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);
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);
561 output_packet(dp, buffer, out_port);
566 alloc_openflow_buffer(struct datapath *dp, size_t openflow_len, uint8_t type,
567 const struct sender *sender, struct buffer **bufferp)
569 struct buffer *buffer;
570 struct ofp_header *oh;
572 buffer = *bufferp = buffer_new(openflow_len);
573 oh = buffer_put_uninit(buffer, openflow_len);
574 oh->version = OFP_VERSION;
576 oh->length = 0; /* Filled in by send_openflow_buffer(). */
577 oh->xid = sender ? sender->xid : 0;
582 send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
583 const struct sender *sender)
585 struct remote *remote = sender ? sender->remote : dp->controller;
586 struct rconn *rconn = remote->rconn;
587 struct ofp_header *oh;
590 oh = buffer_at_assert(buffer, 0, sizeof *oh);
591 oh->length = htons(buffer->size);
593 retval = rconn_send(rconn, buffer);
595 VLOG_WARN("send to %s failed: %s",
596 rconn_get_name(rconn), strerror(retval));
597 buffer_delete(buffer);
602 /* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
603 * packet can be saved in a buffer, then only the first max_len bytes of
604 * 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
605 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
606 * the caller wants to be sent; a value of 0 indicates the entire packet should
609 dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
610 size_t max_len, int reason)
612 struct ofp_packet_in *opi;
616 buffer_id = save_buffer(buffer);
617 total_len = buffer->size;
618 if (buffer_id != UINT32_MAX && buffer->size > max_len) {
619 buffer->size = max_len;
622 opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
623 opi->header.version = OFP_VERSION;
624 opi->header.type = OFPT_PACKET_IN;
625 opi->header.length = htons(buffer->size);
626 opi->header.xid = htonl(0);
627 opi->buffer_id = htonl(buffer_id);
628 opi->total_len = htons(total_len);
629 opi->in_port = htons(in_port);
630 opi->reason = reason;
632 send_openflow_buffer(dp, buffer, NULL);
635 static void fill_port_desc(struct datapath *dp, struct sw_port *p,
636 struct ofp_phy_port *desc)
638 desc->port_no = htons(port_no(dp, p));
639 strncpy((char *) desc->name, netdev_get_name(p->netdev),
641 desc->name[sizeof desc->name - 1] = '\0';
642 memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
643 desc->flags = htonl(p->flags);
644 desc->features = htonl(netdev_get_features(p->netdev));
645 desc->speed = htonl(netdev_get_speed(p->netdev));
649 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
651 struct buffer *buffer;
652 struct ofp_switch_features *ofr;
655 ofr = alloc_openflow_buffer(dp, sizeof *ofr, OFPT_FEATURES_REPLY,
657 ofr->datapath_id = htonll(dp->id);
658 ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
659 ofr->n_compression = 0; /* Not supported */
660 ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
661 ofr->buffer_mb = htonl(UINT32_MAX);
662 ofr->n_buffers = htonl(N_PKT_BUFFERS);
663 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
664 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
665 LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
666 struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
667 memset(opp, 0, sizeof *opp);
668 fill_port_desc(dp, p, opp);
670 send_openflow_buffer(dp, buffer, sender);
674 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
676 int port_no = ntohs(opp->port_no);
677 if (port_no < OFPP_MAX) {
678 struct sw_port *p = &dp->ports[port_no];
680 /* Make sure the port id hasn't changed since this was sent */
681 if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
682 ETH_ADDR_LEN) != 0) {
685 p->flags = htonl(opp->flags);
690 send_port_status(struct sw_port *p, uint8_t status)
692 struct buffer *buffer;
693 struct ofp_port_status *ops;
694 ops = alloc_openflow_buffer(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
696 ops->reason = status;
697 memset(ops->pad, 0, sizeof ops->pad);
698 fill_port_desc(p->dp, p, &ops->desc);
700 send_openflow_buffer(p->dp, buffer, NULL);
704 send_flow_expired(struct datapath *dp, struct sw_flow *flow)
706 struct buffer *buffer;
707 struct ofp_flow_expired *ofe;
708 ofe = alloc_openflow_buffer(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, NULL,
