2 * Copyright (c) 2008, 2009 Nicira Networks.
4 * Permission to use, copy, modify, and/or distribute this software for any
5 * purpose with or without fee is hereby granted, provided that the above
6 * copyright notice and this permission notice appear in all copies.
8 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18 #include "vconn-provider.h"
22 #include <netinet/in.h>
27 #include "dynamic-string.h"
29 #include "ofp-print.h"
31 #include "openflow/nicira-ext.h"
32 #include "openflow/openflow.h"
34 #include "poll-loop.h"
38 #define THIS_MODULE VLM_vconn
41 /* State of an active vconn.*/
43 /* This is the ordinary progression of states. */
44 VCS_CONNECTING, /* Underlying vconn is not connected. */
45 VCS_SEND_HELLO, /* Waiting to send OFPT_HELLO message. */
46 VCS_RECV_HELLO, /* Waiting to receive OFPT_HELLO message. */
47 VCS_CONNECTED, /* Connection established. */
49 /* These states are entered only when something goes wrong. */
50 VCS_SEND_ERROR, /* Sending OFPT_ERROR message. */
51 VCS_DISCONNECTED /* Connection failed or connection closed. */
54 static struct vconn_class *vconn_classes[] = {
62 static struct pvconn_class *pvconn_classes[] = {
70 /* Rate limit for individual OpenFlow messages going over the vconn, output at
71 * DBG level. This is very high because, if these are enabled, it is because
72 * we really need to see them. */
73 static struct vlog_rate_limit ofmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
75 /* Rate limit for OpenFlow message parse errors. These always indicate a bug
76 * in the peer and so there's not much point in showing a lot of them. */
77 static struct vlog_rate_limit bad_ofmsg_rl = VLOG_RATE_LIMIT_INIT(1, 5);
79 static int do_recv(struct vconn *, struct ofpbuf **);
80 static int do_send(struct vconn *, struct ofpbuf *);
82 /* Check the validity of the vconn class structures. */
84 check_vconn_classes(void)
89 for (i = 0; i < ARRAY_SIZE(vconn_classes); i++) {
90 struct vconn_class *class = vconn_classes[i];
91 assert(class->name != NULL);
92 assert(class->open != NULL);
93 if (class->close || class->recv || class->send || class->wait) {
94 assert(class->close != NULL);
95 assert(class->recv != NULL);
96 assert(class->send != NULL);
97 assert(class->wait != NULL);
99 /* This class delegates to another one. */
103 for (i = 0; i < ARRAY_SIZE(pvconn_classes); i++) {
104 struct pvconn_class *class = pvconn_classes[i];
105 assert(class->name != NULL);
106 assert(class->listen != NULL);
107 if (class->close || class->accept || class->wait) {
108 assert(class->close != NULL);
109 assert(class->accept != NULL);
110 assert(class->wait != NULL);
112 /* This class delegates to another one. */
118 /* Prints information on active (if 'active') and passive (if 'passive')
119 * connection methods supported by the vconn. If 'bootstrap' is true, also
120 * advertises options to bootstrap the CA certificate. */
122 vconn_usage(bool active, bool passive, bool bootstrap UNUSED)
124 /* Really this should be implemented via callbacks into the vconn
125 * providers, but that seems too heavy-weight to bother with at the
130 printf("Active OpenFlow connection methods:\n");
131 printf(" tcp:HOST[:PORT] "
132 "PORT (default: %d) on remote TCP HOST\n", OFP_TCP_PORT);
134 printf(" ssl:HOST[:PORT] "
135 "SSL PORT (default: %d) on remote HOST\n", OFP_SSL_PORT);
137 printf(" unix:FILE Unix domain socket named FILE\n");
141 printf("Passive OpenFlow connection methods:\n");
142 printf(" ptcp:[PORT] "
143 "listen to TCP PORT (default: %d)\n",
146 printf(" pssl:[PORT] "
147 "listen for SSL on PORT (default: %d)\n",
150 printf(" punix:FILE "
151 "listen on Unix domain socket FILE\n");
155 printf("PKI configuration (required to use SSL):\n"
156 " -p, --private-key=FILE file with private key\n"
157 " -c, --certificate=FILE file with certificate for private key\n"
158 " -C, --ca-cert=FILE file with peer CA certificate\n");
160 printf(" --bootstrap-ca-cert=FILE file with peer CA certificate "
161 "to read or create\n");
166 /* Attempts to connect to an OpenFlow device. 'name' is a connection name in
167 * the form "TYPE:ARGS", where TYPE is an active vconn class's name and ARGS
168 * are vconn class-specific.
170 * The vconn will automatically negotiate an OpenFlow protocol version
171 * acceptable to both peers on the connection. The version negotiated will be
172 * no lower than 'min_version' and no higher than OFP_VERSION.
