2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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.
24 #include <netinet/in.h>
25 #include <sys/socket.h>
30 #include <sys/ioctl.h>
36 #include "dpif-provider.h"
38 #include "dynamic-string.h"
43 #include "netdev-vport.h"
45 #include "odp-execute.h"
47 #include "ofp-print.h"
50 #include "poll-loop.h"
59 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
61 /* Configuration parameters. */
62 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
63 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
65 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
66 * headers to be aligned on a 4-byte boundary. */
67 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
70 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
71 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
72 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
73 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
75 struct dp_netdev_upcall {
76 struct dpif_upcall upcall; /* Queued upcall information. */
77 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
80 struct dp_netdev_queue {
81 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
82 unsigned int head, tail;
85 /* Datapath based on the network device interface from netdev.h. */
87 const struct dpif_class *class;
91 int max_mtu; /* Maximum MTU of any port added so far. */
93 struct dp_netdev_queue queues[N_QUEUES];
94 struct hmap flow_table; /* Flow table. */
97 long long int n_hit; /* Number of flow table matches. */
98 long long int n_missed; /* Number of flow table misses. */
99 long long int n_lost; /* Number of misses not passed to client. */
102 struct dp_netdev_port *ports[MAX_PORTS];
103 struct list port_list;
107 /* A port in a netdev-based datapath. */
108 struct dp_netdev_port {
109 odp_port_t port_no; /* Index into dp_netdev's 'ports'. */
110 struct list node; /* Element in dp_netdev's 'port_list'. */
111 struct netdev *netdev;
112 struct netdev_saved_flags *sf;
113 struct netdev_rx *rx;
114 char *type; /* Port type as requested by user. */
117 /* A flow in dp_netdev's 'flow_table'. */
118 struct dp_netdev_flow {
119 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
123 long long int used; /* Last used time, in monotonic msecs. */
124 long long int packet_count; /* Number of packets matched. */
125 long long int byte_count; /* Number of bytes matched. */
126 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
129 struct nlattr *actions;
133 /* Interface to netdev-based datapath. */
136 struct dp_netdev *dp;
137 unsigned int dp_serial;
140 /* All netdev-based datapaths. */
141 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
143 /* Global lock for all data. */
144 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
146 static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
147 struct dp_netdev_port **portp);
148 static int get_port_by_name(struct dp_netdev *, const char *devname,
149 struct dp_netdev_port **portp);
150 static void dp_netdev_free(struct dp_netdev *);
151 static void dp_netdev_flow_flush(struct dp_netdev *);
152 static int do_add_port(struct dp_netdev *, const char *devname,
153 const char *type, odp_port_t port_no);
154 static int do_del_port(struct dp_netdev *, odp_port_t port_no);
155 static int dpif_netdev_open(const struct dpif_class *, const char *name,
156 bool create, struct dpif **);
157 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
158 int queue_no, const struct flow *,
159 const struct nlattr *userdata);
160 static void dp_netdev_execute_actions(struct dp_netdev *,
161 struct ofpbuf *, struct flow *,
162 const struct nlattr *actions,
164 static void dp_netdev_port_input(struct dp_netdev *dp,
165 struct dp_netdev_port *port,
166 struct ofpbuf *packet, uint32_t skb_priority,
167 uint32_t skb_mark, const struct flow_tnl *tnl);
169 static struct dpif_netdev *
170 dpif_netdev_cast(const struct dpif *dpif)
172 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
173 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
176 static struct dp_netdev *
177 get_dp_netdev(const struct dpif *dpif)
179 return dpif_netdev_cast(dpif)->dp;
183 dpif_netdev_enumerate(struct sset *all_dps)
185 struct shash_node *node;
187 ovs_mutex_lock(&dp_netdev_mutex);
188 SHASH_FOR_EACH(node, &dp_netdevs) {
189 sset_add(all_dps, node->name);
191 ovs_mutex_unlock(&dp_netdev_mutex);
197 dpif_netdev_class_is_dummy(const struct dpif_class *class)
199 return class != &dpif_netdev_class;
203 dpif_netdev_class_is_planetlab(const struct dpif_class *class)
205 return class == &dpif_planetlab_class;
209 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
211 return strcmp(type, "internal") ? type
212 : dpif_netdev_class_is_planetlab(class) ? "pltap"
213 : dpif_netdev_class_is_dummy(class) ? "dummy"
218 create_dpif_netdev(struct dp_netdev *dp)
220 uint16_t netflow_id = hash_string(dp->name, 0);
221 struct dpif_netdev *dpif;
225 dpif = xmalloc(sizeof *dpif);
226 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
228 dpif->dp_serial = dp->serial;
233 /* Choose an unused, non-zero port number and return it on success.