710 flow_fill_match(&ofe->match, &flow->key);
712 memset(ofe->pad, 0, sizeof ofe->pad);
713 ofe->priority = htons(flow->priority);
715 ofe->duration = htonl(flow->timeout - flow->max_idle - flow->created);
716 ofe->packet_count = htonll(flow->packet_count);
717 ofe->byte_count = htonll(flow->byte_count);
718 send_openflow_buffer(dp, buffer, NULL);
722 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
723 uint16_t type, uint16_t code, const uint8_t *data, size_t len)
725 struct buffer *buffer;
726 struct ofp_error_msg *oem;
727 oem = alloc_openflow_buffer(dp, sizeof(*oem)+len, OFPT_ERROR_MSG,
729 oem->type = htons(type);
730 oem->code = htons(code);
731 memcpy(oem->data, data, len);
732 send_openflow_buffer(dp, buffer, sender);
736 fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
737 int table_idx, time_t now)
739 struct ofp_flow_stats *ofs;
740 int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
741 ofs = buffer_put_uninit(buffer, length);
742 ofs->length = htons(length);
743 ofs->table_id = table_idx;
745 ofs->match.wildcards = htons(flow->key.wildcards);
746 ofs->match.in_port = flow->key.flow.in_port;
747 memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
748 memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
749 ofs->match.dl_vlan = flow->key.flow.dl_vlan;
750 ofs->match.dl_type = flow->key.flow.dl_type;
751 ofs->match.nw_src = flow->key.flow.nw_src;
752 ofs->match.nw_dst = flow->key.flow.nw_dst;
753 ofs->match.nw_proto = flow->key.flow.nw_proto;
754 memset(ofs->match.pad, 0, sizeof ofs->match.pad);
755 ofs->match.tp_src = flow->key.flow.tp_src;
756 ofs->match.tp_dst = flow->key.flow.tp_dst;
757 ofs->duration = htonl(now - flow->created);
758 ofs->packet_count = htonll(flow->packet_count);
759 ofs->byte_count = htonll(flow->byte_count);
760 ofs->priority = htons(flow->priority);
761 ofs->max_idle = htons(flow->max_idle);
762 memcpy(ofs->actions, flow->actions,
763 sizeof *ofs->actions * flow->n_actions);
767 /* 'buffer' was received on 'in_port', a physical switch port between 0 and
768 * OFPP_MAX. Process it according to 'chain'. */
769 void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
771 struct sw_flow_key key;
772 struct sw_flow *flow;
775 flow_extract(buffer, in_port, &key.flow);
776 flow = chain_lookup(dp->chain, &key);
778 flow_used(flow, buffer);
779 execute_actions(dp, buffer, in_port, &key,
780 flow->actions, flow->n_actions);
782 dp_output_control(dp, buffer, in_port, ntohs(dp->config.miss_send_len),
788 do_output(struct datapath *dp, struct buffer *buffer, int in_port,
789 size_t max_len, int out_port)
791 if (out_port != OFPP_CONTROLLER) {
792 dp_output_port(dp, buffer, in_port, out_port);
794 dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
799 execute_actions(struct datapath *dp, struct buffer *buffer,
800 int in_port, const struct sw_flow_key *key,
801 const struct ofp_action *actions, int n_actions)
803 /* Every output action needs a separate clone of 'buffer', but the common
804 * case is just a single output action, so that doing a clone and then
805 * freeing the original buffer is wasteful. So the following code is
806 * slightly obscure just to avoid that. */
808 size_t max_len=0; /* Initialze to make compiler happy */
813 eth_proto = ntohs(key->flow.dl_type);
815 for (i = 0; i < n_actions; i++) {
816 const struct ofp_action *a = &actions[i];
817 struct eth_header *eh = buffer->l2;
819 if (prev_port != -1) {
820 do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
824 switch (ntohs(a->type)) {
826 prev_port = ntohs(a->arg.output.port);
827 max_len = ntohs(a->arg.output.max_len);
830 case OFPAT_SET_DL_VLAN:
831 modify_vlan(buffer, key, a);
834 case OFPAT_SET_DL_SRC:
835 memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
838 case OFPAT_SET_DL_DST:
839 memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
842 case OFPAT_SET_NW_SRC:
843 case OFPAT_SET_NW_DST:
844 modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
847 case OFPAT_SET_TP_SRC:
848 case OFPAT_SET_TP_DST:
849 modify_th(buffer, eth_proto, key->flow.nw_proto, a);
857 do_output(dp, buffer, in_port, max_len, prev_port);
859 buffer_delete(buffer);
862 static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
863 uint8_t nw_proto, const struct ofp_action *a)
