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>
45 #include "poll-loop.h"
51 #define THIS_MODULE VLM_datapath
54 #define BRIDGE_PORT_NO_FLOOD 0x00000001
56 /* Capabilities supported by this implementation. */
57 #define OFP_SUPPORTED_CAPABILITIES (OFPC_MULTI_PHY_TX)
59 /* Actions supported by this implementation. */
60 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
61 | (1 << OFPAT_SET_DL_VLAN) \
62 | (1 << OFPAT_SET_DL_SRC) \
63 | (1 << OFPAT_SET_DL_DST) \
64 | (1 << OFPAT_SET_NW_SRC) \
65 | (1 << OFPAT_SET_NW_DST) \
66 | (1 << OFPAT_SET_TP_SRC) \
67 | (1 << OFPAT_SET_TP_DST) )
72 struct netdev *netdev;
73 struct list node; /* Element in datapath.ports. */
74 unsigned long long int rx_count, tx_count, drop_count;
77 /* A connection to a controller or a management device. */
83 /* The origin of a received OpenFlow message, to enable sending a reply. */
85 struct remote *remote; /* The device that sent the message. */
86 uint32_t xid; /* The OpenFlow transaction ID. */
90 /* Remote connections. */
91 struct remote *controller; /* Connection to controller. */
92 struct list remotes; /* All connections (including controller). */
93 struct vconn *listen_vconn;
97 /* Unique identifier for this datapath */
100 struct sw_chain *chain; /* Forwarding rules. */
102 struct ofp_switch_config config;
105 struct sw_port ports[OFPP_MAX];
106 struct list port_list; /* List of ports, for flooding. */
109 static struct remote *remote_create(struct datapath *, struct rconn *);
110 static void remote_run(struct datapath *, struct remote *);
111 static void remote_wait(struct remote *);
112 static void remote_destroy(struct remote *);
114 void dp_output_port(struct datapath *, struct buffer *,
115 int in_port, int out_port);
116 void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
117 void dp_output_control(struct datapath *, struct buffer *, int in_port,
118 size_t max_len, int reason);
119 static void send_flow_expired(struct datapath *, struct sw_flow *);
120 static void send_port_status(struct sw_port *p, uint8_t status);
121 static void del_switch_port(struct sw_port *p);
122 static void execute_actions(struct datapath *, struct buffer *,
123 int in_port, const struct sw_flow_key *,
124 const struct ofp_action *, int n_actions);
125 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
126 const struct ofp_action *a);
127 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
128 uint8_t nw_proto, const struct ofp_action *a);
129 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
130 uint8_t nw_proto, const struct ofp_action *a);
132 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
133 * into a buffer number (low bits) and a cookie (high bits). The buffer number
134 * is an index into an array of buffers. The cookie distinguishes between
135 * different packets that have occupied a single buffer. Thus, the more
136 * buffers we have, the lower-quality the cookie... */
137 #define PKT_BUFFER_BITS 8
138 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
139 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
141 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
143 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
144 int fwd_control_input(struct datapath *, const struct sender *,
145 const void *, size_t);
147 uint32_t save_buffer(struct buffer *);
148 static struct buffer *retrieve_buffer(uint32_t id);
149 static void discard_buffer(uint32_t id);
151 static int port_no(struct datapath *dp, struct sw_port *p)
153 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
154 return p - dp->ports;
157 /* Generates a unique datapath id. It incorporates the datapath index
158 * and a hardware address, if available. If not, it generates a random
162 gen_datapath_id(void)
164 /* Choose a random datapath id. */
170 for (i = 0; i < ETH_ADDR_LEN; i++) {
171 id |= (uint64_t)(rand() & 0xff) << (8*(ETH_ADDR_LEN-1 - i));
178 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
182 dp = calloc(1, sizeof *dp);
187 dp->last_timeout = time(0);
188 list_init(&dp->remotes);
189 dp->controller = remote_create(dp, rconn);
190 dp->listen_vconn = NULL;
191 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
192 dp->chain = chain_create();
194 VLOG_ERR("could not create chain");
199 list_init(&dp->port_list);
200 dp->config.flags = 0;
201 dp->config.miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
207 dp_add_port(struct datapath *dp, const char *name)
