2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
25 #include <netinet/in.h>
26 #include <sys/socket.h>
30 #include <sys/ioctl.h>
35 #include "dpif-provider.h"
41 #include "ofp-print.h"
44 #include "poll-loop.h"
50 VLOG_DEFINE_THIS_MODULE(dpif_netdev)
52 /* Configuration parameters. */
53 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
54 enum { MAX_QUEUE_LEN = 100 }; /* Maximum number of packets per queue. */
55 enum { N_GROUPS = 16 }; /* Number of port groups. */
56 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
57 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
59 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
60 * headers to be aligned on a 4-byte boundary. */
61 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
63 /* Datapath based on the network device interface from netdev.h. */
70 bool drop_frags; /* Drop all IP fragments, if true. */
71 struct ovs_queue queues[N_QUEUES]; /* Messages queued for dpif_recv(). */
72 struct hmap flow_table; /* Flow table. */
73 struct odp_port_group groups[N_GROUPS];
76 long long int n_frags; /* Number of dropped IP fragments. */
77 long long int n_hit; /* Number of flow table matches. */
78 long long int n_missed; /* Number of flow table misses. */
79 long long int n_lost; /* Number of misses not passed to client. */
83 struct dp_netdev_port *ports[MAX_PORTS];
84 struct list port_list;
88 /* A port in a netdev-based datapath. */
89 struct dp_netdev_port {
90 int port_no; /* Index into dp_netdev's 'ports'. */
91 struct list node; /* Element in dp_netdev's 'port_list'. */
92 struct netdev *netdev;
93 bool internal; /* Internal port (as ODP_PORT_INTERNAL)? */
96 /* A flow in dp_netdev's 'flow_table'. */
97 struct dp_netdev_flow {
98 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
102 struct timespec used; /* Last used time. */
103 long long int packet_count; /* Number of packets matched. */
104 long long int byte_count; /* Number of bytes matched. */
105 uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */
108 union odp_action *actions;
109 unsigned int n_actions;
112 /* Interface to netdev-based datapath. */
115 struct dp_netdev *dp;
117 unsigned int dp_serial;
120 /* All netdev-based datapaths. */
121 static struct dp_netdev *dp_netdevs[256];
122 struct list dp_netdev_list = LIST_INITIALIZER(&dp_netdev_list);
123 enum { N_DP_NETDEVS = ARRAY_SIZE(dp_netdevs) };
125 /* Maximum port MTU seen so far. */
126 static int max_mtu = ETH_PAYLOAD_MAX;
128 static int get_port_by_number(struct dp_netdev *, uint16_t port_no,
129 struct dp_netdev_port **portp);
130 static int get_port_by_name(struct dp_netdev *, const char *devname,
131 struct dp_netdev_port **portp);
132 static void dp_netdev_free(struct dp_netdev *);
133 static void dp_netdev_flow_flush(struct dp_netdev *);
134 static int do_add_port(struct dp_netdev *, const char *devname, uint16_t flags,
136 static int do_del_port(struct dp_netdev *, uint16_t port_no);
137 static int dp_netdev_output_control(struct dp_netdev *, const struct ofpbuf *,
138 int queue_no, int port_no, uint32_t arg);
139 static int dp_netdev_execute_actions(struct dp_netdev *,
140 struct ofpbuf *, const flow_t *,
141 const union odp_action *, int n);
143 static struct dpif_netdev *
144 dpif_netdev_cast(const struct dpif *dpif)
146 dpif_assert_class(dpif, &dpif_netdev_class);
147 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
