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"
53 #define THIS_MODULE VLM_datapath
56 #define BRIDGE_PORT_NO_FLOOD 0x00000001
58 /* Capabilities supported by this implementation. */
59 #define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
64 /* Actions supported by this implementation. */
65 #define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
66 | (1 << OFPAT_SET_DL_VLAN) \
67 | (1 << OFPAT_SET_DL_SRC) \
68 | (1 << OFPAT_SET_DL_DST) \
69 | (1 << OFPAT_SET_NW_SRC) \
70 | (1 << OFPAT_SET_NW_DST) \
71 | (1 << OFPAT_SET_TP_SRC) \
72 | (1 << OFPAT_SET_TP_DST) )
77 struct netdev *netdev;
78 struct list node; /* Element in datapath.ports. */
79 unsigned long long int rx_count, tx_count, drop_count;
82 /* The origin of a received OpenFlow message, to enable sending a reply. */
84 struct remote *remote; /* The device that sent the message. */
85 uint32_t xid; /* The OpenFlow transaction ID. */
88 /* A connection to a controller or a management device. */
93 /* Support for reliable, multi-message replies to requests.
95 * If an incoming request needs to have a reliable reply that might
96 * require multiple messages, it can use remote_start_dump() to set up
97 * a callback that will be called as buffer space for replies. */
98 int (*cb_dump)(struct datapath *, void *aux);
99 void (*cb_done)(void *aux);
104 /* Remote connections. */
105 struct remote *controller; /* Connection to controller. */
106 struct list remotes; /* All connections (including controller). */
107 struct vconn *listen_vconn;
111 /* Unique identifier for this datapath */
114 struct sw_chain *chain; /* Forwarding rules. */
116 struct ofp_switch_config config;
119 struct sw_port ports[OFPP_MAX];
120 struct list port_list; /* List of ports, for flooding. */
123 static struct remote *remote_create(struct datapath *, struct rconn *);
124 static void remote_run(struct datapath *, struct remote *);
125 static void remote_wait(struct remote *);
126 static void remote_destroy(struct remote *);
128 void dp_output_port(struct datapath *, struct buffer *,
129 int in_port, int out_port);
130 void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
131 void dp_output_control(struct datapath *, struct buffer *, int in_port,
132 size_t max_len, int reason);
133 static void send_flow_expired(struct datapath *, struct sw_flow *);
134 static void send_port_status(struct sw_port *p, uint8_t status);
135 static void del_switch_port(struct sw_port *p);
136 static void execute_actions(struct datapath *, struct buffer *,
137 int in_port, const struct sw_flow_key *,
138 const struct ofp_action *, int n_actions);
139 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
140 const struct ofp_action *a);
141 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
142 uint8_t nw_proto, const struct ofp_action *a);
143 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
144 uint8_t nw_proto, const struct ofp_action *a);
146 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
147 * into a buffer number (low bits) and a cookie (high bits). The buffer number
148 * is an index into an array of buffers. The cookie distinguishes between
149 * different packets that have occupied a single buffer. Thus, the more
150 * buffers we have, the lower-quality the cookie... */
151 #define PKT_BUFFER_BITS 8
152 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
153 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
155 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
157 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
158 int fwd_control_input(struct datapath *, const struct sender *,
159 const void *, size_t);
161 uint32_t save_buffer(struct buffer *);
162 static struct buffer *retrieve_buffer(uint32_t id);
163 static void discard_buffer(uint32_t id);
165 static int port_no(struct datapath *dp, struct sw_port *p)
167 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
168 return p - dp->ports;
171 /* Generates and returns a random datapath id. */
173 gen_datapath_id(void)
175 uint8_t ea[ETH_ADDR_LEN];
177 return eth_addr_to_uint64(ea);
181 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
185 dp = calloc(1, sizeof *dp);
190 dp->last_timeout = time(0);
191 list_init(&dp->remotes);
192 dp->controller = remote_create(dp, rconn);
193 dp->listen_vconn = NULL;
194 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
195 dp->chain = chain_create();
