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>
34 #include "classifier.h"
37 #include "dpif-provider.h"
39 #include "dynamic-string.h"
44 #include "netdev-vport.h"
46 #include "odp-execute.h"
48 #include "ofp-print.h"
51 #include "poll-loop.h"
61 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
63 /* By default, choose a priority in the middle. */
64 #define NETDEV_RULE_PRIORITY 0x8000
66 /* Configuration parameters. */
67 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
68 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
70 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
71 * headers to be aligned on a 4-byte boundary. */
72 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
75 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
76 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
77 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
78 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
80 struct dp_netdev_upcall {
81 struct dpif_upcall upcall; /* Queued upcall information. */
82 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
85 struct dp_netdev_queue {
86 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
87 unsigned int head, tail;
90 /* Datapath based on the network device interface from netdev.h. */
92 const struct dpif_class *class;
96 int max_mtu; /* Maximum MTU of any port added so far. */
98 struct dp_netdev_queue queues[N_QUEUES];
99 struct classifier cls; /* Classifier. */
100 struct hmap flow_table; /* Flow table. */
101 struct seq *queue_seq; /* Incremented whenever a packet is queued. */
104 long long int n_hit; /* Number of flow table matches. */
105 long long int n_missed; /* Number of flow table misses. */
106 long long int n_lost; /* Number of misses not passed to client. */
109 struct dp_netdev_port *ports[MAX_PORTS];
110 struct list port_list;
111 struct seq *port_seq; /* Incremented whenever a port changes. */
114 /* A port in a netdev-based datapath. */
115 struct dp_netdev_port {
116 odp_port_t port_no; /* Index into dp_netdev's 'ports'. */
117 struct list node; /* Element in dp_netdev's 'port_list'. */
118 struct netdev *netdev;
119 struct netdev_saved_flags *sf;
120 struct netdev_rx *rx;
121 char *type; /* Port type as requested by user. */
124 /* A flow in dp_netdev's 'flow_table'. */
125 struct dp_netdev_flow {
126 /* Packet classification. */
127 struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
129 /* Hash table index by unmasked flow.*/
130 struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
131 struct flow flow; /* The flow that created this entry. */
134 long long int used; /* Last used time, in monotonic msecs. */
135 long long int packet_count; /* Number of packets matched. */
136 long long int byte_count; /* Number of bytes matched. */
137 uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
140 struct nlattr *actions;
144 /* Interface to netdev-based datapath. */
147 struct dp_netdev *dp;
148 uint64_t last_port_seq;
151 /* All netdev-based datapaths. */
152 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
154 /* Global lock for all data. */
155 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
157 static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
158 struct dp_netdev_port **portp);
159 static int get_port_by_name(struct dp_netdev *, const char *devname,
160 struct dp_netdev_port **portp);
161 static void dp_netdev_free(struct dp_netdev *);
162 static void dp_netdev_flow_flush(struct dp_netdev *);
163 static int do_add_port(struct dp_netdev *, const char *devname,
164 const char *type, odp_port_t port_no);
165 static int do_del_port(struct dp_netdev *, odp_port_t port_no);
166 static int dpif_netdev_open(const struct dpif_class *, const char *name,
167 bool create, struct dpif **);
168 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
169 int queue_no, const struct flow *,
170 const struct nlattr *userdata);
171 static void dp_netdev_execute_actions(struct dp_netdev *, const struct flow *,
173 const struct nlattr *actions,
175 static void dp_netdev_port_input(struct dp_netdev *dp,
176 struct dp_netdev_port *port,
177 struct ofpbuf *packet, uint32_t skb_priority,
178 uint32_t pkt_mark, const struct flow_tnl *tnl);
180 static struct dpif_netdev *
181 dpif_netdev_cast(const struct dpif *dpif)
183 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
184 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
187 static struct dp_netdev *
188 get_dp_netdev(const struct dpif *dpif)
190 return dpif_netdev_cast(dpif)->dp;
194 dpif_netdev_enumerate(struct sset *all_dps)
196 struct shash_node *node;
198 ovs_mutex_lock(&dp_netdev_mutex);
199 SHASH_FOR_EACH(node, &dp_netdevs) {
200 sset_add(all_dps, node->name);
202 ovs_mutex_unlock(&dp_netdev_mutex);
208 dpif_netdev_class_is_dummy(const struct dpif_class *class)
210 return class != &dpif_netdev_class;
214 dpif_netdev_class_is_planetlab(const struct dpif_class *class)
216 return class == &dpif_planetlab_class;
220 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
222 return strcmp(type, "internal") ? type
223 : dpif_netdev_class_is_planetlab(class) ? "pltap"
224 : dpif_netdev_class_is_dummy(class) ? "dummy"
229 create_dpif_netdev(struct dp_netdev *dp)
231 uint16_t netflow_id = hash_string(dp->name, 0);
232 struct dpif_netdev *dpif;
236 dpif = xmalloc(sizeof *dpif);
237 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
239 dpif->last_port_seq = seq_read(dp->port_seq);
244 /* Choose an unused, non-zero port number and return it on success.