150 static struct dp_netdev *
151 get_dp_netdev(const struct dpif *dpif)
153 return dpif_netdev_cast(dpif)->dp;
157 name_to_dp_idx(const char *name)
159 if (!strncmp(name, "dp", 2) && isdigit((unsigned char)name[2])) {
160 int dp_idx = atoi(name + 2);
161 if (dp_idx >= 0 && dp_idx < N_DP_NETDEVS) {
168 static struct dp_netdev *
169 find_dp_netdev(const char *name)
174 dp_idx = name_to_dp_idx(name);
176 return dp_netdevs[dp_idx];
179 for (i = 0; i < N_DP_NETDEVS; i++) {
180 struct dp_netdev *dp = dp_netdevs[i];
182 struct dp_netdev_port *port;
183 if (!get_port_by_name(dp, name, &port)) {
192 create_dpif_netdev(struct dp_netdev *dp)
194 struct dpif_netdev *dpif;
199 dpname = xasprintf("dp%d", dp->dp_idx);
200 dpif = xmalloc(sizeof *dpif);
201 dpif_init(&dpif->dpif, &dpif_netdev_class, dpname, dp->dp_idx, dp->dp_idx);
203 dpif->listen_mask = 0;
204 dpif->dp_serial = dp->serial;
211 create_dp_netdev(const char *name, int dp_idx, struct dpif **dpifp)
213 struct dp_netdev *dp;
217 if (dp_netdevs[dp_idx]) {
221 /* Create datapath. */
222 dp_netdevs[dp_idx] = dp = xzalloc(sizeof *dp);
223 list_push_back(&dp_netdev_list, &dp->node);
226 dp->drop_frags = false;
227 for (i = 0; i < N_QUEUES; i++) {
228 queue_init(&dp->queues[i]);
230 hmap_init(&dp->flow_table);
231 for (i = 0; i < N_GROUPS; i++) {
232 dp->groups[i].ports = NULL;
233 dp->groups[i].n_ports = 0;
234 dp->groups[i].group = i;
236 list_init(&dp->port_list);
237 error = do_add_port(dp, name, ODP_PORT_INTERNAL, ODPP_LOCAL);
243 *dpifp = create_dpif_netdev(dp);
248 dpif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create,
252 if (find_dp_netdev(name)) {
255 int dp_idx = name_to_dp_idx(name);
257 return create_dp_netdev(name, dp_idx, dpifp);
259 /* Scan for unused dp_idx number. */
260 for (dp_idx = 0; dp_idx < N_DP_NETDEVS; dp_idx++) {
261 int error = create_dp_netdev(name, dp_idx, dpifp);
262 if (error != EBUSY) {
267 /* All datapath numbers in use. */
272 struct dp_netdev *dp = find_dp_netdev(name);
274 *dpifp = create_dpif_netdev(dp);
283 dp_netdev_free(struct dp_netdev *dp)
287 dp_netdev_flow_flush(dp);
288 while (dp->n_ports > 0) {
289 struct dp_netdev_port *port = CONTAINER_OF(
290 dp->port_list.next, struct dp_netdev_port, node);
291 do_del_port(dp, port->port_no);
293 for (i = 0; i < N_QUEUES; i++) {
294 queue_destroy(&dp->queues[i]);
296 hmap_destroy(&dp->flow_table);
297 for (i = 0; i < N_GROUPS; i++) {
298 free(dp->groups[i].ports);
300 dp_netdevs[dp->dp_idx] = NULL;
301 list_remove(&dp->node);
306 dpif_netdev_close(struct dpif *dpif)
308 struct dp_netdev *dp = get_dp_netdev(dpif);
309 assert(dp->open_cnt > 0);
310 if (--dp->open_cnt == 0 && dp->destroyed) {
317 dpif_netdev_destroy(struct dpif *dpif)
319 struct dp_netdev *dp = get_dp_netdev(dpif);
320 dp->destroyed = true;
325 dpif_netdev_get_stats(const struct dpif *dpif, struct odp_stats *stats)
327 struct dp_netdev *dp = get_dp_netdev(dpif);
328 memset(stats, 0, sizeof *stats);
329 stats->n_flows = hmap_count(&dp->flow_table);
330 stats->cur_capacity = hmap_capacity(&dp->flow_table);
331 stats->max_capacity = MAX_FLOWS;
332 stats->n_ports = dp->n_ports;
333 stats->max_ports = MAX_PORTS;
334 stats->max_groups = N_GROUPS;
335 stats->n_frags = dp->n_frags;
336 stats->n_hit = dp->n_hit;
337 stats->n_missed = dp->n_missed;
338 stats->n_lost = dp->n_lost;
339 stats->max_miss_queue = MAX_QUEUE_LEN;
340 stats->max_action_queue = MAX_QUEUE_LEN;