197 VLOG_ERR("could not create chain");
202 list_init(&dp->port_list);
203 dp->config.flags = 0;
204 dp->config.miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
210 dp_add_port(struct datapath *dp, const char *name)
212 struct netdev *netdev;
218 error = netdev_open(name, NETDEV_ETH_TYPE_ANY, &netdev);
222 error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false);
224 VLOG_ERR("Couldn't set promiscuous mode on %s device", name);
225 netdev_close(netdev);
228 if (netdev_get_in4(netdev, &in4)) {
229 VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
231 if (netdev_get_in6(netdev, &in6)) {
232 char in6_name[INET6_ADDRSTRLEN + 1];
233 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
234 VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
237 for (p = dp->ports; ; p++) {
238 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
240 } else if (!p->netdev) {
250 list_push_back(&dp->port_list, &p->node);
252 /* Notify the ctlpath that this port has been added */
253 send_port_status(p, OFPPR_ADD);
259 dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
261 assert(!dp->listen_vconn);
262 dp->listen_vconn = listen_vconn;
266 dp_run(struct datapath *dp)
268 time_t now = time(0);
269 struct sw_port *p, *pn;
270 struct remote *r, *rn;
271 struct buffer *buffer = NULL;
273 if (now != dp->last_timeout) {
274 struct list deleted = LIST_INITIALIZER(&deleted);
275 struct sw_flow *f, *n;
277 chain_timeout(dp->chain, &deleted);
278 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
279 send_flow_expired(dp, f);
280 list_remove(&f->node);
283 dp->last_timeout = now;
285 poll_timer_wait(1000);
287 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
291 /* Allocate buffer with some headroom to add headers in forwarding
292 * to the controller or adding a vlan tag, plus an extra 2 bytes to
293 * allow IP headers to be aligned on a 4-byte boundary. */
294 const int headroom = 128 + 2;
295 const int hard_header = VLAN_ETH_HEADER_LEN;
296 const int mtu = netdev_get_mtu(p->netdev);
297 buffer = buffer_new(headroom + hard_header + mtu);
298 buffer->data += headroom;
300 error = netdev_recv(p->netdev, buffer);
303 fwd_port_input(dp, buffer, port_no(dp, p));
305 } else if (error != EAGAIN) {
306 VLOG_ERR("Error receiving data from %s: %s",
307 netdev_get_name(p->netdev), strerror(error));
311 buffer_delete(buffer);
313 /* Talk to remotes. */
314 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
317 if (dp->listen_vconn) {
319 struct vconn *new_vconn;
322 retval = vconn_accept(dp->listen_vconn, &new_vconn);
324 if (retval != EAGAIN) {
325 VLOG_WARN("accept failed (%s)", strerror(retval));
329 remote_create(dp, rconn_new_from_vconn("passive", 128, new_vconn));
335 remote_run(struct datapath *dp, struct remote *r)
341 /* Do some remote processing, but cap it at a reasonable amount so that
342 * other processing doesn't starve. */
343 for (i = 0; i < 50; i++) {
345 struct buffer *buffer;
346 struct ofp_header *oh;
348 buffer = rconn_recv(r->rconn);
353 if (buffer->size >= sizeof *oh) {
354 struct sender sender;
358 sender.xid = oh->xid;
359 fwd_control_input(dp, &sender, buffer->data, buffer->size);
361 VLOG_WARN("received too-short OpenFlow message");
363 buffer_delete(buffer);
365 if (!rconn_is_full(r->rconn)) {
366 int error = r->cb_dump(dp, r->cb_aux);
369 VLOG_WARN("dump callback error: %s", strerror(-error));
371 r->cb_done(r->cb_aux);
380 if (!rconn_is_alive(r->rconn)) {
386 remote_wait(struct remote *r)
388 rconn_run_wait(r->rconn);
389 rconn_recv_wait(r->rconn);
393 remote_destroy(struct remote *r)
396 if (r->cb_dump && r->cb_done) {
397 r->cb_done(r->cb_aux);
399 list_remove(&r->node);
400 rconn_destroy(r->rconn);
405 static struct remote *
406 remote_create(struct datapath *dp, struct rconn *rconn)
408 struct remote *remote = xmalloc(sizeof *remote);
409 list_push_back(&dp->remotes, &remote->node);
410 remote->rconn = rconn;
411 remote->cb_dump = NULL;
415 /* Starts a callback-based, reliable, possibly multi-message reply to a
416 * request made by 'remote'.
418 * 'dump' designates a function that will be called when the 'remote' send
419 * queue has an empty slot. It should compose a message and send it on
420 * 'remote'. On success, it should return 1 if it should be called again when
421 * another send queue slot opens up, 0 if its transmissions are complete, or a
422 * negative errno value on failure.