865 if (eth_proto == ETH_TYPE_IP) {
866 struct ip_header *nh = buffer->l3;
867 uint32_t new, *field;
869 new = a->arg.nw_addr;
870 field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
871 if (nw_proto == IP_TYPE_TCP) {
872 struct tcp_header *th = buffer->l4;
873 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
874 } else if (nw_proto == IP_TYPE_UDP) {
875 struct udp_header *th = buffer->l4;
877 th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
879 th->udp_csum = 0xffff;
883 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
888 static void modify_th(struct buffer *buffer, uint16_t eth_proto,
889 uint8_t nw_proto, const struct ofp_action *a)
891 if (eth_proto == ETH_TYPE_IP) {
892 uint16_t new, *field;
896 if (nw_proto == IP_TYPE_TCP) {
897 struct tcp_header *th = buffer->l4;
898 field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
899 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
901 } else if (nw_proto == IP_TYPE_UDP) {
902 struct udp_header *th = buffer->l4;
903 field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
904 th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
911 modify_vlan(struct buffer *buffer,
912 const struct sw_flow_key *key, const struct ofp_action *a)
914 uint16_t new_id = a->arg.vlan_id;
915 struct vlan_eth_header *veh;
917 if (new_id != htons(OFP_VLAN_NONE)) {
918 if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
919 /* Modify vlan id, but maintain other TCI values */
921 veh->veth_tci &= ~htons(VLAN_VID);
922 veh->veth_tci |= new_id;
924 /* Insert new vlan id. */
925 struct eth_header *eh = buffer->l2;
926 struct vlan_eth_header tmp;
927 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
928 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
929 tmp.veth_type = htons(ETH_TYPE_VLAN);
930 tmp.veth_tci = new_id;
931 tmp.veth_next_type = eh->eth_type;
933 veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
934 memcpy(veh, &tmp, sizeof tmp);
935 buffer->l2 -= VLAN_HEADER_LEN;
938 /* Remove an existing vlan header if it exists */
940 if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
941 struct eth_header tmp;
943 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
944 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
945 tmp.eth_type = veh->veth_next_type;
947 buffer->size -= VLAN_HEADER_LEN;
948 buffer->data += VLAN_HEADER_LEN;
949 buffer->l2 += VLAN_HEADER_LEN;
950 memcpy(buffer->data, &tmp, sizeof tmp);
956 recv_features_request(struct datapath *dp, const struct sender *sender,
959 dp_send_features_reply(dp, sender);
964 recv_get_config_request(struct datapath *dp, const struct sender *sender,
967 struct buffer *buffer;
968 struct ofp_switch_config *osc;
970 osc = alloc_openflow_buffer(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY,
973 assert(sizeof *osc == sizeof dp->config);
974 memcpy(((char *)osc) + sizeof osc->header,
975 ((char *)&dp->config) + sizeof dp->config.header,
976 sizeof dp->config - sizeof dp->config.header);
978 return send_openflow_buffer(dp, buffer, sender);
982 recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
985 const struct ofp_switch_config *osc = msg;
991 recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
994 const struct ofp_packet_out *opo = msg;
996 if (ntohl(opo->buffer_id) == (uint32_t) -1) {
997 /* FIXME: can we avoid copying data here? */
998 int data_len = ntohs(opo->header.length) - sizeof *opo;
999 struct buffer *buffer = buffer_new(data_len);
1000 buffer_put(buffer, opo->u.data, data_len);
1001 dp_output_port(dp, buffer,
1002 ntohs(opo->in_port), ntohs(opo->out_port));
1004 struct sw_flow_key key;
1005 struct buffer *buffer;
1008 buffer = retrieve_buffer(ntohl(opo->buffer_id));
1013 n_acts = (ntohs(opo->header.length) - sizeof *opo)
1014 / sizeof *opo->u.actions;
1015 flow_extract(buffer, ntohs(opo->in_port), &key.flow);
1016 execute_actions(dp, buffer, ntohs(opo->in_port),
1017 &key, opo->u.actions, n_acts);
1023 recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
1026 const struct ofp_port_mod *opm = msg;
1028 dp_update_port_flags(dp, &opm->desc);
1034 add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
1036 int error = -ENOMEM;
1039 struct sw_flow *flow;
1042 /* To prevent loops, make sure there's no action to send to the
1043 * OFP_TABLE virtual port.