345 dpif_netdev_get_drop_frags(const struct dpif *dpif, bool *drop_fragsp)
347 struct dp_netdev *dp = get_dp_netdev(dpif);
348 *drop_fragsp = dp->drop_frags;
353 dpif_netdev_set_drop_frags(struct dpif *dpif, bool drop_frags)
355 struct dp_netdev *dp = get_dp_netdev(dpif);
356 dp->drop_frags = drop_frags;
361 do_add_port(struct dp_netdev *dp, const char *devname, uint16_t flags,
364 bool internal = (flags & ODP_PORT_INTERNAL) != 0;
365 struct dp_netdev_port *port;
366 struct netdev_options netdev_options;
367 struct netdev *netdev;
371 /* XXX reject devices already in some dp_netdev. */
373 /* Open and validate network device. */
374 memset(&netdev_options, 0, sizeof netdev_options);
375 netdev_options.name = devname;
376 netdev_options.ethertype = NETDEV_ETH_TYPE_ANY;
378 netdev_options.type = "tap";
381 error = netdev_open(&netdev_options, &netdev);
385 /* XXX reject loopback devices */
386 /* XXX reject non-Ethernet devices */
388 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
390 netdev_close(netdev);
394 port = xmalloc(sizeof *port);
395 port->port_no = port_no;
396 port->netdev = netdev;
397 port->internal = internal;
399 netdev_get_mtu(netdev, &mtu);
404 list_push_back(&dp->port_list, &port->node);
405 dp->ports[port_no] = port;
413 dpif_netdev_port_add(struct dpif *dpif, const char *devname, uint16_t flags,
416 struct dp_netdev *dp = get_dp_netdev(dpif);
419 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
420 if (!dp->ports[port_no]) {
422 return do_add_port(dp, devname, flags, port_no);
429 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
431 struct dp_netdev *dp = get_dp_netdev(dpif);
432 return port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
436 is_valid_port_number(uint16_t port_no)
438 return port_no < MAX_PORTS;
442 get_port_by_number(struct dp_netdev *dp,
443 uint16_t port_no, struct dp_netdev_port **portp)
445 if (!is_valid_port_number(port_no)) {
449 *portp = dp->ports[port_no];
450 return *portp ? 0 : ENOENT;
455 get_port_by_name(struct dp_netdev *dp,
456 const char *devname, struct dp_netdev_port **portp)
458 struct dp_netdev_port *port;
460 LIST_FOR_EACH (port, node, &dp->port_list) {
461 if (!strcmp(netdev_get_name(port->netdev), devname)) {
470 do_del_port(struct dp_netdev *dp, uint16_t port_no)
472 struct dp_netdev_port *port;
476 error = get_port_by_number(dp, port_no, &port);
481 list_remove(&port->node);
482 dp->ports[port->port_no] = NULL;
486 name = xstrdup(netdev_get_name(port->netdev));
487 netdev_close(port->netdev);
496 answer_port_query(const struct dp_netdev_port *port, struct odp_port *odp_port)
498 memset(odp_port, 0, sizeof *odp_port);
499 ovs_strlcpy(odp_port->devname, netdev_get_name(port->netdev),
500 sizeof odp_port->devname);
501 odp_port->port = port->port_no;
502 odp_port->flags = port->internal ? ODP_PORT_INTERNAL : 0;
506 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
507 struct odp_port *odp_port)
509 struct dp_netdev *dp = get_dp_netdev(dpif);
510 struct dp_netdev_port *port;
513 error = get_port_by_number(dp, port_no, &port);
515 answer_port_query(port, odp_port);
521 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
522 struct odp_port *odp_port)
524 struct dp_netdev *dp = get_dp_netdev(dpif);
525 struct dp_netdev_port *port;
528 error = get_port_by_name(dp, devname, &port);
530 answer_port_query(port, odp_port);