234 * Return ODPP_NONE on failure. */
236 choose_port(struct dp_netdev *dp, const char *name)
240 if (dp->class != &dpif_netdev_class &&
241 dp->class != &dpif_planetlab_class) {
245 /* If the port name begins with "br", start the number search at
246 * 100 to make writing tests easier. */
247 if (!strncmp(name, "br", 2)) {
251 /* If the port name contains a number, try to assign that port number.
252 * This can make writing unit tests easier because port numbers are
254 for (p = name; *p != '\0'; p++) {
255 if (isdigit((unsigned char) *p)) {
256 port_no = start_no + strtol(p, NULL, 10);
257 if (port_no > 0 && port_no < MAX_PORTS
258 && !dp->ports[port_no]) {
259 return u32_to_odp(port_no);
266 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
267 if (!dp->ports[port_no]) {
268 return u32_to_odp(port_no);
276 create_dp_netdev(const char *name, const struct dpif_class *class,
277 struct dp_netdev **dpp)
279 struct dp_netdev *dp;
283 dp = xzalloc(sizeof *dp);
285 dp->name = xstrdup(name);
287 dp->max_mtu = ETH_PAYLOAD_MAX;
288 for (i = 0; i < N_QUEUES; i++) {
289 dp->queues[i].head = dp->queues[i].tail = 0;
291 hmap_init(&dp->flow_table);
292 list_init(&dp->port_list);
294 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
300 shash_add(&dp_netdevs, name, dp);
307 dpif_netdev_open(const struct dpif_class *class, const char *name,
308 bool create, struct dpif **dpifp)
310 struct dp_netdev *dp;
313 ovs_mutex_lock(&dp_netdev_mutex);
314 dp = shash_find_data(&dp_netdevs, name);
316 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
318 error = (dp->class != class ? EINVAL
323 *dpifp = create_dpif_netdev(dp);
325 ovs_mutex_unlock(&dp_netdev_mutex);
331 dp_netdev_purge_queues(struct dp_netdev *dp)
335 for (i = 0; i < N_QUEUES; i++) {
336 struct dp_netdev_queue *q = &dp->queues[i];
338 while (q->tail != q->head) {
339 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
340 ofpbuf_uninit(&u->buf);
346 dp_netdev_free(struct dp_netdev *dp)
348 struct dp_netdev_port *port, *next;
350 dp_netdev_flow_flush(dp);
351 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
352 do_del_port(dp, port->port_no);
354 dp_netdev_purge_queues(dp);
355 hmap_destroy(&dp->flow_table);
361 dpif_netdev_close(struct dpif *dpif)
363 struct dp_netdev *dp = get_dp_netdev(dpif);
365 ovs_mutex_lock(&dp_netdev_mutex);
367 ovs_assert(dp->open_cnt > 0);
368 if (--dp->open_cnt == 0 && dp->destroyed) {
369 shash_find_and_delete(&dp_netdevs, dp->name);
374 ovs_mutex_unlock(&dp_netdev_mutex);
378 dpif_netdev_destroy(struct dpif *dpif)
380 struct dp_netdev *dp = get_dp_netdev(dpif);
382 ovs_mutex_lock(&dp_netdev_mutex);
383 dp->destroyed = true;
384 ovs_mutex_unlock(&dp_netdev_mutex);
390 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
392 struct dp_netdev *dp = get_dp_netdev(dpif);
394 ovs_mutex_lock(&dp_netdev_mutex);
395 stats->n_flows = hmap_count(&dp->flow_table);
396 stats->n_hit = dp->n_hit;
397 stats->n_missed = dp->n_missed;
398 stats->n_lost = dp->n_lost;
399 ovs_mutex_unlock(&dp_netdev_mutex);
405 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
408 struct netdev_saved_flags *sf;
409 struct dp_netdev_port *port;
410 struct netdev *netdev;
411 struct netdev_rx *rx;
412 const char *open_type;
416 /* XXX reject devices already in some dp_netdev. */
418 /* Open and validate network device. */
419 open_type = dpif_netdev_port_open_type(dp->class, type);
420 error = netdev_open(devname, open_type, &netdev);
424 /* XXX reject loopback devices */
425 /* XXX reject non-Ethernet devices */
427 error = netdev_rx_open(netdev, &rx);
429 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
