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
57 #define BRIDGE_PORT_NO_FLOOD 0x00000001
63 /* Capabilities supported by this implementation. */
64 #define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
69 /* Actions supported by this implementation. */
70 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
71 | (1 << OFPAT_SET_DL_VLAN) \
72 | (1 << OFPAT_SET_DL_SRC) \
73 | (1 << OFPAT_SET_DL_DST) \
74 | (1 << OFPAT_SET_NW_SRC) \
75 | (1 << OFPAT_SET_NW_DST) \
76 | (1 << OFPAT_SET_TP_SRC) \
77 | (1 << OFPAT_SET_TP_DST) )
82 struct netdev *netdev;
83 struct list node; /* Element in datapath.ports. */
84 unsigned long long int rx_packets, tx_packets;
85 unsigned long long int rx_bytes, tx_bytes;
86 unsigned long long int tx_dropped;
89 /* The origin of a received OpenFlow message, to enable sending a reply. */
91 struct remote *remote; /* The device that sent the message. */
92 uint32_t xid; /* The OpenFlow transaction ID. */
95 /* A connection to a controller or a management device. */
99 #define TXQ_LIMIT 128 /* Max number of packets to queue for tx. */
100 int n_txq; /* Number of packets queued for tx on rconn. */
102 /* Support for reliable, multi-message replies to requests.
104 * If an incoming request needs to have a reliable reply that might
105 * require multiple messages, it can use remote_start_dump() to set up
106 * a callback that will be called as buffer space for replies. */
107 int (*cb_dump)(struct datapath *, void *aux);
108 void (*cb_done)(void *aux);
113 /* Remote connections. */
114 struct remote *controller; /* Connection to controller. */
115 struct list remotes; /* All connections (including controller). */
116 struct vconn *listen_vconn;
120 /* Unique identifier for this datapath */
123 struct sw_chain *chain; /* Forwarding rules. */
125 /* Configuration set from controller. */
127 uint16_t miss_send_len;
130 struct sw_port ports[OFPP_MAX];
131 struct list port_list; /* List of ports, for flooding. */
134 static struct remote *remote_create(struct datapath *, struct rconn *);
135 static void remote_run(struct datapath *, struct remote *);
136 static void remote_wait(struct remote *);
137 static void remote_destroy(struct remote *);
139 void dp_output_port(struct datapath *, struct buffer *,
140 int in_port, int out_port);
141 void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
142 void dp_output_control(struct datapath *, struct buffer *, int in_port,
143 size_t max_len, int reason);
144 static void send_flow_expired(struct datapath *, struct sw_flow *,
145 enum ofp_flow_expired_reason);
146 static void send_port_status(struct sw_port *p, uint8_t status);
147 static void del_switch_port(struct sw_port *p);
148 static void execute_actions(struct datapath *, struct buffer *,
149 int in_port, const struct sw_flow_key *,
150 const struct ofp_action *, int n_actions);
151 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
152 const struct ofp_action *a);
153 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
154 uint8_t nw_proto, const struct ofp_action *a);
155 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
156 uint8_t nw_proto, const struct ofp_action *a);
158 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
159 * into a buffer number (low bits) and a cookie (high bits). The buffer number
160 * is an index into an array of buffers. The cookie distinguishes between
161 * different packets that have occupied a single buffer. Thus, the more
162 * buffers we have, the lower-quality the cookie... */
163 #define PKT_BUFFER_BITS 8
164 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
165 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
167 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
169 int run_flow_through_tables(struct datapath *, struct buffer *, int in_port);
170 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
171 int fwd_control_input(struct datapath *, const struct sender *,
172 const void *, size_t);
174 uint32_t save_buffer(struct buffer *);
175 static struct buffer *retrieve_buffer(uint32_t id);
176 static void discard_buffer(uint32_t id);
178 static int port_no(struct datapath *dp, struct sw_port *p)
180 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
181 return p - dp->ports;
184 /* Generates and returns a random datapath id. */
186 gen_datapath_id(void)
188 uint8_t ea[ETH_ADDR_LEN];
190 return eth_addr_to_uint64(ea);
194 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
198 dp = calloc(1, sizeof *dp);
203 dp->last_timeout = time_now();
204 list_init(&dp->remotes);
205 dp->controller = remote_create(dp, rconn);
206 dp->listen_vconn = NULL;
207 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
