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_class_is_planetlab(const struct dpif_class *class)
207 return class == &dpif_planetlab_class;
211 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
213 return strcmp(type, "internal") ? type
214 : dpif_netdev_class_is_planetlab(class) ? "pltap"
215 : dpif_netdev_class_is_dummy(class) ? "dummy"
220 create_dpif_netdev(struct dp_netdev *dp)
222 uint16_t netflow_id = hash_string(dp->name, 0);
223 struct dpif_netdev *dpif;
227 dpif = xmalloc(sizeof *dpif);
228 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
230 dpif->last_port_seq = seq_read(dp->port_seq);
235 /* Choose an unused, non-zero port number and return it on success.
236 * Return ODPP_NONE on failure. */
238 choose_port(struct dp_netdev *dp, const char *name)
242 if (dp->class != &dpif_netdev_class &&
243 dp->class != &dpif_planetlab_class) {
247 /* If the port name begins with "br", start the number search at
248 * 100 to make writing tests easier. */
249 if (!strncmp(name, "br", 2)) {
253 /* If the port name contains a number, try to assign that port number.
254 * This can make writing unit tests easier because port numbers are
256 for (p = name; *p != '\0'; p++) {
257 if (isdigit((unsigned char) *p)) {
258 port_no = start_no + strtol(p, NULL, 10);
259 if (port_no > 0 && port_no < MAX_PORTS
260 && !dp->ports[port_no]) {
261 return u32_to_odp(port_no);
268 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
269 if (!dp->ports[port_no]) {
270 return u32_to_odp(port_no);
278 create_dp_netdev(const char *name, const struct dpif_class *class,
279 struct dp_netdev **dpp)
281 struct dp_netdev *dp;
285 dp = xzalloc(sizeof *dp);
287 dp->name = xstrdup(name);
289 dp->max_mtu = ETH_PAYLOAD_MAX;
290 for (i = 0; i < N_QUEUES; i++) {
291 dp->queues[i].head = dp->queues[i].tail = 0;
293 dp->queue_seq = seq_create();
294 hmap_init(&dp->flow_table);
295 list_init(&dp->port_list);
296 dp->port_seq = seq_create();
298 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
304 shash_add(&dp_netdevs, name, dp);
311 dpif_netdev_open(const struct dpif_class *class, const char *name,
312 bool create, struct dpif **dpifp)
314 struct dp_netdev *dp;
317 ovs_mutex_lock(&dp_netdev_mutex);
318 dp = shash_find_data(&dp_netdevs, name);
320 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
322 error = (dp->class != class ? EINVAL
327 *dpifp = create_dpif_netdev(dp);
329 ovs_mutex_unlock(&dp_netdev_mutex);
335 dp_netdev_purge_queues(struct dp_netdev *dp)
339 for (i = 0; i < N_QUEUES; i++) {
340 struct dp_netdev_queue *q = &dp->queues[i];
342 while (q->tail != q->head) {
343 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
344 ofpbuf_uninit(&u->buf);
350 dp_netdev_free(struct dp_netdev *dp)
352 struct dp_netdev_port *port, *next;
354 dp_netdev_flow_flush(dp);
355 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
356 do_del_port(dp, port->port_no);
358 dp_netdev_purge_queues(dp);
359 seq_destroy(dp->queue_seq);
360 hmap_destroy(&dp->flow_table);
361 seq_destroy(dp->port_seq);
367 dpif_netdev_close(struct dpif *dpif)
369 struct dp_netdev *dp = get_dp_netdev(dpif);
371 ovs_mutex_lock(&dp_netdev_mutex);
373 ovs_assert(dp->open_cnt > 0);
374 if (--dp->open_cnt == 0 && dp->destroyed) {
375 shash_find_and_delete(&dp_netdevs, dp->name);
380 ovs_mutex_unlock(&dp_netdev_mutex);
384 dpif_netdev_destroy(struct dpif *dpif)
386 struct dp_netdev *dp = get_dp_netdev(dpif);
388 ovs_mutex_lock(&dp_netdev_mutex);
389 dp->destroyed = true;
390 ovs_mutex_unlock(&dp_netdev_mutex);
396 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
398 struct dp_netdev *dp = get_dp_netdev(dpif);
400 ovs_mutex_lock(&dp_netdev_mutex);
401 stats->n_flows = hmap_count(&dp->flow_table);
402 stats->n_hit = dp->n_hit;
403 stats->n_missed = dp->n_missed;
404 stats->n_lost = dp->n_lost;
405 stats->n_masks = UINT64_MAX;
406 stats->n_mask_hit = UINT64_MAX;
407 ovs_mutex_unlock(&dp_netdev_mutex);
