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 /* Configuration set from controller. */
118 uint16_t miss_send_len;
121 struct sw_port ports[OFPP_MAX];
122 struct list port_list; /* List of ports, for flooding. */
125 static struct remote *remote_create(struct datapath *, struct rconn *);
126 static void remote_run(struct datapath *, struct remote *);
127 static void remote_wait(struct remote *);
128 static void remote_destroy(struct remote *);
130 void dp_output_port(struct datapath *, struct buffer *,
131 int in_port, int out_port);
132 void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
133 void dp_output_control(struct datapath *, struct buffer *, int in_port,
134 size_t max_len, int reason);
135 static void send_flow_expired(struct datapath *, struct sw_flow *,
136 enum ofp_flow_expired_reason);
137 static void send_port_status(struct sw_port *p, uint8_t status);
138 static void del_switch_port(struct sw_port *p);
139 static void execute_actions(struct datapath *, struct buffer *,
140 int in_port, const struct sw_flow_key *,
141 const struct ofp_action *, int n_actions);
142 static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
143 const struct ofp_action *a);
144 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
145 uint8_t nw_proto, const struct ofp_action *a);
146 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
147 uint8_t nw_proto, const struct ofp_action *a);
149 /* Buffers are identified to userspace by a 31-bit opaque ID. We divide the ID
150 * into a buffer number (low bits) and a cookie (high bits). The buffer number
151 * is an index into an array of buffers. The cookie distinguishes between
152 * different packets that have occupied a single buffer. Thus, the more
153 * buffers we have, the lower-quality the cookie... */
154 #define PKT_BUFFER_BITS 8
155 #define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
156 #define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)
158 #define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
160 int run_flow_through_tables(struct datapath *, struct buffer *, int in_port);
161 void fwd_port_input(struct datapath *, struct buffer *, int in_port);
162 int fwd_control_input(struct datapath *, const struct sender *,
163 const void *, size_t);
165 uint32_t save_buffer(struct buffer *);
166 static struct buffer *retrieve_buffer(uint32_t id);
167 static void discard_buffer(uint32_t id);
169 static int port_no(struct datapath *dp, struct sw_port *p)
171 assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
172 return p - dp->ports;
175 /* Generates and returns a random datapath id. */
177 gen_datapath_id(void)
179 uint8_t ea[ETH_ADDR_LEN];
181 return eth_addr_to_uint64(ea);
185 dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
189 dp = calloc(1, sizeof *dp);
194 dp->last_timeout = time(0);
195 list_init(&dp->remotes);
196 dp->controller = remote_create(dp, rconn);
197 dp->listen_vconn = NULL;
198 dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
199 dp->chain = chain_create();
201 VLOG_ERR("could not create chain");
206 list_init(&dp->port_list);
208 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
214 dp_add_port(struct datapath *dp, const char *name)
216 struct netdev *netdev;
222 error = netdev_open(name, NETDEV_ETH_TYPE_ANY, &netdev);
226 error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false);
228 VLOG_ERR("Couldn't set promiscuous mode on %s device", name);
229 netdev_close(netdev);
232 if (netdev_get_in4(netdev, &in4)) {
233 VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
235 if (netdev_get_in6(netdev, &in6)) {
236 char in6_name[INET6_ADDRSTRLEN + 1];
237 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
238 VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
241 for (p = dp->ports; ; p++) {
242 if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
244 } else if (!p->netdev) {
254 list_push_back(&dp->port_list, &p->node);
256 /* Notify the ctlpath that this port has been added */
257 send_port_status(p, OFPPR_ADD);
263 dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
265 assert(!dp->listen_vconn);
266 dp->listen_vconn = listen_vconn;
270 dp_run(struct datapath *dp)
272 time_t now = time(0);
273 struct sw_port *p, *pn;
274 struct remote *r, *rn;
275 struct buffer *buffer = NULL;