208 dp->chain = chain_create();
210 VLOG_ERR("could not create chain");
215 list_init(&dp->port_list);
217 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
223 dp_add_port(struct datapath *dp, const char *name)
225 struct netdev *netdev;
231 error = netdev_open(name, NETDEV_ETH_TYPE_ANY, &netdev);
235 error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false);
237 VLOG_ERR("Couldn't set promiscuous mode on %s device", name);
238 netdev_close(netdev);
241 if (netdev_get_in4(netdev, &in4)) {
242 VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
244 if (netdev_get_in6(netdev, &in6)) {
245 char in6_name[INET6_ADDRSTRLEN + 1];
246 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
247 VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
250 for (p = dp->ports; ; p++) {
251 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
253 } else if (!p->netdev) {
258 memset(p, '\0', sizeof *p);
262 list_push_back(&dp->port_list, &p->node);
264 /* Notify the ctlpath that this port has been added */
265 send_port_status(p, OFPPR_ADD);
271 dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
273 assert(!dp->listen_vconn);
274 dp->listen_vconn = listen_vconn;
278 dp_run(struct datapath *dp)
280 time_t now = time_now();
281 struct sw_port *p, *pn;
282 struct remote *r, *rn;
283 struct buffer *buffer = NULL;
285 if (now != dp->last_timeout) {
286 struct list deleted = LIST_INITIALIZER(&deleted);
287 struct sw_flow *f, *n;
289 chain_timeout(dp->chain, &deleted);
290 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
291 send_flow_expired(dp, f, f->reason);
292 list_remove(&f->node);
295 dp->last_timeout = now;
297 poll_timer_wait(1000);
299 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
303 /* Allocate buffer with some headroom to add headers in forwarding
304 * to the controller or adding a vlan tag, plus an extra 2 bytes to
305 * allow IP headers to be aligned on a 4-byte boundary. */
306 const int headroom = 128 + 2;
307 const int hard_header = VLAN_ETH_HEADER_LEN;
308 const int mtu = netdev_get_mtu(p->netdev);
309 buffer = buffer_new(headroom + hard_header + mtu);
310 buffer->data += headroom;
312 error = netdev_recv(p->netdev, buffer);
315 p->rx_bytes += buffer->size;
316 fwd_port_input(dp, buffer, port_no(dp, p));
318 } else if (error != EAGAIN) {
319 VLOG_ERR("Error receiving data from %s: %s",
320 netdev_get_name(p->netdev), strerror(error));
324 buffer_delete(buffer);
326 /* Talk to remotes. */
327 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
330 if (dp->listen_vconn) {
332 struct vconn *new_vconn;
335 retval = vconn_accept(dp->listen_vconn, &new_vconn);
337 if (retval != EAGAIN) {
338 VLOG_WARN("accept failed (%s)", strerror(retval));
342 remote_create(dp, rconn_new_from_vconn("passive", new_vconn));
348 remote_run(struct datapath *dp, struct remote *r)
354 /* Do some remote processing, but cap it at a reasonable amount so that
355 * other processing doesn't starve. */
356 for (i = 0; i < 50; i++) {
358 struct buffer *buffer;
359 struct ofp_header *oh;
361 buffer = rconn_recv(r->rconn);
366 if (buffer->size >= sizeof *oh) {
367 struct sender sender;
371 sender.xid = oh->xid;
372 fwd_control_input(dp, &sender, buffer->data, buffer->size);
374 VLOG_WARN("received too-short OpenFlow message");
376 buffer_delete(buffer);
378 if (r->n_txq < TXQ_LIMIT) {
379 int error = r->cb_dump(dp, r->cb_aux);
382 VLOG_WARN("dump callback error: %s", strerror(-error));
384 r->cb_done(r->cb_aux);
393 if (!rconn_is_alive(r->rconn)) {
399 remote_wait(struct remote *r)
401 rconn_run_wait(r->rconn);
402 rconn_recv_wait(r->rconn);
406 remote_destroy(struct remote *r)
409 if (r->cb_dump && r->cb_done) {
410 r->cb_done(r->cb_aux);
412 list_remove(&r->node);
413 rconn_destroy(r->rconn);
418 static struct remote *
419 remote_create(struct datapath *dp, struct rconn *rconn)
421 struct remote *remote = xmalloc(sizeof *remote);
422 list_push_back(&dp->remotes, &remote->node);
423 remote->rconn = rconn;
424 remote->cb_dump = NULL;
428 /* Starts a callback-based, reliable, possibly multi-message reply to a
429 * request made by 'remote'.
431 * 'dump' designates a function that will be called when the 'remote' send
432 * queue has an empty slot. It should compose a message and send it on
433 * 'remote'. On success, it should return 1 if it should be called again when
434 * another send queue slot opens up, 0 if its transmissions are complete, or a
435 * negative errno value on failure.
437 * 'done' designates a function to clean up any resources allocated for the
438 * dump. It must handle being called before the dump is complete (which will
439 * happen if 'remote' is closed unexpectedly).