245 * Return ODPP_NONE on failure. */
247 choose_port(struct dp_netdev *dp, const char *name)
251 if (dp->class != &dpif_netdev_class &&
252 dp->class != &dpif_planetlab_class) {
256 /* If the port name begins with "br", start the number search at
257 * 100 to make writing tests easier. */
258 if (!strncmp(name, "br", 2)) {
262 /* If the port name contains a number, try to assign that port number.
263 * This can make writing unit tests easier because port numbers are
265 for (p = name; *p != '\0'; p++) {
266 if (isdigit((unsigned char) *p)) {
267 port_no = start_no + strtol(p, NULL, 10);
268 if (port_no > 0 && port_no < MAX_PORTS
269 && !dp->ports[port_no]) {
270 return u32_to_odp(port_no);
277 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
278 if (!dp->ports[port_no]) {
279 return u32_to_odp(port_no);
287 create_dp_netdev(const char *name, const struct dpif_class *class,
288 struct dp_netdev **dpp)
290 struct dp_netdev *dp;
294 dp = xzalloc(sizeof *dp);
296 dp->name = xstrdup(name);
298 dp->max_mtu = ETH_PAYLOAD_MAX;
299 for (i = 0; i < N_QUEUES; i++) {
300 dp->queues[i].head = dp->queues[i].tail = 0;
302 dp->queue_seq = seq_create();
303 classifier_init(&dp->cls, NULL);
304 hmap_init(&dp->flow_table);
305 list_init(&dp->port_list);
306 dp->port_seq = seq_create();
308 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
314 shash_add(&dp_netdevs, name, dp);
321 dpif_netdev_open(const struct dpif_class *class, const char *name,
322 bool create, struct dpif **dpifp)
324 struct dp_netdev *dp;
327 ovs_mutex_lock(&dp_netdev_mutex);
328 dp = shash_find_data(&dp_netdevs, name);
330 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
332 error = (dp->class != class ? EINVAL
337 *dpifp = create_dpif_netdev(dp);
339 ovs_mutex_unlock(&dp_netdev_mutex);
345 dp_netdev_purge_queues(struct dp_netdev *dp)
349 for (i = 0; i < N_QUEUES; i++) {
350 struct dp_netdev_queue *q = &dp->queues[i];
352 while (q->tail != q->head) {
353 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
354 ofpbuf_uninit(&u->buf);
360 dp_netdev_free(struct dp_netdev *dp)
362 struct dp_netdev_port *port, *next;
364 dp_netdev_flow_flush(dp);
365 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
366 do_del_port(dp, port->port_no);
368 dp_netdev_purge_queues(dp);
369 seq_destroy(dp->queue_seq);
370 classifier_destroy(&dp->cls);
371 hmap_destroy(&dp->flow_table);
372 seq_destroy(dp->port_seq);
378 dpif_netdev_close(struct dpif *dpif)
380 struct dp_netdev *dp = get_dp_netdev(dpif);
382 ovs_mutex_lock(&dp_netdev_mutex);
384 ovs_assert(dp->open_cnt > 0);
385 if (--dp->open_cnt == 0 && dp->destroyed) {
386 shash_find_and_delete(&dp_netdevs, dp->name);
391 ovs_mutex_unlock(&dp_netdev_mutex);
395 dpif_netdev_destroy(struct dpif *dpif)
397 struct dp_netdev *dp = get_dp_netdev(dpif);
399 ovs_mutex_lock(&dp_netdev_mutex);
400 dp->destroyed = true;
401 ovs_mutex_unlock(&dp_netdev_mutex);
407 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
409 struct dp_netdev *dp = get_dp_netdev(dpif);
411 ovs_mutex_lock(&dp_netdev_mutex);
412 stats->n_flows = hmap_count(&dp->flow_table);
413 stats->n_hit = dp->n_hit;
414 stats->n_missed = dp->n_missed;
415 stats->n_lost = dp->n_lost;
416 stats->n_masks = UINT64_MAX;
417 stats->n_mask_hit = UINT64_MAX;
418 ovs_mutex_unlock(&dp_netdev_mutex);
424 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