536 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
538 hmap_remove(&dp->flow_table, &flow->node);
544 dp_netdev_flow_flush(struct dp_netdev *dp)
546 struct dp_netdev_flow *flow, *next;
548 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
549 dp_netdev_free_flow(dp, flow);
554 dpif_netdev_flow_flush(struct dpif *dpif)
556 struct dp_netdev *dp = get_dp_netdev(dpif);
557 dp_netdev_flow_flush(dp);
562 dpif_netdev_port_list(const struct dpif *dpif, struct odp_port *ports, int n)
564 struct dp_netdev *dp = get_dp_netdev(dpif);
565 struct dp_netdev_port *port;
569 LIST_FOR_EACH (port, node, &dp->port_list) {
570 struct odp_port *odp_port = &ports[i];
574 answer_port_query(port, odp_port);
581 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
583 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
584 if (dpif->dp_serial != dpif->dp->serial) {
585 dpif->dp_serial = dpif->dp->serial;
593 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
595 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
596 if (dpif->dp_serial != dpif->dp->serial) {
597 poll_immediate_wake();
602 get_port_group(const struct dpif *dpif, int group_no,
603 struct odp_port_group **groupp)
605 struct dp_netdev *dp = get_dp_netdev(dpif);
607 if (group_no >= 0 && group_no < N_GROUPS) {
608 *groupp = &dp->groups[group_no];
617 dpif_netdev_port_group_get(const struct dpif *dpif, int group_no,
618 uint16_t ports[], int n)
620 struct odp_port_group *group;
627 error = get_port_group(dpif, group_no, &group);
629 memcpy(ports, group->ports, MIN(n, group->n_ports) * sizeof *ports);
630 return group->n_ports;
637 dpif_netdev_port_group_set(struct dpif *dpif, int group_no,
638 const uint16_t ports[], int n)
640 struct odp_port_group *group;
643 if (n < 0 || n > MAX_PORTS) {
647 error = get_port_group(dpif, group_no, &group);
650 group->ports = xmemdup(ports, n * sizeof *group->ports);
652 group->group = group_no;
657 static struct dp_netdev_flow *
658 dp_netdev_lookup_flow(const struct dp_netdev *dp, const flow_t *key)
660 struct dp_netdev_flow *flow;
662 assert(!key->reserved[0] && !key->reserved[1] && !key->reserved[2]);
663 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
664 if (flow_equal(&flow->key, key)) {
672 answer_flow_query(struct dp_netdev_flow *flow, uint32_t query_flags,
673 struct odp_flow *odp_flow)
676 odp_flow->key = flow->key;
677 odp_flow->stats.n_packets = flow->packet_count;
678 odp_flow->stats.n_bytes = flow->byte_count;
679 odp_flow->stats.used_sec = flow->used.tv_sec;
680 odp_flow->stats.used_nsec = flow->used.tv_nsec;
681 odp_flow->stats.tcp_flags = TCP_FLAGS(flow->tcp_ctl);
682 odp_flow->stats.reserved = 0;
683 odp_flow->stats.error = 0;
684 if (odp_flow->n_actions > 0) {
685 unsigned int n = MIN(odp_flow->n_actions, flow->n_actions);
686 memcpy(odp_flow->actions, flow->actions,
687 n * sizeof *odp_flow->actions);
688 odp_flow->n_actions = flow->n_actions;
691 if (query_flags & ODPFF_ZERO_TCP_FLAGS) {
696 odp_flow->stats.error = ENOENT;
701 dpif_netdev_flow_get(const struct dpif *dpif, struct odp_flow flows[], int n)
703 struct dp_netdev *dp = get_dp_netdev(dpif);
706 for (i = 0; i < n; i++) {
707 struct odp_flow *odp_flow = &flows[i];
708 answer_flow_query(dp_netdev_lookup_flow(dp, &odp_flow->key),
709 odp_flow->flags, odp_flow);
715 dpif_netdev_validate_actions(const union odp_action *actions, int n_actions,
721 for (i = 0; i < n_actions; i++) {
