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>
46 #include "poll-loop.h"
52 #define THIS_MODULE VLM_datapath
55 #define BRIDGE_PORT_NO_FLOOD 0x00000001
57 /* Capabilities supported by this implementation. */
58 #define OFP_SUPPORTED_CAPABILITIES (OFPC_MULTI_PHY_TX)
60 /* Actions supported by this implementation. */
61 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
62 | (1 << OFPAT_SET_DL_VLAN) \
63 | (1 << OFPAT_SET_DL_SRC) \
64 | (1 << OFPAT_SET_DL_DST) \
65 | (1 << OFPAT_SET_NW_SRC) \
66 | (1 << OFPAT_SET_NW_DST) \
67 | (1 << OFPAT_SET_TP_SRC) \
68 | (1 << OFPAT_SET_TP_DST) )
73 struct netdev *netdev;
74 struct list node; /* Element in datapath.ports. */
75 unsigned long long int rx_count, tx_count, drop_count;
78 /* The origin of a received OpenFlow message, to enable sending a reply. */
80 struct remote *remote; /* The device that sent the message. */
81 uint32_t xid; /* The OpenFlow transaction ID. */
84 /* A connection to a controller or a management device. */
89 /* Support for reliable, multi-message replies to requests.
91 * If an incoming request needs to have a reliable reply that might
92 * require multiple messages, it can use remote_start_dump() to set up
93 * a callback that will be called as buffer space for replies. */
94 int (*cb_dump)(struct datapath *, void *aux);
95 void (*cb_done)(void *aux);
100 /* Remote connections. */
101 struct remote *controller; /* Connection to controller. */
102 struct list remotes; /* All connections (including controller). */
103 struct vconn *listen_vconn;
107 /* Unique identifier for this datapath */
110 struct sw_chain *chain; /* Forwarding rules. */
112 struct ofp_switch_config config;
115 struct sw_port ports[OFPP_MAX];
116 struct list port_list; /* List of ports, for flooding. */
119 static struct remote *remote_create(struct datapath *, struct rconn *);
120 static void remote_run(struct datapath *, struct remote *);
121 static void remote_wait(struct remote *);
122 static void remote_destroy(struct remote *);
124 void dp_output_port(struct datapath *, struct buffer *,
125 int in_port, int out_port);
126 void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
127 void dp_output_control(struct datapath *, struct buffer *, int in_port,
128 size_t max_len, int reason);
129 static void send_flow_expired(struct datapath *, struct sw_flow *);
130 static void send_port_status(struct sw_port *p, uint8_t status);
131 static void del_switch_port(struct sw_port *p);
132 static void execute_actions(struct datapath *, struct buffer *,
133 int in_port, const struct sw_flow_key *,
134 const struct ofp_action *, int n_actions);
135 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
136 const struct ofp_action *a);
137 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
138 uint8_t nw_proto, const struct ofp_action *a);
139 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
140 uint8_t nw_proto, const struct ofp_action *a);
142 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
143 * into a buffer number (low bits) and a cookie (high bits). The buffer number
144 * is an index into an array of buffers. The cookie distinguishes between
145 * different packets that have occupied a single buffer. Thus, the more
146 * buffers we have, the lower-quality the cookie... */
147 #define PKT_BUFFER_BITS 8
148 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
149 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
151 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
153 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
154 int fwd_control_input(struct datapath *, const struct sender *,
155 const void *, size_t);
157 uint32_t save_buffer(struct buffer *);
158 static struct buffer *retrieve_buffer(uint32_t id);
159 static void discard_buffer(uint32_t id);
161 static int port_no(struct datapath *dp, struct sw_port *p)
163 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
164 return p - dp->ports;
167 /* Generates a unique datapath id. It incorporates the datapath index
168 * and a hardware address, if available. If not, it generates a random
172 gen_datapath_id(void)
174 /* Choose a random datapath id. */
180 for (i = 0; i < ETH_ADDR_LEN; i++) {
181 id |= (uint64_t)(rand() & 0xff) << (8*(ETH_ADDR_LEN-1 - i));
188 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
192 dp = calloc(1, sizeof *dp);
197 dp->last_timeout = time(0);
198 list_init(&dp->remotes);
199 dp->controller = remote_create(dp, rconn);
200 dp->listen_vconn = NULL;