430 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
431 devname, ovs_strerror(errno));
432 netdev_close(netdev);
436 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
439 netdev_close(netdev);
443 port = xmalloc(sizeof *port);
444 port->port_no = port_no;
445 port->netdev = netdev;
448 port->type = xstrdup(type);
450 error = netdev_get_mtu(netdev, &mtu);
451 if (!error && mtu > dp->max_mtu) {
455 list_push_back(&dp->port_list, &port->node);
456 dp->ports[odp_to_u32(port_no)] = port;
463 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
464 odp_port_t *port_nop)
466 struct dp_netdev *dp = get_dp_netdev(dpif);
467 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
468 const char *dpif_port;
472 ovs_mutex_lock(&dp_netdev_mutex);
473 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
474 if (*port_nop != ODPP_NONE) {
475 uint32_t port_idx = odp_to_u32(*port_nop);
476 if (port_idx >= MAX_PORTS) {
478 } else if (dp->ports[port_idx]) {
485 port_no = choose_port(dp, dpif_port);
486 error = port_no == ODPP_NONE ? EFBIG : 0;
490 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
492 ovs_mutex_unlock(&dp_netdev_mutex);
498 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
500 struct dp_netdev *dp = get_dp_netdev(dpif);
503 ovs_mutex_lock(&dp_netdev_mutex);
504 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
505 ovs_mutex_unlock(&dp_netdev_mutex);
511 is_valid_port_number(odp_port_t port_no)
513 return odp_to_u32(port_no) < MAX_PORTS;
517 get_port_by_number(struct dp_netdev *dp,
518 odp_port_t port_no, struct dp_netdev_port **portp)
520 if (!is_valid_port_number(port_no)) {
524 *portp = dp->ports[odp_to_u32(port_no)];
525 return *portp ? 0 : ENOENT;
530 get_port_by_name(struct dp_netdev *dp,
531 const char *devname, struct dp_netdev_port **portp)
533 struct dp_netdev_port *port;
535 LIST_FOR_EACH (port, node, &dp->port_list) {
536 if (!strcmp(netdev_get_name(port->netdev), devname)) {
545 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
547 struct dp_netdev_port *port;
550 error = get_port_by_number(dp, port_no, &port);
555 list_remove(&port->node);
556 dp->ports[odp_to_u32(port_no)] = NULL;
559 netdev_close(port->netdev);
560 netdev_restore_flags(port->sf);
561 netdev_rx_close(port->rx);
569 answer_port_query(const struct dp_netdev_port *port,
570 struct dpif_port *dpif_port)
572 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
573 dpif_port->type = xstrdup(port->type);
574 dpif_port->port_no = port->port_no;
578 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
579 struct dpif_port *dpif_port)
581 struct dp_netdev *dp = get_dp_netdev(dpif);
582 struct dp_netdev_port *port;
585 ovs_mutex_lock(&dp_netdev_mutex);
586 error = get_port_by_number(dp, port_no, &port);
587 if (!error && dpif_port) {
588 answer_port_query(port, dpif_port);
590 ovs_mutex_unlock(&dp_netdev_mutex);
596 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
597 struct dpif_port *dpif_port)
599 struct dp_netdev *dp = get_dp_netdev(dpif);
600 struct dp_netdev_port *port;
603 ovs_mutex_lock(&dp_netdev_mutex);
604 error = get_port_by_name(dp, devname, &port);
605 if (!error && dpif_port) {
606 answer_port_query(port, dpif_port);
608 ovs_mutex_unlock(&dp_netdev_mutex);
614 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
616 return u32_to_odp(MAX_PORTS);
620 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
622 hmap_remove(&dp->flow_table, &flow->node);
628 dp_netdev_flow_flush(struct dp_netdev *dp)
630 struct dp_netdev_flow *flow, *next;
632 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
633 dp_netdev_free_flow(dp, flow);