427 struct netdev_saved_flags *sf;
428 struct dp_netdev_port *port;
429 struct netdev *netdev;
430 struct netdev_rx *rx;
431 enum netdev_flags flags;
432 const char *open_type;
436 /* XXX reject devices already in some dp_netdev. */
438 /* Open and validate network device. */
439 open_type = dpif_netdev_port_open_type(dp->class, type);
440 error = netdev_open(devname, open_type, &netdev);
444 /* XXX reject non-Ethernet devices */
446 netdev_get_flags(netdev, &flags);
447 if (flags & NETDEV_LOOPBACK) {
448 VLOG_ERR("%s: cannot add a loopback device", devname);
449 netdev_close(netdev);
453 error = netdev_rx_open(netdev, &rx);
455 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
456 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
457 devname, ovs_strerror(errno));
458 netdev_close(netdev);
462 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
465 netdev_close(netdev);
469 port = xmalloc(sizeof *port);
470 port->port_no = port_no;
471 port->netdev = netdev;
474 port->type = xstrdup(type);
476 error = netdev_get_mtu(netdev, &mtu);
477 if (!error && mtu > dp->max_mtu) {
481 list_push_back(&dp->port_list, &port->node);
482 dp->ports[odp_to_u32(port_no)] = port;
483 seq_change(dp->port_seq);
489 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
490 odp_port_t *port_nop)
492 struct dp_netdev *dp = get_dp_netdev(dpif);
493 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
494 const char *dpif_port;
498 ovs_mutex_lock(&dp_netdev_mutex);
499 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
500 if (*port_nop != ODPP_NONE) {
501 uint32_t port_idx = odp_to_u32(*port_nop);
502 if (port_idx >= MAX_PORTS) {
504 } else if (dp->ports[port_idx]) {
511 port_no = choose_port(dp, dpif_port);
512 error = port_no == ODPP_NONE ? EFBIG : 0;
516 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
518 ovs_mutex_unlock(&dp_netdev_mutex);
524 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
526 struct dp_netdev *dp = get_dp_netdev(dpif);
529 ovs_mutex_lock(&dp_netdev_mutex);
530 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
531 ovs_mutex_unlock(&dp_netdev_mutex);
537 is_valid_port_number(odp_port_t port_no)
539 return odp_to_u32(port_no) < MAX_PORTS;
543 get_port_by_number(struct dp_netdev *dp,
544 odp_port_t port_no, struct dp_netdev_port **portp)
546 if (!is_valid_port_number(port_no)) {
550 *portp = dp->ports[odp_to_u32(port_no)];
551 return *portp ? 0 : ENOENT;
556 get_port_by_name(struct dp_netdev *dp,
557 const char *devname, struct dp_netdev_port **portp)
559 struct dp_netdev_port *port;
561 LIST_FOR_EACH (port, node, &dp->port_list) {
562 if (!strcmp(netdev_get_name(port->netdev), devname)) {
571 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
573 struct dp_netdev_port *port;
576 error = get_port_by_number(dp, port_no, &port);
581 list_remove(&port->node);
582 dp->ports[odp_to_u32(port_no)] = NULL;
583 seq_change(dp->port_seq);
585 netdev_close(port->netdev);
586 netdev_restore_flags(port->sf);
587 netdev_rx_close(port->rx);
595 answer_port_query(const struct dp_netdev_port *port,
596 struct dpif_port *dpif_port)
598 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
599 dpif_port->type = xstrdup(port->type);
600 dpif_port->port_no = port->port_no;
604 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
605 struct dpif_port *dpif_port)
607 struct dp_netdev *dp = get_dp_netdev(dpif);