201 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
202 dp->chain = chain_create();
204 VLOG_ERR("could not create chain");
209 list_init(&dp->port_list);
210 dp->config.flags = 0;
211 dp->config.miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
217 dp_add_port(struct datapath *dp, const char *name)
219 struct netdev *netdev;
225 error = netdev_open(name, &netdev);
229 if (netdev_get_in4(netdev, &in4)) {
230 VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
232 if (netdev_get_in6(netdev, &in6)) {
233 char in6_name[INET6_ADDRSTRLEN + 1];
234 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
235 VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
238 for (p = dp->ports; ; p++) {
239 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
241 } else if (!p->netdev) {
251 list_push_back(&dp->port_list, &p->node);
253 /* Notify the ctlpath that this port has been added */
254 send_port_status(p, OFPPR_ADD);
260 dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
262 assert(!dp->listen_vconn);
263 dp->listen_vconn = listen_vconn;
267 dp_run(struct datapath *dp)
269 time_t now = time(0);
270 struct sw_port *p, *pn;
271 struct remote *r, *rn;
272 struct buffer *buffer = NULL;
274 if (now != dp->last_timeout) {
275 struct list deleted = LIST_INITIALIZER(&deleted);
276 struct sw_flow *f, *n;
278 chain_timeout(dp->chain, &deleted);
279 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
280 send_flow_expired(dp, f);
281 list_remove(&f->node);
284 dp->last_timeout = now;
286 poll_timer_wait(1000);
288 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
292 /* Allocate buffer with some headroom to add headers in forwarding
293 * to the controller or adding a vlan tag, plus an extra 2 bytes to
294 * allow IP headers to be aligned on a 4-byte boundary. */
295 const int headroom = 128 + 2;
296 const int hard_header = VLAN_ETH_HEADER_LEN;
297 const int mtu = netdev_get_mtu(p->netdev);
298 buffer = buffer_new(headroom + hard_header + mtu);
299 buffer->data += headroom;
301 error = netdev_recv(p->netdev, buffer);
304 fwd_port_input(dp, buffer, port_no(dp, p));
306 } else if (error != EAGAIN) {
307 VLOG_ERR("Error receiving data from %s: %s",
308 netdev_get_name(p->netdev), strerror(error));
312 buffer_delete(buffer);
314 /* Talk to remotes. */
315 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
318 if (dp->listen_vconn) {
320 struct vconn *new_vconn;
323 retval = vconn_accept(dp->listen_vconn, &new_vconn);
325 if (retval != EAGAIN) {
326 VLOG_WARN("accept failed (%s)", strerror(retval));
330 remote_create(dp, rconn_new_from_vconn("passive", 128, new_vconn));
336 remote_run(struct datapath *dp, struct remote *r)
342 /* Do some remote processing, but cap it at a reasonable amount so that
343 * other processing doesn't starve. */
344 for (i = 0; i < 50; i++) {
346 struct buffer *buffer;
347 struct ofp_header *oh;
349 buffer = rconn_recv(r->rconn);
354 if (buffer->size >= sizeof *oh) {
355 struct sender sender;
359 sender.xid = oh->xid;
360 fwd_control_input(dp, &sender, buffer->data, buffer->size);
362 VLOG_WARN("received too-short OpenFlow message");
364 buffer_delete(buffer);
366 if (!rconn_is_full(r->rconn)) {
367 int error = r->cb_dump(dp, r->cb_aux);
370 VLOG_WARN("dump callback error: %s", strerror(-error));
372 r->cb_done(r->cb_aux);
381 if (!rconn_is_alive(r->rconn)) {
387 remote_wait(struct remote *r)
389 rconn_run_wait(r->rconn);
390 rconn_recv_wait(r->rconn);
394 remote_destroy(struct remote *r)
397 if (r->cb_dump && r->cb_done) {
398 r->cb_done(r->cb_aux);
400 list_remove(&r->node);
401 rconn_destroy(r->rconn);
406 static struct remote *
407 remote_create(struct datapath *dp, struct rconn *rconn)
409 struct remote *remote = xmalloc(sizeof *remote);
410 list_push_back(&dp->remotes, &remote->node);
411 remote->rconn = rconn;
412 remote->cb_dump = NULL;
416 /* Starts a callback-based, reliable, possibly multi-message reply to a
417 * request made by 'remote'.
419 * 'dump' designates a function that will be called when the 'remote' send
420 * queue has an empty slot. It should compose a message and send it on
421 * 'remote'. On success, it should return 1 if it should be called again when
422 * another send queue slot opens up, 0 if its transmissions are complete, or a
423 * negative errno value on failure.
425 * 'done' designates a function to clean up any resources allocated for the
426 * dump. It must handle being called before the dump is complete (which will
427 * happen if 'remote' is closed unexpectedly).
