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"
60 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
62 /* Configuration parameters. */
63 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
64 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
66 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
67 * headers to be aligned on a 4-byte boundary. */
68 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
71 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
72 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
73 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
74 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
76 struct dp_netdev_upcall {
77 struct dpif_upcall upcall; /* Queued upcall information. */
78 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
81 struct dp_netdev_queue {
82 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
83 unsigned int head, tail;
86 /* Datapath based on the network device interface from netdev.h. */
88 const struct dpif_class *class;
92 int max_mtu; /* Maximum MTU of any port added so far. */
94 struct dp_netdev_queue queues[N_QUEUES];
95 struct hmap flow_table; /* Flow table. */
96 struct seq *queue_seq; /* Incremented whenever a packet is queued. */
99 long long int n_hit; /* Number of flow table matches. */
100 long long int n_missed; /* Number of flow table misses. */
101 long long int n_lost; /* Number of misses not passed to client. */
104 struct dp_netdev_port *ports[MAX_PORTS];
105 struct list port_list;
106 struct seq *port_seq; /* Incremented whenever a port changes. */
109 /* A port in a netdev-based datapath. */
110 struct dp_netdev_port {
111 odp_port_t port_no; /* Index into dp_netdev's 'ports'. */
112 struct list node; /* Element in dp_netdev's 'port_list'. */
113 struct netdev *netdev;
114 struct netdev_saved_flags *sf;
115 struct netdev_rx *rx;
116 char *type; /* Port type as requested by user. */
119 /* A flow in dp_netdev's 'flow_table'. */
120 struct dp_netdev_flow {
121 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
125 long long int used; /* Last used time, in monotonic msecs. */
126 long long int packet_count; /* Number of packets matched. */
127 long long int byte_count; /* Number of bytes matched. */
128 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
131 struct nlattr *actions;
135 /* Interface to netdev-based datapath. */
138 struct dp_netdev *dp;
139 uint64_t last_port_seq;
142 /* All netdev-based datapaths. */
143 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
145 /* Global lock for all data. */
146 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
148 static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
149 struct dp_netdev_port **portp);
150 static int get_port_by_name(struct dp_netdev *, const char *devname,
151 struct dp_netdev_port **portp);
152 static void dp_netdev_free(struct dp_netdev *);
153 static void dp_netdev_flow_flush(struct dp_netdev *);
154 static int do_add_port(struct dp_netdev *, const char *devname,
155 const char *type, odp_port_t port_no);
156 static int do_del_port(struct dp_netdev *, odp_port_t port_no);
157 static int dpif_netdev_open(const struct dpif_class *, const char *name,
158 bool create, struct dpif **);
159 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
160 int queue_no, const struct flow *,
161 const struct nlattr *userdata);
162 static void dp_netdev_execute_actions(struct dp_netdev *,
163 struct ofpbuf *, struct flow *,
164 const struct nlattr *actions,
166 static void dp_netdev_port_input(struct dp_netdev *dp,
167 struct dp_netdev_port *port,
168 struct ofpbuf *packet, uint32_t skb_priority,
169 uint32_t pkt_mark, const struct flow_tnl *tnl);
171 static struct dpif_netdev *
172 dpif_netdev_cast(const struct dpif *dpif)
174 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
175 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
178 static struct dp_netdev *
179 get_dp_netdev(const struct dpif *dpif)
181 return dpif_netdev_cast(dpif)->dp;
185 dpif_netdev_enumerate(struct sset *all_dps)
187 struct shash_node *node;
189 ovs_mutex_lock(&dp_netdev_mutex);
190 SHASH_FOR_EACH(node, &dp_netdevs) {
191 sset_add(all_dps, node->name);
193 ovs_mutex_unlock(&dp_netdev_mutex);
199 dpif_netdev_class_is_dummy(const struct dpif_class *class)
201 return class != &dpif_netdev_class;
205 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
207 return strcmp(type, "internal") ? type
208 : dpif_netdev_class_is_dummy(class) ? "dummy"
213 create_dpif_netdev(struct dp_netdev *dp)
215 uint16_t netflow_id = hash_string(dp->name, 0);
216 struct dpif_netdev *dpif;
220 dpif = xmalloc(sizeof *dpif);
221 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
223 dpif->last_port_seq = seq_read(dp->port_seq);
228 /* Choose an unused, non-zero port number and return it on success.