424 * 'done' designates a function to clean up any resources allocated for the
425 * dump. It must handle being called before the dump is complete (which will
426 * happen if 'remote' is closed unexpectedly).
428 * 'aux' is passed to 'dump' and 'done'. */
430 remote_start_dump(struct remote *remote,
431 int (*dump)(struct datapath *, void *),
432 void (*done)(void *),
435 assert(!remote->cb_dump);
436 remote->cb_dump = dump;
437 remote->cb_done = done;
438 remote->cb_aux = aux;
442 dp_wait(struct datapath *dp)
447 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
448 netdev_recv_wait(p->netdev);
450 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
453 if (dp->listen_vconn) {
454 vconn_accept_wait(dp->listen_vconn);
458 /* Delete 'p' from switch. */
460 del_switch_port(struct sw_port *p)
462 send_port_status(p, OFPPR_DELETE);
463 netdev_close(p->netdev);
465 list_remove(&p->node);
469 dp_destroy(struct datapath *dp)
471 struct sw_port *p, *n;
477 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
480 chain_destroy(dp->chain);
484 /* Send packets out all the ports except the originating one. If the
485 * "flood" argument is set, don't send out ports with flooding disabled.
488 output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
494 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
495 if (port_no(dp, p) == in_port) {
498 if (flood && p->flags & BRIDGE_PORT_NO_FLOOD) {
501 if (prev_port != -1) {
502 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
504 prev_port = port_no(dp, p);
507 dp_output_port(dp, buffer, in_port, prev_port);
509 buffer_delete(buffer);
515 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
517 if (out_port >= 0 && out_port < OFPP_MAX) {
518 struct sw_port *p = &dp->ports[out_port];
519 if (p->netdev != NULL) {
520 if (!netdev_send(p->netdev, buffer)) {
529 buffer_delete(buffer);
530 /* FIXME: ratelimit */
531 VLOG_DBG("can't forward to bad port %d\n", out_port);
534 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
537 dp_output_port(struct datapath *dp, struct buffer *buffer,
538 int in_port, int out_port)
542 if (out_port == OFPP_FLOOD) {
543 output_all(dp, buffer, in_port, 1);
544 } else if (out_port == OFPP_ALL) {
545 output_all(dp, buffer, in_port, 0);
546 } else if (out_port == OFPP_CONTROLLER) {
547 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
548 } else if (out_port == OFPP_TABLE) {
549 struct sw_flow_key key;
550 struct sw_flow *flow;
553 flow_extract(buffer, in_port, &key.flow);
554 flow = chain_lookup(dp->chain, &key);
556 flow_used(flow, buffer);
557 execute_actions(dp, buffer, in_port, &key,
558 flow->actions, flow->n_actions);
560 buffer_delete(buffer);
563 output_packet(dp, buffer, out_port);
568 make_openflow_reply(size_t openflow_len, uint8_t type,
569 const struct sender *sender, struct buffer **bufferp)
571 return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
576 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
577 const struct sender *sender)
579 struct remote *remote = sender ? sender->remote : dp->controller;
580 struct rconn *rconn = remote->rconn;
583 update_openflow_length(buffer);
584 retval = rconn_send(rconn, buffer);
586 VLOG_WARN("send to %s failed: %s",
587 rconn_get_name(rconn), strerror(retval));
588 buffer_delete(buffer);
593 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
594 * packet can be saved in a buffer, then only the first max_len bytes of
595 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
596 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
597 * the caller wants to be sent; a value of 0 indicates the entire packet should
600 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
601 size_t max_len, int reason)
603 struct ofp_packet_in *opi;
607 buffer_id = save_buffer(buffer);
608 total_len = buffer->size;
609 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
610 buffer->size = max_len;
613 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
614 opi->header.version = OFP_VERSION;
615 opi->header.type = OFPT_PACKET_IN;
616 opi->header.length = htons(buffer->size);
617 opi->header.xid = htonl(0);
618 opi->buffer_id = htonl(buffer_id);
619 opi->total_len = htons(total_len);
620 opi->in_port = htons(in_port);
621 opi->reason = reason;
623 send_openflow_buffer(dp, buffer, NULL);
626 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
627 struct ofp_phy_port *desc)
629 desc->port_no = htons(port_no(dp, p));
630 strncpy((char *) desc->name, netdev_get_name(p->netdev),
632 desc->name[sizeof desc->name - 1] = '\0';
633 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
634 desc->flags = htonl(p->flags);
635 desc->features = htonl(netdev_get_features(p->netdev));
636 desc->speed = htonl(netdev_get_speed(p->netdev));
640 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
642 struct buffer *buffer;
643 struct ofp_switch_features *ofr;
646 ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
648 ofr->datapath_id = htonll(dp->id);