441 * 'aux' is passed to 'dump' and 'done'. */
443 remote_start_dump(struct remote *remote,
444 int (*dump)(struct datapath *, void *),
445 void (*done)(void *),
448 assert(!remote->cb_dump);
449 remote->cb_dump = dump;
450 remote->cb_done = done;
451 remote->cb_aux = aux;
455 dp_wait(struct datapath *dp)
460 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
461 netdev_recv_wait(p->netdev);
463 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
466 if (dp->listen_vconn) {
467 vconn_accept_wait(dp->listen_vconn);
471 /* Delete 'p' from switch. */
473 del_switch_port(struct sw_port *p)
475 send_port_status(p, OFPPR_DELETE);
476 netdev_close(p->netdev);
478 list_remove(&p->node);
482 dp_destroy(struct datapath *dp)
484 struct sw_port *p, *n;
490 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
493 chain_destroy(dp->chain);
497 /* Send packets out all the ports except the originating one. If the
498 * "flood" argument is set, don't send out ports with flooding disabled.
501 output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
507 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
508 if (port_no(dp, p) == in_port) {
511 if (flood && p->flags & BRIDGE_PORT_NO_FLOOD) {
514 if (prev_port != -1) {
515 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
517 prev_port = port_no(dp, p);
520 dp_output_port(dp, buffer, in_port, prev_port);
522 buffer_delete(buffer);
528 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
530 if (out_port >= 0 && out_port < OFPP_MAX) {
531 struct sw_port *p = &dp->ports[out_port];
532 if (p->netdev != NULL) {
533 if (!netdev_send(p->netdev, buffer)) {
535 p->tx_bytes += buffer->size;
543 buffer_delete(buffer);
544 /* FIXME: ratelimit */
545 VLOG_DBG("can't forward to bad port %d\n", out_port);
548 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
551 dp_output_port(struct datapath *dp, struct buffer *buffer,
552 int in_port, int out_port)
556 if (out_port == OFPP_FLOOD) {
557 output_all(dp, buffer, in_port, 1);
558 } else if (out_port == OFPP_ALL) {
559 output_all(dp, buffer, in_port, 0);
560 } else if (out_port == OFPP_CONTROLLER) {
561 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
562 } else if (out_port == OFPP_IN_PORT) {
563 output_packet(dp, buffer, in_port);
564 } else if (out_port == OFPP_TABLE) {
565 if (run_flow_through_tables(dp, buffer, in_port)) {
566 buffer_delete(buffer);
569 if (in_port == out_port) {
570 /* FIXME: ratelimit */
571 VLOG_DBG("can't directly forward to input port");
574 output_packet(dp, buffer, out_port);
579 make_openflow_reply(size_t openflow_len, uint8_t type,
580 const struct sender *sender, struct buffer **bufferp)
582 return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
587 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
588 const struct sender *sender)
590 struct remote *remote = sender ? sender->remote : dp->controller;
591 struct rconn *rconn = remote->rconn;
594 update_openflow_length(buffer);
595 retval = (remote->n_txq < TXQ_LIMIT
596 ? rconn_send(rconn, buffer, &remote->n_txq)
599 VLOG_WARN("send to %s failed: %s",
600 rconn_get_name(rconn), strerror(retval));
601 buffer_delete(buffer);
606 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
607 * packet can be saved in a buffer, then only the first max_len bytes of
608 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
609 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
610 * the caller wants to be sent; a value of 0 indicates the entire packet should
613 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
614 size_t max_len, int reason)
616 struct ofp_packet_in *opi;
620 buffer_id = save_buffer(buffer);
621 total_len = buffer->size;
622 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
623 buffer->size = max_len;
626 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
627 opi->header.version = OFP_VERSION;
628 opi->header.type = OFPT_PACKET_IN;
629 opi->header.length = htons(buffer->size);
630 opi->header.xid = htonl(0);
631 opi->buffer_id = htonl(buffer_id);
632 opi->total_len = htons(total_len);
633 opi->in_port = htons(in_port);
634 opi->reason = reason;
636 send_openflow_buffer(dp, buffer, NULL);
639 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
640 struct ofp_phy_port *desc)
642 desc->port_no = htons(port_no(dp, p));
643 strncpy((char *) desc->name, netdev_get_name(p->netdev),
645 desc->name[sizeof desc->name - 1] = '\0';
646 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
647 desc->flags = htonl(p->flags);
648 desc->features = htonl(netdev_get_features(p->netdev));
649 desc->speed = htonl(netdev_get_speed(p->netdev));
653 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
655 struct buffer *buffer;
656 struct ofp_switch_features *ofr;
659 ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
661 ofr->datapath_id = htonll(dp->id);
662 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
663 ofr->n_compression = 0; /* Not supported */
664 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
665 ofr->buffer_mb = htonl(UINT32_MAX);
666 ofr->n_buffers = htonl(N_PKT_BUFFERS);
667 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
668 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
669 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