277 if (now != dp->last_timeout) {
278 struct list deleted = LIST_INITIALIZER(&deleted);
279 struct sw_flow *f, *n;
281 chain_timeout(dp->chain, &deleted);
282 LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
283 send_flow_expired(dp, f, f->reason);
284 list_remove(&f->node);
287 dp->last_timeout = now;
289 poll_timer_wait(1000);
291 LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
295 /* Allocate buffer with some headroom to add headers in forwarding
296 * to the controller or adding a vlan tag, plus an extra 2 bytes to
297 * allow IP headers to be aligned on a 4-byte boundary. */
298 const int headroom = 128 + 2;
299 const int hard_header = VLAN_ETH_HEADER_LEN;
300 const int mtu = netdev_get_mtu(p->netdev);
301 buffer = buffer_new(headroom + hard_header + mtu);
302 buffer->data += headroom;
304 error = netdev_recv(p->netdev, buffer);
307 fwd_port_input(dp, buffer, port_no(dp, p));
309 } else if (error != EAGAIN) {
310 VLOG_ERR("Error receiving data from %s: %s",
311 netdev_get_name(p->netdev), strerror(error));
315 buffer_delete(buffer);
317 /* Talk to remotes. */
318 LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
321 if (dp->listen_vconn) {
323 struct vconn *new_vconn;
326 retval = vconn_accept(dp->listen_vconn, &new_vconn);
328 if (retval != EAGAIN) {
329 VLOG_WARN("accept failed (%s)", strerror(retval));
333 remote_create(dp, rconn_new_from_vconn("passive", 128, new_vconn));
339 remote_run(struct datapath *dp, struct remote *r)
345 /* Do some remote processing, but cap it at a reasonable amount so that
346 * other processing doesn't starve. */
347 for (i = 0; i < 50; i++) {
349 struct buffer *buffer;
350 struct ofp_header *oh;
352 buffer = rconn_recv(r->rconn);
357 if (buffer->size >= sizeof *oh) {
358 struct sender sender;
362 sender.xid = oh->xid;
363 fwd_control_input(dp, &sender, buffer->data, buffer->size);
365 VLOG_WARN("received too-short OpenFlow message");
367 buffer_delete(buffer);
369 if (!rconn_is_full(r->rconn)) {
370 int error = r->cb_dump(dp, r->cb_aux);
373 VLOG_WARN("dump callback error: %s", strerror(-error));
375 r->cb_done(r->cb_aux);
384 if (!rconn_is_alive(r->rconn)) {
390 remote_wait(struct remote *r)
392 rconn_run_wait(r->rconn);
393 rconn_recv_wait(r->rconn);
397 remote_destroy(struct remote *r)
400 if (r->cb_dump && r->cb_done) {
401 r->cb_done(r->cb_aux);
403 list_remove(&r->node);
404 rconn_destroy(r->rconn);
409 static struct remote *
410 remote_create(struct datapath *dp, struct rconn *rconn)
412 struct remote *remote = xmalloc(sizeof *remote);
413 list_push_back(&dp->remotes, &remote->node);
414 remote->rconn = rconn;
415 remote->cb_dump = NULL;
419 /* Starts a callback-based, reliable, possibly multi-message reply to a
420 * request made by 'remote'.
422 * 'dump' designates a function that will be called when the 'remote' send
423 * queue has an empty slot. It should compose a message and send it on
424 * 'remote'. On success, it should return 1 if it should be called again when
425 * another send queue slot opens up, 0 if its transmissions are complete, or a
426 * negative errno value on failure.
428 * 'done' designates a function to clean up any resources allocated for the
429 * dump. It must handle being called before the dump is complete (which will
430 * happen if 'remote' is closed unexpectedly).
432 * 'aux' is passed to 'dump' and 'done'. */
434 remote_start_dump(struct remote *remote,
435 int (*dump)(struct datapath *, void *),
436 void (*done)(void *),
439 assert(!remote->cb_dump);
440 remote->cb_dump = dump;
441 remote->cb_done = done;
442 remote->cb_aux = aux;
446 dp_wait(struct datapath *dp)
451 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
452 netdev_recv_wait(p->netdev);
454 LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
457 if (dp->listen_vconn) {
458 vconn_accept_wait(dp->listen_vconn);
462 /* Delete 'p' from switch. */
464 del_switch_port(struct sw_port *p)
466 send_port_status(p, OFPPR_DELETE);
467 netdev_close(p->netdev);
469 list_remove(&p->node);
473 dp_destroy(struct datapath *dp)
475 struct sw_port *p, *n;
481 LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
484 chain_destroy(dp->chain);
488 /* Send packets out all the ports except the originating one. If the
489 * "flood" argument is set, don't send out ports with flooding disabled.