429 * 'aux' is passed to 'dump' and 'done'. */
431 remote_start_dump(struct remote *remote,
432 int (*dump)(struct datapath *, void *),
433 void (*done)(void *),
436 assert(!remote->cb_dump);
437 remote->cb_dump = dump;
438 remote->cb_done = done;
439 remote->cb_aux = aux;
443 dp_wait(struct datapath *dp)
448 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
449 netdev_recv_wait(p->netdev);
451 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
454 if (dp->listen_vconn) {
455 vconn_accept_wait(dp->listen_vconn);
459 /* Delete 'p' from switch. */
461 del_switch_port(struct sw_port *p)
463 send_port_status(p, OFPPR_DELETE);
464 netdev_close(p->netdev);
466 list_remove(&p->node);
470 dp_destroy(struct datapath *dp)
472 struct sw_port *p, *n;
478 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
481 chain_destroy(dp->chain);
485 /* Send packets out all the ports except the originating one. If the
486 * "flood" argument is set, don't send out ports with flooding disabled.
489 output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
495 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
496 if (port_no(dp, p) == in_port) {
499 if (flood && p->flags & BRIDGE_PORT_NO_FLOOD) {
502 if (prev_port != -1) {
503 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
505 prev_port = port_no(dp, p);
508 dp_output_port(dp, buffer, in_port, prev_port);
510 buffer_delete(buffer);
516 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
518 if (out_port >= 0 && out_port < OFPP_MAX) {
519 struct sw_port *p = &dp->ports[out_port];
520 if (p->netdev != NULL) {
521 if (!netdev_send(p->netdev, buffer)) {
530 buffer_delete(buffer);
531 /* FIXME: ratelimit */
532 VLOG_DBG("can't forward to bad port %d\n", out_port);
535 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
538 dp_output_port(struct datapath *dp, struct buffer *buffer,
539 int in_port, int out_port)
543 if (out_port == OFPP_FLOOD) {
544 output_all(dp, buffer, in_port, 1);
545 } else if (out_port == OFPP_ALL) {
546 output_all(dp, buffer, in_port, 0);
547 } else if (out_port == OFPP_CONTROLLER) {
548 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
549 } else if (out_port == OFPP_TABLE) {
550 struct sw_flow_key key;
551 struct sw_flow *flow;
554 flow_extract(buffer, in_port, &key.flow);
555 flow = chain_lookup(dp->chain, &key);
557 flow_used(flow, buffer);
558 execute_actions(dp, buffer, in_port, &key,
559 flow->actions, flow->n_actions);
562 output_packet(dp, buffer, out_port);
567 alloc_openflow_buffer(struct datapath *dp, size_t openflow_len, uint8_t type,
568 const struct sender *sender, struct buffer **bufferp)
570 struct buffer *buffer;
571 struct ofp_header *oh;
573 buffer = *bufferp = buffer_new(openflow_len);
574 oh = buffer_put_uninit(buffer, openflow_len);
575 oh->version = OFP_VERSION;
577 oh->length = 0; /* Filled in by send_openflow_buffer(). */
578 oh->xid = sender ? sender->xid : 0;
583 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
584 const struct sender *sender)
586 struct remote *remote = sender ? sender->remote : dp->controller;
587 struct rconn *rconn = remote->rconn;
588 struct ofp_header *oh;
591 oh = buffer_at_assert(buffer, 0, sizeof *oh);
592 oh->length = htons(buffer->size);
594 retval = rconn_send(rconn, buffer);
596 VLOG_WARN("send to %s failed: %s",
597 rconn_get_name(rconn), strerror(retval));
598 buffer_delete(buffer);
603 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
604 * packet can be saved in a buffer, then only the first max_len bytes of
605 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
606 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
607 * the caller wants to be sent; a value of 0 indicates the entire packet should
610 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
611 size_t max_len, int reason)
613 struct ofp_packet_in *opi;
617 buffer_id = save_buffer(buffer);
618 total_len = buffer->size;
619 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
620 buffer->size = max_len;
623 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
624 opi->header.version = OFP_VERSION;
625 opi->header.type = OFPT_PACKET_IN;
626 opi->header.length = htons(buffer->size);
627 opi->header.xid = htonl(0);
628 opi->buffer_id = htonl(buffer_id);
629 opi->total_len = htons(total_len);
630 opi->in_port = htons(in_port);
631 opi->reason = reason;
633 send_openflow_buffer(dp, buffer, NULL);
636 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
637 struct ofp_phy_port *desc)