1045 n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
1046 for (i=0; i<n_acts; i++) {
1047 const struct ofp_action *a = &ofm->actions[i];
1049 if (a->type == htons(OFPAT_OUTPUT)
1050 && (a->arg.output.port == htons(OFPP_TABLE)
1051 || a->arg.output.port == htons(OFPP_NONE))) {
1052 /* xxx Send fancy new error message? */
1057 /* Allocate memory. */
1058 flow = flow_alloc(n_acts);
1062 /* Fill out flow. */
1063 flow_extract_match(&flow->key, &ofm->match);
1064 flow->max_idle = ntohs(ofm->max_idle);
1065 flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
1066 flow->timeout = time(0) + flow->max_idle; /* FIXME */
1067 flow->n_actions = n_acts;
1068 flow->created = time(0); /* FIXME */
1069 flow->byte_count = 0;
1070 flow->packet_count = 0;
1071 memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
1074 error = chain_insert(dp->chain, flow);
1076 goto error_free_flow;
1079 if (ntohl(ofm->buffer_id) != UINT32_MAX) {
1080 struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
1082 struct sw_flow_key key;
1083 uint16_t in_port = ntohs(ofm->match.in_port);
1084 flow_used(flow, buffer);
1085 flow_extract(buffer, in_port, &key.flow);
1086 execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
1096 if (ntohl(ofm->buffer_id) != (uint32_t) -1)
1097 discard_buffer(ntohl(ofm->buffer_id));
1102 recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
1105 const struct ofp_flow_mod *ofm = msg;
1106 uint16_t command = ntohs(ofm->command);
1108 if (command == OFPFC_ADD) {
1109 return add_flow(dp, ofm);
1110 } else if (command == OFPFC_DELETE) {
1111 struct sw_flow_key key;
1112 flow_extract_match(&key, &ofm->match);
1113 return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
1114 } else if (command == OFPFC_DELETE_STRICT) {
1115 struct sw_flow_key key;
1117 flow_extract_match(&key, &ofm->match);
1118 priority = key.wildcards ? ntohs(ofm->priority) : -1;
1119 return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
1125 struct flow_stats_state {
1127 struct sw_table_position position;
1128 struct ofp_flow_stats_request rq;
1131 struct buffer *buffer;
1134 #define MAX_FLOW_STATS_BYTES 4096
1136 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1139 const struct ofp_flow_stats_request *fsr = body;
1140 struct flow_stats_state *s = xmalloc(sizeof *s);
1141 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1142 memset(&s->position, 0, sizeof s->position);
1148 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1150 struct flow_stats_state *s = private;
1151 fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
1152 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1155 static int flow_stats_dump(struct datapath *dp, void *state,
1156 struct buffer *buffer)
1158 struct flow_stats_state *s = state;
1159 struct sw_flow_key match_key;
1161 flow_extract_match(&match_key, &s->rq.match);
1164 while (s->table_idx < dp->chain->n_tables
1165 && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
1167 struct sw_table *table = dp->chain->tables[s->table_idx];
1169 if (table->iterate(table, &match_key, &s->position,
1170 flow_stats_dump_callback, s))
1174 memset(&s->position, 0, sizeof s->position);
1176 return s->buffer->size >= MAX_FLOW_STATS_BYTES;
1179 static void flow_stats_done(void *state)
1184 struct aggregate_stats_state {
1185 struct ofp_aggregate_stats_request rq;
1188 static int aggregate_stats_init(struct datapath *dp,
1189 const void *body, int body_len,
1192 const struct ofp_aggregate_stats_request *rq = body;
1193 struct aggregate_stats_state *s = xmalloc(sizeof *s);
1199 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1201 struct ofp_aggregate_stats_reply *rpy = private;
1202 rpy->packet_count += flow->packet_count;
1203 rpy->byte_count += flow->byte_count;
1208 static int aggregate_stats_dump(struct datapath *dp, void *state,
1209 struct buffer *buffer)
1211 struct aggregate_stats_state *s = state;
1212 struct ofp_aggregate_stats_request *rq = &s->rq;
1213 struct ofp_aggregate_stats_reply *rpy;
1214 struct sw_table_position position;
1215 struct sw_flow_key match_key;
1218 rpy = buffer_put_uninit(buffer, sizeof *rpy);
1219 memset(rpy, 0, sizeof *rpy);
1221 flow_extract_match(&match_key, &rq->match);
1222 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1223 memset(&position, 0, sizeof position);
1224 while (table_idx < dp->chain->n_tables
1225 && (rq->table_id == 0xff || rq->table_id == table_idx))
1227 struct sw_table *table = dp->chain->tables[table_idx];
1230 error = table->iterate(table, &match_key, &position,
1231 aggregate_stats_dump_callback, rpy);
1236 memset(&position, 0, sizeof position);
1239 rpy->packet_count = htonll(rpy->packet_count);