670 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
671 memset(opp, 0, sizeof *opp);
672 fill_port_desc(dp, p, opp);
674 send_openflow_buffer(dp, buffer, sender);
678 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
680 int port_no = ntohs(opp->port_no);
681 if (port_no < OFPP_MAX) {
682 struct sw_port *p = &dp->ports[port_no];
684 /* Make sure the port id hasn't changed since this was sent */
685 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
686 ETH_ADDR_LEN) != 0) {
689 p->flags = htonl(opp->flags);
694 send_port_status(struct sw_port *p, uint8_t status)
696 struct buffer *buffer;
697 struct ofp_port_status *ops;
698 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
699 ops->reason = status;
700 memset(ops->pad, 0, sizeof ops->pad);
701 fill_port_desc(p->dp, p, &ops->desc);
703 send_openflow_buffer(p->dp, buffer, NULL);
707 send_flow_expired(struct datapath *dp, struct sw_flow *flow,
708 enum ofp_flow_expired_reason reason)
710 struct buffer *buffer;
711 struct ofp_flow_expired *ofe;
712 ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
713 flow_fill_match(&ofe->match, &flow->key);
715 ofe->priority = htons(flow->priority);
716 ofe->reason = reason;
717 memset(ofe->pad, 0, sizeof ofe->pad);
719 ofe->duration = htonl(time_now() - flow->created);
720 memset(ofe->pad2, 0, sizeof ofe->pad2);
721 ofe->packet_count = htonll(flow->packet_count);
722 ofe->byte_count = htonll(flow->byte_count);
723 send_openflow_buffer(dp, buffer, NULL);
727 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
728 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
730 struct buffer *buffer;
731 struct ofp_error_msg *oem;
732 oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG,
734 oem->type = htons(type);
735 oem->code = htons(code);
736 memcpy(oem->data, data, len);
737 send_openflow_buffer(dp, buffer, sender);
741 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
742 int table_idx, time_t now)
744 struct ofp_flow_stats *ofs;
745 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
746 ofs = buffer_put_uninit(buffer, length);
747 ofs->length = htons(length);
748 ofs->table_id = table_idx;
750 ofs->match.wildcards = htonl(flow->key.wildcards);
751 ofs->match.in_port = flow->key.flow.in_port;
752 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
753 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
754 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
755 ofs->match.dl_type = flow->key.flow.dl_type;
756 ofs->match.nw_src = flow->key.flow.nw_src;
757 ofs->match.nw_dst = flow->key.flow.nw_dst;
758 ofs->match.nw_proto = flow->key.flow.nw_proto;
760 ofs->match.tp_src = flow->key.flow.tp_src;
761 ofs->match.tp_dst = flow->key.flow.tp_dst;
762 ofs->duration = htonl(now - flow->created);
763 ofs->priority = htons(flow->priority);
764 ofs->idle_timeout = htons(flow->idle_timeout);
765 ofs->hard_timeout = htons(flow->hard_timeout);
766 memset(ofs->pad2, 0, sizeof ofs->pad2);
767 ofs->packet_count = htonll(flow->packet_count);
768 ofs->byte_count = htonll(flow->byte_count);
769 memcpy(ofs->actions, flow->actions,
770 sizeof *ofs->actions * flow->n_actions);
774 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
775 * OFPP_MAX. Process it according to 'dp''s flow table. Returns 0 if
776 * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no
777 * matching flow, in which case 'buffer' still belongs to the caller. */
778 int run_flow_through_tables(struct datapath *dp, struct buffer *buffer,
781 struct sw_flow_key key;
782 struct sw_flow *flow;
785 if (flow_extract(buffer, in_port, &key.flow)
786 && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
788 buffer_delete(buffer);
792 flow = chain_lookup(dp->chain, &key);
794 flow_used(flow, buffer);
795 execute_actions(dp, buffer, in_port, &key,
796 flow->actions, flow->n_actions);
803 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
804 * OFPP_MAX. Process it according to 'dp''s flow table, sending it up to the
805 * controller if no flow matches. Takes ownership of 'buffer'. */
806 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
808 if (run_flow_through_tables(dp, buffer, in_port)) {
809 dp_output_control(dp, buffer, in_port, dp->miss_send_len,
815 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
816 size_t max_len, int out_port)
818 if (out_port != OFPP_CONTROLLER) {
819 dp_output_port(dp, buffer, in_port, out_port);
821 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
826 execute_actions(struct datapath *dp, struct buffer *buffer,
827 int in_port, const struct sw_flow_key *key,
828 const struct ofp_action *actions, int n_actions)
830 /* Every output action needs a separate clone of 'buffer', but the common
831 * case is just a single output action, so that doing a clone and then
832 * freeing the original buffer is wasteful. So the following code is
833 * slightly obscure just to avoid that. */
835 size_t max_len=0; /* Initialze to make compiler happy */
840 eth_proto = ntohs(key->flow.dl_type);
842 for (i = 0; i < n_actions; i++) {
843 const struct ofp_action *a = &actions[i];
844 struct eth_header *eh = buffer->l2;
846 if (prev_port != -1) {
847 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
851 switch (ntohs(a->type)) {
