq_len);
1585 cb->sender = *sender;
1586 cb->s = &stats[type];
1589 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1590 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1591 VLOG_WARN_RL(&rl, "stats request type %d with bad body length %d",
1598 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1601 "failed initialization of stats request type %d: %s",
1602 type, strerror(-err));
1607 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1617 recv_echo_request(struct datapath *dp, const struct sender *sender,
1620 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1624 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1625 const void *oh UNUSED)
1630 /* 'msg', which is 'length' bytes long, was received from the control path.
1631 * Apply it to 'chain'. */
1633 fwd_control_input(struct datapath *dp, const struct sender *sender,
1634 const void *msg, size_t length)
1636 struct openflow_packet {
1638 int (*handler)(struct datapath *, const struct sender *, const void *);
1641 static const struct openflow_packet packets[] = {
1642 [OFPT_FEATURES_REQUEST] = {
1643 sizeof (struct ofp_header),
1644 recv_features_request,
1646 [OFPT_GET_CONFIG_REQUEST] = {
1647 sizeof (struct ofp_header),
1648 recv_get_config_request,
1650 [OFPT_SET_CONFIG] = {
1651 sizeof (struct ofp_switch_config),
1654 [OFPT_PACKET_OUT] = {
1655 sizeof (struct ofp_packet_out),
1659 sizeof (struct ofp_flow_mod),
1663 sizeof (struct ofp_port_mod),
1666 [OFPT_STATS_REQUEST] = {
1667 sizeof (struct ofp_stats_request),
1670 [OFPT_ECHO_REQUEST] = {
1671 sizeof (struct ofp_header),
1674 [OFPT_ECHO_REPLY] = {
1675 sizeof (struct ofp_header),
1680 const struct openflow_packet *pkt;
1681 struct ofp_header *oh;
1683 oh = (struct ofp_header *) msg;
1684 assert(oh->version == OFP_VERSION);
1685 if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length)
1688 pkt = &packets[oh->type];
1691 if (length < pkt->min_size)
1694 return pkt->handler(dp, sender, msg);
1697 /* Packet buffering. */
1699 #define OVERWRITE_SECS 1
1701 struct packet_buffer {
1702 struct buffer *buffer;
1707 static struct packet_buffer buffers[N_PKT_BUFFERS];
1708 static unsigned int buffer_idx;
1710 uint32_t save_buffer(struct buffer *buffer)
1712 struct packet_buffer *p;
1715 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1716 p = &buffers[buffer_idx];
1718 /* Don't buffer packet if existing entry is less than
1719 * OVERWRITE_SECS old. */
1720 if (time_now() < p->timeout) { /* FIXME */
1723 buffer_delete(p->buffer);
1726 /* Don't use maximum cookie value since the all-bits-1 id is
1728 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1730 p->buffer = buffer_clone(buffer); /* FIXME */
1731 p->timeout = time_now() + OVERWRITE_SECS; /* FIXME */
1732 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1737 static struct buffer *retrieve_buffer(uint32_t id)
1739 struct buffer *buffer = NULL;
1740 struct packet_buffer *p;
1742 p = &buffers[id & PKT_BUFFER_MASK];
1743 if (p->cookie == id >> PKT_BUFFER_BITS) {
1747 printf("cookie mismatch: %x != %x\n",
1748 id >> PKT_BUFFER_BITS, p->cookie);
1754 static void discard_buffer(uint32_t id)
1756 struct packet_buffer *p;
1758 p = &buffers[id & PKT_BUFFER_MASK];
1759 if (p->cookie == id >> PKT_BUFFER_BITS) {
1760 buffer_delete(p->buffer);