608 struct dp_netdev_port *port;
611 ovs_mutex_lock(&dp_netdev_mutex);
612 error = get_port_by_number(dp, port_no, &port);
613 if (!error && dpif_port) {
614 answer_port_query(port, dpif_port);
616 ovs_mutex_unlock(&dp_netdev_mutex);
622 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
623 struct dpif_port *dpif_port)
625 struct dp_netdev *dp = get_dp_netdev(dpif);
626 struct dp_netdev_port *port;
629 ovs_mutex_lock(&dp_netdev_mutex);
630 error = get_port_by_name(dp, devname, &port);
631 if (!error && dpif_port) {
632 answer_port_query(port, dpif_port);
634 ovs_mutex_unlock(&dp_netdev_mutex);
640 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
646 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
648 ovs_rwlock_wrlock(&dp->cls.rwlock);
649 classifier_remove(&dp->cls, &netdev_flow->cr);
650 ovs_rwlock_unlock(&dp->cls.rwlock);
651 cls_rule_destroy(&netdev_flow->cr);
653 hmap_remove(&dp->flow_table, &netdev_flow->node);
654 free(netdev_flow->actions);
659 dp_netdev_flow_flush(struct dp_netdev *dp)
661 struct dp_netdev_flow *netdev_flow, *next;
663 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
664 dp_netdev_free_flow(dp, netdev_flow);
669 dpif_netdev_flow_flush(struct dpif *dpif)
671 struct dp_netdev *dp = get_dp_netdev(dpif);
673 ovs_mutex_lock(&dp_netdev_mutex);
674 dp_netdev_flow_flush(dp);
675 ovs_mutex_unlock(&dp_netdev_mutex);
680 struct dp_netdev_port_state {
686 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
688 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
693 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
694 struct dpif_port *dpif_port)
696 struct dp_netdev_port_state *state = state_;
697 struct dp_netdev *dp = get_dp_netdev(dpif);
700 ovs_mutex_lock(&dp_netdev_mutex);
701 for (port_idx = odp_to_u32(state->port_no);
702 port_idx < MAX_PORTS; port_idx++) {
703 struct dp_netdev_port *port = dp->ports[port_idx];
706 state->name = xstrdup(netdev_get_name(port->netdev));
707 dpif_port->name = state->name;
708 dpif_port->type = port->type;
709 dpif_port->port_no = port->port_no;
710 state->port_no = u32_to_odp(port_idx + 1);
711 ovs_mutex_unlock(&dp_netdev_mutex);
716 ovs_mutex_unlock(&dp_netdev_mutex);
722 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
724 struct dp_netdev_port_state *state = state_;
731 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
733 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
734 uint64_t new_port_seq;
737 ovs_mutex_lock(&dp_netdev_mutex);
738 new_port_seq = seq_read(dpif->dp->port_seq);
739 if (dpif->last_port_seq != new_port_seq) {
740 dpif->last_port_seq = new_port_seq;
745 ovs_mutex_unlock(&dp_netdev_mutex);
751 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
753 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
755 ovs_mutex_lock(&dp_netdev_mutex);
756 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
757 ovs_mutex_unlock(&dp_netdev_mutex);
760 static struct dp_netdev_flow *
761 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
765 ovs_rwlock_wrlock(&dp->cls.rwlock);
766 cr = classifier_lookup(&dp->cls, flow, NULL);
767 ovs_rwlock_unlock(&dp->cls.rwlock);
770 ? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
774 static struct dp_netdev_flow *
775 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
777 struct dp_netdev_flow *netdev_flow;
779 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