639 desc->port_no = htons(port_no(dp, p));
640 strncpy((char *) desc->name, netdev_get_name(p->netdev),
642 desc->name[sizeof desc->name - 1] = '\0';
643 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
644 desc->flags = htonl(p->flags);
645 desc->features = htonl(netdev_get_features(p->netdev));
646 desc->speed = htonl(netdev_get_speed(p->netdev));
650 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
652 struct buffer *buffer;
653 struct ofp_switch_features *ofr;
656 ofr = alloc_openflow_buffer(dp, sizeof *ofr, OFPT_FEATURES_REPLY,
658 ofr->datapath_id = htonll(dp->id);
659 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
660 ofr->n_compression = 0; /* Not supported */
661 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
662 ofr->buffer_mb = htonl(UINT32_MAX);
663 ofr->n_buffers = htonl(N_PKT_BUFFERS);
664 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
665 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
666 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
667 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
668 memset(opp, 0, sizeof *opp);
669 fill_port_desc(dp, p, opp);
671 send_openflow_buffer(dp, buffer, sender);
675 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
677 int port_no = ntohs(opp->port_no);
678 if (port_no < OFPP_MAX) {
679 struct sw_port *p = &dp->ports[port_no];
681 /* Make sure the port id hasn't changed since this was sent */
682 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
683 ETH_ADDR_LEN) != 0) {
686 p->flags = htonl(opp->flags);
691 send_port_status(struct sw_port *p, uint8_t status)
693 struct buffer *buffer;
694 struct ofp_port_status *ops;
695 ops = alloc_openflow_buffer(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
697 ops->reason = status;
698 memset(ops->pad, 0, sizeof ops->pad);
699 fill_port_desc(p->dp, p, &ops->desc);
701 send_openflow_buffer(p->dp, buffer, NULL);
705 send_flow_expired(struct datapath *dp, struct sw_flow *flow)
707 struct buffer *buffer;
708 struct ofp_flow_expired *ofe;
709 ofe = alloc_openflow_buffer(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, NULL,
711 flow_fill_match(&ofe->match, &flow->key);
713 memset(ofe->pad, 0, sizeof ofe->pad);
714 ofe->priority = htons(flow->priority);
716 ofe->duration = htonl(flow->timeout - flow->max_idle - flow->created);
717 ofe->packet_count = htonll(flow->packet_count);
718 ofe->byte_count = htonll(flow->byte_count);
719 send_openflow_buffer(dp, buffer, NULL);
723 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
724 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
726 struct buffer *buffer;
727 struct ofp_error_msg *oem;
728 oem = alloc_openflow_buffer(dp, sizeof(*oem)+len, OFPT_ERROR_MSG,
730 oem->type = htons(type);
731 oem->code = htons(code);
732 memcpy(oem->data, data, len);
733 send_openflow_buffer(dp, buffer, sender);
737 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
738 int table_idx, time_t now)
740 struct ofp_flow_stats *ofs;
741 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
742 ofs = buffer_put_uninit(buffer, length);
743 ofs->length = htons(length);
744 ofs->table_id = table_idx;
746 ofs->match.wildcards = htons(flow->key.wildcards);
747 ofs->match.in_port = flow->key.flow.in_port;
748 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
749 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
750 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
751 ofs->match.dl_type = flow->key.flow.dl_type;
752 ofs->match.nw_src = flow->key.flow.nw_src;
753 ofs->match.nw_dst = flow->key.flow.nw_dst;
754 ofs->match.nw_proto = flow->key.flow.nw_proto;
755 memset(ofs->match.pad, 0, sizeof ofs->match.pad);
756 ofs->match.tp_src = flow->key.flow.tp_src;
757 ofs->match.tp_dst = flow->key.flow.tp_dst;
758 ofs->duration = htonl(now - flow->created);
759 ofs->packet_count = htonll(flow->packet_count);
760 ofs->byte_count = htonll(flow->byte_count);
761 ofs->priority = htons(flow->priority);
762 ofs->max_idle = htons(flow->max_idle);
763 memcpy(ofs->actions, flow->actions,
764 sizeof *ofs->actions * flow->n_actions);
768 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
769 * OFPP_MAX. Process it according to 'chain'. */
770 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
772 struct sw_flow_key key;
773 struct sw_flow *flow;
776 flow_extract(buffer, in_port, &key.flow);
777 flow = chain_lookup(dp->chain, &key);
779 flow_used(flow, buffer);
780 execute_actions(dp, buffer, in_port, &key,
781 flow->actions, flow->n_actions);