209 struct netdev *netdev;
213 error = netdev_open(name, &netdev);
218 for (p = dp->ports; ; p++) {
219 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
221 } else if (!p->netdev) {
231 list_push_back(&dp->port_list, &p->node);
233 /* Notify the ctlpath that this port has been added */
234 send_port_status(p, OFPPR_ADD);
240 dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
242 assert(!dp->listen_vconn);
243 dp->listen_vconn = listen_vconn;
247 dp_run(struct datapath *dp)
249 time_t now = time(0);
250 struct sw_port *p, *pn;
251 struct remote *r, *rn;
252 struct buffer *buffer = NULL;
254 if (now != dp->last_timeout) {
255 struct list deleted = LIST_INITIALIZER(&deleted);
256 struct sw_flow *f, *n;
258 chain_timeout(dp->chain, &deleted);
259 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
260 send_flow_expired(dp, f);
261 list_remove(&f->node);
264 dp->last_timeout = now;
266 poll_timer_wait(1000);
268 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
272 /* Allocate buffer with some headroom to add headers in forwarding
273 * to the controller or adding a vlan tag, plus an extra 2 bytes to
274 * allow IP headers to be aligned on a 4-byte boundary. */
275 const int headroom = 128 + 2;
276 const int hard_header = VLAN_ETH_HEADER_LEN;
277 const int mtu = netdev_get_mtu(p->netdev);
278 buffer = buffer_new(headroom + hard_header + mtu);
279 buffer->data += headroom;
281 error = netdev_recv(p->netdev, buffer);
284 fwd_port_input(dp, buffer, port_no(dp, p));
286 } else if (error != EAGAIN) {
287 VLOG_ERR("Error receiving data from %s: %s",
288 netdev_get_name(p->netdev), strerror(error));
292 buffer_delete(buffer);
294 /* Talk to remotes. */
295 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
298 if (dp->listen_vconn) {
300 struct vconn *new_vconn;
303 retval = vconn_accept(dp->listen_vconn, &new_vconn);
305 if (retval != EAGAIN) {
306 VLOG_WARN("accept failed (%s)", strerror(retval));
310 remote_create(dp, rconn_new_from_vconn("passive", 128, new_vconn));
316 remote_run(struct datapath *dp, struct remote *r)
322 /* Process a number of commands from the remote, but cap them at a
323 * reasonable number so that other processing doesn't starve. */
324 for (i = 0; i < 50; i++) {
325 struct buffer *buffer;
326 struct ofp_header *oh;
328 buffer = rconn_recv(r->rconn);
333 if (buffer->size >= sizeof *oh) {
334 struct sender sender;
338 sender.xid = oh->xid;
339 fwd_control_input(dp, &sender, buffer->data, buffer->size);
341 VLOG_WARN("received too-short OpenFlow message");
343 buffer_delete(buffer);
346 if (!rconn_is_alive(r->rconn)) {
352 remote_wait(struct remote *r)
354 rconn_run_wait(r->rconn);
355 rconn_recv_wait(r->rconn);
359 remote_destroy(struct remote *r)
362 list_remove(&r->node);
363 rconn_destroy(r->rconn);
368 static struct remote *
369 remote_create(struct datapath *dp, struct rconn *rconn)
371 struct remote *remote = xmalloc(sizeof *remote);
372 list_push_back(&dp->remotes, &remote->node);
373 remote->rconn = rconn;
378 dp_wait(struct datapath *dp)
383 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
384 netdev_recv_wait(p->netdev);
386 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
389 if (dp->listen_vconn) {
390 vconn_accept_wait(dp->listen_vconn);
394 /* Delete 'p' from switch. */
396 del_switch_port(struct sw_port *p)
398 send_port_status(p, OFPPR_DELETE);
399 netdev_close(p->netdev);
401 list_remove(&p->node);
405 dp_destroy(struct datapath *dp)
407 struct sw_port *p, *n;
413 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
416 chain_destroy(dp->chain);
421 flood(struct datapath *dp, struct buffer *buffer, int in_port)
427 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
428 if (port_no(dp, p) == in_port || p->flags & BRIDGE_PORT_NO_FLOOD) {
431 if (prev_port != -1) {
432 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
434 prev_port = port_no(dp, p);
437 dp_output_port(dp, buffer, in_port, prev_port);
439 buffer_delete(buffer);
445 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
447 if (out_port >= 0 && out_port < OFPP_MAX) {
448 struct sw_port *p = &dp->ports[out_port];
449 if (p->netdev != NULL) {
450 if (!netdev_send(p->netdev, buffer)) {
459 buffer_delete(buffer);
460 /* FIXME: ratelimit */
461 VLOG_DBG("can't forward to bad port %d\n", out_port);
464 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
467 dp_output_port(struct datapath *dp, struct buffer *buffer,