649 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
650 ofr->n_compression = 0; /* Not supported */
651 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
652 ofr->buffer_mb = htonl(UINT32_MAX);
653 ofr->n_buffers = htonl(N_PKT_BUFFERS);
654 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
655 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
656 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
657 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
658 memset(opp, 0, sizeof *opp);
659 fill_port_desc(dp, p, opp);
661 send_openflow_buffer(dp, buffer, sender);
665 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
667 int port_no = ntohs(opp->port_no);
668 if (port_no < OFPP_MAX) {
669 struct sw_port *p = &dp->ports[port_no];
671 /* Make sure the port id hasn't changed since this was sent */
672 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
673 ETH_ADDR_LEN) != 0) {
676 p->flags = htonl(opp->flags);
681 send_port_status(struct sw_port *p, uint8_t status)
683 struct buffer *buffer;
684 struct ofp_port_status *ops;
685 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
686 ops->reason = status;
687 memset(ops->pad, 0, sizeof ops->pad);
688 fill_port_desc(p->dp, p, &ops->desc);
690 send_openflow_buffer(p->dp, buffer, NULL);
694 send_flow_expired(struct datapath *dp, struct sw_flow *flow)
696 struct buffer *buffer;
697 struct ofp_flow_expired *ofe;
698 ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
699 flow_fill_match(&ofe->match, &flow->key);
701 memset(ofe->pad, 0, sizeof ofe->pad);
702 ofe->priority = htons(flow->priority);
704 ofe->duration = htonl(flow->timeout - flow->max_idle - flow->created);
705 ofe->packet_count = htonll(flow->packet_count);
706 ofe->byte_count = htonll(flow->byte_count);
707 send_openflow_buffer(dp, buffer, NULL);
711 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
712 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
714 struct buffer *buffer;
715 struct ofp_error_msg *oem;
716 oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG,
718 oem->type = htons(type);
719 oem->code = htons(code);
720 memcpy(oem->data, data, len);
721 send_openflow_buffer(dp, buffer, sender);
725 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
726 int table_idx, time_t now)
728 struct ofp_flow_stats *ofs;
729 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
730 ofs = buffer_put_uninit(buffer, length);
731 ofs->length = htons(length);
732 ofs->table_id = table_idx;
734 ofs->match.wildcards = htons(flow->key.wildcards);
735 ofs->match.in_port = flow->key.flow.in_port;
736 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
737 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
738 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
739 ofs->match.dl_type = flow->key.flow.dl_type;
740 ofs->match.nw_src = flow->key.flow.nw_src;
741 ofs->match.nw_dst = flow->key.flow.nw_dst;
742 ofs->match.nw_proto = flow->key.flow.nw_proto;
743 memset(ofs->match.pad, 0, sizeof ofs->match.pad);
744 ofs->match.tp_src = flow->key.flow.tp_src;
745 ofs->match.tp_dst = flow->key.flow.tp_dst;
746 ofs->duration = htonl(now - flow->created);
747 ofs->packet_count = htonll(flow->packet_count);
748 ofs->byte_count = htonll(flow->byte_count);
749 ofs->priority = htons(flow->priority);
750 ofs->max_idle = htons(flow->max_idle);
751 memcpy(ofs->actions, flow->actions,
752 sizeof *ofs->actions * flow->n_actions);
756 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
757 * OFPP_MAX. Process it according to 'chain'. */
758 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
760 struct sw_flow_key key;
761 struct sw_flow *flow;
764 flow_extract(buffer, in_port, &key.flow);
765 flow = chain_lookup(dp->chain, &key);
767 flow_used(flow, buffer);
768 execute_actions(dp, buffer, in_port, &key,
769 flow->actions, flow->n_actions);
771 dp_output_control(dp, buffer, in_port, ntohs(dp->config.miss_send_len),
777 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
778 size_t max_len, int out_port)
780 if (out_port != OFPP_CONTROLLER) {
781 dp_output_port(dp, buffer, in_port, out_port);
783 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
788 execute_actions(struct datapath *dp, struct buffer *buffer,
789 int in_port, const struct sw_flow_key *key,
790 const struct ofp_action *actions, int n_actions)
792 /* Every output action needs a separate clone of 'buffer', but the common
793 * case is just a single output action, so that doing a clone and then
794 * freeing the original buffer is wasteful. So the following code is
795 * slightly obscure just to avoid that. */
797 size_t max_len=0; /* Initialze to make compiler happy */
802 eth_proto = ntohs(key->flow.dl_type);
804 for (i = 0; i < n_actions; i++) {
805 const struct ofp_action *a = &actions[i];
806 struct eth_header *eh = buffer->l2;
808 if (prev_port != -1) {
809 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
813 switch (ntohs(a->type)) {