229 * Return ODPP_NONE on failure. */
231 choose_port(struct dp_netdev *dp, const char *name)
235 if (dp->class != &dpif_netdev_class) {
239 /* If the port name begins with "br", start the number search at
240 * 100 to make writing tests easier. */
241 if (!strncmp(name, "br", 2)) {
245 /* If the port name contains a number, try to assign that port number.
246 * This can make writing unit tests easier because port numbers are
248 for (p = name; *p != '\0'; p++) {
249 if (isdigit((unsigned char) *p)) {
250 port_no = start_no + strtol(p, NULL, 10);
251 if (port_no > 0 && port_no < MAX_PORTS
252 && !dp->ports[port_no]) {
253 return u32_to_odp(port_no);
260 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
261 if (!dp->ports[port_no]) {
262 return u32_to_odp(port_no);
270 create_dp_netdev(const char *name, const struct dpif_class *class,
271 struct dp_netdev **dpp)
273 struct dp_netdev *dp;
277 dp = xzalloc(sizeof *dp);
279 dp->name = xstrdup(name);
281 dp->max_mtu = ETH_PAYLOAD_MAX;
282 for (i = 0; i < N_QUEUES; i++) {
283 dp->queues[i].head = dp->queues[i].tail = 0;
285 dp->queue_seq = seq_create();
286 hmap_init(&dp->flow_table);
287 list_init(&dp->port_list);
288 dp->port_seq = seq_create();
290 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
296 shash_add(&dp_netdevs, name, dp);
303 dpif_netdev_open(const struct dpif_class *class, const char *name,
304 bool create, struct dpif **dpifp)
306 struct dp_netdev *dp;
309 ovs_mutex_lock(&dp_netdev_mutex);
310 dp = shash_find_data(&dp_netdevs, name);
312 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
314 error = (dp->class != class ? EINVAL
319 *dpifp = create_dpif_netdev(dp);
321 ovs_mutex_unlock(&dp_netdev_mutex);
327 dp_netdev_purge_queues(struct dp_netdev *dp)
331 for (i = 0; i < N_QUEUES; i++) {
332 struct dp_netdev_queue *q = &dp->queues[i];
334 while (q->tail != q->head) {
335 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
336 ofpbuf_uninit(&u->buf);
342 dp_netdev_free(struct dp_netdev *dp)
344 struct dp_netdev_port *port, *next;
346 dp_netdev_flow_flush(dp);
347 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
348 do_del_port(dp, port->port_no);
350 dp_netdev_purge_queues(dp);
351 seq_destroy(dp->queue_seq);
352 hmap_destroy(&dp->flow_table);
353 seq_destroy(dp->port_seq);
359 dpif_netdev_close(struct dpif *dpif)
361 struct dp_netdev *dp = get_dp_netdev(dpif);
363 ovs_mutex_lock(&dp_netdev_mutex);
365 ovs_assert(dp->open_cnt > 0);
366 if (--dp->open_cnt == 0 && dp->destroyed) {
367 shash_find_and_delete(&dp_netdevs, dp->name);
372 ovs_mutex_unlock(&dp_netdev_mutex);
376 dpif_netdev_destroy(struct dpif *dpif)
378 struct dp_netdev *dp = get_dp_netdev(dpif);
380 ovs_mutex_lock(&dp_netdev_mutex);
381 dp->destroyed = true;
382 ovs_mutex_unlock(&dp_netdev_mutex);
388 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
390 struct dp_netdev *dp = get_dp_netdev(dpif);
392 ovs_mutex_lock(&dp_netdev_mutex);
393 stats->n_flows = hmap_count(&dp->flow_table);
394 stats->n_hit = dp->n_hit;
395 stats->n_missed = dp->n_missed;
396 stats->n_lost = dp->n_lost;
397 stats->n_masks = UINT64_MAX;
398 stats->n_mask_hit = UINT64_MAX;
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 enum netdev_flags flags;
413 const char *open_type;
417 /* XXX reject devices already in some dp_netdev. */
419 /* Open and validate network device. */
420 open_type = dpif_netdev_port_open_type(dp->class, type);
421 error = netdev_open(devname, open_type, &netdev);
425 /* XXX reject non-Ethernet devices */
427 netdev_get_flags(netdev, &flags);
428 if (flags & NETDEV_LOOPBACK) {
429 VLOG_ERR("%s: cannot add a loopback device", devname);