468 int in_port, int out_port)
472 if (out_port == OFPP_FLOOD) {
473 flood(dp, buffer, in_port);
474 } else if (out_port == OFPP_CONTROLLER) {
475 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
477 output_packet(dp, buffer, out_port);
482 alloc_openflow_buffer(struct datapath *dp, size_t openflow_len, uint8_t type,
483 const struct sender *sender, struct buffer **bufferp)
485 struct buffer *buffer;
486 struct ofp_header *oh;
488 buffer = *bufferp = buffer_new(openflow_len);
489 oh = buffer_put_uninit(buffer, openflow_len);
490 oh->version = OFP_VERSION;
492 oh->length = 0; /* Filled in by send_openflow_buffer(). */
493 oh->xid = sender ? sender->xid : 0;
498 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
499 const struct sender *sender)
501 struct remote *remote = sender ? sender->remote : dp->controller;
502 struct rconn *rconn = remote->rconn;
503 struct ofp_header *oh;
506 oh = buffer_at_assert(buffer, 0, sizeof *oh);
507 oh->length = htons(buffer->size);
509 retval = rconn_send(rconn, buffer);
511 VLOG_WARN("send to %s failed: %s",
512 rconn_get_name(rconn), strerror(retval));
513 buffer_delete(buffer);
518 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
519 * packet can be saved in a buffer, then only the first max_len bytes of
520 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
521 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
522 * the caller wants to be sent; a value of 0 indicates the entire packet should
525 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
526 size_t max_len, int reason)
528 struct ofp_packet_in *opi;
532 buffer_id = save_buffer(buffer);
533 total_len = buffer->size;
534 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
535 buffer->size = max_len;
538 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
539 opi->header.version = OFP_VERSION;
540 opi->header.type = OFPT_PACKET_IN;
541 opi->header.length = htons(buffer->size);
542 opi->header.xid = htonl(0);
543 opi->buffer_id = htonl(buffer_id);
544 opi->total_len = htons(total_len);
545 opi->in_port = htons(in_port);
546 opi->reason = reason;
548 send_openflow_buffer(dp, buffer, NULL);
551 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
552 struct ofp_phy_port *desc)
554 desc->port_no = htons(port_no(dp, p));
555 strncpy((char *) desc->name, netdev_get_name(p->netdev),
557 desc->name[sizeof desc->name - 1] = '\0';
558 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
559 desc->flags = htonl(p->flags);
560 desc->features = htonl(netdev_get_features(p->netdev));
561 desc->speed = htonl(netdev_get_speed(p->netdev));
565 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
567 struct buffer *buffer;
568 struct ofp_switch_features *ofr;
571 ofr = alloc_openflow_buffer(dp, sizeof *ofr, OFPT_FEATURES_REPLY,
573 ofr->datapath_id = htonll(dp->id);
574 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
575 ofr->n_mac_only = htonl(TABLE_MAC_MAX_FLOWS);
576 ofr->n_compression = 0; /* Not supported */
577 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
578 ofr->buffer_mb = htonl(UINT32_MAX);
579 ofr->n_buffers = htonl(N_PKT_BUFFERS);
580 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
581 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
582 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
583 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
584 memset(opp, 0, sizeof *opp);
585 fill_port_desc(dp, p, opp);
587 send_openflow_buffer(dp, buffer, sender);
591 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
595 p = &dp->ports[htons(opp->port_no)];
597 /* Make sure the port id hasn't changed since this was sent */
598 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
602 p->flags = htonl(opp->flags);
606 send_port_status(struct sw_port *p, uint8_t status)
608 struct buffer *buffer;
609 struct ofp_port_status *ops;
610 ops = alloc_openflow_buffer(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
612 ops->reason = status;
613 fill_port_desc(p->dp, p, &ops->desc);
614 send_openflow_buffer(p->dp, buffer, NULL);
618 send_flow_expired(struct datapath *dp, struct sw_flow *flow)
620 struct buffer *buffer;
621 struct ofp_flow_expired *ofe;
622 ofe = alloc_openflow_buffer(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, NULL,
624 flow_fill_match(&ofe->match, &flow->key);
625 ofe->duration = htonl(flow->timeout - flow->max_idle - flow->created);
626 ofe->packet_count = htonll(flow->packet_count);