413 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
416 struct netdev_saved_flags *sf;
417 struct dp_netdev_port *port;
418 struct netdev *netdev;
419 struct netdev_rx *rx;
420 enum netdev_flags flags;
421 const char *open_type;
425 /* XXX reject devices already in some dp_netdev. */
427 /* Open and validate network device. */
428 open_type = dpif_netdev_port_open_type(dp->class, type);
429 error = netdev_open(devname, open_type, &netdev);
433 /* XXX reject non-Ethernet devices */
435 netdev_get_flags(netdev, &flags);
436 if (flags & NETDEV_LOOPBACK) {
437 VLOG_ERR("%s: cannot add a loopback device", devname);
438 netdev_close(netdev);
442 error = netdev_rx_open(netdev, &rx);
444 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
445 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
446 devname, ovs_strerror(errno));
447 netdev_close(netdev);
451 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
454 netdev_close(netdev);
458 port = xmalloc(sizeof *port);
459 port->port_no = port_no;
460 port->netdev = netdev;
463 port->type = xstrdup(type);
465 error = netdev_get_mtu(netdev, &mtu);
466 if (!error && mtu > dp->max_mtu) {
470 list_push_back(&dp->port_list, &port->node);
471 dp->ports[odp_to_u32(port_no)] = port;
472 seq_change(dp->port_seq);
478 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
479 odp_port_t *port_nop)
481 struct dp_netdev *dp = get_dp_netdev(dpif);
482 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
483 const char *dpif_port;
487 ovs_mutex_lock(&dp_netdev_mutex);
488 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
489 if (*port_nop != ODPP_NONE) {
490 uint32_t port_idx = odp_to_u32(*port_nop);
491 if (port_idx >= MAX_PORTS) {
493 } else if (dp->ports[port_idx]) {
500 port_no = choose_port(dp, dpif_port);
501 error = port_no == ODPP_NONE ? EFBIG : 0;
505 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
507 ovs_mutex_unlock(&dp_netdev_mutex);
513 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
515 struct dp_netdev *dp = get_dp_netdev(dpif);
518 ovs_mutex_lock(&dp_netdev_mutex);
519 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
520 ovs_mutex_unlock(&dp_netdev_mutex);
526 is_valid_port_number(odp_port_t port_no)
528 return odp_to_u32(port_no) < MAX_PORTS;
532 get_port_by_number(struct dp_netdev *dp,
533 odp_port_t port_no, struct dp_netdev_port **portp)
535 if (!is_valid_port_number(port_no)) {
539 *portp = dp->ports[odp_to_u32(port_no)];
540 return *portp ? 0 : ENOENT;
545 get_port_by_name(struct dp_netdev *dp,
546 const char *devname, struct dp_netdev_port **portp)
548 struct dp_netdev_port *port;
550 LIST_FOR_EACH (port, node, &dp->port_list) {
551 if (!strcmp(netdev_get_name(port->netdev), devname)) {
560 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
562 struct dp_netdev_port *port;
565 error = get_port_by_number(dp, port_no, &port);
570 list_remove(&port->node);
571 dp->ports[odp_to_u32(port_no)] = NULL;
572 seq_change(dp->port_seq);
574 netdev_close(port->netdev);
575 netdev_restore_flags(port->sf);
576 netdev_rx_close(port->rx);
584 answer_port_query(const struct dp_netdev_port *port,
585 struct dpif_port *dpif_port)
587 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
588 dpif_port->type = xstrdup(port->type);
589 dpif_port->port_no = port->port_no;
593 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
594 struct dpif_port *dpif_port)
596 struct dp_netdev *dp = get_dp_netdev(dpif);
597 struct dp_netdev_port *port;
600 ovs_mutex_lock(&dp_netdev_mutex);
601 error = get_port_by_number(dp, port_no, &port);
602 if (!error && dpif_port) {
603 answer_port_query(port, dpif_port);
605 ovs_mutex_unlock(&dp_netdev_mutex);
611 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
612 struct dpif_port *dpif_port)
614 struct dp_netdev *dp = get_dp_netdev(dpif);
615 struct dp_netdev_port *port;
618 ovs_mutex_lock(&dp_netdev_mutex);
619 error = get_port_by_name(dp, devname, &port);