1240 rpy->byte_count = htonll(rpy->byte_count);
1241 rpy->flow_count = htonl(rpy->flow_count);
1245 static void aggregate_stats_done(void *state)
1250 static int table_stats_dump(struct datapath *dp, void *state,
1251 struct buffer *buffer)
1254 for (i = 0; i < dp->chain->n_tables; i++) {
1255 struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
1256 struct sw_table_stats stats;
1257 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1258 strncpy(ots->name, stats.name, sizeof ots->name);
1260 memset(ots->pad, 0, sizeof ots->pad);
1261 ots->max_entries = htonl(stats.max_flows);
1262 ots->active_count = htonl(stats.n_flows);
1263 ots->matched_count = htonll(0); /* FIXME */
1268 struct port_stats_state {
1272 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1275 struct port_stats_state *s = xmalloc(sizeof *s);
1281 static int port_stats_dump(struct datapath *dp, void *state,
1282 struct buffer *buffer)
1284 struct port_stats_state *s = state;
1287 for (i = s->port; i < OFPP_MAX; i++) {
1288 struct sw_port *p = &dp->ports[i];
1289 struct ofp_port_stats *ops;
1293 ops = buffer_put_uninit(buffer, sizeof *ops);
1294 ops->port_no = htons(port_no(dp, p));
1295 memset(ops->pad, 0, sizeof ops->pad);
1296 ops->rx_count = htonll(p->rx_count);
1297 ops->tx_count = htonll(p->tx_count);
1298 ops->drop_count = htonll(p->drop_count);
1305 static void port_stats_done(void *state)
1311 /* Minimum and maximum acceptable number of bytes in body member of
1312 * struct ofp_stats_request. */
1313 size_t min_body, max_body;
1315 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1316 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1317 * Returns zero if successful, otherwise a negative error code.
1318 * May initialize '*state' to state information. May be null if no
1319 * initialization is required.*/
1320 int (*init)(struct datapath *dp, const void *body, int body_len,
1323 /* Appends statistics for 'dp' to 'buffer', which initially contains a
1324 * struct ofp_stats_reply. On success, it should return 1 if it should be
1325 * called again later with another buffer, 0 if it is done, or a negative
1326 * errno value on failure. */
1327 int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
1329 /* Cleans any state created by the init or dump functions. May be null
1330 * if no cleanup is required. */
1331 void (*done)(void *state);
1334 static const struct stats_type stats[] = {
1336 sizeof(struct ofp_flow_stats_request),
1337 sizeof(struct ofp_flow_stats_request),
1342 [OFPST_AGGREGATE] = {
1343 sizeof(struct ofp_aggregate_stats_request),
1344 sizeof(struct ofp_aggregate_stats_request),
1345 aggregate_stats_init,
1346 aggregate_stats_dump,
1347 aggregate_stats_done
1365 struct stats_dump_cb {
1367 struct ofp_stats_request *rq;
1368 struct sender sender;
1369 const struct stats_type *s;
1374 stats_dump(struct datapath *dp, void *cb_)
1376 struct stats_dump_cb *cb = cb_;
1377 struct ofp_stats_reply *osr;
1378 struct buffer *buffer;
1385 osr = alloc_openflow_buffer(dp, sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
1387 osr->type = htons(cb->s - stats);
1390 err = cb->s->dump(dp, cb->state, buffer);
1396 /* Buffer might have been reallocated, so find our data again. */
1397 osr = buffer_at_assert(buffer, 0, sizeof *osr);
1398 osr->flags = ntohs(OFPSF_REPLY_MORE);
1400 err2 = send_openflow_buffer(dp, buffer, &cb->sender);
1410 stats_done(void *cb_)
1412 struct stats_dump_cb *cb = cb_;
1415 cb->s->done(cb->state);
1422 recv_stats_request(struct datapath *dp, const struct sender *sender,
1425 const struct ofp_stats_request *rq = oh;
1426 size_t rq_len = ntohs(rq->header.length);
1427 struct stats_dump_cb *cb;
1431 type = ntohs(rq->type);
1432 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1433 VLOG_WARN("received stats request of unknown type %d", type);
1437 cb = xmalloc(sizeof *cb);
1439 cb->rq = xmemdup(rq, rq_len);
1440 cb->sender = *sender;
1441 cb->s = &stats[type];
1444 body_len = rq_len - offsetof(struct ofp_stats_request, body);
1445 if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
1446 VLOG_WARN("stats request type %d with bad body length %d",
1453 err = cb->s->init(dp, rq->body, body_len, &cb->state);
1455 VLOG_WARN("failed initialization of stats request type %d: %s",
1456 type, strerror(-err));
1461 remote_start_dump(sender->remote, stats_dump, stats_done, cb);
1471 recv_echo_request(struct datapath *dp, const struct sender *sender,
1474 return send_openflow_buffer(dp, make_echo_reply(oh), sender);
1478 recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
1479 const void *oh UNUSED)
1484 /* 'msg', which is 'length' bytes long, was received from the control path.