853 prev_port = ntohs(a->arg.output.port);
854 max_len = ntohs(a->arg.output.max_len);
857 case OFPAT_SET_DL_VLAN:
858 modify_vlan(buffer, key, a);
861 case OFPAT_SET_DL_SRC:
862 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
865 case OFPAT_SET_DL_DST:
866 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
869 case OFPAT_SET_NW_SRC:
870 case OFPAT_SET_NW_DST:
871 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
874 case OFPAT_SET_TP_SRC:
875 case OFPAT_SET_TP_DST:
876 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
884 do_output(dp, buffer, in_port, max_len, prev_port);
886 buffer_delete(buffer);
889 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
890 uint8_t nw_proto, const struct ofp_action *a)
892 if (eth_proto == ETH_TYPE_IP) {
893 struct ip_header *nh = buffer->l3;
894 uint32_t new, *field;
896 new = a->arg.nw_addr;
897 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
898 if (nw_proto == IP_TYPE_TCP) {
899 struct tcp_header *th = buffer->l4;
900 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
901 } else if (nw_proto == IP_TYPE_UDP) {
902 struct udp_header *th = buffer->l4;
904 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
906 th->udp_csum = 0xffff;
910 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
915 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
916 uint8_t nw_proto, const struct ofp_action *a)
918 if (eth_proto == ETH_TYPE_IP) {
919 uint16_t new, *field;
923 if (nw_proto == IP_TYPE_TCP) {
924 struct tcp_header *th = buffer->l4;
925 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
926 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
928 } else if (nw_proto == IP_TYPE_UDP) {
929 struct udp_header *th = buffer->l4;
930 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
931 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
938 modify_vlan(struct buffer *buffer,
939 const struct sw_flow_key *key, const struct ofp_action *a)
941 uint16_t new_id = a->arg.vlan_id;
942 struct vlan_eth_header *veh;
944 if (new_id != htons(OFP_VLAN_NONE)) {
945 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
946 /* Modify vlan id, but maintain other TCI values */
948 veh->veth_tci &= ~htons(VLAN_VID);
949 veh->veth_tci |= new_id;
951 /* Insert new vlan id. */
952 struct eth_header *eh = buffer->l2;
953 struct vlan_eth_header tmp;
954 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
955 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
956 tmp.veth_type = htons(ETH_TYPE_VLAN);
957 tmp.veth_tci = new_id;
958 tmp.veth_next_type = eh->eth_type;
960 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
961 memcpy(veh, &tmp, sizeof tmp);
962 buffer->l2 -= VLAN_HEADER_LEN;
965 /* Remove an existing vlan header if it exists */
967 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
968 struct eth_header tmp;
970 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
971 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
972 tmp.eth_type = veh->veth_next_type;
974 buffer->size -= VLAN_HEADER_LEN;
975 buffer->data += VLAN_HEADER_LEN;
976 buffer->l2 += VLAN_HEADER_LEN;
977 memcpy(buffer->data, &tmp, sizeof tmp);
983 recv_features_request(struct datapath *dp, const struct sender *sender,
986 dp_send_features_reply(dp, sender);
991 recv_get_config_request(struct datapath *dp, const struct sender *sender,
994 struct buffer *buffer;
995 struct ofp_switch_config *osc;
997 osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
1000 osc->flags = htons(dp->flags);
1001 osc->miss_send_len = htons(dp->miss_send_len);
1003 return send_openflow_buffer(dp, buffer, sender);
1007 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
1010 const struct ofp_switch_config *osc = msg;
1013 flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
1014 if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
1015 && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
1016 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
1019 dp->miss_send_len = ntohs(osc->miss_send_len);
1024 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1027 const struct ofp_packet_out *opo = msg;
1028 struct sw_flow_key key;
1029 struct buffer *buffer;
1030 int n_actions = ntohs(opo->n_actions);
1031 int act_len = n_actions * sizeof opo->actions[0];
1033 if (act_len > (ntohs(opo->header.length) - sizeof *opo)) {
1034 VLOG_DBG("message too short for number of actions");
1038 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1039 /* FIXME: can we avoid copying data here? */
1040 int data_len = ntohs(opo->header.length) - sizeof *opo - act_len;
1041 buffer = buffer_new(data_len);
1042 buffer_put(buffer, &opo->actions[n_actions], data_len);
1044 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1050 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1051 execute_actions(dp, buffer, ntohs(opo->in_port),
1052 &key, opo->actions, n_actions);
1058 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1061 const struct ofp_port_mod *opm = msg;
1063 dp_update_port_flags(dp, &opm->desc);
1069 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1071 int error = -ENOMEM;
1074 struct sw_flow *flow;
1077 /* To prevent loops, make sure there's no action to send to the
1078 * OFP_TABLE virtual port.
