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 uint16_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 *netdev_flow)
629 hmap_remove(&dp->flow_table, &netdev_flow->node);
630 free(netdev_flow->actions);
635 dp_netdev_flow_flush(struct dp_netdev *dp)
637 struct dp_netdev_flow *netdev_flow, *next;
639 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
640 dp_netdev_free_flow(dp, netdev_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 *netdev_flow;
741 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(key, 0),
743 if (flow_equal(&netdev_flow->key, key)) {
751 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
752 struct dpif_flow_stats *stats)
754 stats->n_packets = netdev_flow->packet_count;
755 stats->n_bytes = netdev_flow->byte_count;
756 stats->used = netdev_flow->used;
757 stats->tcp_flags = netdev_flow->tcp_flags;
761 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
766 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
767 /* This should not happen: it indicates that odp_flow_key_from_flow()
768 * and odp_flow_key_to_flow() disagree on the acceptable form of a
769 * flow. Log the problem as an error, with enough details to enable
771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
773 if (!VLOG_DROP_ERR(&rl)) {
777 odp_flow_key_format(key, key_len, &s);
778 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
785 in_port = flow->in_port.odp_port;
786 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
794 dpif_netdev_flow_get(const struct dpif *dpif,
795 const struct nlattr *nl_key, size_t nl_key_len,
796 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
798 struct dp_netdev *dp = get_dp_netdev(dpif);
799 struct dp_netdev_flow *netdev_flow;
803 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
808 ovs_mutex_lock(&dp_netdev_mutex);
809 netdev_flow = dp_netdev_lookup_flow(dp, &key);
812 get_dpif_flow_stats(netdev_flow, stats);
815 *actionsp = ofpbuf_clone_data(netdev_flow->actions,
816 netdev_flow->actions_len);
821 ovs_mutex_unlock(&dp_netdev_mutex);
827 set_flow_actions(struct dp_netdev_flow *netdev_flow,
828 const struct nlattr *actions, size_t actions_len)
830 netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
831 netdev_flow->actions_len = actions_len;
832 memcpy(netdev_flow->actions, actions, actions_len);
837 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
838 const struct nlattr *actions, size_t actions_len)
840 struct dp_netdev_flow *netdev_flow;
843 netdev_flow = xzalloc(sizeof *netdev_flow);
844 netdev_flow->key = *key;
846 error = set_flow_actions(netdev_flow, actions, actions_len);
852 hmap_insert(&dp->flow_table, &netdev_flow->node,
853 flow_hash(&netdev_flow->key, 0));
858 clear_stats(struct dp_netdev_flow *netdev_flow)
860 netdev_flow->used = 0;
861 netdev_flow->packet_count = 0;
862 netdev_flow->byte_count = 0;
863 netdev_flow->tcp_flags = 0;
867 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
869 struct dp_netdev *dp = get_dp_netdev(dpif);
870 struct dp_netdev_flow *netdev_flow;
874 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
879 ovs_mutex_lock(&dp_netdev_mutex);
880 netdev_flow = dp_netdev_lookup_flow(dp, &key);
882 if (put->flags & DPIF_FP_CREATE) {
883 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
885 memset(put->stats, 0, sizeof *put->stats);
887 error = dp_netdev_flow_add(dp, &key, put->actions,
896 if (put->flags & DPIF_FP_MODIFY) {
897 error = set_flow_actions(netdev_flow, put->actions,
901 get_dpif_flow_stats(netdev_flow, put->stats);
903 if (put->flags & DPIF_FP_ZERO_STATS) {
904 clear_stats(netdev_flow);
911 ovs_mutex_unlock(&dp_netdev_mutex);
917 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
919 struct dp_netdev *dp = get_dp_netdev(dpif);
920 struct dp_netdev_flow *netdev_flow;
924 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
929 ovs_mutex_lock(&dp_netdev_mutex);
930 netdev_flow = dp_netdev_lookup_flow(dp, &key);
933 get_dpif_flow_stats(netdev_flow, del->stats);
935 dp_netdev_free_flow(dp, netdev_flow);
939 ovs_mutex_unlock(&dp_netdev_mutex);
944 struct dp_netdev_flow_state {
947 struct nlattr *actions;
948 struct odputil_keybuf keybuf;
949 struct dpif_flow_stats stats;
953 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
955 struct dp_netdev_flow_state *state;
957 *statep = state = xmalloc(sizeof *state);
960 state->actions = NULL;
965 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
966 const struct nlattr **key, size_t *key_len,
967 const struct nlattr **mask, size_t *mask_len,
968 const struct nlattr **actions, size_t *actions_len,
969 const struct dpif_flow_stats **stats)
971 struct dp_netdev_flow_state *state = state_;
972 struct dp_netdev *dp = get_dp_netdev(dpif);
973 struct dp_netdev_flow *netdev_flow;
974 struct hmap_node *node;
976 ovs_mutex_lock(&dp_netdev_mutex);
977 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
979 ovs_mutex_unlock(&dp_netdev_mutex);