492 output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
498 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
499 if (port_no(dp, p) == in_port) {
502 if (flood && p->flags & BRIDGE_PORT_NO_FLOOD) {
505 if (prev_port != -1) {
506 dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
508 prev_port = port_no(dp, p);
511 dp_output_port(dp, buffer, in_port, prev_port);
513 buffer_delete(buffer);
519 output_packet(struct datapath *dp, struct buffer *buffer, int out_port)
521 if (out_port >= 0 && out_port < OFPP_MAX) {
522 struct sw_port *p = &dp->ports[out_port];
523 if (p->netdev != NULL) {
524 if (!netdev_send(p->netdev, buffer)) {
533 buffer_delete(buffer);
534 /* FIXME: ratelimit */
535 VLOG_DBG("can't forward to bad port %d\n", out_port);
538 /* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
541 dp_output_port(struct datapath *dp, struct buffer *buffer,
542 int in_port, int out_port)
546 if (out_port == OFPP_FLOOD) {
547 output_all(dp, buffer, in_port, 1);
548 } else if (out_port == OFPP_ALL) {
549 output_all(dp, buffer, in_port, 0);
550 } else if (out_port == OFPP_CONTROLLER) {
551 dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
552 } else if (out_port == OFPP_TABLE) {
553 if (run_flow_through_tables(dp, buffer, in_port)) {
554 buffer_delete(buffer);
557 output_packet(dp, buffer, out_port);
562 make_openflow_reply(size_t openflow_len, uint8_t type,
563 const struct sender *sender, struct buffer **bufferp)
565 return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
570 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
571 const struct sender *sender)
573 struct remote *remote = sender ? sender->remote : dp->controller;
574 struct rconn *rconn = remote->rconn;
577 update_openflow_length(buffer);
578 retval = rconn_send(rconn, buffer);
580 VLOG_WARN("send to %s failed: %s",
581 rconn_get_name(rconn), strerror(retval));
582 buffer_delete(buffer);
587 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
588 * packet can be saved in a buffer, then only the first max_len bytes of
589 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
590 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
591 * the caller wants to be sent; a value of 0 indicates the entire packet should
594 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
595 size_t max_len, int reason)
597 struct ofp_packet_in *opi;
601 buffer_id = save_buffer(buffer);
602 total_len = buffer->size;
603 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
604 buffer->size = max_len;
607 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
608 opi->header.version = OFP_VERSION;
609 opi->header.type = OFPT_PACKET_IN;
610 opi->header.length = htons(buffer->size);
611 opi->header.xid = htonl(0);
612 opi->buffer_id = htonl(buffer_id);
613 opi->total_len = htons(total_len);
614 opi->in_port = htons(in_port);
615 opi->reason = reason;
617 send_openflow_buffer(dp, buffer, NULL);
620 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
621 struct ofp_phy_port *desc)
623 desc->port_no = htons(port_no(dp, p));
624 strncpy((char *) desc->name, netdev_get_name(p->netdev),
626 desc->name[sizeof desc->name - 1] = '\0';
627 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
628 desc->flags = htonl(p->flags);
629 desc->features = htonl(netdev_get_features(p->netdev));
630 desc->speed = htonl(netdev_get_speed(p->netdev));
634 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
636 struct buffer *buffer;
637 struct ofp_switch_features *ofr;
640 ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
642 ofr->datapath_id = htonll(dp->id);
643 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
644 ofr->n_compression = 0; /* Not supported */
645 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
646 ofr->buffer_mb = htonl(UINT32_MAX);
647 ofr->n_buffers = htonl(N_PKT_BUFFERS);
648 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
649 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
650 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
651 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
652 memset(opp, 0, sizeof *opp);
653 fill_port_desc(dp, p, opp);
655 send_openflow_buffer(dp, buffer, sender);