1080 n_actions = (ntohs(ofm->header.length) - sizeof *ofm)
1081 / sizeof *ofm->actions;
1082 for (i=0; i<n_actions; i++) {
1083 const struct ofp_action *a = &ofm->actions[i];
1085 if (a->type == htons(OFPAT_OUTPUT)
1086 && (a->arg.output.port == htons(OFPP_TABLE)
1087 || a->arg.output.port == htons(OFPP_NONE)
1088 || a->arg.output.port == ofm->match.in_port)) {
1089 /* xxx Send fancy new error message? */
1094 /* Allocate memory. */
1095 flow = flow_alloc(n_actions);
1099 /* Fill out flow. */
1100 flow_extract_match(&flow->key, &ofm->match);
1101 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1102 flow->idle_timeout = ntohs(ofm->idle_timeout);
1103 flow->hard_timeout = ntohs(ofm->hard_timeout);
1104 flow->used = flow->created = time_now();
1105 flow->n_actions = n_actions;
1106 flow->byte_count = 0;
1107 flow->packet_count = 0;
1108 memcpy(flow->actions, ofm->actions, n_actions * sizeof *flow->actions);
1111 error = chain_insert(dp->chain, flow);
1113 goto error_free_flow;
1116 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1117 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1119 struct sw_flow_key key;
1120 uint16_t in_port = ntohs(ofm->match.in_port);
1121 flow_used(flow, buffer);
1122 flow_extract(buffer, in_port, &key.flow);
1123 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_actions);
1133 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1134 discard_buffer(ntohl(ofm->buffer_id));
1139 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1142 const struct ofp_flow_mod *ofm = msg;
1143 uint16_t command = ntohs(ofm->command);
1145 if (command == OFPFC_ADD) {
1146 return add_flow(dp, ofm);
1147 } else if (command == OFPFC_DELETE) {
1148 struct sw_flow_key key;
1149 flow_extract_match(&key, &ofm->match);
1150 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1151 } else if (command == OFPFC_DELETE_STRICT) {
1152 struct sw_flow_key key;
1154 flow_extract_match(&key, &ofm->match);
1155 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1156 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1162 static int version_stats_dump(struct datapath *dp, void *state,
1163 struct buffer *buffer)
1165 struct ofp_version_stats *ovs = buffer_put_uninit(buffer, sizeof *ovs);
1167 strncpy(ovs->mfr_desc, &mfr_desc, sizeof ovs->mfr_desc);
1168 strncpy(ovs->hw_desc, &hw_desc, sizeof ovs->hw_desc);
1169 strncpy(ovs->sw_desc, &sw_desc, sizeof ovs->sw_desc);
1174 struct flow_stats_state {
1176 struct sw_table_position position;
1177 struct ofp_flow_stats_request rq;
1180 struct buffer *buffer;
1183 #define MAX_FLOW_STATS_BYTES 4096
1185 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1188 const struct ofp_flow_stats_request *fsr = body;
1189 struct flow_stats_state *s = xmalloc(sizeof *s);
1190 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1191 memset(&s->position, 0, sizeof s->position);
1197 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1199 struct flow_stats_state *s = private;
1200 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1201 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1204 static int flow_stats_dump(struct datapath *dp, void *state,
1205 struct buffer *buffer)
1207 struct flow_stats_state *s = state;
1208 struct sw_flow_key match_key;
1210 flow_extract_match(&match_key, &s->rq.match);
1212 s->now = time_now();
1213 while (s->table_idx < dp->chain->n_tables
1214 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1216 struct sw_table *table = dp->chain->tables[s->table_idx];
1218 if (table->iterate(table, &match_key, &s->position,
1219 flow_stats_dump_callback, s))