815 prev_port = ntohs(a->arg.output.port);
816 max_len = ntohs(a->arg.output.max_len);
819 case OFPAT_SET_DL_VLAN:
820 modify_vlan(buffer, key, a);
823 case OFPAT_SET_DL_SRC:
824 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
827 case OFPAT_SET_DL_DST:
828 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
831 case OFPAT_SET_NW_SRC:
832 case OFPAT_SET_NW_DST:
833 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
836 case OFPAT_SET_TP_SRC:
837 case OFPAT_SET_TP_DST:
838 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
846 do_output(dp, buffer, in_port, max_len, prev_port);
848 buffer_delete(buffer);
851 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
852 uint8_t nw_proto, const struct ofp_action *a)
854 if (eth_proto == ETH_TYPE_IP) {
855 struct ip_header *nh = buffer->l3;
856 uint32_t new, *field;
858 new = a->arg.nw_addr;
859 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
860 if (nw_proto == IP_TYPE_TCP) {
861 struct tcp_header *th = buffer->l4;
862 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
863 } else if (nw_proto == IP_TYPE_UDP) {
864 struct udp_header *th = buffer->l4;
866 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
868 th->udp_csum = 0xffff;
872 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
877 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
878 uint8_t nw_proto, const struct ofp_action *a)
880 if (eth_proto == ETH_TYPE_IP) {
881 uint16_t new, *field;
885 if (nw_proto == IP_TYPE_TCP) {
886 struct tcp_header *th = buffer->l4;
887 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
888 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
890 } else if (nw_proto == IP_TYPE_UDP) {
891 struct udp_header *th = buffer->l4;
892 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
893 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
900 modify_vlan(struct buffer *buffer,
901 const struct sw_flow_key *key, const struct ofp_action *a)
903 uint16_t new_id = a->arg.vlan_id;
904 struct vlan_eth_header *veh;
906 if (new_id != htons(OFP_VLAN_NONE)) {
907 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
908 /* Modify vlan id, but maintain other TCI values */
910 veh->veth_tci &= ~htons(VLAN_VID);
911 veh->veth_tci |= new_id;
913 /* Insert new vlan id. */
914 struct eth_header *eh = buffer->l2;
915 struct vlan_eth_header tmp;
916 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
917 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
918 tmp.veth_type = htons(ETH_TYPE_VLAN);
919 tmp.veth_tci = new_id;
920 tmp.veth_next_type = eh->eth_type;
922 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
923 memcpy(veh, &tmp, sizeof tmp);
924 buffer->l2 -= VLAN_HEADER_LEN;
927 /* Remove an existing vlan header if it exists */
929 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
930 struct eth_header tmp;
932 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
933 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
934 tmp.eth_type = veh->veth_next_type;
936 buffer->size -= VLAN_HEADER_LEN;
937 buffer->data += VLAN_HEADER_LEN;
938 buffer->l2 += VLAN_HEADER_LEN;
939 memcpy(buffer->data, &tmp, sizeof tmp);
945 recv_features_request(struct datapath *dp, const struct sender *sender,
948 dp_send_features_reply(dp, sender);
953 recv_get_config_request(struct datapath *dp, const struct sender *sender,
956 struct buffer *buffer;
957 struct ofp_switch_config *osc;
959 osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
962 assert(sizeof *osc == sizeof dp->config);
963 memcpy(((char *)osc) + sizeof osc->header,
964 ((char *)&dp->config) + sizeof dp->config.header,
965 sizeof dp->config - sizeof dp->config.header);
967 return send_openflow_buffer(dp, buffer, sender);
971 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
974 const struct ofp_switch_config *osc = msg;
980 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
983 const struct ofp_packet_out *opo = msg;
985 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
986 /* FIXME: can we avoid copying data here? */
987 int data_len = ntohs(opo->header.length) - sizeof *opo;
988 struct buffer *buffer = buffer_new(data_len);
989 buffer_put(buffer, opo->u.data, data_len);
990 dp_output_port(dp, buffer,
991 ntohs(opo->in_port), ntohs(opo->out_port));
993 struct sw_flow_key key;
994 struct buffer *buffer;
997 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1002 n_acts = (ntohs(opo->header.length) - sizeof *opo)
1003 / sizeof *opo->u.actions;
1004 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1005 execute_actions(dp, buffer, ntohs(opo->in_port),
1006 &key, opo->u.actions, n_acts);
1012 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1015 const struct ofp_port_mod *opm = msg;
1017 dp_update_port_flags(dp, &opm->desc);
1023 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1025 int error = -ENOMEM;
1028 struct sw_flow *flow;
1031 /* To prevent loops, make sure there's no action to send to the
1032 * OFP_TABLE virtual port.