627 ofe->byte_count = htonll(flow->byte_count);
628 send_openflow_buffer(dp, buffer, NULL);
632 fill_flow_stats(struct ofp_flow_stats *ofs, struct sw_flow *flow,
633 int table_idx, time_t now)
635 ofs->match.wildcards = htons(flow->key.wildcards);
636 ofs->match.in_port = flow->key.flow.in_port;
637 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
638 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
639 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
640 ofs->match.dl_type = flow->key.flow.dl_type;
641 ofs->match.nw_src = flow->key.flow.nw_src;
642 ofs->match.nw_dst = flow->key.flow.nw_dst;
643 ofs->match.nw_proto = flow->key.flow.nw_proto;
644 memset(ofs->match.pad, 0, sizeof ofs->match.pad);
645 ofs->match.tp_src = flow->key.flow.tp_src;
646 ofs->match.tp_dst = flow->key.flow.tp_dst;
647 ofs->duration = htonl(now - flow->created);
648 ofs->table_id = htons(table_idx);
649 ofs->packet_count = htonll(flow->packet_count);
650 ofs->byte_count = htonll(flow->byte_count);
654 dp_send_flow_stats(struct datapath *dp, const struct sender *sender,
655 const struct ofp_match *match)
657 struct buffer *buffer;
658 struct ofp_flow_stat_reply *fsr;
659 size_t header_size, fudge, flow_size;
660 struct sw_flow_key match_key;
661 int table_idx, n_flows, max_flows;
664 header_size = offsetof(struct ofp_flow_stat_reply, flows);
666 flow_size = sizeof fsr->flows[0];
667 max_flows = (65536 - header_size - fudge) / flow_size;
668 fsr = alloc_openflow_buffer(dp, header_size,
669 OFPT_FLOW_STAT_REPLY, sender, &buffer);
672 flow_extract_match(&match_key, match);
674 for (table_idx = 0; table_idx < dp->chain->n_tables; table_idx++) {
675 struct sw_table *table = dp->chain->tables[table_idx];
676 struct swt_iterator iter;
678 if (n_flows >= max_flows) {
682 if (!table->iterator(table, &iter)) {
683 printf("iterator failed for table %d\n", table_idx);
687 for (; iter.flow; table->iterator_next(&iter)) {
688 if (flow_matches(&match_key, &iter.flow->key)) {
689 struct ofp_flow_stats *ofs = buffer_put_uninit(buffer,
691 fill_flow_stats(ofs, iter.flow, table_idx, now);
692 if (++n_flows >= max_flows) {
697 table->iterator_destroy(&iter);
699 return send_openflow_buffer(dp, buffer, sender);
703 dp_send_port_stats(struct datapath *dp, const struct sender *sender)
705 struct buffer *buffer;
706 struct ofp_port_stat_reply *psr;
709 psr = alloc_openflow_buffer(dp, offsetof(struct ofp_port_stat_reply,
711 OFPT_PORT_STAT_REPLY, sender, &buffer);
712 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
713 struct ofp_port_stats *ps = buffer_put_uninit(buffer, sizeof *ps);
714 ps->port_no = htons(port_no(dp, p));
715 memset(ps->pad, 0, sizeof ps->pad);
716 ps->rx_count = htonll(p->rx_count);
717 ps->tx_count = htonll(p->tx_count);
718 ps->drop_count = htonll(p->drop_count);
720 return send_openflow_buffer(dp, buffer, sender);
724 dp_send_table_stats(struct datapath *dp, const struct sender *sender)
726 struct buffer *buffer;
727 struct ofp_table_stat_reply *tsr;
730 tsr = alloc_openflow_buffer(dp, offsetof(struct ofp_table_stat_reply,
732 OFPT_TABLE_STAT_REPLY, sender, &buffer);
733 for (i = 0; i < dp->chain->n_tables; i++) {
734 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
735 struct sw_table_stats stats;
736 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
737 strncpy(ots->name, stats.name, sizeof ots->name);
738 ots->table_id = htons(i);
739 ots->pad[0] = ots->pad[1] = 0;
740 ots->max_entries = htonl(stats.max_flows);
741 ots->active_count = htonl(stats.n_flows);
742 ots->matched_count = htonll(0); /* FIXME */
744 return send_openflow_buffer(dp, buffer, sender);
747 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
748 * OFPP_MAX. Process it according to 'chain'. */
749 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
751 struct sw_flow_key key;
752 struct sw_flow *flow;
755 flow_extract(buffer, in_port, &key.flow);
756 flow = chain_lookup(dp->chain, &key);
758 flow_used(flow, buffer);
759 execute_actions(dp, buffer, in_port, &key,
760 flow->actions, flow->n_actions);
762 dp_output_control(dp, buffer, in_port, ntohs(dp->config.miss_send_len),
768 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
769 size_t max_len, int out_port)
771 if (out_port != OFPP_CONTROLLER) {
772 dp_output_port(dp, buffer, in_port, out_port);
774 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
779 execute_actions(struct datapath *dp, struct buffer *buffer,
780 int in_port, const struct sw_flow_key *key,