983 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
988 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
989 odp_flow_key_from_flow(&buf, &netdev_flow->key,
990 netdev_flow->key.in_port.odp_port);
1002 free(state->actions);
1003 state->actions = xmemdup(netdev_flow->actions,
1004 netdev_flow->actions_len);
1006 *actions = state->actions;
1007 *actions_len = netdev_flow->actions_len;
1011 get_dpif_flow_stats(netdev_flow, &state->stats);
1012 *stats = &state->stats;
1015 ovs_mutex_unlock(&dp_netdev_mutex);
1020 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1022 struct dp_netdev_flow_state *state = state_;
1024 free(state->actions);
1030 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
1032 struct dp_netdev *dp = get_dp_netdev(dpif);
1037 if (execute->packet->size < ETH_HEADER_LEN ||
1038 execute->packet->size > UINT16_MAX) {
1042 /* Make a deep copy of 'packet', because we might modify its data. */
1043 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
1044 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
1045 ofpbuf_put(©, execute->packet->data, execute->packet->size);
1047 flow_extract(©, 0, 0, NULL, NULL, &key);
1048 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
1051 ovs_mutex_lock(&dp_netdev_mutex);
1052 dp_netdev_execute_actions(dp, ©, &key,
1053 execute->actions, execute->actions_len);
1054 ovs_mutex_unlock(&dp_netdev_mutex);
1057 ofpbuf_uninit(©);
1062 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1068 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1069 uint32_t queue_id, uint32_t *priority)
1071 *priority = queue_id;
1075 static struct dp_netdev_queue *
1076 find_nonempty_queue(struct dpif *dpif)
1078 struct dp_netdev *dp = get_dp_netdev(dpif);
1081 for (i = 0; i < N_QUEUES; i++) {
1082 struct dp_netdev_queue *q = &dp->queues[i];
1083 if (q->head != q->tail) {
1091 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1094 struct dp_netdev_queue *q;
1097 ovs_mutex_lock(&dp_netdev_mutex);
1098 q = find_nonempty_queue(dpif);
1100 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1102 *upcall = u->upcall;
1103 upcall->packet = buf;
1112 ovs_mutex_unlock(&dp_netdev_mutex);
1118 dpif_netdev_recv_wait(struct dpif *dpif)
1120 struct dp_netdev *dp = get_dp_netdev(dpif);
1123 ovs_mutex_lock(&dp_netdev_mutex);
1124 seq = seq_read(dp->queue_seq);
1125 if (find_nonempty_queue(dpif)) {
1126 poll_immediate_wake();
1128 seq_wait(dp->queue_seq, seq);
1130 ovs_mutex_unlock(&dp_netdev_mutex);
1134 dpif_netdev_recv_purge(struct dpif *dpif)
1136 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1137 ovs_mutex_lock(&dp_netdev_mutex);
1138 dp_netdev_purge_queues(dpif_netdev->dp);
1139 ovs_mutex_unlock(&dp_netdev_mutex);
1143 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1144 const struct ofpbuf *packet)
1146 netdev_flow->used = time_msec();
1147 netdev_flow->packet_count++;
1148 netdev_flow->byte_count += packet->size;
1149 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->key);
1153 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1154 struct ofpbuf *packet, uint32_t skb_priority,
1155 uint32_t pkt_mark, const struct flow_tnl *tnl)
1157 struct dp_netdev_flow *netdev_flow;
1159 union flow_in_port in_port_;
1161 if (packet->size < ETH_HEADER_LEN) {
1164 in_port_.odp_port = port->port_no;
1165 flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
1166 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1168 dp_netdev_flow_used(netdev_flow, packet);
1169 dp_netdev_execute_actions(dp, packet, &key,
1170 netdev_flow->actions,
1171 netdev_flow->actions_len);
1175 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1180 dpif_netdev_run(struct dpif *dpif)
1182 struct dp_netdev_port *port;
1183 struct dp_netdev *dp;
1184 struct ofpbuf packet;
1186 ovs_mutex_lock(&dp_netdev_mutex);
1187 dp = get_dp_netdev(dpif);
1188 ofpbuf_init(&packet,
1189 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1191 LIST_FOR_EACH (port, node, &dp->port_list) {
1194 /* Reset packet contents. */
1195 ofpbuf_clear(&packet);
1196 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1198 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1200 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1201 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1202 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1204 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1205 netdev_get_name(port->netdev), ovs_strerror(error));
1208 ofpbuf_uninit(&packet);
1209 ovs_mutex_unlock(&dp_netdev_mutex);
1213 dpif_netdev_wait(struct dpif *dpif)
1215 struct dp_netdev_port *port;
1217 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1218 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1219 * A makes it to poll_block().
1221 * But I think it doesn't matter:
1223 * - In the dpif_port_add() case, A will not wake up when a packet
1224 * arrives on the new port, but this would also happen if the
1225 * ordering were reversed.