174 * Returns 0 if successful, otherwise a positive errno value. If successful,
175 * stores a pointer to the new connection in '*vconnp', otherwise a null
178 vconn_open(const char *name, int min_version, struct vconn **vconnp)
183 COVERAGE_INC(vconn_open);
184 check_vconn_classes();
187 prefix_len = strcspn(name, ":");
188 if (prefix_len == strlen(name)) {
191 for (i = 0; i < ARRAY_SIZE(vconn_classes); i++) {
192 struct vconn_class *class = vconn_classes[i];
193 if (strlen(class->name) == prefix_len
194 && !memcmp(class->name, name, prefix_len)) {
196 char *suffix_copy = xstrdup(name + prefix_len + 1);
197 int retval = class->open(name, suffix_copy, &vconn);
200 assert(vconn->state != VCS_CONNECTING
201 || vconn->class->connect);
202 vconn->min_version = min_version;
212 vconn_open_block(const char *name, int min_version, struct vconn **vconnp)
217 error = vconn_open(name, min_version, &vconn);
218 while (error == EAGAIN) {
219 vconn_connect_wait(vconn);
221 error = vconn_connect(vconn);
222 assert(error != EINPROGRESS);
233 /* Closes 'vconn'. */
235 vconn_close(struct vconn *vconn)
238 char *name = vconn->name;
239 (vconn->class->close)(vconn);
244 /* Returns the name of 'vconn', that is, the string passed to vconn_open(). */
246 vconn_get_name(const struct vconn *vconn)
251 /* Returns the IP address of the peer, or 0 if the peer is not connected over
252 * an IP-based protocol or if its IP address is not yet known. */
254 vconn_get_ip(const struct vconn *vconn)
260 vcs_connecting(struct vconn *vconn)
262 int retval = (vconn->class->connect)(vconn);
263 assert(retval != EINPROGRESS);
265 vconn->state = VCS_SEND_HELLO;
266 } else if (retval != EAGAIN) {
267 vconn->state = VCS_DISCONNECTED;
268 vconn->error = retval;
273 vcs_send_hello(struct vconn *vconn)
278 make_openflow(sizeof(struct ofp_header), OFPT_HELLO, &b);
279 retval = do_send(vconn, b);
281 vconn->state = VCS_RECV_HELLO;
284 if (retval != EAGAIN) {
285 vconn->state = VCS_DISCONNECTED;
286 vconn->error = retval;
292 vcs_recv_hello(struct vconn *vconn)
297 retval = do_recv(vconn, &b);
299 struct ofp_header *oh = b->data;
301 if (oh->type == OFPT_HELLO) {
302 if (b->size > sizeof *oh) {
303 struct ds msg = DS_EMPTY_INITIALIZER;
304 ds_put_format(&msg, "%s: extra-long hello:\n", vconn->name);
305 ds_put_hex_dump(&msg, b->data, b->size, 0, true);
306 VLOG_WARN_RL(&bad_ofmsg_rl, "%s", ds_cstr(&msg));
310 vconn->version = MIN(OFP_VERSION, oh->version);
311 if (vconn->version < vconn->min_version) {
312 VLOG_WARN_RL(&bad_ofmsg_rl,
313 "%s: version negotiation failed: we support "
314 "versions 0x%02x to 0x%02x inclusive but peer "
315 "supports no later than version 0x%02"PRIx8,
316 vconn->name, vconn->min_version, OFP_VERSION,
318 vconn->state = VCS_SEND_ERROR;
320 VLOG_DBG("%s: negotiated OpenFlow version 0x%02x "
321 "(we support versions 0x%02x to 0x%02x inclusive, "
322 "peer no later than version 0x%02"PRIx8")",
323 vconn->name, vconn->version, vconn->min_version,
324 OFP_VERSION, oh->version);
325 vconn->state = VCS_CONNECTED;
330 char *s = ofp_to_string(b->data, b->size, 1);
331 VLOG_WARN_RL(&bad_ofmsg_rl,
332 "%s: received message while expecting hello: %s",
340 if (retval != EAGAIN) {
341 vconn->state = VCS_DISCONNECTED;
342 vconn->error = retval;
347 vcs_send_error(struct vconn *vconn)
349 struct ofp_error_msg *error;
354 snprintf(s, sizeof s, "We support versions 0x%02x to 0x%02x inclusive but "
355 "you support no later than version 0x%02"PRIx8".",
356 vconn->min_version, OFP_VERSION, vconn->version);
357 error = make_openflow(sizeof *error, OFPT_ERROR, &b);
358 error->type = htons(OFPET_HELLO_FAILED);
359 error->code = htons(OFPHFC_INCOMPATIBLE);
360 ofpbuf_put(b, s, strlen(s));
361 update_openflow_length(b);
362 retval = do_send(vconn, b);
366 if (retval != EAGAIN) {
367 vconn->state = VCS_DISCONNECTED;
368 vconn->error = retval ? retval : EPROTO;
372 /* Tries to complete the connection on 'vconn', which must be an active
373 * vconn. If 'vconn''s connection is complete, returns 0 if the connection
374 * was successful or a positive errno value if it failed. If the
375 * connection is still in progress, returns EAGAIN. */
377 vconn_connect(struct vconn *vconn)
379 enum vconn_state last_state;
381 assert(vconn->min_version >= 0);
383 last_state = vconn->state;
384 switch (vconn->state) {
386 vcs_connecting(vconn);
390 vcs_send_hello(vconn);
394 vcs_recv_hello(vconn);
401 vcs_send_error(vconn);
404 case VCS_DISCONNECTED:
410 } while (vconn->state != last_state);
415 /* Tries to receive an OpenFlow message from 'vconn', which must be an active
416 * vconn. If successful, stores the received message into '*msgp' and returns
417 * 0. The caller is responsible for destroying the message with
418 * ofpbuf_delete(). On failure, returns a positive errno value and stores a
419 * null pointer into '*msgp'. On normal connection close, returns EOF.