638 dpif_netdev_flow_flush(struct dpif *dpif)
640 struct dp_netdev *dp = get_dp_netdev(dpif);
642 ovs_mutex_lock(&dp_netdev_mutex);
643 dp_netdev_flow_flush(dp);
644 ovs_mutex_unlock(&dp_netdev_mutex);
649 struct dp_netdev_port_state {
655 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
657 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
662 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
663 struct dpif_port *dpif_port)
665 struct dp_netdev_port_state *state = state_;
666 struct dp_netdev *dp = get_dp_netdev(dpif);
669 ovs_mutex_lock(&dp_netdev_mutex);
670 for (port_idx = odp_to_u32(state->port_no);
671 port_idx < MAX_PORTS; port_idx++) {
672 struct dp_netdev_port *port = dp->ports[port_idx];
675 state->name = xstrdup(netdev_get_name(port->netdev));
676 dpif_port->name = state->name;
677 dpif_port->type = port->type;
678 dpif_port->port_no = port->port_no;
679 state->port_no = u32_to_odp(port_idx + 1);
680 ovs_mutex_unlock(&dp_netdev_mutex);
685 ovs_mutex_unlock(&dp_netdev_mutex);
691 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
693 struct dp_netdev_port_state *state = state_;
700 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
702 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
705 ovs_mutex_lock(&dp_netdev_mutex);
706 if (dpif->dp_serial != dpif->dp->serial) {
707 dpif->dp_serial = dpif->dp->serial;
712 ovs_mutex_unlock(&dp_netdev_mutex);
718 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
720 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
722 /* XXX In a multithreaded process, there is a race window between this
723 * function and the poll_block() in one thread and a change in
724 * dpif->dp->serial in another thread. */
726 ovs_mutex_lock(&dp_netdev_mutex);
727 if (dpif->dp_serial != dpif->dp->serial) {
728 poll_immediate_wake();
730 ovs_mutex_unlock(&dp_netdev_mutex);
733 static struct dp_netdev_flow *
734 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
736 struct dp_netdev_flow *flow;
738 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
739 if (flow_equal(&flow->key, key)) {
747 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
749 stats->n_packets = flow->packet_count;
750 stats->n_bytes = flow->byte_count;
751 stats->used = flow->used;
752 stats->tcp_flags = flow->tcp_flags;
756 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
761 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
762 /* This should not happen: it indicates that odp_flow_key_from_flow()
763 * and odp_flow_key_to_flow() disagree on the acceptable form of a
764 * flow. Log the problem as an error, with enough details to enable
766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
768 if (!VLOG_DROP_ERR(&rl)) {
772 odp_flow_key_format(key, key_len, &s);
773 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
780 in_port = flow->in_port.odp_port;
781 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
789 dpif_netdev_flow_get(const struct dpif *dpif,
790 const struct nlattr *nl_key, size_t nl_key_len,
791 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
793 struct dp_netdev *dp = get_dp_netdev(dpif);
794 struct dp_netdev_flow *flow;
798 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
803 ovs_mutex_lock(&dp_netdev_mutex);
804 flow = dp_netdev_lookup_flow(dp, &key);
807 get_dpif_flow_stats(flow, stats);
810 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
815 ovs_mutex_unlock(&dp_netdev_mutex);
821 set_flow_actions(struct dp_netdev_flow *flow,
822 const struct nlattr *actions, size_t actions_len)
824 flow->actions = xrealloc(flow->actions, actions_len);
825 flow->actions_len = actions_len;