430 netdev_close(netdev);
434 error = netdev_rx_open(netdev, &rx);
436 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
437 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
438 devname, ovs_strerror(errno));
439 netdev_close(netdev);
443 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
446 netdev_close(netdev);
450 port = xmalloc(sizeof *port);
451 port->port_no = port_no;
452 port->netdev = netdev;
455 port->type = xstrdup(type);
457 error = netdev_get_mtu(netdev, &mtu);
458 if (!error && mtu > dp->max_mtu) {
462 list_push_back(&dp->port_list, &port->node);
463 dp->ports[odp_to_u32(port_no)] = port;
464 seq_change(dp->port_seq);
470 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
471 odp_port_t *port_nop)
473 struct dp_netdev *dp = get_dp_netdev(dpif);
474 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
475 const char *dpif_port;
479 ovs_mutex_lock(&dp_netdev_mutex);
480 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
481 if (*port_nop != ODPP_NONE) {
482 uint32_t port_idx = odp_to_u32(*port_nop);
483 if (port_idx >= MAX_PORTS) {
485 } else if (dp->ports[port_idx]) {
492 port_no = choose_port(dp, dpif_port);
493 error = port_no == ODPP_NONE ? EFBIG : 0;
497 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
499 ovs_mutex_unlock(&dp_netdev_mutex);
505 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
507 struct dp_netdev *dp = get_dp_netdev(dpif);
510 ovs_mutex_lock(&dp_netdev_mutex);
511 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
512 ovs_mutex_unlock(&dp_netdev_mutex);
518 is_valid_port_number(odp_port_t port_no)
520 return odp_to_u32(port_no) < MAX_PORTS;
524 get_port_by_number(struct dp_netdev *dp,
525 odp_port_t port_no, struct dp_netdev_port **portp)
527 if (!is_valid_port_number(port_no)) {
531 *portp = dp->ports[odp_to_u32(port_no)];
532 return *portp ? 0 : ENOENT;
537 get_port_by_name(struct dp_netdev *dp,
538 const char *devname, struct dp_netdev_port **portp)
540 struct dp_netdev_port *port;
542 LIST_FOR_EACH (port, node, &dp->port_list) {
543 if (!strcmp(netdev_get_name(port->netdev), devname)) {
552 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
554 struct dp_netdev_port *port;
557 error = get_port_by_number(dp, port_no, &port);
562 list_remove(&port->node);
563 dp->ports[odp_to_u32(port_no)] = NULL;
564 seq_change(dp->port_seq);
566 netdev_close(port->netdev);
567 netdev_restore_flags(port->sf);
568 netdev_rx_close(port->rx);
576 answer_port_query(const struct dp_netdev_port *port,
577 struct dpif_port *dpif_port)
579 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
580 dpif_port->type = xstrdup(port->type);
581 dpif_port->port_no = port->port_no;
585 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
586 struct dpif_port *dpif_port)
588 struct dp_netdev *dp = get_dp_netdev(dpif);
589 struct dp_netdev_port *port;
592 ovs_mutex_lock(&dp_netdev_mutex);
593 error = get_port_by_number(dp, port_no, &port);
594 if (!error && dpif_port) {
595 answer_port_query(port, dpif_port);
597 ovs_mutex_unlock(&dp_netdev_mutex);
603 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
604 struct dpif_port *dpif_port)
606 struct dp_netdev *dp = get_dp_netdev(dpif);
607 struct dp_netdev_port *port;
610 ovs_mutex_lock(&dp_netdev_mutex);
611 error = get_port_by_name(dp, devname, &port);
612 if (!error && dpif_port) {
613 answer_port_query(port, dpif_port);
615 ovs_mutex_unlock(&dp_netdev_mutex);
621 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
627 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
629 hmap_remove(&dp->flow_table, &flow->node);