620 if (!error && dpif_port) {
621 answer_port_query(port, dpif_port);
623 ovs_mutex_unlock(&dp_netdev_mutex);
629 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
635 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
637 hmap_remove(&dp->flow_table, &netdev_flow->node);
638 free(netdev_flow->actions);
643 dp_netdev_flow_flush(struct dp_netdev *dp)
645 struct dp_netdev_flow *netdev_flow, *next;
647 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
648 dp_netdev_free_flow(dp, netdev_flow);
653 dpif_netdev_flow_flush(struct dpif *dpif)
655 struct dp_netdev *dp = get_dp_netdev(dpif);
657 ovs_mutex_lock(&dp_netdev_mutex);
658 dp_netdev_flow_flush(dp);
659 ovs_mutex_unlock(&dp_netdev_mutex);
664 struct dp_netdev_port_state {
670 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
672 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
677 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
678 struct dpif_port *dpif_port)
680 struct dp_netdev_port_state *state = state_;
681 struct dp_netdev *dp = get_dp_netdev(dpif);
684 ovs_mutex_lock(&dp_netdev_mutex);
685 for (port_idx = odp_to_u32(state->port_no);
686 port_idx < MAX_PORTS; port_idx++) {
687 struct dp_netdev_port *port = dp->ports[port_idx];
690 state->name = xstrdup(netdev_get_name(port->netdev));
691 dpif_port->name = state->name;
692 dpif_port->type = port->type;
693 dpif_port->port_no = port->port_no;
694 state->port_no = u32_to_odp(port_idx + 1);
695 ovs_mutex_unlock(&dp_netdev_mutex);
700 ovs_mutex_unlock(&dp_netdev_mutex);
706 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
708 struct dp_netdev_port_state *state = state_;
715 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
717 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
718 uint64_t new_port_seq;
721 ovs_mutex_lock(&dp_netdev_mutex);
722 new_port_seq = seq_read(dpif->dp->port_seq);
723 if (dpif->last_port_seq != new_port_seq) {
724 dpif->last_port_seq = new_port_seq;
729 ovs_mutex_unlock(&dp_netdev_mutex);
735 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
737 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
739 ovs_mutex_lock(&dp_netdev_mutex);
740 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
741 ovs_mutex_unlock(&dp_netdev_mutex);
744 static struct dp_netdev_flow *
745 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
747 struct dp_netdev_flow *netdev_flow;
749 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(key, 0),
751 if (flow_equal(&netdev_flow->key, key)) {
759 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
760 struct dpif_flow_stats *stats)
762 stats->n_packets = netdev_flow->packet_count;
763 stats->n_bytes = netdev_flow->byte_count;
764 stats->used = netdev_flow->used;
765 stats->tcp_flags = netdev_flow->tcp_flags;
769 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
774 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
775 /* This should not happen: it indicates that odp_flow_key_from_flow()
776 * and odp_flow_key_to_flow() disagree on the acceptable form of a
777 * flow. Log the problem as an error, with enough details to enable
779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
781 if (!VLOG_DROP_ERR(&rl)) {
785 odp_flow_key_format(key, key_len, &s);
786 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
793 in_port = flow->in_port.odp_port;
794 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
802 dpif_netdev_flow_get(const struct dpif *dpif,
803 const struct nlattr *nl_key, size_t nl_key_len,
804 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
806 struct dp_netdev *dp = get_dp_netdev(dpif);
807 struct dp_netdev_flow *netdev_flow;
811 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
816 ovs_mutex_lock(&dp_netdev_mutex);
817 netdev_flow = dp_netdev_lookup_flow(dp, &key);
820 get_dpif_flow_stats(netdev_flow, stats);
823 *actionsp = ofpbuf_clone_data(netdev_flow->actions,
824 netdev_flow->actions_len);
829 ovs_mutex_unlock(&dp_netdev_mutex);