781 const struct ofp_action *actions, int n_actions)
783 /* Every output action needs a separate clone of 'buffer', but the common
784 * case is just a single output action, so that doing a clone and then
785 * freeing the original buffer is wasteful. So the following code is
786 * slightly obscure just to avoid that. */
788 size_t max_len=0; /* Initialze to make compiler happy */
793 eth_proto = ntohs(key->flow.dl_type);
795 for (i = 0; i < n_actions; i++) {
796 const struct ofp_action *a = &actions[i];
797 struct eth_header *eh = buffer->l2;
799 if (prev_port != -1) {
800 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
804 switch (ntohs(a->type)) {
806 prev_port = ntohs(a->arg.output.port);
807 max_len = ntohs(a->arg.output.max_len);
810 case OFPAT_SET_DL_VLAN:
811 modify_vlan(buffer, key, a);
814 case OFPAT_SET_DL_SRC:
815 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
818 case OFPAT_SET_DL_DST:
819 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
822 case OFPAT_SET_NW_SRC:
823 case OFPAT_SET_NW_DST:
824 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
827 case OFPAT_SET_TP_SRC:
828 case OFPAT_SET_TP_DST:
829 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
837 do_output(dp, buffer, in_port, max_len, prev_port);
839 buffer_delete(buffer);
842 /* Returns the new checksum for a packet in which the checksum field previously
843 * contained 'old_csum' and in which a field that contained 'old_u16' was
844 * changed to contain 'new_u16'. */
846 recalc_csum16(uint16_t old_csum, uint16_t old_u16, uint16_t new_u16)
848 /* Ones-complement arithmetic is endian-independent, so this code does not
849 * use htons() or ntohs().
851 * See RFC 1624 for formula and explanation. */
852 uint16_t hc_complement = ~old_csum;
853 uint16_t m_complement = ~old_u16;
854 uint16_t m_prime = new_u16;
855 uint32_t sum = hc_complement + m_complement + m_prime;
856 uint16_t hc_prime_complement = sum + (sum >> 16);
857 return ~hc_prime_complement;
860 /* Returns the new checksum for a packet in which the checksum field previously
861 * contained 'old_csum' and in which a field that contained 'old_u32' was
862 * changed to contain 'new_u32'. */
864 recalc_csum32(uint16_t old_csum, uint32_t old_u32, uint32_t new_u32)
866 return recalc_csum16(recalc_csum16(old_csum, old_u32, new_u32),
867 old_u32 >> 16, new_u32 >> 16);
870 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
871 uint8_t nw_proto, const struct ofp_action *a)
873 if (eth_proto == ETH_TYPE_IP) {
874 struct ip_header *nh = buffer->l3;
875 uint32_t new, *field;
877 new = a->arg.nw_addr;
878 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
879 if (nw_proto == IP_TYPE_TCP) {
880 struct tcp_header *th = buffer->l4;
881 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
882 } else if (nw_proto == IP_TYPE_UDP) {
883 struct udp_header *th = buffer->l4;
885 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
887 th->udp_csum = 0xffff;
891 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
896 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
897 uint8_t nw_proto, const struct ofp_action *a)
899 if (eth_proto == ETH_TYPE_IP) {
900 uint16_t new, *field;
904 if (nw_proto == IP_TYPE_TCP) {
905 struct tcp_header *th = buffer->l4;
906 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
907 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
909 } else if (nw_proto == IP_TYPE_UDP) {
910 struct udp_header *th = buffer->l4;
911 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
912 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
919 modify_vlan(struct buffer *buffer,
920 const struct sw_flow_key *key, const struct ofp_action *a)
922 uint16_t new_id = a->arg.vlan_id;
923 struct vlan_eth_header *veh;
925 if (new_id != OFP_VLAN_NONE) {
926 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
927 /* Modify vlan id, but maintain other TCI values */
929 veh->veth_tci &= ~htons(VLAN_VID);
930 veh->veth_tci |= htons(new_id);
932 /* Insert new vlan id. */
933 struct eth_header *eh = buffer->l2;
934 struct vlan_eth_header tmp;
935 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
936 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
937 tmp.veth_type = htons(ETH_TYPE_VLAN);
938 tmp.veth_tci = new_id;
939 tmp.veth_next_type = eh->eth_type;
941 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
942 memcpy(veh, &tmp, sizeof tmp);
943 buffer->l2 -= VLAN_HEADER_LEN;