635 dp_netdev_flow_flush(struct dp_netdev *dp)
637 struct dp_netdev_flow *flow, *next;
639 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
640 dp_netdev_free_flow(dp, flow);
645 dpif_netdev_flow_flush(struct dpif *dpif)
647 struct dp_netdev *dp = get_dp_netdev(dpif);
649 ovs_mutex_lock(&dp_netdev_mutex);
650 dp_netdev_flow_flush(dp);
651 ovs_mutex_unlock(&dp_netdev_mutex);
656 struct dp_netdev_port_state {
662 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
664 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
669 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
670 struct dpif_port *dpif_port)
672 struct dp_netdev_port_state *state = state_;
673 struct dp_netdev *dp = get_dp_netdev(dpif);
676 ovs_mutex_lock(&dp_netdev_mutex);
677 for (port_idx = odp_to_u32(state->port_no);
678 port_idx < MAX_PORTS; port_idx++) {
679 struct dp_netdev_port *port = dp->ports[port_idx];
682 state->name = xstrdup(netdev_get_name(port->netdev));
683 dpif_port->name = state->name;
684 dpif_port->type = port->type;
685 dpif_port->port_no = port->port_no;
686 state->port_no = u32_to_odp(port_idx + 1);
687 ovs_mutex_unlock(&dp_netdev_mutex);
692 ovs_mutex_unlock(&dp_netdev_mutex);
698 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
700 struct dp_netdev_port_state *state = state_;
707 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
709 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
710 uint64_t new_port_seq;
713 ovs_mutex_lock(&dp_netdev_mutex);
714 new_port_seq = seq_read(dpif->dp->port_seq);
715 if (dpif->last_port_seq != new_port_seq) {
716 dpif->last_port_seq = new_port_seq;
721 ovs_mutex_unlock(&dp_netdev_mutex);
727 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
729 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
731 ovs_mutex_lock(&dp_netdev_mutex);
732 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
733 ovs_mutex_unlock(&dp_netdev_mutex);
736 static struct dp_netdev_flow *
737 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
739 struct dp_netdev_flow *flow;
741 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
742 if (flow_equal(&flow->key, key)) {
750 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
752 stats->n_packets = flow->packet_count;
753 stats->n_bytes = flow->byte_count;
754 stats->used = flow->used;
755 stats->tcp_flags = flow->tcp_flags;
759 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
764 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
765 /* This should not happen: it indicates that odp_flow_key_from_flow()
766 * and odp_flow_key_to_flow() disagree on the acceptable form of a
767 * flow. Log the problem as an error, with enough details to enable
769 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
771 if (!VLOG_DROP_ERR(&rl)) {
775 odp_flow_key_format(key, key_len, &s);
776 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
783 in_port = flow->in_port.odp_port;
784 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
792 dpif_netdev_flow_get(const struct dpif *dpif,
793 const struct nlattr *nl_key, size_t nl_key_len,
794 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
796 struct dp_netdev *dp = get_dp_netdev(dpif);
797 struct dp_netdev_flow *flow;
801 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
806 ovs_mutex_lock(&dp_netdev_mutex);
807 flow = dp_netdev_lookup_flow(dp, &key);
810 get_dpif_flow_stats(flow, stats);
813 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
818 ovs_mutex_unlock(&dp_netdev_mutex);
824 set_flow_actions(struct dp_netdev_flow *flow,
825 const struct nlattr *actions, size_t actions_len)
827 flow->actions = xrealloc(flow->actions, actions_len);