783 dp_output_control(dp, buffer, in_port, ntohs(dp->config.miss_send_len),
789 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
790 size_t max_len, int out_port)
792 if (out_port != OFPP_CONTROLLER) {
793 dp_output_port(dp, buffer, in_port, out_port);
795 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
800 execute_actions(struct datapath *dp, struct buffer *buffer,
801 int in_port, const struct sw_flow_key *key,
802 const struct ofp_action *actions, int n_actions)
804 /* Every output action needs a separate clone of 'buffer', but the common
805 * case is just a single output action, so that doing a clone and then
806 * freeing the original buffer is wasteful. So the following code is
807 * slightly obscure just to avoid that. */
809 size_t max_len=0; /* Initialze to make compiler happy */
814 eth_proto = ntohs(key->flow.dl_type);
816 for (i = 0; i < n_actions; i++) {
817 const struct ofp_action *a = &actions[i];
818 struct eth_header *eh = buffer->l2;
820 if (prev_port != -1) {
821 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
825 switch (ntohs(a->type)) {
827 prev_port = ntohs(a->arg.output.port);
828 max_len = ntohs(a->arg.output.max_len);
831 case OFPAT_SET_DL_VLAN:
832 modify_vlan(buffer, key, a);
835 case OFPAT_SET_DL_SRC:
836 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
839 case OFPAT_SET_DL_DST:
840 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
843 case OFPAT_SET_NW_SRC:
844 case OFPAT_SET_NW_DST:
845 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
848 case OFPAT_SET_TP_SRC:
849 case OFPAT_SET_TP_DST:
850 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
858 do_output(dp, buffer, in_port, max_len, prev_port);
860 buffer_delete(buffer);
863 /* Returns the new checksum for a packet in which the checksum field previously
864 * contained 'old_csum' and in which a field that contained 'old_u16' was
865 * changed to contain 'new_u16'. */
867 recalc_csum16(uint16_t old_csum, uint16_t old_u16, uint16_t new_u16)
869 /* Ones-complement arithmetic is endian-independent, so this code does not
870 * use htons() or ntohs().
872 * See RFC 1624 for formula and explanation. */
873 uint16_t hc_complement = ~old_csum;
874 uint16_t m_complement = ~old_u16;
875 uint16_t m_prime = new_u16;
876 uint32_t sum = hc_complement + m_complement + m_prime;
877 uint16_t hc_prime_complement = sum + (sum >> 16);
878 return ~hc_prime_complement;
881 /* Returns the new checksum for a packet in which the checksum field previously
882 * contained 'old_csum' and in which a field that contained 'old_u32' was
883 * changed to contain 'new_u32'. */
885 recalc_csum32(uint16_t old_csum, uint32_t old_u32, uint32_t new_u32)
887 return recalc_csum16(recalc_csum16(old_csum, old_u32, new_u32),
888 old_u32 >> 16, new_u32 >> 16);
891 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
892 uint8_t nw_proto, const struct ofp_action *a)
894 if (eth_proto == ETH_TYPE_IP) {
895 struct ip_header *nh = buffer->l3;
896 uint32_t new, *field;
898 new = a->arg.nw_addr;
899 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
900 if (nw_proto == IP_TYPE_TCP) {
901 struct tcp_header *th = buffer->l4;
902 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
903 } else if (nw_proto == IP_TYPE_UDP) {
904 struct udp_header *th = buffer->l4;
906 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
908 th->udp_csum = 0xffff;
912 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
917 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
918 uint8_t nw_proto, const struct ofp_action *a)
920 if (eth_proto == ETH_TYPE_IP) {
921 uint16_t new, *field;
925 if (nw_proto == IP_TYPE_TCP) {
926 struct tcp_header *th = buffer->l4;
927 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
928 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
930 } else if (nw_proto == IP_TYPE_UDP) {
931 struct udp_header *th = buffer->l4;
932 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
933 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
940 modify_vlan(struct buffer *buffer,
941 const struct sw_flow_key *key, const struct ofp_action *a)
943 uint16_t new_id = a->arg.vlan_id;
944 struct vlan_eth_header *veh;
946 if (new_id != OFP_VLAN_NONE) {
947 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
948 /* Modify vlan id, but maintain other TCI values */
950 veh->veth_tci &= ~htons(VLAN_VID);
951 veh->veth_tci |= htons(new_id);
953 /* Insert new vlan id. */
954 struct eth_header *eh = buffer->l2;
955 struct vlan_eth_header tmp;