659 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
661 int port_no = ntohs(opp->port_no);
662 if (port_no < OFPP_MAX) {
663 struct sw_port *p = &dp->ports[port_no];
665 /* Make sure the port id hasn't changed since this was sent */
666 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
667 ETH_ADDR_LEN) != 0) {
670 p->flags = htonl(opp->flags);
675 send_port_status(struct sw_port *p, uint8_t status)
677 struct buffer *buffer;
678 struct ofp_port_status *ops;
679 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
680 ops->reason = status;
681 memset(ops->pad, 0, sizeof ops->pad);
682 fill_port_desc(p->dp, p, &ops->desc);
684 send_openflow_buffer(p->dp, buffer, NULL);
688 send_flow_expired(struct datapath *dp, struct sw_flow *flow,
689 enum ofp_flow_expired_reason reason)
691 struct buffer *buffer;
692 struct ofp_flow_expired *ofe;
693 ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
694 flow_fill_match(&ofe->match, &flow->key);
696 ofe->priority = htons(flow->priority);
697 ofe->reason = reason;
698 memset(ofe->pad, 0, sizeof ofe->pad);
700 ofe->duration = htonl(time(0) - flow->created);
701 memset(ofe->pad2, 0, sizeof ofe->pad2);
702 ofe->packet_count = htonll(flow->packet_count);
703 ofe->byte_count = htonll(flow->byte_count);
704 send_openflow_buffer(dp, buffer, NULL);
708 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
709 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
711 struct buffer *buffer;
712 struct ofp_error_msg *oem;
713 oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG,
715 oem->type = htons(type);
716 oem->code = htons(code);
717 memcpy(oem->data, data, len);
718 send_openflow_buffer(dp, buffer, sender);
722 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
723 int table_idx, time_t now)
725 struct ofp_flow_stats *ofs;
726 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
727 ofs = buffer_put_uninit(buffer, length);
728 ofs->length = htons(length);
729 ofs->table_id = table_idx;
731 ofs->match.wildcards = htons(flow->key.wildcards);
732 ofs->match.in_port = flow->key.flow.in_port;
733 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
734 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
735 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
736 ofs->match.dl_type = flow->key.flow.dl_type;
737 ofs->match.nw_src = flow->key.flow.nw_src;
738 ofs->match.nw_dst = flow->key.flow.nw_dst;
739 ofs->match.nw_proto = flow->key.flow.nw_proto;
740 memset(ofs->match.pad, 0, sizeof ofs->match.pad);
741 ofs->match.tp_src = flow->key.flow.tp_src;
742 ofs->match.tp_dst = flow->key.flow.tp_dst;
743 ofs->duration = htonl(now - flow->created);
744 ofs->priority = htons(flow->priority);
745 ofs->idle_timeout = htons(flow->idle_timeout);
746 ofs->hard_timeout = htons(flow->hard_timeout);
747 memset(ofs->pad2, 0, sizeof ofs->pad2);
748 ofs->packet_count = htonll(flow->packet_count);
749 ofs->byte_count = htonll(flow->byte_count);
750 memcpy(ofs->actions, flow->actions,
751 sizeof *ofs->actions * flow->n_actions);
755 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
756 * OFPP_MAX. Process it according to 'dp''s flow table. Returns 0 if
757 * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no
758 * matching flow, in which case 'buffer' still belongs to the caller. */
759 int run_flow_through_tables(struct datapath *dp, struct buffer *buffer,
762 struct sw_flow_key key;
763 struct sw_flow *flow;
766 if (flow_extract(buffer, in_port, &key.flow)
767 && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
769 buffer_delete(buffer);
773 flow = chain_lookup(dp->chain, &key);
775 flow_used(flow, buffer);
776 execute_actions(dp, buffer, in_port, &key,
777 flow->actions, flow->n_actions);
784 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
785 * OFPP_MAX. Process it according to 'dp''s flow table, sending it up to the
786 * controller if no flow matches. Takes ownership of 'buffer'. */
787 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
789 if (run_flow_through_tables(dp, buffer, in_port)) {
790 dp_output_control(dp, buffer, in_port, dp->miss_send_len,
796 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