828 flow->actions_len = actions_len;
829 memcpy(flow->actions, actions, actions_len);
834 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
835 const struct nlattr *actions, size_t actions_len)
837 struct dp_netdev_flow *flow;
840 flow = xzalloc(sizeof *flow);
843 error = set_flow_actions(flow, actions, actions_len);
849 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
854 clear_stats(struct dp_netdev_flow *flow)
857 flow->packet_count = 0;
858 flow->byte_count = 0;
863 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
865 struct dp_netdev *dp = get_dp_netdev(dpif);
866 struct dp_netdev_flow *flow;
870 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
875 ovs_mutex_lock(&dp_netdev_mutex);
876 flow = dp_netdev_lookup_flow(dp, &key);
878 if (put->flags & DPIF_FP_CREATE) {
879 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
881 memset(put->stats, 0, sizeof *put->stats);
883 error = dp_netdev_flow_add(dp, &key, put->actions,
892 if (put->flags & DPIF_FP_MODIFY) {
893 error = set_flow_actions(flow, put->actions, put->actions_len);
896 get_dpif_flow_stats(flow, put->stats);
898 if (put->flags & DPIF_FP_ZERO_STATS) {
906 ovs_mutex_unlock(&dp_netdev_mutex);
912 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
914 struct dp_netdev *dp = get_dp_netdev(dpif);
915 struct dp_netdev_flow *flow;
919 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
924 ovs_mutex_lock(&dp_netdev_mutex);
925 flow = dp_netdev_lookup_flow(dp, &key);
928 get_dpif_flow_stats(flow, del->stats);
930 dp_netdev_free_flow(dp, flow);
934 ovs_mutex_unlock(&dp_netdev_mutex);
939 struct dp_netdev_flow_state {
942 struct nlattr *actions;
943 struct odputil_keybuf keybuf;
944 struct dpif_flow_stats stats;
948 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
950 struct dp_netdev_flow_state *state;
952 *statep = state = xmalloc(sizeof *state);
955 state->actions = NULL;
960 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
961 const struct nlattr **key, size_t *key_len,
962 const struct nlattr **mask, size_t *mask_len,
963 const struct nlattr **actions, size_t *actions_len,
964 const struct dpif_flow_stats **stats)
966 struct dp_netdev_flow_state *state = state_;
967 struct dp_netdev *dp = get_dp_netdev(dpif);
968 struct dp_netdev_flow *flow;
969 struct hmap_node *node;
971 ovs_mutex_lock(&dp_netdev_mutex);
972 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
974 ovs_mutex_unlock(&dp_netdev_mutex);
978 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
983 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
984 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port.odp_port);
996 free(state->actions);
997 state->actions = xmemdup(flow->actions, flow->actions_len);
999 *actions = state->actions;
1000 *actions_len = flow->actions_len;
1004 get_dpif_flow_stats(flow, &state->stats);
1005 *stats = &state->stats;
1008 ovs_mutex_unlock(&dp_netdev_mutex);
1013 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1015 struct dp_netdev_flow_state *state = state_;
1017 free(state->actions);
1023 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
1025 struct dp_netdev *dp = get_dp_netdev(dpif);
1030 if (execute->packet->size < ETH_HEADER_LEN ||
1031 execute->packet->size > UINT16_MAX) {
1035 /* Make a deep copy of 'packet', because we might modify its data. */
1036 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
1037 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
1038 ofpbuf_put(©, execute->packet->data, execute->packet->size);
1040 flow_extract(©, 0, 0, NULL, NULL, &key);
1041 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