1485 * Apply it to 'chain'. */
1487 fwd_control_input(struct datapath *dp, const struct sender *sender,
1488 const void *msg, size_t length)
1490 struct openflow_packet {
1492 int (*handler)(struct datapath *, const struct sender *, const void *);
1495 static const struct openflow_packet packets[] = {
1496 [OFPT_FEATURES_REQUEST] = {
1497 sizeof (struct ofp_header),
1498 recv_features_request,
1500 [OFPT_GET_CONFIG_REQUEST] = {
1501 sizeof (struct ofp_header),
1502 recv_get_config_request,
1504 [OFPT_SET_CONFIG] = {
1505 sizeof (struct ofp_switch_config),
1508 [OFPT_PACKET_OUT] = {
1509 sizeof (struct ofp_packet_out),
1513 sizeof (struct ofp_flow_mod),
1517 sizeof (struct ofp_port_mod),
1520 [OFPT_STATS_REQUEST] = {
1521 sizeof (struct ofp_stats_request),
1524 [OFPT_ECHO_REQUEST] = {
1525 sizeof (struct ofp_header),
1528 [OFPT_ECHO_REPLY] = {
1529 sizeof (struct ofp_header),
1534 const struct openflow_packet *pkt;
1535 struct ofp_header *oh;
1537 oh = (struct ofp_header *) msg;
1538 if (oh->version != OFP_VERSION || oh->type >= ARRAY_SIZE(packets)
1539 || ntohs(oh->length) > length)
1542 pkt = &packets[oh->type];
1545 if (length < pkt->min_size)
1548 return pkt->handler(dp, sender, msg);
1551 /* Packet buffering. */
1553 #define OVERWRITE_SECS 1
1555 struct packet_buffer {
1556 struct buffer *buffer;
1561 static struct packet_buffer buffers[N_PKT_BUFFERS];
1562 static unsigned int buffer_idx;
1564 uint32_t save_buffer(struct buffer *buffer)
1566 struct packet_buffer *p;
1569 buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
1570 p = &buffers[buffer_idx];
1572 /* Don't buffer packet if existing entry is less than
1573 * OVERWRITE_SECS old. */
1574 if (time(0) < p->timeout) { /* FIXME */
1577 buffer_delete(p->buffer);
1580 /* Don't use maximum cookie value since the all-bits-1 id is
1582 if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
1584 p->buffer = buffer_clone(buffer); /* FIXME */
1585 p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
1586 id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
1591 static struct buffer *retrieve_buffer(uint32_t id)
1593 struct buffer *buffer = NULL;
1594 struct packet_buffer *p;
1596 p = &buffers[id & PKT_BUFFER_MASK];
1597 if (p->cookie == id >> PKT_BUFFER_BITS) {
1601 printf("cookie mismatch: %x != %x\n",
1602 id >> PKT_BUFFER_BITS, p->cookie);
1608 static void discard_buffer(uint32_t id)
1610 struct packet_buffer *p;
1612 p = &buffers[id & PKT_BUFFER_MASK];
1613 if (p->cookie == id >> PKT_BUFFER_BITS) {
1614 buffer_delete(p->buffer);