421 * vconn_recv will not block waiting for a packet to arrive. If no packets
422 * have been received, it returns EAGAIN immediately. */
424 vconn_recv(struct vconn *vconn, struct ofpbuf **msgp)
426 int retval = vconn_connect(vconn);
428 retval = do_recv(vconn, msgp);
434 do_recv(struct vconn *vconn, struct ofpbuf **msgp)
439 retval = (vconn->class->recv)(vconn, msgp);
441 struct ofp_header *oh;
443 COVERAGE_INC(vconn_received);
444 if (VLOG_IS_DBG_ENABLED()) {
445 char *s = ofp_to_string((*msgp)->data, (*msgp)->size, 1);
446 VLOG_DBG_RL(&ofmsg_rl, "%s: received: %s", vconn->name, s);
450 oh = ofpbuf_at_assert(*msgp, 0, sizeof *oh);
451 if (oh->version != vconn->version
452 && oh->type != OFPT_HELLO
453 && oh->type != OFPT_ERROR
454 && oh->type != OFPT_ECHO_REQUEST
455 && oh->type != OFPT_ECHO_REPLY
456 && oh->type != OFPT_VENDOR)
458 if (vconn->version < 0) {
459 if (oh->type == OFPT_PACKET_IN
460 || oh->type == OFPT_FLOW_EXPIRED
461 || oh->type == OFPT_PORT_STATUS) {
462 /* The kernel datapath is stateless and doesn't really
463 * support version negotiation, so it can end up sending
464 * these asynchronous message before version negotiation
465 * is complete. Just ignore them.
467 * (After we move OFPT_PORT_STATUS messages from the kernel
468 * into secchan, we won't get those here, since secchan
469 * does proper version negotiation.) */
470 ofpbuf_delete(*msgp);
473 VLOG_ERR_RL(&bad_ofmsg_rl,
474 "%s: received OpenFlow message type %"PRIu8" "
475 "before version negotiation complete",
476 vconn->name, oh->type);
478 VLOG_ERR_RL(&bad_ofmsg_rl,
479 "%s: received OpenFlow version 0x%02"PRIx8" "
481 vconn->name, oh->version, vconn->version);
483 ofpbuf_delete(*msgp);
493 /* Tries to queue 'msg' for transmission on 'vconn', which must be an active
494 * vconn. If successful, returns 0, in which case ownership of 'msg' is
495 * transferred to the vconn. Success does not guarantee that 'msg' has been or
496 * ever will be delivered to the peer, only that it has been queued for
499 * Returns a positive errno value on failure, in which case the caller
500 * retains ownership of 'msg'.
502 * vconn_send will not block. If 'msg' cannot be immediately accepted for
503 * transmission, it returns EAGAIN immediately. */
505 vconn_send(struct vconn *vconn, struct ofpbuf *msg)
507 int retval = vconn_connect(vconn);
509 retval = do_send(vconn, msg);
515 do_send(struct vconn *vconn, struct ofpbuf *msg)
519 assert(msg->size >= sizeof(struct ofp_header));
520 assert(((struct ofp_header *) msg->data)->length == htons(msg->size));
521 if (!VLOG_IS_DBG_ENABLED()) {
522 COVERAGE_INC(vconn_sent);
523 retval = (vconn->class->send)(vconn, msg);
525 char *s = ofp_to_string(msg->data, msg->size, 1);
526 retval = (vconn->class->send)(vconn, msg);
527 if (retval != EAGAIN) {
528 VLOG_DBG_RL(&ofmsg_rl, "%s: sent (%s): %s",
529 vconn->name, strerror(retval), s);
536 /* Same as vconn_send, except that it waits until 'msg' can be transmitted. */
538 vconn_send_block(struct vconn *vconn, struct ofpbuf *msg)
541 while ((retval = vconn_send(vconn, msg)) == EAGAIN) {
542 vconn_send_wait(vconn);
548 /* Same as vconn_recv, except that it waits until a message is received. */
550 vconn_recv_block(struct vconn *vconn, struct ofpbuf **msgp)
553 while ((retval = vconn_recv(vconn, msgp)) == EAGAIN) {
554 vconn_recv_wait(vconn);
560 /* Waits until a message with a transaction ID matching 'xid' is recived on
561 * 'vconn'. Returns 0 if successful, in which case the reply is stored in
562 * '*replyp' for the caller to examine and free. Otherwise returns a positive
563 * errno value, or EOF, and sets '*replyp' to null.