1044 ovs_mutex_lock(&dp_netdev_mutex);
1045 dp_netdev_execute_actions(dp, ©, &key,
1046 execute->actions, execute->actions_len);
1047 ovs_mutex_unlock(&dp_netdev_mutex);
1050 ofpbuf_uninit(©);
1055 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1061 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1062 uint32_t queue_id, uint32_t *priority)
1064 *priority = queue_id;
1068 static struct dp_netdev_queue *
1069 find_nonempty_queue(struct dpif *dpif)
1071 struct dp_netdev *dp = get_dp_netdev(dpif);
1074 for (i = 0; i < N_QUEUES; i++) {
1075 struct dp_netdev_queue *q = &dp->queues[i];
1076 if (q->head != q->tail) {
1084 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1087 struct dp_netdev_queue *q;
1090 ovs_mutex_lock(&dp_netdev_mutex);
1091 q = find_nonempty_queue(dpif);
1093 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1095 *upcall = u->upcall;
1096 upcall->packet = buf;
1105 ovs_mutex_unlock(&dp_netdev_mutex);
1111 dpif_netdev_recv_wait(struct dpif *dpif)
1113 struct dp_netdev *dp = get_dp_netdev(dpif);
1116 ovs_mutex_lock(&dp_netdev_mutex);
1117 seq = seq_read(dp->queue_seq);
1118 if (find_nonempty_queue(dpif)) {
1119 poll_immediate_wake();
1121 seq_wait(dp->queue_seq, seq);
1123 ovs_mutex_unlock(&dp_netdev_mutex);
1127 dpif_netdev_recv_purge(struct dpif *dpif)
1129 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1130 ovs_mutex_lock(&dp_netdev_mutex);
1131 dp_netdev_purge_queues(dpif_netdev->dp);
1132 ovs_mutex_unlock(&dp_netdev_mutex);
1136 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1138 flow->used = time_msec();
1139 flow->packet_count++;
1140 flow->byte_count += packet->size;
1141 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1145 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1146 struct ofpbuf *packet, uint32_t skb_priority,
1147 uint32_t pkt_mark, const struct flow_tnl *tnl)
1149 struct dp_netdev_flow *flow;
1151 union flow_in_port in_port_;
1153 if (packet->size < ETH_HEADER_LEN) {
1156 in_port_.odp_port = port->port_no;
1157 flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
1158 flow = dp_netdev_lookup_flow(dp, &key);
1160 dp_netdev_flow_used(flow, packet);
1161 dp_netdev_execute_actions(dp, packet, &key,
1162 flow->actions, flow->actions_len);
1166 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1171 dpif_netdev_run(struct dpif *dpif)
1173 struct dp_netdev_port *port;
1174 struct dp_netdev *dp;
1175 struct ofpbuf packet;
1177 ovs_mutex_lock(&dp_netdev_mutex);
1178 dp = get_dp_netdev(dpif);
1179 ofpbuf_init(&packet,
1180 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1182 LIST_FOR_EACH (port, node, &dp->port_list) {
1185 /* Reset packet contents. */
1186 ofpbuf_clear(&packet);
1187 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1189 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1191 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1192 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1193 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1195 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1196 netdev_get_name(port->netdev), ovs_strerror(error));
1199 ofpbuf_uninit(&packet);
1200 ovs_mutex_unlock(&dp_netdev_mutex);
1204 dpif_netdev_wait(struct dpif *dpif)
1206 struct dp_netdev_port *port;
1208 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1209 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1210 * A makes it to poll_block().
1212 * But I think it doesn't matter:
1214 * - In the dpif_port_add() case, A will not wake up when a packet
1215 * arrives on the new port, but this would also happen if the
1216 * ordering were reversed.