826 memcpy(flow->actions, actions, actions_len);
831 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
832 const struct nlattr *actions, size_t actions_len)
834 struct dp_netdev_flow *flow;
837 flow = xzalloc(sizeof *flow);
840 error = set_flow_actions(flow, actions, actions_len);
846 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
851 clear_stats(struct dp_netdev_flow *flow)
854 flow->packet_count = 0;
855 flow->byte_count = 0;
860 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
862 struct dp_netdev *dp = get_dp_netdev(dpif);
863 struct dp_netdev_flow *flow;
867 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
872 ovs_mutex_lock(&dp_netdev_mutex);
873 flow = dp_netdev_lookup_flow(dp, &key);
875 if (put->flags & DPIF_FP_CREATE) {
876 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
878 memset(put->stats, 0, sizeof *put->stats);
880 error = dp_netdev_flow_add(dp, &key, put->actions,
889 if (put->flags & DPIF_FP_MODIFY) {
890 error = set_flow_actions(flow, put->actions, put->actions_len);
893 get_dpif_flow_stats(flow, put->stats);
895 if (put->flags & DPIF_FP_ZERO_STATS) {
903 ovs_mutex_unlock(&dp_netdev_mutex);
909 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
911 struct dp_netdev *dp = get_dp_netdev(dpif);
912 struct dp_netdev_flow *flow;
916 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
921 ovs_mutex_lock(&dp_netdev_mutex);
922 flow = dp_netdev_lookup_flow(dp, &key);
925 get_dpif_flow_stats(flow, del->stats);
927 dp_netdev_free_flow(dp, flow);
931 ovs_mutex_unlock(&dp_netdev_mutex);
936 struct dp_netdev_flow_state {
939 struct nlattr *actions;
940 struct odputil_keybuf keybuf;
941 struct dpif_flow_stats stats;
945 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
947 struct dp_netdev_flow_state *state;
949 *statep = state = xmalloc(sizeof *state);
952 state->actions = NULL;
957 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
958 const struct nlattr **key, size_t *key_len,
959 const struct nlattr **mask, size_t *mask_len,
960 const struct nlattr **actions, size_t *actions_len,
961 const struct dpif_flow_stats **stats)
963 struct dp_netdev_flow_state *state = state_;
964 struct dp_netdev *dp = get_dp_netdev(dpif);
965 struct dp_netdev_flow *flow;
966 struct hmap_node *node;
968 ovs_mutex_lock(&dp_netdev_mutex);
969 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
971 ovs_mutex_unlock(&dp_netdev_mutex);
975 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
980 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
981 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port.odp_port);
993 free(state->actions);
994 state->actions = xmemdup(flow->actions, flow->actions_len);
996 *actions = state->actions;
997 *actions_len = flow->actions_len;
1001 get_dpif_flow_stats(flow, &state->stats);
1002 *stats = &state->stats;
1005 ovs_mutex_unlock(&dp_netdev_mutex);
1010 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1012 struct dp_netdev_flow_state *state = state_;
1014 free(state->actions);
1020 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
1022 struct dp_netdev *dp = get_dp_netdev(dpif);
1027 if (execute->packet->size < ETH_HEADER_LEN ||
1028 execute->packet->size > UINT16_MAX) {
1032 /* Make a deep copy of 'packet', because we might modify its data. */
1033 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
1034 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
1035 ofpbuf_put(©, execute->packet->data, execute->packet->size);
1037 flow_extract(©, 0, 0, NULL, NULL, &key);
1038 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
1041 ovs_mutex_lock(&dp_netdev_mutex);