797 size_t max_len, int out_port)
799 if (out_port != OFPP_CONTROLLER) {
800 dp_output_port(dp, buffer, in_port, out_port);
802 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
807 execute_actions(struct datapath *dp, struct buffer *buffer,
808 int in_port, const struct sw_flow_key *key,
809 const struct ofp_action *actions, int n_actions)
811 /* Every output action needs a separate clone of 'buffer', but the common
812 * case is just a single output action, so that doing a clone and then
813 * freeing the original buffer is wasteful. So the following code is
814 * slightly obscure just to avoid that. */
816 size_t max_len=0; /* Initialze to make compiler happy */
821 eth_proto = ntohs(key->flow.dl_type);
823 for (i = 0; i < n_actions; i++) {
824 const struct ofp_action *a = &actions[i];
825 struct eth_header *eh = buffer->l2;
827 if (prev_port != -1) {
828 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
832 switch (ntohs(a->type)) {
834 prev_port = ntohs(a->arg.output.port);
835 max_len = ntohs(a->arg.output.max_len);
838 case OFPAT_SET_DL_VLAN:
839 modify_vlan(buffer, key, a);
842 case OFPAT_SET_DL_SRC:
843 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
846 case OFPAT_SET_DL_DST:
847 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
850 case OFPAT_SET_NW_SRC:
851 case OFPAT_SET_NW_DST:
852 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
855 case OFPAT_SET_TP_SRC:
856 case OFPAT_SET_TP_DST:
857 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
865 do_output(dp, buffer, in_port, max_len, prev_port);
867 buffer_delete(buffer);
870 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
871 uint8_t nw_proto, const struct ofp_action *a)
873 if (eth_proto == ETH_TYPE_IP) {
874 struct ip_header *nh = buffer->l3;
875 uint32_t new, *field;
877 new = a->arg.nw_addr;
878 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
879 if (nw_proto == IP_TYPE_TCP) {
880 struct tcp_header *th = buffer->l4;
881 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
882 } else if (nw_proto == IP_TYPE_UDP) {
883 struct udp_header *th = buffer->l4;
885 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
887 th->udp_csum = 0xffff;
891 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
896 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
897 uint8_t nw_proto, const struct ofp_action *a)
899 if (eth_proto == ETH_TYPE_IP) {
900 uint16_t new, *field;
904 if (nw_proto == IP_TYPE_TCP) {
905 struct tcp_header *th = buffer->l4;
906 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
907 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
909 } else if (nw_proto == IP_TYPE_UDP) {
910 struct udp_header *th = buffer->l4;
911 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
912 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
919 modify_vlan(struct buffer *buffer,
920 const struct sw_flow_key *key, const struct ofp_action *a)
922 uint16_t new_id = a->arg.vlan_id;
923 struct vlan_eth_header *veh;
925 if (new_id != htons(OFP_VLAN_NONE)) {
926 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
927 /* Modify vlan id, but maintain other TCI values */
929 veh->veth_tci &= ~htons(VLAN_VID);
930 veh->veth_tci |= new_id;
932 /* Insert new vlan id. */
933 struct eth_header *eh = buffer->l2;
934 struct vlan_eth_header tmp;
935 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
936 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
937 tmp.veth_type = htons(ETH_TYPE_VLAN);
938 tmp.veth_tci = new_id;
939 tmp.veth_next_type = eh->eth_type;
941 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
942 memcpy(veh, &tmp, sizeof tmp);
943 buffer->l2 -= VLAN_HEADER_LEN;
946 /* Remove an existing vlan header if it exists */
948 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
949 struct eth_header tmp;
951 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
952 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
953 tmp.eth_type = veh->veth_next_type;
955 buffer->size -= VLAN_HEADER_LEN;
956 buffer->data += VLAN_HEADER_LEN;
957 buffer->l2 += VLAN_HEADER_LEN;
958 memcpy(buffer->data, &tmp, sizeof tmp);
964 recv_features_request(struct datapath *dp, const struct sender *sender,