1218 * - In the dpif_port_remove() case, A might wake up spuriously, but
1219 * that is harmless. */
1221 ovs_mutex_lock(&dp_netdev_mutex);
1222 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1224 netdev_rx_wait(port->rx);
1227 ovs_mutex_unlock(&dp_netdev_mutex);
1231 dp_netdev_output_port(void *dp_, struct ofpbuf *packet,
1232 const struct flow *flow OVS_UNUSED, odp_port_t out_port)
1234 struct dp_netdev *dp = dp_;
1235 struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
1237 netdev_send(p->netdev, packet);
1242 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1243 int queue_no, const struct flow *flow,
1244 const struct nlattr *userdata)
1246 struct dp_netdev_queue *q = &dp->queues[queue_no];
1247 if (q->head - q->tail < MAX_QUEUE_LEN) {
1248 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1249 struct dpif_upcall *upcall = &u->upcall;
1250 struct ofpbuf *buf = &u->buf;
1253 upcall->type = queue_no;
1255 /* Allocate buffer big enough for everything. */
1256 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1258 buf_size += NLA_ALIGN(userdata->nla_len);
1260 ofpbuf_init(buf, buf_size);
1263 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1264 upcall->key = buf->data;
1265 upcall->key_len = buf->size;
1269 upcall->userdata = ofpbuf_put(buf, userdata,
1270 NLA_ALIGN(userdata->nla_len));
1275 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1276 * is visible in 'upcall->packet'. The ODP flow and (if present)
1277 * userdata become part of the headroom. */
1278 ofpbuf_put_zeros(buf, 2);
1279 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1280 buf->size = packet->size;
1281 upcall->packet = buf;
1283 seq_change(dp->queue_seq);
1293 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1294 const struct flow *key,
1295 const struct nlattr *a)
1297 const struct nlattr *userdata;
1299 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1300 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1304 dp_netdev_execute_actions(struct dp_netdev *dp,
1305 struct ofpbuf *packet, struct flow *key,
1306 const struct nlattr *actions,
1309 odp_execute_actions(dp, packet, key, actions, actions_len,
1310 dp_netdev_output_port, dp_netdev_action_userspace);
1313 const struct dpif_class dpif_netdev_class = {
1315 dpif_netdev_enumerate,
1316 dpif_netdev_port_open_type,
1319 dpif_netdev_destroy,
1322 dpif_netdev_get_stats,
1323 dpif_netdev_port_add,
1324 dpif_netdev_port_del,
1325 dpif_netdev_port_query_by_number,
1326 dpif_netdev_port_query_by_name,
1327 dpif_netdev_get_max_ports,
1328 NULL, /* port_get_pid */
1329 dpif_netdev_port_dump_start,
1330 dpif_netdev_port_dump_next,
1331 dpif_netdev_port_dump_done,
1332 dpif_netdev_port_poll,
1333 dpif_netdev_port_poll_wait,
1334 dpif_netdev_flow_get,
1335 dpif_netdev_flow_put,
1336 dpif_netdev_flow_del,
1337 dpif_netdev_flow_flush,
1338 dpif_netdev_flow_dump_start,
1339 dpif_netdev_flow_dump_next,
1340 dpif_netdev_flow_dump_done,
1341 dpif_netdev_execute,
1343 dpif_netdev_recv_set,
1344 dpif_netdev_queue_to_priority,
1346 dpif_netdev_recv_wait,
1347 dpif_netdev_recv_purge,
1351 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1352 const char *argv[], void *aux OVS_UNUSED)
1354 struct dp_netdev_port *port;
1355 struct dp_netdev *dp;
1358 dp = shash_find_data(&dp_netdevs, argv[1]);
1359 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1360 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1364 if (get_port_by_name(dp, argv[2], &port)) {
1365 unixctl_command_reply_error(conn, "unknown port");
1369 port_no = atoi(argv[3]);
1370 if (port_no <= 0 || port_no >= MAX_PORTS) {
1371 unixctl_command_reply_error(conn, "bad port number");
1374 if (dp->ports[port_no]) {
1375 unixctl_command_reply_error(conn, "port number already in use");
1378 dp->ports[odp_to_u32(port->port_no)] = NULL;
1379 dp->ports[port_no] = port;
1380 port->port_no = u32_to_odp(port_no);
1381 seq_change(dp->port_seq);
1382 unixctl_command_reply(conn, NULL);
1386 dpif_dummy_register__(const char *type)
1388 struct dpif_class *class;
1390 class = xmalloc(sizeof *class);
1391 *class = dpif_netdev_class;
1392 class->type = xstrdup(type);
1393 dp_register_provider(class);
1397 dpif_dummy_register(bool override)
1404 dp_enumerate_types(&types);
1405 SSET_FOR_EACH (type, &types) {
1406 if (!dp_unregister_provider(type)) {
1407 dpif_dummy_register__(type);
1410 sset_destroy(&types);
1413 dpif_dummy_register__("dummy");
1415 unixctl_command_register("dpif-dummy/change-port-number",
1416 "DP PORT NEW-NUMBER",
1417 3, 3, dpif_dummy_change_port_number, NULL);