1042 dp_netdev_execute_actions(dp, ©, &key,
1043 execute->actions, execute->actions_len);
1044 ovs_mutex_unlock(&dp_netdev_mutex);
1047 ofpbuf_uninit(©);
1052 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1058 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1059 uint32_t queue_id, uint32_t *priority)
1061 *priority = queue_id;
1065 static struct dp_netdev_queue *
1066 find_nonempty_queue(struct dpif *dpif)
1068 struct dp_netdev *dp = get_dp_netdev(dpif);
1071 for (i = 0; i < N_QUEUES; i++) {
1072 struct dp_netdev_queue *q = &dp->queues[i];
1073 if (q->head != q->tail) {
1081 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1084 struct dp_netdev_queue *q;
1087 ovs_mutex_lock(&dp_netdev_mutex);
1088 q = find_nonempty_queue(dpif);
1090 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1092 *upcall = u->upcall;
1093 upcall->packet = buf;
1102 ovs_mutex_unlock(&dp_netdev_mutex);
1108 dpif_netdev_recv_wait(struct dpif *dpif)
1110 /* XXX In a multithreaded process, there is a race window between this
1111 * function and the poll_block() in one thread and a packet being queued in
1112 * another thread. */
1114 ovs_mutex_lock(&dp_netdev_mutex);
1115 if (find_nonempty_queue(dpif)) {
1116 poll_immediate_wake();
1118 ovs_mutex_unlock(&dp_netdev_mutex);
1122 dpif_netdev_recv_purge(struct dpif *dpif)
1124 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1125 ovs_mutex_lock(&dp_netdev_mutex);
1126 dp_netdev_purge_queues(dpif_netdev->dp);
1127 ovs_mutex_unlock(&dp_netdev_mutex);
1131 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1133 flow->used = time_msec();
1134 flow->packet_count++;
1135 flow->byte_count += packet->size;
1136 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1140 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1141 struct ofpbuf *packet, uint32_t skb_priority,
1142 uint32_t skb_mark, const struct flow_tnl *tnl)
1144 struct dp_netdev_flow *flow;
1146 union flow_in_port in_port_;
1148 if (packet->size < ETH_HEADER_LEN) {
1151 in_port_.odp_port = port->port_no;
1152 flow_extract(packet, skb_priority, skb_mark, tnl, &in_port_, &key);
1153 flow = dp_netdev_lookup_flow(dp, &key);
1155 dp_netdev_flow_used(flow, packet);
1156 dp_netdev_execute_actions(dp, packet, &key,
1157 flow->actions, flow->actions_len);
1161 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1166 dpif_netdev_run(struct dpif *dpif)
1168 struct dp_netdev_port *port;
1169 struct dp_netdev *dp;
1170 struct ofpbuf packet;
1172 ovs_mutex_lock(&dp_netdev_mutex);
1173 dp = get_dp_netdev(dpif);
1174 ofpbuf_init(&packet,
1175 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1177 LIST_FOR_EACH (port, node, &dp->port_list) {
1180 /* Reset packet contents. */
1181 ofpbuf_clear(&packet);
1182 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1184 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1186 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1187 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1188 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1190 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1191 netdev_get_name(port->netdev), ovs_strerror(error));
1194 ofpbuf_uninit(&packet);
1195 ovs_mutex_unlock(&dp_netdev_mutex);
1199 dpif_netdev_wait(struct dpif *dpif)
1201 struct dp_netdev_port *port;
1203 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1204 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1205 * A makes it to poll_block().
1207 * But I think it doesn't matter:
1209 * - In the dpif_port_add() case, A will not wake up when a packet
1210 * arrives on the new port, but this would also happen if the
1211 * ordering were reversed.