946 /* Remove an existing vlan header if it exists */
948 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
949 struct eth_header tmp;
951 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
952 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
953 tmp.eth_type = veh->veth_next_type;
955 buffer->size -= VLAN_HEADER_LEN;
956 buffer->data += VLAN_HEADER_LEN;
957 buffer->l2 += VLAN_HEADER_LEN;
958 memcpy(buffer->data, &tmp, sizeof tmp);
964 recv_features_request(struct datapath *dp, const struct sender *sender,
967 dp_send_features_reply(dp, sender);
972 recv_get_config_request(struct datapath *dp, const struct sender *sender,
975 struct buffer *buffer;
976 struct ofp_switch_config *osc;
978 osc = alloc_openflow_buffer(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY,
981 assert(sizeof *osc == sizeof dp->config);
982 memcpy(((char *)osc) + sizeof osc->header,
983 ((char *)&dp->config) + sizeof dp->config.header,
984 sizeof dp->config - sizeof dp->config.header);
986 return send_openflow_buffer(dp, buffer, sender);
990 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
993 const struct ofp_switch_config *osc = msg;
999 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1002 const struct ofp_packet_out *opo = msg;
1004 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1005 /* FIXME: can we avoid copying data here? */
1006 int data_len = ntohs(opo->header.length) - sizeof *opo;
1007 struct buffer *buffer = buffer_new(data_len);
1008 buffer_put(buffer, opo->u.data, data_len);
1009 dp_output_port(dp, buffer,
1010 ntohs(opo->in_port), ntohs(opo->out_port));
1012 struct sw_flow_key key;
1013 struct buffer *buffer;
1016 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1021 n_acts = (ntohs(opo->header.length) - sizeof *opo)
1022 / sizeof *opo->u.actions;
1023 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1024 execute_actions(dp, buffer, ntohs(opo->in_port),
1025 &key, opo->u.actions, n_acts);
1031 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1034 const struct ofp_port_mod *opm = msg;
1036 dp_update_port_flags(dp, &opm->desc);
1042 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1044 int error = -ENOMEM;
1046 struct sw_flow *flow;
1049 /* Check number of actions. */
1050 n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
1051 if (n_acts > MAX_ACTIONS) {
1056 /* Allocate memory. */
1057 flow = flow_alloc(n_acts);
1061 /* Fill out flow. */
1062 flow_extract_match(&flow->key, &ofm->match);
1063 flow->group_id = ntohl(ofm->group_id);
1064 flow->max_idle = ntohs(ofm->max_idle);
1065 flow->timeout = time(0) + flow->max_idle; /* FIXME */
1066 flow->n_actions = n_acts;
1067 flow->created = time(0); /* FIXME */
1068 flow->byte_count = 0;
1069 flow->packet_count = 0;
1070 memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
1073 error = chain_insert(dp->chain, flow);
1075 goto error_free_flow;
1078 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1079 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1081 struct sw_flow_key key;
1082 uint16_t in_port = ntohs(ofm->match.in_port);
1083 flow_used(flow, buffer);
1084 flow_extract(buffer, in_port, &key.flow);
1085 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
1095 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1096 discard_buffer(ntohl(ofm->buffer_id));
1101 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1104 const struct ofp_flow_mod *ofm = msg;
1105 uint16_t command = ntohs(ofm->command);
1107 if (command == OFPFC_ADD) {
1108 return add_flow(dp, ofm);
1109 } else if (command == OFPFC_DELETE) {
1110 struct sw_flow_key key;
1111 flow_extract_match(&key, &ofm->match);
1112 return chain_delete(dp->chain, &key, 0) ? 0 : -ESRCH;
1113 } else if (command == OFPFC_DELETE_STRICT) {
1114 struct sw_flow_key key;
1115 flow_extract_match(&key, &ofm->match);
1116 return chain_delete(dp->chain, &key, 1) ? 0 : -ESRCH;
1123 recv_flow_status_request(struct datapath *dp, const struct sender *sender,
1126 const struct ofp_flow_stat_request *fsr = msg;
1127 if (fsr->type == OFPFS_INDIV) {
1128 return dp_send_flow_stats(dp, sender, &fsr->match);
1136 recv_port_status_request(struct datapath *dp, const struct sender *sender,
1139 return dp_send_port_stats(dp, sender);
1143 recv_table_status_request(struct datapath *dp, const struct sender *sender,
1146 return dp_send_table_stats(dp, sender);
1149 /* 'msg', which is 'length' bytes long, was received from the control path.