722 const union odp_action *a = &actions[i];
725 if (a->output.port >= MAX_PORTS) {
730 case ODPAT_OUTPUT_GROUP:
732 if (a->output_group.group >= N_GROUPS) {
737 case ODPAT_CONTROLLER:
740 case ODPAT_SET_VLAN_VID:
742 if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) {
747 case ODPAT_SET_VLAN_PCP:
749 if (a->vlan_pcp.vlan_pcp & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) {
754 case ODPAT_SET_NW_TOS:
756 if (a->nw_tos.nw_tos & IP_ECN_MASK) {
761 case ODPAT_STRIP_VLAN:
762 case ODPAT_SET_DL_SRC:
763 case ODPAT_SET_DL_DST:
764 case ODPAT_SET_NW_SRC:
765 case ODPAT_SET_NW_DST:
766 case ODPAT_SET_TP_SRC:
767 case ODPAT_SET_TP_DST:
779 set_flow_actions(struct dp_netdev_flow *flow, struct odp_flow *odp_flow)
785 if (odp_flow->n_actions >= 4096 / sizeof *odp_flow->actions) {
788 error = dpif_netdev_validate_actions(odp_flow->actions,
789 odp_flow->n_actions, &mutates);
794 n_bytes = odp_flow->n_actions * sizeof *flow->actions;
795 flow->actions = xrealloc(flow->actions, n_bytes);
796 flow->n_actions = odp_flow->n_actions;
797 memcpy(flow->actions, odp_flow->actions, n_bytes);
802 add_flow(struct dpif *dpif, struct odp_flow *odp_flow)
804 struct dp_netdev *dp = get_dp_netdev(dpif);
805 struct dp_netdev_flow *flow;
808 flow = xzalloc(sizeof *flow);
809 flow->key = odp_flow->key;
810 memset(flow->key.reserved, 0, sizeof flow->key.reserved);
812 error = set_flow_actions(flow, odp_flow);
818 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
823 clear_stats(struct dp_netdev_flow *flow)
825 flow->used.tv_sec = 0;
826 flow->used.tv_nsec = 0;
827 flow->packet_count = 0;
828 flow->byte_count = 0;
833 dpif_netdev_flow_put(struct dpif *dpif, struct odp_flow_put *put)
835 struct dp_netdev *dp = get_dp_netdev(dpif);
836 struct dp_netdev_flow *flow;
838 flow = dp_netdev_lookup_flow(dp, &put->flow.key);
840 if (put->flags & ODPPF_CREATE) {
841 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
842 return add_flow(dpif, &put->flow);
850 if (put->flags & ODPPF_MODIFY) {
851 int error = set_flow_actions(flow, &put->flow);
852 if (!error && put->flags & ODPPF_ZERO_STATS) {
864 dpif_netdev_flow_del(struct dpif *dpif, struct odp_flow *odp_flow)
866 struct dp_netdev *dp = get_dp_netdev(dpif);
867 struct dp_netdev_flow *flow;
869 flow = dp_netdev_lookup_flow(dp, &odp_flow->key);
871 answer_flow_query(flow, 0, odp_flow);
872 dp_netdev_free_flow(dp, flow);
880 dpif_netdev_flow_list(const struct dpif *dpif, struct odp_flow flows[], int n)
882 struct dp_netdev *dp = get_dp_netdev(dpif);
883 struct dp_netdev_flow *flow;
887 HMAP_FOR_EACH (flow, node, &dp->flow_table) {
891 answer_flow_query(flow, 0, &flows[i++]);
893 return hmap_count(&dp->flow_table);
897 dpif_netdev_execute(struct dpif *dpif, uint16_t in_port,
898 const union odp_action actions[], int n_actions,
899 const struct ofpbuf *packet)
901 struct dp_netdev *dp = get_dp_netdev(dpif);
907 if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
911 error = dpif_netdev_validate_actions(actions, n_actions, &mutates);
917 /* We need a deep copy of 'packet' since we're going to modify its
919 ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size);
920 copy.data = (char*)copy.base + DP_NETDEV_HEADROOM;
921 ofpbuf_put(©, packet->data, packet->size);
923 /* We still need a shallow copy of 'packet', even though we won't
924 * modify its data, because flow_extract() modifies packet->l2, etc.