1034 n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
1035 for (i=0; i<n_acts; i++) {
1036 const struct ofp_action *a = &ofm->actions[i];
1038 if (a->type == htons(OFPAT_OUTPUT)
1039 && (a->arg.output.port == htons(OFPP_TABLE)
1040 || a->arg.output.port == htons(OFPP_NONE))) {
1041 /* xxx Send fancy new error message? */
1046 /* Allocate memory. */
1047 flow = flow_alloc(n_acts);
1051 /* Fill out flow. */
1052 flow_extract_match(&flow->key, &ofm->match);
1053 flow->max_idle = ntohs(ofm->max_idle);
1054 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1055 flow->timeout = time(0) + flow->max_idle; /* FIXME */
1056 flow->n_actions = n_acts;
1057 flow->created = time(0); /* FIXME */
1058 flow->byte_count = 0;
1059 flow->packet_count = 0;
1060 memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
1063 error = chain_insert(dp->chain, flow);
1065 goto error_free_flow;
1068 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1069 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1071 struct sw_flow_key key;
1072 uint16_t in_port = ntohs(ofm->match.in_port);
1073 flow_used(flow, buffer);
1074 flow_extract(buffer, in_port, &key.flow);
1075 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
1085 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1086 discard_buffer(ntohl(ofm->buffer_id));
1091 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1094 const struct ofp_flow_mod *ofm = msg;
1095 uint16_t command = ntohs(ofm->command);
1097 if (command == OFPFC_ADD) {
1098 return add_flow(dp, ofm);
1099 } else if (command == OFPFC_DELETE) {
1100 struct sw_flow_key key;
1101 flow_extract_match(&key, &ofm->match);
1102 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1103 } else if (command == OFPFC_DELETE_STRICT) {
1104 struct sw_flow_key key;
1106 flow_extract_match(&key, &ofm->match);
1107 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1108 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1114 struct flow_stats_state {
1116 struct sw_table_position position;
1117 struct ofp_flow_stats_request rq;
1120 struct buffer *buffer;
1123 #define MAX_FLOW_STATS_BYTES 4096
1125 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1128 const struct ofp_flow_stats_request *fsr = body;
1129 struct flow_stats_state *s = xmalloc(sizeof *s);
1130 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1131 memset(&s->position, 0, sizeof s->position);
1137 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1139 struct flow_stats_state *s = private;
1140 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1141 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1144 static int flow_stats_dump(struct datapath *dp, void *state,
1145 struct buffer *buffer)
1147 struct flow_stats_state *s = state;
1148 struct sw_flow_key match_key;
1150 flow_extract_match(&match_key, &s->rq.match);
1153 while (s->table_idx < dp->chain->n_tables
1154 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1156 struct sw_table *table = dp->chain->tables[s->table_idx];
1158 if (table->iterate(table, &match_key, &s->position,
1159 flow_stats_dump_callback, s))
1163 memset(&s->position, 0, sizeof s->position);
1165 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1168 static void flow_stats_done(void *state)
1173 struct aggregate_stats_state {
1174 struct ofp_aggregate_stats_request rq;
1177 static int aggregate_stats_init(struct datapath *dp,
1178 const void *body, int body_len,
1181 const struct ofp_aggregate_stats_request *rq = body;
1182 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1188 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1190 struct ofp_aggregate_stats_reply *rpy = private;
1191 rpy->packet_count += flow->packet_count;
1192 rpy->byte_count += flow->byte_count;
1197 static int aggregate_stats_dump(struct datapath *dp, void *state,
1198 struct buffer *buffer)
1200 struct aggregate_stats_state *s = state;
1201 struct ofp_aggregate_stats_request *rq = &s->rq;