956 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
957 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
958 tmp.veth_type = htons(ETH_TYPE_VLAN);
959 tmp.veth_tci = new_id;
960 tmp.veth_next_type = eh->eth_type;
962 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
963 memcpy(veh, &tmp, sizeof tmp);
964 buffer->l2 -= VLAN_HEADER_LEN;
967 /* Remove an existing vlan header if it exists */
969 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
970 struct eth_header tmp;
972 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
973 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
974 tmp.eth_type = veh->veth_next_type;
976 buffer->size -= VLAN_HEADER_LEN;
977 buffer->data += VLAN_HEADER_LEN;
978 buffer->l2 += VLAN_HEADER_LEN;
979 memcpy(buffer->data, &tmp, sizeof tmp);
985 recv_features_request(struct datapath *dp, const struct sender *sender,
988 dp_send_features_reply(dp, sender);
993 recv_get_config_request(struct datapath *dp, const struct sender *sender,
996 struct buffer *buffer;
997 struct ofp_switch_config *osc;
999 osc = alloc_openflow_buffer(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY,
1002 assert(sizeof *osc == sizeof dp->config);
1003 memcpy(((char *)osc) + sizeof osc->header,
1004 ((char *)&dp->config) + sizeof dp->config.header,
1005 sizeof dp->config - sizeof dp->config.header);
1007 return send_openflow_buffer(dp, buffer, sender);
1011 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
1014 const struct ofp_switch_config *osc = msg;
1020 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1023 const struct ofp_packet_out *opo = msg;
1025 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1026 /* FIXME: can we avoid copying data here? */
1027 int data_len = ntohs(opo->header.length) - sizeof *opo;
1028 struct buffer *buffer = buffer_new(data_len);
1029 buffer_put(buffer, opo->u.data, data_len);
1030 dp_output_port(dp, buffer,
1031 ntohs(opo->in_port), ntohs(opo->out_port));
1033 struct sw_flow_key key;
1034 struct buffer *buffer;
1037 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1042 n_acts = (ntohs(opo->header.length) - sizeof *opo)
1043 / sizeof *opo->u.actions;
1044 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1045 execute_actions(dp, buffer, ntohs(opo->in_port),
1046 &key, opo->u.actions, n_acts);
1052 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1055 const struct ofp_port_mod *opm = msg;
1057 dp_update_port_flags(dp, &opm->desc);
1063 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1065 int error = -ENOMEM;
1068 struct sw_flow *flow;
1071 /* To prevent loops, make sure there's no action to send to the
1072 * OFP_TABLE virtual port.
1074 n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
1075 for (i=0; i<n_acts; i++) {
1076 const struct ofp_action *a = &ofm->actions[i];
1078 if (a->type == htons(OFPAT_OUTPUT)
1079 && a->arg.output.port == htons(OFPP_TABLE)) {
1080 /* xxx Send fancy new error message? */
1085 /* Allocate memory. */
1086 flow = flow_alloc(n_acts);
1090 /* Fill out flow. */
1091 flow_extract_match(&flow->key, &ofm->match);
1092 flow->max_idle = ntohs(ofm->max_idle);
1093 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1094 flow->timeout = time(0) + flow->max_idle; /* FIXME */
1095 flow->n_actions = n_acts;
1096 flow->created = time(0); /* FIXME */
1097 flow->byte_count = 0;
1098 flow->packet_count = 0;
1099 memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
1102 error = chain_insert(dp->chain, flow);
1104 goto error_free_flow;
1107 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1108 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1110 struct sw_flow_key key;
1111 uint16_t in_port = ntohs(ofm->match.in_port);
1112 flow_used(flow, buffer);
1113 flow_extract(buffer, in_port, &key.flow);
1114 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
1124 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1125 discard_buffer(ntohl(ofm->buffer_id));
1130 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1133 const struct ofp_flow_mod *ofm = msg;
1134 uint16_t command = ntohs(ofm->command);
1136 if (command == OFPFC_ADD) {
1137 return add_flow(dp, ofm);
1138 } else if (command == OFPFC_DELETE) {
1139 struct sw_flow_key key;
1140 flow_extract_match(&key, &ofm->match);
1141 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1142 } else if (command == OFPFC_DELETE_STRICT) {
1143 struct sw_flow_key key;
1145 flow_extract_match(&key, &ofm->match);