925 * We could probably get away with modifying those but it's more polite
929 flow_extract(©, 0, in_port, &flow);
930 error = dp_netdev_execute_actions(dp, ©, &flow, actions, n_actions);
932 ofpbuf_uninit(©);
938 dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask)
940 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
941 *listen_mask = dpif_netdev->listen_mask;
946 dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
948 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
949 if (!(listen_mask & ~ODPL_ALL)) {
950 dpif_netdev->listen_mask = listen_mask;
957 static struct ovs_queue *
958 find_nonempty_queue(struct dpif *dpif)
960 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
961 struct dp_netdev *dp = get_dp_netdev(dpif);
962 int mask = dpif_netdev->listen_mask;
965 for (i = 0; i < N_QUEUES; i++) {
966 struct ovs_queue *q = &dp->queues[i];
967 if (q->n && mask & (1u << i)) {
975 dpif_netdev_recv(struct dpif *dpif, struct ofpbuf **bufp)
977 struct ovs_queue *q = find_nonempty_queue(dpif);
979 *bufp = queue_pop_head(q);
987 dpif_netdev_recv_wait(struct dpif *dpif)
989 struct ovs_queue *q = find_nonempty_queue(dpif);
991 poll_immediate_wake();
993 /* No messages ready to be received, and dp_wait() will ensure that we
994 * wake up to queue new messages, so there is nothing to do. */
999 dp_netdev_flow_used(struct dp_netdev_flow *flow, const flow_t *key,
1000 const struct ofpbuf *packet)
1002 time_timespec(&flow->used);
1003 flow->packet_count++;
1004 flow->byte_count += packet->size;
1005 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
1006 struct tcp_header *th = packet->l4;
1007 flow->tcp_ctl |= th->tcp_ctl;
1012 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1013 struct ofpbuf *packet)
1015 struct dp_netdev_flow *flow;
1018 if (packet->size < ETH_HEADER_LEN) {
1021 if (flow_extract(packet, 0, port->port_no, &key) && dp->drop_frags) {
1026 flow = dp_netdev_lookup_flow(dp, &key);
1028 dp_netdev_flow_used(flow, &key, packet);
1029 dp_netdev_execute_actions(dp, packet, &key,
1030 flow->actions, flow->n_actions);
1034 dp_netdev_output_control(dp, packet, _ODPL_MISS_NR, port->port_no, 0);
1041 struct ofpbuf packet;
1042 struct dp_netdev *dp;
1044 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + max_mtu);
1045 LIST_FOR_EACH (dp, node, &dp_netdev_list) {
1046 struct dp_netdev_port *port;
1048 LIST_FOR_EACH (port, node, &dp->port_list) {
1051 /* Reset packet contents. */
1052 packet.data = (char*)packet.base + DP_NETDEV_HEADROOM;
1055 error = netdev_recv(port->netdev, &packet);
1057 dp_netdev_port_input(dp, port, &packet);
1058 } else if (error != EAGAIN) {
1059 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1060 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1061 netdev_get_name(port->netdev), strerror(error));
1065 ofpbuf_uninit(&packet);
1069 dp_netdev_wait(void)
1071 struct dp_netdev *dp;
1073 LIST_FOR_EACH (dp, node, &dp_netdev_list) {
1074 struct dp_netdev_port *port;
1075 LIST_FOR_EACH (port, node, &dp->port_list) {
1076 netdev_recv_wait(port->netdev);
1082 /* Modify the TCI field of 'packet'. If a VLAN tag is not present, one
1083 * is added with the TCI field set to 'tci'. If a VLAN tag is present,
1084 * then 'mask' bits are cleared before 'tci' is logically OR'd into the
1087 * Note that the function does not ensure that 'tci' does not affect
1088 * bits outside of 'mask'.