1227 * - In the dpif_port_remove() case, A might wake up spuriously, but
1228 * that is harmless. */
1230 ovs_mutex_lock(&dp_netdev_mutex);
1231 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1233 netdev_rx_wait(port->rx);
1236 ovs_mutex_unlock(&dp_netdev_mutex);
1240 dp_netdev_output_port(void *dp_, struct ofpbuf *packet,
1241 const struct flow *flow OVS_UNUSED, odp_port_t out_port)
1243 struct dp_netdev *dp = dp_;
1244 struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
1246 netdev_send(p->netdev, packet);
1251 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1252 int queue_no, const struct flow *flow,
1253 const struct nlattr *userdata)
1255 struct dp_netdev_queue *q = &dp->queues[queue_no];
1256 if (q->head - q->tail < MAX_QUEUE_LEN) {
1257 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1258 struct dpif_upcall *upcall = &u->upcall;
1259 struct ofpbuf *buf = &u->buf;
1262 upcall->type = queue_no;
1264 /* Allocate buffer big enough for everything. */
1265 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1267 buf_size += NLA_ALIGN(userdata->nla_len);
1269 ofpbuf_init(buf, buf_size);
1272 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1273 upcall->key = buf->data;
1274 upcall->key_len = buf->size;
1278 upcall->userdata = ofpbuf_put(buf, userdata,
1279 NLA_ALIGN(userdata->nla_len));
1284 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1285 * is visible in 'upcall->packet'. The ODP flow and (if present)
1286 * userdata become part of the headroom. */
1287 ofpbuf_put_zeros(buf, 2);
1288 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1289 buf->size = packet->size;
1290 upcall->packet = buf;
1292 seq_change(dp->queue_seq);
1302 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1303 const struct flow *key,
1304 const struct nlattr *a)
1306 const struct nlattr *userdata;
1308 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1309 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1313 dp_netdev_execute_actions(struct dp_netdev *dp,
1314 struct ofpbuf *packet, struct flow *key,
1315 const struct nlattr *actions,
1318 odp_execute_actions(dp, packet, key, actions, actions_len,
1319 dp_netdev_output_port, dp_netdev_action_userspace);
1322 const struct dpif_class dpif_netdev_class = {
1324 dpif_netdev_enumerate,
1325 dpif_netdev_port_open_type,
1328 dpif_netdev_destroy,
1331 dpif_netdev_get_stats,
1332 dpif_netdev_port_add,
1333 dpif_netdev_port_del,
1334 dpif_netdev_port_query_by_number,
1335 dpif_netdev_port_query_by_name,
1336 dpif_netdev_get_max_ports,
1337 NULL, /* port_get_pid */
1338 dpif_netdev_port_dump_start,
1339 dpif_netdev_port_dump_next,
1340 dpif_netdev_port_dump_done,
1341 dpif_netdev_port_poll,
1342 dpif_netdev_port_poll_wait,
1343 dpif_netdev_flow_get,
1344 dpif_netdev_flow_put,
1345 dpif_netdev_flow_del,
1346 dpif_netdev_flow_flush,
1347 dpif_netdev_flow_dump_start,
1348 dpif_netdev_flow_dump_next,
1349 dpif_netdev_flow_dump_done,
1350 dpif_netdev_execute,
1352 dpif_netdev_recv_set,
1353 dpif_netdev_queue_to_priority,
1355 dpif_netdev_recv_wait,
1356 dpif_netdev_recv_purge,
1360 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1361 const char *argv[], void *aux OVS_UNUSED)
1363 struct dp_netdev_port *port;
1364 struct dp_netdev *dp;
1367 dp = shash_find_data(&dp_netdevs, argv[1]);
1368 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1369 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1373 if (get_port_by_name(dp, argv[2], &port)) {
1374 unixctl_command_reply_error(conn, "unknown port");
1378 port_no = atoi(argv[3]);
1379 if (port_no <= 0 || port_no >= MAX_PORTS) {
1380 unixctl_command_reply_error(conn, "bad port number");
1383 if (dp->ports[port_no]) {
1384 unixctl_command_reply_error(conn, "port number already in use");
1387 dp->ports[odp_to_u32(port->port_no)] = NULL;
1388 dp->ports[port_no] = port;
1389 port->port_no = u32_to_odp(port_no);
1390 seq_change(dp->port_seq);
1391 unixctl_command_reply(conn, NULL);
1395 dpif_dummy_register__(const char *type)
1397 struct dpif_class *class;
1399 class = xmalloc(sizeof *class);
1400 *class = dpif_netdev_class;
1401 class->type = xstrdup(type);
1402 dp_register_provider(class);
1406 dpif_dummy_register(bool override)
1413 dp_enumerate_types(&types);
1414 SSET_FOR_EACH (type, &types) {
1415 if (!dp_unregister_provider(type)) {
1416 dpif_dummy_register__(type);
1419 sset_destroy(&types);
1422 dpif_dummy_register__("dummy");
1424 unixctl_command_register("dpif-dummy/change-port-number",
1425 "DP PORT NEW-NUMBER",
1426 3, 3, dpif_dummy_change_port_number, NULL);