565 * 'request' is always destroyed, regardless of the return value. */
567 vconn_recv_xid(struct vconn *vconn, uint32_t xid, struct ofpbuf **replyp)
571 struct ofpbuf *reply;
574 error = vconn_recv_block(vconn, &reply);
579 recv_xid = ((struct ofp_header *) reply->data)->xid;
580 if (xid == recv_xid) {
585 VLOG_DBG_RL(&bad_ofmsg_rl, "%s: received reply with xid %08"PRIx32
586 " != expected %08"PRIx32, vconn->name, recv_xid, xid);
587 ofpbuf_delete(reply);
591 /* Sends 'request' to 'vconn' and blocks until it receives a reply with a
592 * matching transaction ID. Returns 0 if successful, in which case the reply
593 * is stored in '*replyp' for the caller to examine and free. Otherwise
594 * returns a positive errno value, or EOF, and sets '*replyp' to null.
596 * 'request' is always destroyed, regardless of the return value. */
598 vconn_transact(struct vconn *vconn, struct ofpbuf *request,
599 struct ofpbuf **replyp)
601 uint32_t send_xid = ((struct ofp_header *) request->data)->xid;
605 error = vconn_send_block(vconn, request);
607 ofpbuf_delete(request);
609 return error ? error : vconn_recv_xid(vconn, send_xid, replyp);
613 vconn_wait(struct vconn *vconn, enum vconn_wait_type wait)
615 assert(wait == WAIT_CONNECT || wait == WAIT_RECV || wait == WAIT_SEND);
617 switch (vconn->state) {
634 case VCS_DISCONNECTED:
635 poll_immediate_wake();
638 (vconn->class->wait)(vconn, wait);
642 vconn_connect_wait(struct vconn *vconn)
644 vconn_wait(vconn, WAIT_CONNECT);
648 vconn_recv_wait(struct vconn *vconn)
650 vconn_wait(vconn, WAIT_RECV);
654 vconn_send_wait(struct vconn *vconn)
656 vconn_wait(vconn, WAIT_SEND);
659 /* Attempts to start listening for OpenFlow connections. 'name' is a
660 * connection name in the form "TYPE:ARGS", where TYPE is an passive vconn
661 * class's name and ARGS are vconn class-specific.
663 * Returns 0 if successful, otherwise a positive errno value. If successful,
664 * stores a pointer to the new connection in '*pvconnp', otherwise a null
667 pvconn_open(const char *name, struct pvconn **pvconnp)
672 check_vconn_classes();
675 prefix_len = strcspn(name, ":");
676 if (prefix_len == strlen(name)) {
679 for (i = 0; i < ARRAY_SIZE(pvconn_classes); i++) {
680 struct pvconn_class *class = pvconn_classes[i];
681 if (strlen(class->name) == prefix_len
682 && !memcmp(class->name, name, prefix_len)) {
683 char *suffix_copy = xstrdup(name + prefix_len + 1);
684 int retval = class->listen(name, suffix_copy, pvconnp);
695 /* Returns the name that was used to open 'pvconn'. The caller must not
696 * modify or free the name. */
698 pvconn_get_name(const struct pvconn *pvconn)
703 /* Closes 'pvconn'. */
705 pvconn_close(struct pvconn *pvconn)
707 if (pvconn != NULL) {
708 char *name = pvconn->name;
709 (pvconn->class->close)(pvconn);
714 /* Tries to accept a new connection on 'pvconn'. If successful, stores the new
715 * connection in '*new_vconn' and returns 0. Otherwise, returns a positive
718 * The new vconn will automatically negotiate an OpenFlow protocol version
719 * acceptable to both peers on the connection. The version negotiated will be
720 * no lower than 'min_version' and no higher than OFP_VERSION.
722 * pvconn_accept() will not block waiting for a connection. If no connection
723 * is ready to be accepted, it returns EAGAIN immediately. */
725 pvconn_accept(struct pvconn *pvconn, int min_version, struct vconn **new_vconn)
727 int retval = (pvconn->class->accept)(pvconn, new_vconn);
731 assert((*new_vconn)->state != VCS_CONNECTING
732 || (*new_vconn)->class->connect);
733 (*new_vconn)->min_version = min_version;
739 pvconn_wait(struct pvconn *pvconn)
741 (pvconn->class->wait)(pvconn);
744 /* XXX we should really use consecutive xids to avoid probabilistic
746 static inline uint32_t
749 return random_uint32();
752 /* Allocates and stores in '*bufferp' a new ofpbuf with a size of
753 * 'openflow_len', starting with an OpenFlow header with the given 'type' and
754 * an arbitrary transaction id. Allocated bytes beyond the header, if any, are
757 * The caller is responsible for freeing '*bufferp' when it is no longer
760 * The OpenFlow header length is initially set to 'openflow_len'; if the
761 * message is later extended, the length should be updated with
762 * update_openflow_length() before sending.
764 * Returns the header. */
766 make_openflow(size_t openflow_len, uint8_t type, struct ofpbuf **bufferp)
768 *bufferp = ofpbuf_new(openflow_len);
769 return put_openflow_xid(openflow_len, type, alloc_xid(), *bufferp);
772 /* Allocates and stores in '*bufferp' a new ofpbuf with a size of
773 * 'openflow_len', starting with an OpenFlow header with the given 'type' and
774 * transaction id 'xid'. Allocated bytes beyond the header, if any, are
777 * The caller is responsible for freeing '*bufferp' when it is no longer
780 * The OpenFlow header length is initially set to 'openflow_len'; if the
781 * message is later extended, the length should be updated with
782 * update_openflow_length() before sending.