1146 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1147 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1153 struct flow_stats_state {
1155 struct sw_table_position position;
1156 struct ofp_flow_stats_request rq;
1159 struct buffer *buffer;
1162 #define MAX_FLOW_STATS_BYTES 4096
1164 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1167 const struct ofp_flow_stats_request *fsr = body;
1168 struct flow_stats_state *s = xmalloc(sizeof *s);
1169 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1170 memset(&s->position, 0, sizeof s->position);
1176 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1178 struct flow_stats_state *s = private;
1179 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1180 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1183 static int flow_stats_dump(struct datapath *dp, void *state,
1184 struct buffer *buffer)
1186 struct flow_stats_state *s = state;
1187 struct sw_flow_key match_key;
1189 flow_extract_match(&match_key, &s->rq.match);
1192 while (s->table_idx < dp->chain->n_tables
1193 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1195 struct sw_table *table = dp->chain->tables[s->table_idx];
1197 if (table->iterate(table, &match_key, &s->position,
1198 flow_stats_dump_callback, s))
1202 memset(&s->position, 0, sizeof s->position);
1204 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1207 static void flow_stats_done(void *state)
1212 struct aggregate_stats_state {
1213 struct ofp_aggregate_stats_request rq;
1216 static int aggregate_stats_init(struct datapath *dp,
1217 const void *body, int body_len,
1220 const struct ofp_aggregate_stats_request *rq = body;
1221 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1227 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1229 struct ofp_aggregate_stats_reply *rpy = private;
1230 rpy->packet_count += flow->packet_count;
1231 rpy->byte_count += flow->byte_count;
1236 static int aggregate_stats_dump(struct datapath *dp, void *state,
1237 struct buffer *buffer)
1239 struct aggregate_stats_state *s = state;
1240 struct ofp_aggregate_stats_request *rq = &s->rq;
1241 struct ofp_aggregate_stats_reply *rpy;
1242 struct sw_table_position position;
1243 struct sw_flow_key match_key;
1246 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1247 memset(rpy, 0, sizeof *rpy);
1249 flow_extract_match(&match_key, &rq->match);
1250 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1251 memset(&position, 0, sizeof position);
1252 while (table_idx < dp->chain->n_tables
1253 && (rq->table_id == 0xff || rq->table_id == table_idx))
1255 struct sw_table *table = dp->chain->tables[table_idx];
1258 error = table->iterate(table, &match_key, &position,
1259 aggregate_stats_dump_callback, rpy);
1264 memset(&position, 0, sizeof position);
1267 rpy->packet_count = htonll(rpy->packet_count);
1268 rpy->byte_count = htonll(rpy->byte_count);
1269 rpy->flow_count = htonl(rpy->flow_count);
1273 static void aggregate_stats_done(void *state)
1278 static int table_stats_dump(struct datapath *dp, void *state,
1279 struct buffer *buffer)
1282 for (i = 0; i < dp->chain->n_tables; i++) {
1283 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1284 struct sw_table_stats stats;
1285 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1286 strncpy(ots->name, stats.name, sizeof ots->name);
1288 memset(ots->pad, 0, sizeof ots->pad);
1289 ots->max_entries = htonl(stats.max_flows);
1290 ots->active_count = htonl(stats.n_flows);
1291 ots->matched_count = htonll(0); /* FIXME */
1296 struct port_stats_state {
1300 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1303 struct port_stats_state *s = xmalloc(sizeof *s);
1309 static int port_stats_dump(struct datapath *dp, void *state,
1310 struct buffer *buffer)
1312 struct port_stats_state *s = state;
1315 for (i = s->port; i < OFPP_MAX; i++) {
1316 struct sw_port *p = &dp->ports[i];
1317 struct ofp_port_stats *ops;
1321 ops = buffer_put_uninit(buffer, sizeof *ops);
1322 ops->port_no = htons(port_no(dp, p));
1323 memset(ops->pad, 0, sizeof ops->pad);
1324 ops->rx_count = htonll(p->rx_count);
1325 ops->tx_count = htonll(p->tx_count);
1326 ops->drop_count = htonll(p->drop_count);
1333 static void port_stats_done(void *state)
1339 /* Minimum and maximum acceptable number of bytes in body member of
1340 * struct ofp_stats_request. */
1341 size_t min_body, max_body;
1343 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1344 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1345 * Returns zero if successful, otherwise a negative error code.