967 dp_send_features_reply(dp, sender);
972 recv_get_config_request(struct datapath *dp, const struct sender *sender,
975 struct buffer *buffer;
976 struct ofp_switch_config *osc;
978 osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
981 osc->flags = htons(dp->flags);
982 osc->miss_send_len = htons(dp->miss_send_len);
984 return send_openflow_buffer(dp, buffer, sender);
988 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
991 const struct ofp_switch_config *osc = msg;
994 flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
995 if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
996 && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
997 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
1000 dp->miss_send_len = ntohs(osc->miss_send_len);
1005 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
1008 const struct ofp_packet_out *opo = msg;
1010 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
1011 /* FIXME: can we avoid copying data here? */
1012 int data_len = ntohs(opo->header.length) - sizeof *opo;
1013 struct buffer *buffer = buffer_new(data_len);
1014 buffer_put(buffer, opo->u.data, data_len);
1015 dp_output_port(dp, buffer,
1016 ntohs(opo->in_port), ntohs(opo->out_port));
1018 struct sw_flow_key key;
1019 struct buffer *buffer;
1022 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1027 n_acts = (ntohs(opo->header.length) - sizeof *opo)
1028 / sizeof *opo->u.actions;
1029 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1030 execute_actions(dp, buffer, ntohs(opo->in_port),
1031 &key, opo->u.actions, n_acts);
1037 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1040 const struct ofp_port_mod *opm = msg;
1042 dp_update_port_flags(dp, &opm->desc);
1048 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1050 int error = -ENOMEM;
1053 struct sw_flow *flow;
1056 /* To prevent loops, make sure there's no action to send to the
1057 * OFP_TABLE virtual port.
1059 n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
1060 for (i=0; i<n_acts; i++) {
1061 const struct ofp_action *a = &ofm->actions[i];
1063 if (a->type == htons(OFPAT_OUTPUT)
1064 && (a->arg.output.port == htons(OFPP_TABLE)
1065 || a->arg.output.port == htons(OFPP_NONE))) {
1066 /* xxx Send fancy new error message? */
1071 /* Allocate memory. */
1072 flow = flow_alloc(n_acts);
1076 /* Fill out flow. */
1077 flow_extract_match(&flow->key, &ofm->match);
1078 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1079 flow->idle_timeout = ntohs(ofm->idle_timeout);
1080 flow->hard_timeout = ntohs(ofm->hard_timeout);
1081 flow->used = flow->created = time(0);
1082 flow->n_actions = n_acts;
1083 flow->byte_count = 0;
1084 flow->packet_count = 0;
1085 memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
1088 error = chain_insert(dp->chain, flow);
1090 goto error_free_flow;
1093 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1094 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1096 struct sw_flow_key key;
1097 uint16_t in_port = ntohs(ofm->match.in_port);
1098 flow_used(flow, buffer);
1099 flow_extract(buffer, in_port, &key.flow);
1100 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
1110 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1111 discard_buffer(ntohl(ofm->buffer_id));
1116 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1119 const struct ofp_flow_mod *ofm = msg;
1120 uint16_t command = ntohs(ofm->command);
1122 if (command == OFPFC_ADD) {
1123 return add_flow(dp, ofm);
1124 } else if (command == OFPFC_DELETE) {
1125 struct sw_flow_key key;
1126 flow_extract_match(&key, &ofm->match);
1127 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1128 } else if (command == OFPFC_DELETE_STRICT) {
1129 struct sw_flow_key key;
1131 flow_extract_match(&key, &ofm->match);
1132 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1133 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1139 struct flow_stats_state {
1141 struct sw_table_position position;
1142 struct ofp_flow_stats_request rq;
1145 struct buffer *buffer;
1148 #define MAX_FLOW_STATS_BYTES 4096
1150 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1153 const struct ofp_flow_stats_request *fsr = body;