1223 memset(&s->position, 0, sizeof s->position);
1225 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1228 static void flow_stats_done(void *state)
1233 struct aggregate_stats_state {
1234 struct ofp_aggregate_stats_request rq;
1237 static int aggregate_stats_init(struct datapath *dp,
1238 const void *body, int body_len,
1241 const struct ofp_aggregate_stats_request *rq = body;
1242 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1248 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1250 struct ofp_aggregate_stats_reply *rpy = private;
1251 rpy->packet_count += flow->packet_count;
1252 rpy->byte_count += flow->byte_count;
1257 static int aggregate_stats_dump(struct datapath *dp, void *state,
1258 struct buffer *buffer)
1260 struct aggregate_stats_state *s = state;
1261 struct ofp_aggregate_stats_request *rq = &s->rq;
1262 struct ofp_aggregate_stats_reply *rpy;
1263 struct sw_table_position position;
1264 struct sw_flow_key match_key;
1267 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1268 memset(rpy, 0, sizeof *rpy);
1270 flow_extract_match(&match_key, &rq->match);
1271 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1272 memset(&position, 0, sizeof position);
1273 while (table_idx < dp->chain->n_tables
1274 && (rq->table_id == 0xff || rq->table_id == table_idx))
1276 struct sw_table *table = dp->chain->tables[table_idx];
1279 error = table->iterate(table, &match_key, &position,
1280 aggregate_stats_dump_callback, rpy);
1285 memset(&position, 0, sizeof position);
1288 rpy->packet_count = htonll(rpy->packet_count);
1289 rpy->byte_count = htonll(rpy->byte_count);
1290 rpy->flow_count = htonl(rpy->flow_count);
1294 static void aggregate_stats_done(void *state)
1299 static int table_stats_dump(struct datapath *dp, void *state,
1300 struct buffer *buffer)
1303 for (i = 0; i < dp->chain->n_tables; i++) {
1304 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1305 struct sw_table_stats stats;
1306 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1307 strncpy(ots->name, stats.name, sizeof ots->name);
1309 memset(ots->pad, 0, sizeof ots->pad);
1310 ots->max_entries = htonl(stats.max_flows);
1311 ots->active_count = htonl(stats.n_flows);
1312 ots->matched_count = htonll(stats.n_matched);
1317 struct port_stats_state {
1321 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1324 struct port_stats_state *s = xmalloc(sizeof *s);
1330 static int port_stats_dump(struct datapath *dp, void *state,
1331 struct buffer *buffer)
1333 struct port_stats_state *s = state;
1336 for (i = s->port; i < OFPP_MAX; i++) {
1337 struct sw_port *p = &dp->ports[i];
1338 struct ofp_port_stats *ops;
1342 ops = buffer_put_uninit(buffer, sizeof *ops);
1343 ops->port_no = htons(port_no(dp, p));
1344 memset(ops->pad, 0, sizeof ops->pad);
1345 ops->rx_packets = htonll(p->rx_packets);
1346 ops->tx_packets = htonll(p->tx_packets);
1347 ops->rx_bytes = htonll(p->rx_bytes);
1348 ops->tx_bytes = htonll(p->tx_bytes);
1349 ops->rx_dropped = htonll(-1);
1350 ops->tx_dropped = htonll(p->tx_dropped);
1351 ops->rx_errors = htonll(-1);
1352 ops->tx_errors = htonll(-1);
1353 ops->rx_frame_err = htonll(-1);
1354 ops->rx_over_err = htonll(-1);
1355 ops->rx_crc_err = htonll(-1);
1356 ops->collisions = htonll(-1);
1363 static void port_stats_done(void *state)
1369 /* Minimum and maximum acceptable number of bytes in body member of
1370 * struct ofp_stats_request. */
1371 size_t min_body, max_body;
1373 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1374 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1375 * Returns zero if successful, otherwise a negative error code.