1346 * May initialize '*state' to state information. May be null if no
1347 * initialization is required.*/
1348 int (*init)(struct datapath *dp, const void *body, int body_len,
1351 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1352 * struct ofp_stats_reply. On success, it should return 1 if it should be
1353 * called again later with another buffer, 0 if it is done, or a negative
1354 * errno value on failure. */
1355 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1357 /* Cleans any state created by the init or dump functions. May be null
1358 * if no cleanup is required. */
1359 void (*done)(void *state);
1362 static const struct stats_type stats[] = {
1364 sizeof(struct ofp_flow_stats_request),
1365 sizeof(struct ofp_flow_stats_request),
1370 [OFPST_AGGREGATE] = {
1371 sizeof(struct ofp_aggregate_stats_request),
1372 sizeof(struct ofp_aggregate_stats_request),
1373 aggregate_stats_init,
1374 aggregate_stats_dump,
1375 aggregate_stats_done
1393 struct stats_dump_cb {
1395 struct ofp_stats_request *rq;
1396 struct sender sender;
1397 const struct stats_type *s;
1402 stats_dump(struct datapath *dp, void *cb_)
1404 struct stats_dump_cb *cb = cb_;
1405 struct ofp_stats_reply *osr;
1406 struct buffer *buffer;
1413 osr = alloc_openflow_buffer(dp, sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1415 osr->type = htons(cb->s - stats);
1418 err = cb->s->dump(dp, cb->state, buffer);
1424 /* Buffer might have been reallocated, so find our data again. */
1425 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1426 osr->flags = ntohs(OFPSF_REPLY_MORE);
1428 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1438 stats_done(void *cb_)
1440 struct stats_dump_cb *cb = cb_;
1443 cb->s->done(cb->state);
1450 recv_stats_request(struct datapath *dp, const struct sender *sender,
1453 const struct ofp_stats_request *rq = oh;
1454 size_t rq_len = ntohs(rq->header.length);
1455 struct stats_dump_cb *cb;
1459 type = ntohs(rq->type);
1460 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1461 VLOG_WARN("received stats request of unknown type %d", type);
1465 cb = xmalloc(sizeof *cb);
1467 cb->rq = xmemdup(rq, rq_len);
1468 cb->sender = *sender;
1469 cb->s = &stats[type];
1472 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1473 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1474 VLOG_WARN("stats request type %d with bad body length %d",
1481 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1483 VLOG_WARN("failed initialization of stats request type %d: %s",
1484 type, strerror(-err));
1489 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1498 /* 'msg', which is 'length' bytes long, was received from the control path.
1499 * Apply it to 'chain'. */
1501 fwd_control_input(struct datapath *dp, const struct sender *sender,
1502 const void *msg, size_t length)
1504 struct openflow_packet {
1506 int (*handler)(struct datapath *, const struct sender *, const void *);
1509 static const struct openflow_packet packets[] = {
1510 [OFPT_FEATURES_REQUEST] = {
1511 sizeof (struct ofp_header),
1512 recv_features_request,
1514 [OFPT_GET_CONFIG_REQUEST] = {
1515 sizeof (struct ofp_header),
1516 recv_get_config_request,
1518 [OFPT_SET_CONFIG] = {
1519 sizeof (struct ofp_switch_config),
1522 [OFPT_PACKET_OUT] = {
1523 sizeof (struct ofp_packet_out),
1527 sizeof (struct ofp_flow_mod),
1531 sizeof (struct ofp_port_mod),
1534 [OFPT_STATS_REQUEST] = {
1535 sizeof (struct ofp_stats_request),
1540 const struct openflow_packet *pkt;
1541 struct ofp_header *oh;
1543 oh = (struct ofp_header *) msg;
1544 if (oh->version != OFP_VERSION || oh->type >= ARRAY_SIZE(packets)
1545 || ntohs(oh->length) > length)
1548 pkt = &packets[oh->type];
1551 if (length < pkt->min_size)
1554 return pkt->handler(dp, sender, msg);
1557 /* Packet buffering. */
1559 #define OVERWRITE_SECS 1
1561 struct packet_buffer {
1562 struct buffer *buffer;
1567 static struct packet_buffer buffers[N_PKT_BUFFERS];
1568 static unsigned int buffer_idx;
1570 uint32_t save_buffer(struct buffer *buffer)
1572 struct packet_buffer *p;
1575 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1576 p = &buffers[buffer_idx];
1578 /* Don't buffer packet if existing entry is less than
1579 * OVERWRITE_SECS old. */
1580 if (time(0) < p->timeout) { /* FIXME */
1583 buffer_delete(p->buffer);
1586 /* Don't use maximum cookie value since the all-bits-1 id is
1588 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1590 p->buffer = buffer_clone(buffer); /* FIXME */
1591 p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
1592 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1597 static struct buffer *retrieve_buffer(uint32_t id)
1599 struct buffer *buffer = NULL;
1600 struct packet_buffer *p;
1602 p = &buffers[id & PKT_BUFFER_MASK];
1603 if (p->cookie == id >> PKT_BUFFER_BITS) {
1607 printf("cookie mismatch: %x != %x\n",
1608 id >> PKT_BUFFER_BITS, p->cookie);
1614 static void discard_buffer(uint32_t id)
1616 struct packet_buffer *p;
1618 p = &buffers[id & PKT_BUFFER_MASK];
1619 if (p->cookie == id >> PKT_BUFFER_BITS) {
1620 buffer_delete(p->buffer);