784 * Returns the header. */
786 make_openflow_xid(size_t openflow_len, uint8_t type, uint32_t xid,
787 struct ofpbuf **bufferp)
789 *bufferp = ofpbuf_new(openflow_len);
790 return put_openflow_xid(openflow_len, type, xid, *bufferp);
793 /* Appends 'openflow_len' bytes to 'buffer', starting with an OpenFlow header
794 * with the given 'type' and an arbitrary transaction id. Allocated bytes
795 * beyond the header, if any, are zeroed.
797 * The OpenFlow header length is initially set to 'openflow_len'; if the
798 * message is later extended, the length should be updated with
799 * update_openflow_length() before sending.
801 * Returns the header. */
803 put_openflow(size_t openflow_len, uint8_t type, struct ofpbuf *buffer)
805 return put_openflow_xid(openflow_len, type, alloc_xid(), buffer);
808 /* Appends 'openflow_len' bytes to 'buffer', starting with an OpenFlow header
809 * with the given 'type' and an transaction id 'xid'. Allocated bytes beyond
810 * the header, if any, are zeroed.
812 * The OpenFlow header length is initially set to 'openflow_len'; if the
813 * message is later extended, the length should be updated with
814 * update_openflow_length() before sending.
816 * Returns the header. */
818 put_openflow_xid(size_t openflow_len, uint8_t type, uint32_t xid,
819 struct ofpbuf *buffer)
821 struct ofp_header *oh;
823 assert(openflow_len >= sizeof *oh);
824 assert(openflow_len <= UINT16_MAX);
826 oh = ofpbuf_put_uninit(buffer, openflow_len);
827 oh->version = OFP_VERSION;
829 oh->length = htons(openflow_len);
831 memset(oh + 1, 0, openflow_len - sizeof *oh);
835 /* Updates the 'length' field of the OpenFlow message in 'buffer' to
838 update_openflow_length(struct ofpbuf *buffer)
840 struct ofp_header *oh = ofpbuf_at_assert(buffer, 0, sizeof *oh);
841 oh->length = htons(buffer->size);
845 make_flow_mod(uint16_t command, const flow_t *flow, size_t actions_len)
847 struct ofp_flow_mod *ofm;
848 size_t size = sizeof *ofm + actions_len;
849 struct ofpbuf *out = ofpbuf_new(size);
850 ofm = ofpbuf_put_zeros(out, sizeof *ofm);
851 ofm->header.version = OFP_VERSION;
852 ofm->header.type = OFPT_FLOW_MOD;
853 ofm->header.length = htons(size);
854 ofm->match.wildcards = htonl(0);
855 ofm->match.in_port = htons(flow->in_port == ODPP_LOCAL ? OFPP_LOCAL
857 memcpy(ofm->match.dl_src, flow->dl_src, sizeof ofm->match.dl_src);
858 memcpy(ofm->match.dl_dst, flow->dl_dst, sizeof ofm->match.dl_dst);
859 ofm->match.dl_vlan = flow->dl_vlan;
860 ofm->match.dl_type = flow->dl_type;
861 ofm->match.nw_src = flow->nw_src;
862 ofm->match.nw_dst = flow->nw_dst;
863 ofm->match.nw_proto = flow->nw_proto;
864 ofm->match.tp_src = flow->tp_src;
865 ofm->match.tp_dst = flow->tp_dst;
866 ofm->command = htons(command);
871 make_add_flow(const flow_t *flow, uint32_t buffer_id,
872 uint16_t idle_timeout, size_t actions_len)
874 struct ofpbuf *out = make_flow_mod(OFPFC_ADD, flow, actions_len);
875 struct ofp_flow_mod *ofm = out->data;
876 ofm->idle_timeout = htons(idle_timeout);
877 ofm->hard_timeout = htons(OFP_FLOW_PERMANENT);
878 ofm->buffer_id = htonl(buffer_id);
883 make_del_flow(const flow_t *flow)
885 struct ofpbuf *out = make_flow_mod(OFPFC_DELETE_STRICT, flow, 0);
886 struct ofp_flow_mod *ofm = out->data;
887 ofm->out_port = htons(OFPP_NONE);
892 make_add_simple_flow(const flow_t *flow,
893 uint32_t buffer_id, uint16_t out_port,
894 uint16_t idle_timeout)
896 struct ofp_action_output *oao;
897 struct ofpbuf *buffer = make_add_flow(flow, buffer_id, idle_timeout,
899 oao = ofpbuf_put_zeros(buffer, sizeof *oao);
900 oao->type = htons(OFPAT_OUTPUT);
901 oao->len = htons(sizeof *oao);
902 oao->port = htons(out_port);
907 make_packet_out(const struct ofpbuf *packet, uint32_t buffer_id,
909 const struct ofp_action_header *actions, size_t n_actions)
911 size_t actions_len = n_actions * sizeof *actions;
912 struct ofp_packet_out *opo;
913 size_t size = sizeof *opo + actions_len + (packet ? packet->size : 0);
914 struct ofpbuf *out = ofpbuf_new(size);
916 opo = ofpbuf_put_uninit(out, sizeof *opo);
917 opo->header.version = OFP_VERSION;
918 opo->header.type = OFPT_PACKET_OUT;