835 set_flow_actions(struct dp_netdev_flow *netdev_flow,
836 const struct nlattr *actions, size_t actions_len)
838 netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
839 netdev_flow->actions_len = actions_len;
840 memcpy(netdev_flow->actions, actions, actions_len);
845 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
846 const struct nlattr *actions, size_t actions_len)
848 struct dp_netdev_flow *netdev_flow;
851 netdev_flow = xzalloc(sizeof *netdev_flow);
852 netdev_flow->key = *key;
854 error = set_flow_actions(netdev_flow, actions, actions_len);
860 hmap_insert(&dp->flow_table, &netdev_flow->node,
861 flow_hash(&netdev_flow->key, 0));
866 clear_stats(struct dp_netdev_flow *netdev_flow)
868 netdev_flow->used = 0;
869 netdev_flow->packet_count = 0;
870 netdev_flow->byte_count = 0;
871 netdev_flow->tcp_flags = 0;
875 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
877 struct dp_netdev *dp = get_dp_netdev(dpif);
878 struct dp_netdev_flow *netdev_flow;
882 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
887 ovs_mutex_lock(&dp_netdev_mutex);
888 netdev_flow = dp_netdev_lookup_flow(dp, &key);
890 if (put->flags & DPIF_FP_CREATE) {
891 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
893 memset(put->stats, 0, sizeof *put->stats);
895 error = dp_netdev_flow_add(dp, &key, put->actions,
904 if (put->flags & DPIF_FP_MODIFY) {
905 error = set_flow_actions(netdev_flow, put->actions,
909 get_dpif_flow_stats(netdev_flow, put->stats);
911 if (put->flags & DPIF_FP_ZERO_STATS) {
912 clear_stats(netdev_flow);
919 ovs_mutex_unlock(&dp_netdev_mutex);
925 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
927 struct dp_netdev *dp = get_dp_netdev(dpif);
928 struct dp_netdev_flow *netdev_flow;
932 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
937 ovs_mutex_lock(&dp_netdev_mutex);
938 netdev_flow = dp_netdev_lookup_flow(dp, &key);
941 get_dpif_flow_stats(netdev_flow, del->stats);
943 dp_netdev_free_flow(dp, netdev_flow);
947 ovs_mutex_unlock(&dp_netdev_mutex);
952 struct dp_netdev_flow_state {
955 struct nlattr *actions;
956 struct odputil_keybuf keybuf;
957 struct dpif_flow_stats stats;
961 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
963 struct dp_netdev_flow_state *state;
965 *statep = state = xmalloc(sizeof *state);
968 state->actions = NULL;
973 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
974 const struct nlattr **key, size_t *key_len,
975 const struct nlattr **mask, size_t *mask_len,
976 const struct nlattr **actions, size_t *actions_len,
977 const struct dpif_flow_stats **stats)
979 struct dp_netdev_flow_state *state = state_;
980 struct dp_netdev *dp = get_dp_netdev(dpif);
981 struct dp_netdev_flow *netdev_flow;
982 struct hmap_node *node;
984 ovs_mutex_lock(&dp_netdev_mutex);
985 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
987 ovs_mutex_unlock(&dp_netdev_mutex);
991 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
996 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
997 odp_flow_key_from_flow(&buf, &netdev_flow->key,
998 netdev_flow->key.in_port.odp_port);
1001 *key_len = buf.size;
1010 free(state->actions);
1011 state->actions = xmemdup(netdev_flow->actions,
1012 netdev_flow->actions_len);
1014 *actions = state->actions;
1015 *actions_len = netdev_flow->actions_len;
1019 get_dpif_flow_stats(netdev_flow, &state->stats);
1020 *stats = &state->stats;
1023 ovs_mutex_unlock(&dp_netdev_mutex);
1028 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1030 struct dp_netdev_flow_state *state = state_;
1032 free(state->actions);
1038 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
1040 struct dp_netdev *dp = get_dp_netdev(dpif);
1045 if (execute->packet->size < ETH_HEADER_LEN ||
1046 execute->packet->size > UINT16_MAX) {
1050 /* Make a deep copy of 'packet', because we might modify its data. */
1051 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