1091 dp_netdev_modify_vlan_tci(struct ofpbuf *packet, uint16_t tci, uint16_t mask)
1093 struct vlan_eth_header *veh;
1094 struct eth_header *eh;
1097 if (packet->size >= sizeof(struct vlan_eth_header)
1098 && eh->eth_type == htons(ETH_TYPE_VLAN)) {
1099 /* Clear 'mask' bits, but maintain other TCI bits. */
1101 veh->veth_tci &= ~htons(mask);
1102 veh->veth_tci |= htons(tci);
1104 /* Insert new 802.1Q header. */
1105 struct vlan_eth_header tmp;
1106 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
1107 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
1108 tmp.veth_type = htons(ETH_TYPE_VLAN);
1109 tmp.veth_tci = htons(tci);
1110 tmp.veth_next_type = eh->eth_type;
1112 veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN);
1113 memcpy(veh, &tmp, sizeof tmp);
1114 packet->l2 = (char*)packet->l2 - VLAN_HEADER_LEN;
1119 dp_netdev_strip_vlan(struct ofpbuf *packet)
1121 struct vlan_eth_header *veh = packet->l2;
1122 if (packet->size >= sizeof *veh
1123 && veh->veth_type == htons(ETH_TYPE_VLAN)) {
1124 struct eth_header tmp;
1126 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1127 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1128 tmp.eth_type = veh->veth_next_type;
1130 packet->size -= VLAN_HEADER_LEN;
1131 packet->data = (char*)packet->data + VLAN_HEADER_LEN;
1132 packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
1133 memcpy(packet->data, &tmp, sizeof tmp);
1138 dp_netdev_set_dl_src(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1140 struct eth_header *eh = packet->l2;
1141 memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src);
1145 dp_netdev_set_dl_dst(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1147 struct eth_header *eh = packet->l2;
1148 memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst);
1152 is_ip(const struct ofpbuf *packet, const flow_t *key)
1154 return key->dl_type == htons(ETH_TYPE_IP) && packet->l4;
1158 dp_netdev_set_nw_addr(struct ofpbuf *packet, const flow_t *key,
1159 const struct odp_action_nw_addr *a)
1161 if (is_ip(packet, key)) {
1162 struct ip_header *nh = packet->l3;
1165 field = a->type == ODPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
1166 if (key->nw_proto == IP_TYPE_TCP && packet->l7) {
1167 struct tcp_header *th = packet->l4;
1168 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, a->nw_addr);
1169 } else if (key->nw_proto == IP_TYPE_UDP && packet->l7) {
1170 struct udp_header *uh = packet->l4;
1172 uh->udp_csum = recalc_csum32(uh->udp_csum, *field, a->nw_addr);
1173 if (!uh->udp_csum) {
1174 uh->udp_csum = 0xffff;
1178 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, a->nw_addr);
1179 *field = a->nw_addr;
1184 dp_netdev_set_nw_tos(struct ofpbuf *packet, const flow_t *key,
1185 const struct odp_action_nw_tos *a)
1187 if (is_ip(packet, key)) {
1188 struct ip_header *nh = packet->l3;
1189 uint8_t *field = &nh->ip_tos;
1191 /* Set the DSCP bits and preserve the ECN bits. */
1192 uint8_t new = a->nw_tos | (nh->ip_tos & IP_ECN_MASK);
1194 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
1195 htons((uint16_t)a->nw_tos));
1201 dp_netdev_set_tp_port(struct ofpbuf *packet, const flow_t *key,
1202 const struct odp_action_tp_port *a)
1204 if (is_ip(packet, key)) {
1206 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1207 struct tcp_header *th = packet->l4;
1208 field = a->type == ODPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
1209 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, a->tp_port);
1210 *field = a->tp_port;
1211 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1212 struct udp_header *uh = packet->l4;
1213 field = a->type == ODPAT_SET_TP_SRC ? &uh->udp_src : &uh->udp_dst;
1214 uh->udp_csum = recalc_csum16(uh->udp_csum, *field, a->tp_port);
1215 *field = a->tp_port;
1223 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1226 struct dp_netdev_port *p = dp->ports[out_port];