919 opo->header.length = htons(size);
920 opo->header.xid = htonl(0);
921 opo->buffer_id = htonl(buffer_id);
922 opo->in_port = htons(in_port == ODPP_LOCAL ? OFPP_LOCAL : in_port);
923 opo->actions_len = htons(actions_len);
924 ofpbuf_put(out, actions, actions_len);
926 ofpbuf_put(out, packet->data, packet->size);
932 make_unbuffered_packet_out(const struct ofpbuf *packet,
933 uint16_t in_port, uint16_t out_port)
935 struct ofp_action_output action;
936 action.type = htons(OFPAT_OUTPUT);
937 action.len = htons(sizeof action);
938 action.port = htons(out_port);
939 return make_packet_out(packet, UINT32_MAX, in_port,
940 (struct ofp_action_header *) &action, 1);
944 make_buffered_packet_out(uint32_t buffer_id,
945 uint16_t in_port, uint16_t out_port)
947 struct ofp_action_output action;
948 action.type = htons(OFPAT_OUTPUT);
949 action.len = htons(sizeof action);
950 action.port = htons(out_port);
951 return make_packet_out(NULL, buffer_id, in_port,
952 (struct ofp_action_header *) &action, 1);
955 /* Creates and returns an OFPT_ECHO_REQUEST message with an empty payload. */
957 make_echo_request(void)
959 struct ofp_header *rq;
960 struct ofpbuf *out = ofpbuf_new(sizeof *rq);
961 rq = ofpbuf_put_uninit(out, sizeof *rq);
962 rq->version = OFP_VERSION;
963 rq->type = OFPT_ECHO_REQUEST;
964 rq->length = htons(sizeof *rq);
969 /* Creates and returns an OFPT_ECHO_REPLY message matching the
970 * OFPT_ECHO_REQUEST message in 'rq'. */
972 make_echo_reply(const struct ofp_header *rq)
974 size_t size = ntohs(rq->length);
975 struct ofpbuf *out = ofpbuf_new(size);
976 struct ofp_header *reply = ofpbuf_put(out, rq, size);
977 reply->type = OFPT_ECHO_REPLY;
982 check_message_type(uint8_t got_type, uint8_t want_type)
984 if (got_type != want_type) {
985 char *want_type_name = ofp_message_type_to_string(want_type);
986 char *got_type_name = ofp_message_type_to_string(got_type);
987 VLOG_WARN_RL(&bad_ofmsg_rl,
988 "received bad message type %s (expected %s)",
989 got_type_name, want_type_name);
990 free(want_type_name);
992 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
997 /* Checks that 'msg' has type 'type' and that it is exactly 'size' bytes long.
998 * Returns 0 if the checks pass, otherwise an OpenFlow error code (produced
999 * with ofp_mkerr()). */
1001 check_ofp_message(const struct ofp_header *msg, uint8_t type, size_t size)
1006 error = check_message_type(msg->type, type);
1011 got_size = ntohs(msg->length);
1012 if (got_size != size) {
1013 char *type_name = ofp_message_type_to_string(type);
1014 VLOG_WARN_RL(&bad_ofmsg_rl,
1015 "received %s message of length %"PRIu16" (expected %zu)",
1016 type_name, got_size, size);
1018 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
1024 /* Checks that 'msg' has type 'type' and that 'msg' is 'size' plus a
1025 * nonnegative integer multiple of 'array_elt_size' bytes long. Returns 0 if
1026 * the checks pass, otherwise an OpenFlow error code (produced with
1029 * If 'n_array_elts' is nonnull, then '*n_array_elts' is set to the number of
1030 * 'array_elt_size' blocks in 'msg' past the first 'min_size' bytes, when
1033 check_ofp_message_array(const struct ofp_header *msg, uint8_t type,
1034 size_t min_size, size_t array_elt_size,
1035 size_t *n_array_elts)
1040 assert(array_elt_size);
1042 error = check_message_type(msg->type, type);
1047 got_size = ntohs(msg->length);
1048 if (got_size < min_size) {
1049 char *type_name = ofp_message_type_to_string(type);
1050 VLOG_WARN_RL(&bad_ofmsg_rl, "received %s message of length %"PRIu16" "
1051 "(expected at least %zu)",
1052 type_name, got_size, min_size);
1054 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
1056 if ((got_size - min_size) % array_elt_size) {
1057 char *type_name = ofp_message_type_to_string(type);
1058 VLOG_WARN_RL(&bad_ofmsg_rl,
1059 "received %s message of bad length %"PRIu16": the "
1060 "excess over %zu (%zu) is not evenly divisible by %zu "
1061 "(remainder is %zu)",
1062 type_name, got_size, min_size, got_size - min_size,
1063 array_elt_size, (got_size - min_size) % array_elt_size);