1213 * - In the dpif_port_remove() case, A might wake up spuriously, but
1214 * that is harmless. */
1216 ovs_mutex_lock(&dp_netdev_mutex);
1217 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1219 netdev_rx_wait(port->rx);
1222 ovs_mutex_unlock(&dp_netdev_mutex);
1226 dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port)
1228 struct dp_netdev *dp = dp_;
1229 struct dp_netdev_port *p = dp->ports[out_port];
1231 netdev_send(p->netdev, packet);
1236 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1237 int queue_no, const struct flow *flow,
1238 const struct nlattr *userdata)
1240 struct dp_netdev_queue *q = &dp->queues[queue_no];
1241 if (q->head - q->tail < MAX_QUEUE_LEN) {
1242 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1243 struct dpif_upcall *upcall = &u->upcall;
1244 struct ofpbuf *buf = &u->buf;
1247 upcall->type = queue_no;
1249 /* Allocate buffer big enough for everything. */
1250 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1252 buf_size += NLA_ALIGN(userdata->nla_len);
1254 ofpbuf_init(buf, buf_size);
1257 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1258 upcall->key = buf->data;
1259 upcall->key_len = buf->size;
1263 upcall->userdata = ofpbuf_put(buf, userdata,
1264 NLA_ALIGN(userdata->nla_len));
1269 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1270 * is visible in 'upcall->packet'. The ODP flow and (if present)
1271 * userdata become part of the headroom. */
1272 ofpbuf_put_zeros(buf, 2);
1273 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1274 buf->size = packet->size;
1275 upcall->packet = buf;
1285 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1286 const struct flow *key,
1287 const struct nlattr *userdata)
1289 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1293 dp_netdev_execute_actions(struct dp_netdev *dp,
1294 struct ofpbuf *packet, struct flow *key,
1295 const struct nlattr *actions,
1298 odp_execute_actions(dp, packet, key, actions, actions_len,
1299 dp_netdev_output_port, dp_netdev_action_userspace);
1302 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1303 dpif_netdev_enumerate, \
1304 dpif_netdev_port_open_type, \
1306 dpif_netdev_close, \
1307 dpif_netdev_destroy, \
1310 dpif_netdev_get_stats, \
1311 dpif_netdev_port_add, \
1312 dpif_netdev_port_del, \
1313 dpif_netdev_port_query_by_number, \
1314 dpif_netdev_port_query_by_name, \
1315 dpif_netdev_get_max_ports, \
1316 NULL, /* port_get_pid */ \
1317 dpif_netdev_port_dump_start, \
1318 dpif_netdev_port_dump_next, \
1319 dpif_netdev_port_dump_done, \
1320 dpif_netdev_port_poll, \
1321 dpif_netdev_port_poll_wait, \
1322 dpif_netdev_flow_get, \
1323 dpif_netdev_flow_put, \
1324 dpif_netdev_flow_del, \
1325 dpif_netdev_flow_flush, \
1326 dpif_netdev_flow_dump_start, \
1327 dpif_netdev_flow_dump_next, \
1328 dpif_netdev_flow_dump_done, \
1329 dpif_netdev_execute, \
1330 NULL, /* operate */ \
1331 dpif_netdev_recv_set, \
1332 dpif_netdev_queue_to_priority, \
1334 dpif_netdev_recv_wait, \
1335 dpif_netdev_recv_purge, \
1337 const struct dpif_class dpif_netdev_class = {
1339 DPIF_NETDEV_CLASS_FUNCTIONS
1342 const struct dpif_class dpif_planetlab_class = {
1344 DPIF_NETDEV_CLASS_FUNCTIONS
1348 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1349 const char *argv[], void *aux OVS_UNUSED)
1351 struct dp_netdev_port *port;
1352 struct dp_netdev *dp;
1355 dp = shash_find_data(&dp_netdevs, argv[1]);
1356 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1357 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1361 if (get_port_by_name(dp, argv[2], &port)) {
1362 unixctl_command_reply_error(conn, "unknown port");
1366 port_no = atoi(argv[3]);
1367 if (port_no <= 0 || port_no >= MAX_PORTS) {
1368 unixctl_command_reply_error(conn, "bad port number");
1371 if (dp->ports[port_no]) {
1372 unixctl_command_reply_error(conn, "port number already in use");
1375 dp->ports[odp_to_u32(port->port_no)] = NULL;
1376 dp->ports[port_no] = port;
1377 port->port_no = u32_to_odp(port_no);
1379 unixctl_command_reply(conn, NULL);
1383 dpif_dummy_register__(const char *type)
1385 struct dpif_class *class;
1387 class = xmalloc(sizeof *class);
1388 *class = dpif_netdev_class;
1389 class->type = xstrdup(type);
1390 dp_register_provider(class);
1394 dpif_dummy_register(bool override)
1401 dp_enumerate_types(&types);
1402 SSET_FOR_EACH (type, &types) {
1403 if (!dp_unregister_provider(type)) {
1404 dpif_dummy_register__(type);
1407 sset_destroy(&types);
1410 dpif_dummy_register__("dummy");
1412 unixctl_command_register("dpif-dummy/change-port-number",
1413 "DP PORT NEW-NUMBER",
1414 3, 3, dpif_dummy_change_port_number, NULL);