1376 * May initialize '*state' to state information. May be null if no
1377 * initialization is required.*/
1378 int (*init)(struct datapath *dp, const void *body, int body_len,
1381 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1382 * struct ofp_stats_reply. On success, it should return 1 if it should be
1383 * called again later with another buffer, 0 if it is done, or a negative
1384 * errno value on failure. */
1385 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1387 /* Cleans any state created by the init or dump functions. May be null
1388 * if no cleanup is required. */
1389 void (*done)(void *state);
1392 static const struct stats_type stats[] = {
1401 sizeof(struct ofp_flow_stats_request),
1402 sizeof(struct ofp_flow_stats_request),
1407 [OFPST_AGGREGATE] = {
1408 sizeof(struct ofp_aggregate_stats_request),
1409 sizeof(struct ofp_aggregate_stats_request),
1410 aggregate_stats_init,
1411 aggregate_stats_dump,
1412 aggregate_stats_done
1430 struct stats_dump_cb {
1432 struct ofp_stats_request *rq;
1433 struct sender sender;
1434 const struct stats_type *s;
1439 stats_dump(struct datapath *dp, void *cb_)
1441 struct stats_dump_cb *cb = cb_;
1442 struct ofp_stats_reply *osr;
1443 struct buffer *buffer;
1450 osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1452 osr->type = htons(cb->s - stats);
1455 err = cb->s->dump(dp, cb->state, buffer);
1461 /* Buffer might have been reallocated, so find our data again. */
1462 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1463 osr->flags = ntohs(OFPSF_REPLY_MORE);
1465 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1475 stats_done(void *cb_)
1477 struct stats_dump_cb *cb = cb_;
1480 cb->s->done(cb->state);
1487 recv_stats_request(struct datapath *dp, const struct sender *sender,
1490 const struct ofp_stats_request *rq = oh;
1491 size_t rq_len = ntohs(rq->header.length);
1492 struct stats_dump_cb *cb;
1496 type = ntohs(rq->type);
1497 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1498 VLOG_WARN("received stats request of unknown type %d", type);
1502 cb = xmalloc(sizeof *cb);
1504 cb->rq = xmemdup(rq, rq_len);
1505 cb->sender = *sender;
1506 cb->s = &stats[type];
1509 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1510 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1511 VLOG_WARN("stats request type %d with bad body length %d",
1518 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1520 VLOG_WARN("failed initialization of stats request type %d: %s",
1521 type, strerror(-err));
1526 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1536 recv_echo_request(struct datapath *dp, const struct sender *sender,
1539 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1543 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1544 const void *oh UNUSED)
1549 /* 'msg', which is 'length' bytes long, was received from the control path.
1550 * Apply it to 'chain'. */
1552 fwd_control_input(struct datapath *dp, const struct sender *sender,
1553 const void *msg, size_t length)
1555 struct openflow_packet {
1557 int (*handler)(struct datapath *, const struct sender *, const void *);
1560 static const struct openflow_packet packets[] = {
1561 [OFPT_FEATURES_REQUEST] = {
1562 sizeof (struct ofp_header),
1563 recv_features_request,
1565 [OFPT_GET_CONFIG_REQUEST] = {
1566 sizeof (struct ofp_header),
1567 recv_get_config_request,
1569 [OFPT_SET_CONFIG] = {
1570 sizeof (struct ofp_switch_config),
1573 [OFPT_PACKET_OUT] = {
1574 sizeof (struct ofp_packet_out),
1578 sizeof (struct ofp_flow_mod),
1582 sizeof (struct ofp_port_mod),
1585 [OFPT_STATS_REQUEST] = {
1586 sizeof (struct ofp_stats_request),
1589 [OFPT_ECHO_REQUEST] = {
1590 sizeof (struct ofp_header),
1593 [OFPT_ECHO_REPLY] = {
1594 sizeof (struct ofp_header),
1599 const struct openflow_packet *pkt;
1600 struct ofp_header *oh;
1602 oh = (struct ofp_header *) msg;
1603 assert(oh->version == OFP_VERSION);
1604 if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length)
1607 pkt = &packets[oh->type];
1610 if (length < pkt->min_size)
1613 return pkt->handler(dp, sender, msg);
1616 /* Packet buffering. */
1618 #define OVERWRITE_SECS 1
1620 struct packet_buffer {
1621 struct buffer *buffer;
1626 static struct packet_buffer buffers[N_PKT_BUFFERS];
1627 static unsigned int buffer_idx;
1629 uint32_t save_buffer(struct buffer *buffer)
1631 struct packet_buffer *p;
1634 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1635 p = &buffers[buffer_idx];
1637 /* Don't buffer packet if existing entry is less than
1638 * OVERWRITE_SECS old. */
1639 if (time_now() < p->timeout) { /* FIXME */
1642 buffer_delete(p->buffer);
1645 /* Don't use maximum cookie value since the all-bits-1 id is
1647 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1649 p->buffer = buffer_clone(buffer); /* FIXME */
1650 p->timeout = time_now() + OVERWRITE_SECS; /* FIXME */
1651 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1656 static struct buffer *retrieve_buffer(uint32_t id)
1658 struct buffer *buffer = NULL;
1659 struct packet_buffer *p;
1661 p = &buffers[id & PKT_BUFFER_MASK];
1662 if (p->cookie == id >> PKT_BUFFER_BITS) {
1666 printf("cookie mismatch: %x != %x\n",
1667 id >> PKT_BUFFER_BITS, p->cookie);
1673 static void discard_buffer(uint32_t id)
1675 struct packet_buffer *p;
1677 p = &buffers[id & PKT_BUFFER_MASK];
1678 if (p->cookie == id >> PKT_BUFFER_BITS) {
1679 buffer_delete(p->buffer);