1150 * Apply it to 'chain'. */
1152 fwd_control_input(struct datapath *dp, const struct sender *sender,
1153 const void *msg, size_t length)
1155 struct openflow_packet {
1157 int (*handler)(struct datapath *, const struct sender *, const void *);
1160 static const struct openflow_packet packets[] = {
1161 [OFPT_FEATURES_REQUEST] = {
1162 sizeof (struct ofp_header),
1163 recv_features_request,
1165 [OFPT_GET_CONFIG_REQUEST] = {
1166 sizeof (struct ofp_header),
1167 recv_get_config_request,
1169 [OFPT_SET_CONFIG] = {
1170 sizeof (struct ofp_switch_config),
1173 [OFPT_PACKET_OUT] = {
1174 sizeof (struct ofp_packet_out),
1178 sizeof (struct ofp_flow_mod),
1182 sizeof (struct ofp_port_mod),
1185 [OFPT_FLOW_STAT_REQUEST] = {
1186 sizeof (struct ofp_flow_stat_request),
1187 recv_flow_status_request,
1189 [OFPT_PORT_STAT_REQUEST] = {
1190 sizeof (struct ofp_port_stat_request),
1191 recv_port_status_request,
1193 [OFPT_TABLE_STAT_REQUEST] = {
1194 sizeof (struct ofp_table_stat_request),
1195 recv_table_status_request,
1199 const struct openflow_packet *pkt;
1200 struct ofp_header *oh;
1202 oh = (struct ofp_header *) msg;
1203 if (oh->version != 1 || oh->type >= ARRAY_SIZE(packets)
1204 || ntohs(oh->length) > length)
1207 pkt = &packets[oh->type];
1210 if (length < pkt->min_size)
1213 return pkt->handler(dp, sender, msg);
1216 /* Packet buffering. */
1218 #define OVERWRITE_SECS 1
1220 struct packet_buffer {
1221 struct buffer *buffer;
1226 static struct packet_buffer buffers[N_PKT_BUFFERS];
1227 static unsigned int buffer_idx;
1229 uint32_t save_buffer(struct buffer *buffer)
1231 struct packet_buffer *p;
1234 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1235 p = &buffers[buffer_idx];
1237 /* Don't buffer packet if existing entry is less than
1238 * OVERWRITE_SECS old. */
1239 if (time(0) < p->timeout) { /* FIXME */
1242 buffer_delete(p->buffer);
1245 /* Don't use maximum cookie value since the all-bits-1 id is
1247 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1249 p->buffer = buffer_clone(buffer); /* FIXME */
1250 p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
1251 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1256 static struct buffer *retrieve_buffer(uint32_t id)
1258 struct buffer *buffer = NULL;
1259 struct packet_buffer *p;
1261 p = &buffers[id & PKT_BUFFER_MASK];
1262 if (p->cookie == id >> PKT_BUFFER_BITS) {
1266 printf("cookie mismatch: %x != %x\n",
1267 id >> PKT_BUFFER_BITS, p->cookie);
1273 static void discard_buffer(uint32_t id)
1275 struct packet_buffer *p;
1277 p = &buffers[id & PKT_BUFFER_MASK];
1278 if (p->cookie == id >> PKT_BUFFER_BITS) {
1279 buffer_delete(p->buffer);