1228 netdev_send(p->netdev, packet);
1233 dp_netdev_output_group(struct dp_netdev *dp, uint16_t group, uint16_t in_port,
1234 struct ofpbuf *packet)
1236 struct odp_port_group *g = &dp->groups[group];
1239 for (i = 0; i < g->n_ports; i++) {
1240 uint16_t out_port = g->ports[i];
1241 if (out_port != in_port) {
1242 dp_netdev_output_port(dp, packet, out_port);
1248 dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet,
1249 int queue_no, int port_no, uint32_t arg)
1251 struct ovs_queue *q = &dp->queues[queue_no];
1252 struct odp_msg *header;
1256 if (q->n >= MAX_QUEUE_LEN) {
1261 msg_size = sizeof *header + packet->size;
1262 msg = ofpbuf_new_with_headroom(msg_size, DPIF_RECV_MSG_PADDING);
1263 header = ofpbuf_put_uninit(msg, sizeof *header);
1264 header->type = queue_no;
1265 header->length = msg_size;
1266 header->port = port_no;
1268 ofpbuf_put(msg, packet->data, packet->size);
1269 queue_push_tail(q, msg);
1274 /* Returns true if 'packet' is an invalid Ethernet+IPv4 ARP packet: one with
1275 * screwy or truncated header fields or one whose inner and outer Ethernet
1276 * address differ. */
1278 dp_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct odp_flow_key *key)
1280 struct arp_eth_header *arp;
1281 struct eth_header *eth;
1284 if (key->dl_type != htons(ETH_TYPE_ARP)) {
1288 l3_size = (char *) ofpbuf_end(packet) - (char *) packet->l3;
1289 if (l3_size < sizeof(struct arp_eth_header)) {
1295 return (arp->ar_hrd != htons(ARP_HRD_ETHERNET)
1296 || arp->ar_pro != htons(ARP_PRO_IP)
1297 || arp->ar_hln != ETH_HEADER_LEN
1299 || !eth_addr_equals(arp->ar_sha, eth->eth_src));
1303 dp_netdev_execute_actions(struct dp_netdev *dp,
1304 struct ofpbuf *packet, const flow_t *key,
1305 const union odp_action *actions, int n_actions)
1308 for (i = 0; i < n_actions; i++) {
1309 const union odp_action *a = &actions[i];
1313 dp_netdev_output_port(dp, packet, a->output.port);
1316 case ODPAT_OUTPUT_GROUP:
1317 dp_netdev_output_group(dp, a->output_group.group, key->in_port,
1321 case ODPAT_CONTROLLER:
1322 dp_netdev_output_control(dp, packet, _ODPL_ACTION_NR,
1323 key->in_port, a->controller.arg);
1326 case ODPAT_SET_VLAN_VID:
1327 dp_netdev_modify_vlan_tci(packet, ntohs(a->vlan_vid.vlan_vid),
1331 case ODPAT_SET_VLAN_PCP:
1332 dp_netdev_modify_vlan_tci(packet,
1333 a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT,
1337 case ODPAT_STRIP_VLAN:
1338 dp_netdev_strip_vlan(packet);
1341 case ODPAT_SET_DL_SRC:
1342 dp_netdev_set_dl_src(packet, a->dl_addr.dl_addr);
1345 case ODPAT_SET_DL_DST:
1346 dp_netdev_set_dl_dst(packet, a->dl_addr.dl_addr);
1349 case ODPAT_SET_NW_SRC:
1350 case ODPAT_SET_NW_DST:
1351 dp_netdev_set_nw_addr(packet, key, &a->nw_addr);
1354 case ODPAT_SET_NW_TOS:
1355 dp_netdev_set_nw_tos(packet, key, &a->nw_tos);
1358 case ODPAT_SET_TP_SRC:
1359 case ODPAT_SET_TP_DST:
1360 dp_netdev_set_tp_port(packet, key, &a->tp_port);
1363 case ODPAT_DROP_SPOOFED_ARP:
1364 if (dp_netdev_is_spoofed_arp(packet, key)) {
1372 const struct dpif_class dpif_netdev_class = {
1376 NULL, /* enumerate */
1379 NULL, /* get_all_names */
1380 dpif_netdev_destroy,
1381 dpif_netdev_get_stats,
1382 dpif_netdev_get_drop_frags,
1383 dpif_netdev_set_drop_frags,
1384 dpif_netdev_port_add,
1385 dpif_netdev_port_del,
1386 dpif_netdev_port_query_by_number,
1387 dpif_netdev_port_query_by_name,
1388 dpif_netdev_port_list,
1389 dpif_netdev_port_poll,
1390 dpif_netdev_port_poll_wait,
1391 dpif_netdev_port_group_get,
1392 dpif_netdev_port_group_set,
1393 dpif_netdev_flow_get,
1394 dpif_netdev_flow_put,
1395 dpif_netdev_flow_del,
1396 dpif_netdev_flow_flush,
1397 dpif_netdev_flow_list,
1398 dpif_netdev_execute,
1399 dpif_netdev_recv_get_mask,
1400 dpif_netdev_recv_set_mask,
1401 NULL, /* get_sflow_probability */
1402 NULL, /* set_sflow_probability */
1403 NULL, /* queue_to_priority */
1405 dpif_netdev_recv_wait,