1052 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
1053 ofpbuf_put(©, execute->packet->data, execute->packet->size);
1055 flow_extract(©, 0, 0, NULL, NULL, &key);
1056 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
1059 ovs_mutex_lock(&dp_netdev_mutex);
1060 dp_netdev_execute_actions(dp, ©, &key,
1061 execute->actions, execute->actions_len);
1062 ovs_mutex_unlock(&dp_netdev_mutex);
1065 ofpbuf_uninit(©);
1070 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1076 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1077 uint32_t queue_id, uint32_t *priority)
1079 *priority = queue_id;
1083 static struct dp_netdev_queue *
1084 find_nonempty_queue(struct dpif *dpif)
1086 struct dp_netdev *dp = get_dp_netdev(dpif);
1089 for (i = 0; i < N_QUEUES; i++) {
1090 struct dp_netdev_queue *q = &dp->queues[i];
1091 if (q->head != q->tail) {
1099 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1102 struct dp_netdev_queue *q;
1105 ovs_mutex_lock(&dp_netdev_mutex);
1106 q = find_nonempty_queue(dpif);
1108 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1110 *upcall = u->upcall;
1111 upcall->packet = buf;
1120 ovs_mutex_unlock(&dp_netdev_mutex);
1126 dpif_netdev_recv_wait(struct dpif *dpif)
1128 struct dp_netdev *dp = get_dp_netdev(dpif);
1131 ovs_mutex_lock(&dp_netdev_mutex);
1132 seq = seq_read(dp->queue_seq);
1133 if (find_nonempty_queue(dpif)) {
1134 poll_immediate_wake();
1136 seq_wait(dp->queue_seq, seq);
1138 ovs_mutex_unlock(&dp_netdev_mutex);
1142 dpif_netdev_recv_purge(struct dpif *dpif)
1144 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1145 ovs_mutex_lock(&dp_netdev_mutex);
1146 dp_netdev_purge_queues(dpif_netdev->dp);
1147 ovs_mutex_unlock(&dp_netdev_mutex);
1151 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1152 const struct ofpbuf *packet)
1154 netdev_flow->used = time_msec();
1155 netdev_flow->packet_count++;
1156 netdev_flow->byte_count += packet->size;
1157 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->key);
1161 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1162 struct ofpbuf *packet, uint32_t skb_priority,
1163 uint32_t pkt_mark, const struct flow_tnl *tnl)
1165 struct dp_netdev_flow *netdev_flow;
1167 union flow_in_port in_port_;
1169 if (packet->size < ETH_HEADER_LEN) {
1172 in_port_.odp_port = port->port_no;
1173 flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
1174 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1176 dp_netdev_flow_used(netdev_flow, packet);
1177 dp_netdev_execute_actions(dp, packet, &key,
1178 netdev_flow->actions,
1179 netdev_flow->actions_len);
1183 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1188 dpif_netdev_run(struct dpif *dpif)
1190 struct dp_netdev_port *port;
1191 struct dp_netdev *dp;
1192 struct ofpbuf packet;
1194 ovs_mutex_lock(&dp_netdev_mutex);
1195 dp = get_dp_netdev(dpif);
1196 ofpbuf_init(&packet,
1197 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1199 LIST_FOR_EACH (port, node, &dp->port_list) {
1202 /* Reset packet contents. */
1203 ofpbuf_clear(&packet);
1204 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1206 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1208 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1209 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1210 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1212 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1213 netdev_get_name(port->netdev), ovs_strerror(error));
1216 ofpbuf_uninit(&packet);
1217 ovs_mutex_unlock(&dp_netdev_mutex);
1221 dpif_netdev_wait(struct dpif *dpif)
1223 struct dp_netdev_port *port;
1225 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1226 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1227 * A makes it to poll_block().
1229 * But I think it doesn't matter:
1231 * - In the dpif_port_add() case, A will not wake up when a packet
1232 * arrives on the new port, but this would also happen if the
1233 * ordering were reversed.