1154 struct flow_stats_state *s = xmalloc(sizeof *s);
1155 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1156 memset(&s->position, 0, sizeof s->position);
1162 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1164 struct flow_stats_state *s = private;
1165 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1166 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1169 static int flow_stats_dump(struct datapath *dp, void *state,
1170 struct buffer *buffer)
1172 struct flow_stats_state *s = state;
1173 struct sw_flow_key match_key;
1175 flow_extract_match(&match_key, &s->rq.match);
1178 while (s->table_idx < dp->chain->n_tables
1179 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1181 struct sw_table *table = dp->chain->tables[s->table_idx];
1183 if (table->iterate(table, &match_key, &s->position,
1184 flow_stats_dump_callback, s))
1188 memset(&s->position, 0, sizeof s->position);
1190 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1193 static void flow_stats_done(void *state)
1198 struct aggregate_stats_state {
1199 struct ofp_aggregate_stats_request rq;
1202 static int aggregate_stats_init(struct datapath *dp,
1203 const void *body, int body_len,
1206 const struct ofp_aggregate_stats_request *rq = body;
1207 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1213 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1215 struct ofp_aggregate_stats_reply *rpy = private;
1216 rpy->packet_count += flow->packet_count;
1217 rpy->byte_count += flow->byte_count;
1222 static int aggregate_stats_dump(struct datapath *dp, void *state,
1223 struct buffer *buffer)
1225 struct aggregate_stats_state *s = state;
1226 struct ofp_aggregate_stats_request *rq = &s->rq;
1227 struct ofp_aggregate_stats_reply *rpy;
1228 struct sw_table_position position;
1229 struct sw_flow_key match_key;
1232 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1233 memset(rpy, 0, sizeof *rpy);
1235 flow_extract_match(&match_key, &rq->match);
1236 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1237 memset(&position, 0, sizeof position);
1238 while (table_idx < dp->chain->n_tables
1239 && (rq->table_id == 0xff || rq->table_id == table_idx))
1241 struct sw_table *table = dp->chain->tables[table_idx];
1244 error = table->iterate(table, &match_key, &position,
1245 aggregate_stats_dump_callback, rpy);
1250 memset(&position, 0, sizeof position);
1253 rpy->packet_count = htonll(rpy->packet_count);
1254 rpy->byte_count = htonll(rpy->byte_count);
1255 rpy->flow_count = htonl(rpy->flow_count);
1259 static void aggregate_stats_done(void *state)
1264 static int table_stats_dump(struct datapath *dp, void *state,
1265 struct buffer *buffer)
1268 for (i = 0; i < dp->chain->n_tables; i++) {
1269 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1270 struct sw_table_stats stats;
1271 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1272 strncpy(ots->name, stats.name, sizeof ots->name);
1274 memset(ots->pad, 0, sizeof ots->pad);
1275 ots->max_entries = htonl(stats.max_flows);
1276 ots->active_count = htonl(stats.n_flows);
1277 ots->matched_count = htonll(stats.n_matched);
1282 struct port_stats_state {
1286 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1289 struct port_stats_state *s = xmalloc(sizeof *s);
1295 static int port_stats_dump(struct datapath *dp, void *state,
1296 struct buffer *buffer)
1298 struct port_stats_state *s = state;
1301 for (i = s->port; i < OFPP_MAX; i++) {
1302 struct sw_port *p = &dp->ports[i];
1303 struct ofp_port_stats *ops;
1307 ops = buffer_put_uninit(buffer, sizeof *ops);
1308 ops->port_no = htons(port_no(dp, p));
1309 memset(ops->pad, 0, sizeof ops->pad);
1310 ops->rx_count = htonll(p->rx_count);
1311 ops->tx_count = htonll(p->tx_count);
1312 ops->drop_count = htonll(p->drop_count);
1319 static void port_stats_done(void *state)
1325 /* Minimum and maximum acceptable number of bytes in body member of
1326 * struct ofp_stats_request. */
1327 size_t min_body, max_body;
1329 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1330 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1331 * Returns zero if successful, otherwise a negative error code.