1065 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
1068 *n_array_elts = (got_size - min_size) / array_elt_size;
1074 check_ofp_packet_out(const struct ofp_header *oh, struct ofpbuf *data,
1075 int *n_actionsp, int max_ports)
1077 const struct ofp_packet_out *opo;
1078 unsigned int actions_len, n_actions;
1083 error = check_ofp_message_array(oh, OFPT_PACKET_OUT,
1084 sizeof *opo, 1, &extra);
1088 opo = (const struct ofp_packet_out *) oh;
1090 actions_len = ntohs(opo->actions_len);
1091 if (actions_len > extra) {
1092 VLOG_WARN_RL(&bad_ofmsg_rl, "packet-out claims %zu bytes of actions "
1093 "but message has room for only %zu bytes",
1094 actions_len, extra);
1095 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
1097 if (actions_len % sizeof(union ofp_action)) {
1098 VLOG_WARN_RL(&bad_ofmsg_rl, "packet-out claims %zu bytes of actions, "
1099 "which is not a multiple of %zu",
1100 actions_len, sizeof(union ofp_action));
1101 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
1104 n_actions = actions_len / sizeof(union ofp_action);
1105 error = validate_actions((const union ofp_action *) opo->actions,
1106 n_actions, max_ports);
1111 data->data = (void *) &opo->actions[n_actions];
1112 data->size = extra - actions_len;
1113 *n_actionsp = n_actions;
1117 const struct ofp_flow_stats *
1118 flow_stats_first(struct flow_stats_iterator *iter,
1119 const struct ofp_stats_reply *osr)
1121 iter->pos = osr->body;
1122 iter->end = osr->body + (ntohs(osr->header.length)
1123 - offsetof(struct ofp_stats_reply, body));
1124 return flow_stats_next(iter);
1127 const struct ofp_flow_stats *
1128 flow_stats_next(struct flow_stats_iterator *iter)
1130 ptrdiff_t bytes_left = iter->end - iter->pos;
1131 const struct ofp_flow_stats *fs;
1134 if (bytes_left < sizeof *fs) {
1135 if (bytes_left != 0) {
1136 VLOG_WARN_RL(&bad_ofmsg_rl,
1137 "%td leftover bytes in flow stats reply", bytes_left);
1142 fs = (const void *) iter->pos;
1143 length = ntohs(fs->length);
1144 if (length < sizeof *fs) {
1145 VLOG_WARN_RL(&bad_ofmsg_rl, "flow stats length %zu is shorter than "
1146 "min %zu", length, sizeof *fs);
1148 } else if (length > bytes_left) {
1149 VLOG_WARN_RL(&bad_ofmsg_rl, "flow stats length %zu but only %td "
1150 "bytes left", length, bytes_left);
1152 } else if ((length - sizeof *fs) % sizeof fs->actions[0]) {
1153 VLOG_WARN_RL(&bad_ofmsg_rl, "flow stats length %zu has %zu bytes "
1154 "left over in final action", length,
1155 (length - sizeof *fs) % sizeof fs->actions[0]);
1158 iter->pos += length;
1162 /* Alignment of ofp_actions. */
1163 #define ACTION_ALIGNMENT 8
1166 check_action_exact_len(const union ofp_action *a, unsigned int len,
1167 unsigned int required_len)
1169 if (len != required_len) {
1170 VLOG_DBG_RL(&bad_ofmsg_rl,
1171 "action %u has invalid length %"PRIu16" (must be %u)\n",
1172 a->type, ntohs(a->header.len), required_len);
1173 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_LEN);
1179 check_action_port(int port, int max_ports)
1187 case OFPP_CONTROLLER:
1192 if (port >= 0 && port < max_ports) {
1195 VLOG_WARN_RL(&bad_ofmsg_rl, "unknown output port %x", port);
1196 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_OUT_PORT);
1201 check_nicira_action(const union ofp_action *a, unsigned int len)
1203 const struct nx_action_header *nah;
1206 VLOG_DBG_RL(&bad_ofmsg_rl,
1207 "Nicira vendor action only %u bytes", len);
1208 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_LEN);
1210 nah = (const struct nx_action_header *) a;
1212 switch (ntohs(nah->subtype)) {
1213 case NXAST_RESUBMIT:
1214 return check_action_exact_len(a, len, 16);
1216 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_VENDOR_TYPE);
1221 check_action(const union ofp_action *a, unsigned int len, int max_ports)
1227 error = check_action_port(ntohs(a->output.port), max_ports);
1231 return check_action_exact_len(a, len, 8);
1233 case OFPAT_SET_VLAN_VID:
1234 case OFPAT_SET_VLAN_PCP:
1235 case OFPAT_STRIP_VLAN:
1236 case OFPAT_SET_NW_SRC:
1237 case OFPAT_SET_NW_DST:
1238 case OFPAT_SET_TP_SRC:
1239 case OFPAT_SET_TP_DST:
1240 return check_action_exact_len(a, len, 8);