1202 struct ofp_aggregate_stats_reply *rpy;
1203 struct sw_table_position position;
1204 struct sw_flow_key match_key;
1207 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1208 memset(rpy, 0, sizeof *rpy);
1210 flow_extract_match(&match_key, &rq->match);
1211 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1212 memset(&position, 0, sizeof position);
1213 while (table_idx < dp->chain->n_tables
1214 && (rq->table_id == 0xff || rq->table_id == table_idx))
1216 struct sw_table *table = dp->chain->tables[table_idx];
1219 error = table->iterate(table, &match_key, &position,
1220 aggregate_stats_dump_callback, rpy);
1225 memset(&position, 0, sizeof position);
1228 rpy->packet_count = htonll(rpy->packet_count);
1229 rpy->byte_count = htonll(rpy->byte_count);
1230 rpy->flow_count = htonl(rpy->flow_count);
1234 static void aggregate_stats_done(void *state)
1239 static int table_stats_dump(struct datapath *dp, void *state,
1240 struct buffer *buffer)
1243 for (i = 0; i < dp->chain->n_tables; i++) {
1244 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1245 struct sw_table_stats stats;
1246 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1247 strncpy(ots->name, stats.name, sizeof ots->name);
1249 memset(ots->pad, 0, sizeof ots->pad);
1250 ots->max_entries = htonl(stats.max_flows);
1251 ots->active_count = htonl(stats.n_flows);
1252 ots->matched_count = htonll(stats.n_matched);
1257 struct port_stats_state {
1261 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1264 struct port_stats_state *s = xmalloc(sizeof *s);
1270 static int port_stats_dump(struct datapath *dp, void *state,
1271 struct buffer *buffer)
1273 struct port_stats_state *s = state;
1276 for (i = s->port; i < OFPP_MAX; i++) {
1277 struct sw_port *p = &dp->ports[i];
1278 struct ofp_port_stats *ops;
1282 ops = buffer_put_uninit(buffer, sizeof *ops);
1283 ops->port_no = htons(port_no(dp, p));
1284 memset(ops->pad, 0, sizeof ops->pad);
1285 ops->rx_count = htonll(p->rx_count);
1286 ops->tx_count = htonll(p->tx_count);
1287 ops->drop_count = htonll(p->drop_count);
1294 static void port_stats_done(void *state)
1300 /* Minimum and maximum acceptable number of bytes in body member of
1301 * struct ofp_stats_request. */
1302 size_t min_body, max_body;
1304 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1305 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1306 * Returns zero if successful, otherwise a negative error code.
1307 * May initialize '*state' to state information. May be null if no
1308 * initialization is required.*/
1309 int (*init)(struct datapath *dp, const void *body, int body_len,
1312 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1313 * struct ofp_stats_reply. On success, it should return 1 if it should be
1314 * called again later with another buffer, 0 if it is done, or a negative
1315 * errno value on failure. */
1316 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1318 /* Cleans any state created by the init or dump functions. May be null
1319 * if no cleanup is required. */
1320 void (*done)(void *state);
1323 static const struct stats_type stats[] = {
1325 sizeof(struct ofp_flow_stats_request),
1326 sizeof(struct ofp_flow_stats_request),
1331 [OFPST_AGGREGATE] = {
1332 sizeof(struct ofp_aggregate_stats_request),
1333 sizeof(struct ofp_aggregate_stats_request),
1334 aggregate_stats_init,
1335 aggregate_stats_dump,
1336 aggregate_stats_done
1354 struct stats_dump_cb {
1356 struct ofp_stats_request *rq;
1357 struct sender sender;
1358 const struct stats_type *s;
1363 stats_dump(struct datapath *dp, void *cb_)
1365 struct stats_dump_cb *cb = cb_;
1366 struct ofp_stats_reply *osr;
1367 struct buffer *buffer;
1374 osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1376 osr->type = htons(cb->s - stats);
1379 err = cb->s->dump(dp, cb->state, buffer);
1385 /* Buffer might have been reallocated, so find our data again. */