1235 * - In the dpif_port_remove() case, A might wake up spuriously, but
1236 * that is harmless. */
1238 ovs_mutex_lock(&dp_netdev_mutex);
1239 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1241 netdev_rx_wait(port->rx);
1244 ovs_mutex_unlock(&dp_netdev_mutex);
1248 dp_netdev_output_port(void *dp_, struct ofpbuf *packet,
1249 const struct flow *flow OVS_UNUSED, odp_port_t out_port)
1251 struct dp_netdev *dp = dp_;
1252 struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
1254 netdev_send(p->netdev, packet);
1259 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1260 int queue_no, const struct flow *flow,
1261 const struct nlattr *userdata)
1263 struct dp_netdev_queue *q = &dp->queues[queue_no];
1264 if (q->head - q->tail < MAX_QUEUE_LEN) {
1265 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1266 struct dpif_upcall *upcall = &u->upcall;
1267 struct ofpbuf *buf = &u->buf;
1270 upcall->type = queue_no;
1272 /* Allocate buffer big enough for everything. */
1273 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1275 buf_size += NLA_ALIGN(userdata->nla_len);
1277 ofpbuf_init(buf, buf_size);
1280 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1281 upcall->key = buf->data;
1282 upcall->key_len = buf->size;
1286 upcall->userdata = ofpbuf_put(buf, userdata,
1287 NLA_ALIGN(userdata->nla_len));
1292 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1293 * is visible in 'upcall->packet'. The ODP flow and (if present)
1294 * userdata become part of the headroom. */
1295 ofpbuf_put_zeros(buf, 2);
1296 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1297 buf->size = packet->size;
1298 upcall->packet = buf;
1300 seq_change(dp->queue_seq);
1310 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1311 const struct flow *key,
1312 const struct nlattr *a)
1314 const struct nlattr *userdata;
1316 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1317 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1321 dp_netdev_execute_actions(struct dp_netdev *dp,
1322 struct ofpbuf *packet, struct flow *key,
1323 const struct nlattr *actions,
1326 odp_execute_actions(dp, packet, key, actions, actions_len,
1327 dp_netdev_output_port, dp_netdev_action_userspace);
1330 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1331 dpif_netdev_enumerate, \
1332 dpif_netdev_port_open_type, \
1334 dpif_netdev_close, \
1335 dpif_netdev_destroy, \
1338 dpif_netdev_get_stats, \
1339 dpif_netdev_port_add, \
1340 dpif_netdev_port_del, \
1341 dpif_netdev_port_query_by_number, \
1342 dpif_netdev_port_query_by_name, \
1343 dpif_netdev_get_max_ports, \
1344 NULL, /* port_get_pid */ \
1345 dpif_netdev_port_dump_start, \
1346 dpif_netdev_port_dump_next, \
1347 dpif_netdev_port_dump_done, \
1348 dpif_netdev_port_poll, \
1349 dpif_netdev_port_poll_wait, \
1350 dpif_netdev_flow_get, \
1351 dpif_netdev_flow_put, \
1352 dpif_netdev_flow_del, \
1353 dpif_netdev_flow_flush, \
1354 dpif_netdev_flow_dump_start, \
1355 dpif_netdev_flow_dump_next, \
1356 dpif_netdev_flow_dump_done, \
1357 dpif_netdev_execute, \
1358 NULL, /* operate */ \
1359 dpif_netdev_recv_set, \
1360 dpif_netdev_queue_to_priority, \
1362 dpif_netdev_recv_wait, \
1363 dpif_netdev_recv_purge, \
1365 const struct dpif_class dpif_netdev_class = {
1367 DPIF_NETDEV_CLASS_FUNCTIONS
1370 const struct dpif_class dpif_planetlab_class = {
1372 DPIF_NETDEV_CLASS_FUNCTIONS
1376 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1377 const char *argv[], void *aux OVS_UNUSED)
1379 struct dp_netdev_port *port;
1380 struct dp_netdev *dp;
1383 dp = shash_find_data(&dp_netdevs, argv[1]);
1384 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1385 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1389 if (get_port_by_name(dp, argv[2], &port)) {
1390 unixctl_command_reply_error(conn, "unknown port");
1394 port_no = atoi(argv[3]);
1395 if (port_no <= 0 || port_no >= MAX_PORTS) {
1396 unixctl_command_reply_error(conn, "bad port number");
1399 if (dp->ports[port_no]) {
1400 unixctl_command_reply_error(conn, "port number already in use");
1403 dp->ports[odp_to_u32(port->port_no)] = NULL;
1404 dp->ports[port_no] = port;
1405 port->port_no = u32_to_odp(port_no);
1406 seq_change(dp->port_seq);
1407 unixctl_command_reply(conn, NULL);
1411 dpif_dummy_register__(const char *type)
1413 struct dpif_class *class;
1415 class = xmalloc(sizeof *class);
1416 *class = dpif_netdev_class;
1417 class->type = xstrdup(type);
1418 dp_register_provider(class);
1422 dpif_dummy_register(bool override)
1429 dp_enumerate_types(&types);
1430 SSET_FOR_EACH (type, &types) {
1431 if (!dp_unregister_provider(type)) {
1432 dpif_dummy_register__(type);
1435 sset_destroy(&types);
1438 dpif_dummy_register__("dummy");
1440 unixctl_command_register("dpif-dummy/change-port-number",
1441 "DP PORT NEW-NUMBER",
1442 3, 3, dpif_dummy_change_port_number, NULL);