1242 case OFPAT_SET_DL_SRC:
1243 case OFPAT_SET_DL_DST:
1244 return check_action_exact_len(a, len, 16);
1247 if (a->vendor.vendor == htonl(NX_VENDOR_ID)) {
1248 return check_nicira_action(a, len);
1250 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_VENDOR);
1255 VLOG_WARN_RL(&bad_ofmsg_rl, "unknown action type %"PRIu16, a->type);
1256 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_TYPE);
1260 VLOG_DBG_RL(&bad_ofmsg_rl, "action has invalid length 0");
1261 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_LEN);
1263 if (len % ACTION_ALIGNMENT) {
1264 VLOG_DBG_RL(&bad_ofmsg_rl, "action length %u is not a multiple of %d",
1265 len, ACTION_ALIGNMENT);
1266 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_LEN);
1272 validate_actions(const union ofp_action *actions, size_t n_actions,
1275 const union ofp_action *a;
1277 for (a = actions; a < &actions[n_actions]; ) {
1278 unsigned int len = ntohs(a->header.len);
1279 unsigned int n_slots = len / ACTION_ALIGNMENT;
1280 unsigned int slots_left = &actions[n_actions] - a;
1283 if (n_slots > slots_left) {
1284 VLOG_DBG_RL(&bad_ofmsg_rl,
1285 "action requires %u slots but only %td remain",
1286 n_slots, slots_left);
1287 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_LEN);
1289 error = check_action(a, len, max_ports);
1298 /* The set of actions must either come from a trusted source or have been
1299 * previously validated with validate_actions(). */
1300 const union ofp_action *
1301 actions_first(struct actions_iterator *iter,
1302 const union ofp_action *oa, size_t n_actions)
1305 iter->end = oa + n_actions;
1306 return actions_next(iter);
1309 const union ofp_action *
1310 actions_next(struct actions_iterator *iter)
1312 if (iter->pos < iter->end) {
1313 const union ofp_action *a = iter->pos;
1314 unsigned int len = ntohs(a->header.len);
1315 iter->pos += len / ACTION_ALIGNMENT;
1323 normalize_match(struct ofp_match *m)
1325 enum { OFPFW_NW = OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK | OFPFW_NW_PROTO };
1326 enum { OFPFW_TP = OFPFW_TP_SRC | OFPFW_TP_DST };
1329 wc = ntohl(m->wildcards) & OFPFW_ALL;
1330 if (wc & OFPFW_DL_TYPE) {
1333 /* Can't sensibly m on network or transport headers if the
1334 * data link type is unknown. */
1335 wc |= OFPFW_NW | OFPFW_TP;
1336 m->nw_src = m->nw_dst = m->nw_proto = 0;
1337 m->tp_src = m->tp_dst = 0;
1338 } else if (m->dl_type == htons(ETH_TYPE_IP)) {
1339 if (wc & OFPFW_NW_PROTO) {
1342 /* Can't sensibly m on transport headers if the network
1343 * protocol is unknown. */
1345 m->tp_src = m->tp_dst = 0;
1346 } else if (m->nw_proto == IPPROTO_TCP ||
1347 m->nw_proto == IPPROTO_UDP ||
1348 m->nw_proto == IPPROTO_ICMP) {
1349 if (wc & OFPFW_TP_SRC) {
1352 if (wc & OFPFW_TP_DST) {
1356 /* Transport layer fields will always be extracted as zeros, so we
1357 * can do an exact-m on those values. */
1359 m->tp_src = m->tp_dst = 0;
1361 if (wc & OFPFW_NW_SRC_MASK) {
1362 m->nw_src &= flow_nw_bits_to_mask(wc, OFPFW_NW_SRC_SHIFT);
1364 if (wc & OFPFW_NW_DST_MASK) {
1365 m->nw_dst &= flow_nw_bits_to_mask(wc, OFPFW_NW_DST_SHIFT);
1368 /* Network and transport layer fields will always be extracted as
1369 * zeros, so we can do an exact-m on those values. */
1370 wc &= ~(OFPFW_NW | OFPFW_TP);
1371 m->nw_proto = m->nw_src = m->nw_dst = 0;
1372 m->tp_src = m->tp_dst = 0;
1374 if (wc & OFPFW_DL_SRC) {
1375 memset(m->dl_src, 0, sizeof m->dl_src);
1377 if (wc & OFPFW_DL_DST) {
1378 memset(m->dl_dst, 0, sizeof m->dl_dst);
1380 m->wildcards = htonl(wc);
1384 vconn_init(struct vconn *vconn, struct vconn_class *class, int connect_status,
1385 uint32_t ip, const char *name, bool reconnectable)
1387 vconn->class = class;
1388 vconn->state = (connect_status == EAGAIN ? VCS_CONNECTING
1389 : !connect_status ? VCS_SEND_HELLO
1390 : VCS_DISCONNECTED);
1391 vconn->error = connect_status;
1392 vconn->version = -1;
1393 vconn->min_version = -1;
1395 vconn->name = xstrdup(name);
1396 vconn->reconnectable = reconnectable;
1400 pvconn_init(struct pvconn *pvconn, struct pvconn_class *class,
1403 pvconn->class = class;
1404 pvconn->name = xstrdup(name);