1332 * May initialize '*state' to state information. May be null if no
1333 * initialization is required.*/
1334 int (*init)(struct datapath *dp, const void *body, int body_len,
1337 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1338 * struct ofp_stats_reply. On success, it should return 1 if it should be
1339 * called again later with another buffer, 0 if it is done, or a negative
1340 * errno value on failure. */
1341 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1343 /* Cleans any state created by the init or dump functions. May be null
1344 * if no cleanup is required. */
1345 void (*done)(void *state);
1348 static const struct stats_type stats[] = {
1350 sizeof(struct ofp_flow_stats_request),
1351 sizeof(struct ofp_flow_stats_request),
1356 [OFPST_AGGREGATE] = {
1357 sizeof(struct ofp_aggregate_stats_request),
1358 sizeof(struct ofp_aggregate_stats_request),
1359 aggregate_stats_init,
1360 aggregate_stats_dump,
1361 aggregate_stats_done
1379 struct stats_dump_cb {
1381 struct ofp_stats_request *rq;
1382 struct sender sender;
1383 const struct stats_type *s;
1388 stats_dump(struct datapath *dp, void *cb_)
1390 struct stats_dump_cb *cb = cb_;
1391 struct ofp_stats_reply *osr;
1392 struct buffer *buffer;
1399 osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1401 osr->type = htons(cb->s - stats);
1404 err = cb->s->dump(dp, cb->state, buffer);
1410 /* Buffer might have been reallocated, so find our data again. */
1411 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1412 osr->flags = ntohs(OFPSF_REPLY_MORE);
1414 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1424 stats_done(void *cb_)
1426 struct stats_dump_cb *cb = cb_;
1429 cb->s->done(cb->state);
1436 recv_stats_request(struct datapath *dp, const struct sender *sender,
1439 const struct ofp_stats_request *rq = oh;
1440 size_t rq_len = ntohs(rq->header.length);
1441 struct stats_dump_cb *cb;
1445 type = ntohs(rq->type);
1446 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1447 VLOG_WARN("received stats request of unknown type %d", type);
1451 cb = xmalloc(sizeof *cb);
1453 cb->rq = xmemdup(rq, rq_len);
1454 cb->sender = *sender;
1455 cb->s = &stats[type];
1458 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1459 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1460 VLOG_WARN("stats request type %d with bad body length %d",
1467 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1469 VLOG_WARN("failed initialization of stats request type %d: %s",
1470 type, strerror(-err));
1475 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1485 recv_echo_request(struct datapath *dp, const struct sender *sender,
1488 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1492 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1493 const void *oh UNUSED)
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),
1538 [OFPT_ECHO_REQUEST] = {
1539 sizeof (struct ofp_header),
1542 [OFPT_ECHO_REPLY] = {
1543 sizeof (struct ofp_header),
1548 const struct openflow_packet *pkt;
1549 struct ofp_header *oh;
1551 oh = (struct ofp_header *) msg;
1552 assert(oh->version == OFP_VERSION);
1553 if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length)
1556 pkt = &packets[oh->type];
1559 if (length < pkt->min_size)
1562 return pkt->handler(dp, sender, msg);
1565 /* Packet buffering. */
1567 #define OVERWRITE_SECS 1
1569 struct packet_buffer {
1570 struct buffer *buffer;
1575 static struct packet_buffer buffers[N_PKT_BUFFERS];
1576 static unsigned int buffer_idx;
1578 uint32_t save_buffer(struct buffer *buffer)
1580 struct packet_buffer *p;
1583 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1584 p = &buffers[buffer_idx];
1586 /* Don't buffer packet if existing entry is less than
1587 * OVERWRITE_SECS old. */
1588 if (time(0) < p->timeout) { /* FIXME */
1591 buffer_delete(p->buffer);
1594 /* Don't use maximum cookie value since the all-bits-1 id is
1596 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1598 p->buffer = buffer_clone(buffer); /* FIXME */
1599 p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
1600 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1605 static struct buffer *retrieve_buffer(uint32_t id)
1607 struct buffer *buffer = NULL;
1608 struct packet_buffer *p;
1610 p = &buffers[id & PKT_BUFFER_MASK];
1611 if (p->cookie == id >> PKT_BUFFER_BITS) {
1615 printf("cookie mismatch: %x != %x\n",
1616 id >> PKT_BUFFER_BITS, p->cookie);
1622 static void discard_buffer(uint32_t id)
1624 struct packet_buffer *p;
1626 p = &buffers[id & PKT_BUFFER_MASK];
1627 if (p->cookie == id >> PKT_BUFFER_BITS) {
1628 buffer_delete(p->buffer);