1386 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1387 osr->flags = ntohs(OFPSF_REPLY_MORE);
1389 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1399 stats_done(void *cb_)
1401 struct stats_dump_cb *cb = cb_;
1404 cb->s->done(cb->state);
1411 recv_stats_request(struct datapath *dp, const struct sender *sender,
1414 const struct ofp_stats_request *rq = oh;
1415 size_t rq_len = ntohs(rq->header.length);
1416 struct stats_dump_cb *cb;
1420 type = ntohs(rq->type);
1421 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1422 VLOG_WARN("received stats request of unknown type %d", type);
1426 cb = xmalloc(sizeof *cb);
1428 cb->rq = xmemdup(rq, rq_len);
1429 cb->sender = *sender;
1430 cb->s = &stats[type];
1433 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1434 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1435 VLOG_WARN("stats request type %d with bad body length %d",
1442 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1444 VLOG_WARN("failed initialization of stats request type %d: %s",
1445 type, strerror(-err));
1450 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1460 recv_echo_request(struct datapath *dp, const struct sender *sender,
1463 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1467 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1468 const void *oh UNUSED)
1473 /* 'msg', which is 'length' bytes long, was received from the control path.
1474 * Apply it to 'chain'. */
1476 fwd_control_input(struct datapath *dp, const struct sender *sender,
1477 const void *msg, size_t length)
1479 struct openflow_packet {
1481 int (*handler)(struct datapath *, const struct sender *, const void *);
1484 static const struct openflow_packet packets[] = {
1485 [OFPT_FEATURES_REQUEST] = {
1486 sizeof (struct ofp_header),
1487 recv_features_request,
1489 [OFPT_GET_CONFIG_REQUEST] = {
1490 sizeof (struct ofp_header),
1491 recv_get_config_request,
1493 [OFPT_SET_CONFIG] = {
1494 sizeof (struct ofp_switch_config),
1497 [OFPT_PACKET_OUT] = {
1498 sizeof (struct ofp_packet_out),
1502 sizeof (struct ofp_flow_mod),
1506 sizeof (struct ofp_port_mod),
1509 [OFPT_STATS_REQUEST] = {
1510 sizeof (struct ofp_stats_request),
1513 [OFPT_ECHO_REQUEST] = {
1514 sizeof (struct ofp_header),
1517 [OFPT_ECHO_REPLY] = {
1518 sizeof (struct ofp_header),
1523 const struct openflow_packet *pkt;
1524 struct ofp_header *oh;
1526 oh = (struct ofp_header *) msg;
1527 assert(oh->version == OFP_VERSION);
1528 if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length)
1531 pkt = &packets[oh->type];
1534 if (length < pkt->min_size)
1537 return pkt->handler(dp, sender, msg);
1540 /* Packet buffering. */
1542 #define OVERWRITE_SECS 1
1544 struct packet_buffer {
1545 struct buffer *buffer;
1550 static struct packet_buffer buffers[N_PKT_BUFFERS];
1551 static unsigned int buffer_idx;
1553 uint32_t save_buffer(struct buffer *buffer)
1555 struct packet_buffer *p;
1558 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1559 p = &buffers[buffer_idx];
1561 /* Don't buffer packet if existing entry is less than
1562 * OVERWRITE_SECS old. */
1563 if (time(0) < p->timeout) { /* FIXME */
1566 buffer_delete(p->buffer);
1569 /* Don't use maximum cookie value since the all-bits-1 id is
1571 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1573 p->buffer = buffer_clone(buffer); /* FIXME */
1574 p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
1575 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1580 static struct buffer *retrieve_buffer(uint32_t id)
1582 struct buffer *buffer = NULL;
1583 struct packet_buffer *p;
1585 p = &buffers[id & PKT_BUFFER_MASK];
1586 if (p->cookie == id >> PKT_BUFFER_BITS) {
1590 printf("cookie mismatch: %x != %x\n",
1591 id >> PKT_BUFFER_BITS, p->cookie);
1597 static void discard_buffer(uint32_t id)
1599 struct packet_buffer *p;
1601 p = &buffers[id & PKT_BUFFER_MASK];
1602 if (p->cookie == id >> PKT_BUFFER_BITS) {
1603 buffer_delete(p->buffer);