781 if (flow_equal(&netdev_flow->flow, flow)) {
789 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
790 struct dpif_flow_stats *stats)
792 stats->n_packets = netdev_flow->packet_count;
793 stats->n_bytes = netdev_flow->byte_count;
794 stats->used = netdev_flow->used;
795 stats->tcp_flags = netdev_flow->tcp_flags;
799 dpif_netdev_flow_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
800 const struct nlattr *mask_key,
801 uint32_t mask_key_len, struct flow *flow,
806 if (odp_flow_key_to_flow(key, key_len, flow)
808 && odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow))) {
809 /* This should not happen: it indicates that odp_flow_key_from_flow()
810 * and odp_flow_key_to_flow() disagree on the acceptable form of a flow
811 * or odp_flow_key_from_mask() and odp_flow_key_to_mask() disagree on
812 * the acceptable form of a mask. Log the problem as an error, with
813 * enough details to enable debugging. */
814 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
816 if (!VLOG_DROP_ERR(&rl)) {
820 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
822 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
830 /* Force unwildcard the in_port. */
831 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
834 in_port = flow->in_port.odp_port;
835 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
843 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
846 return dpif_netdev_flow_mask_from_nlattrs(key, key_len, NULL, 0, flow,
851 dpif_netdev_flow_get(const struct dpif *dpif,
852 const struct nlattr *nl_key, size_t nl_key_len,
853 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
855 struct dp_netdev *dp = get_dp_netdev(dpif);
856 struct dp_netdev_flow *netdev_flow;
860 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
865 ovs_mutex_lock(&dp_netdev_mutex);
866 netdev_flow = dp_netdev_find_flow(dp, &key);
869 get_dpif_flow_stats(netdev_flow, stats);
872 *actionsp = ofpbuf_clone_data(netdev_flow->actions,
873 netdev_flow->actions_len);
878 ovs_mutex_unlock(&dp_netdev_mutex);
884 set_flow_actions(struct dp_netdev_flow *netdev_flow,
885 const struct nlattr *actions, size_t actions_len)
887 netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
888 netdev_flow->actions_len = actions_len;
889 memcpy(netdev_flow->actions, actions, actions_len);
894 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
895 const struct flow_wildcards *wc,
896 const struct nlattr *actions,
899 struct dp_netdev_flow *netdev_flow;
903 netdev_flow = xzalloc(sizeof *netdev_flow);
904 netdev_flow->flow = *flow;
906 match_init(&match, flow, wc);
907 cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
908 ovs_rwlock_wrlock(&dp->cls.rwlock);
909 classifier_insert(&dp->cls, &netdev_flow->cr);
910 ovs_rwlock_unlock(&dp->cls.rwlock);
912 error = set_flow_actions(netdev_flow, actions, actions_len);
914 ovs_rwlock_wrlock(&dp->cls.rwlock);
915 classifier_remove(&dp->cls, &netdev_flow->cr);
916 ovs_rwlock_unlock(&dp->cls.rwlock);
917 cls_rule_destroy(&netdev_flow->cr);
923 hmap_insert(&dp->flow_table, &netdev_flow->node, flow_hash(flow, 0));
928 clear_stats(struct dp_netdev_flow *netdev_flow)
930 netdev_flow->used = 0;
931 netdev_flow->packet_count = 0;
932 netdev_flow->byte_count = 0;
933 netdev_flow->tcp_flags = 0;
937 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
939 struct dp_netdev *dp = get_dp_netdev(dpif);
940 struct dp_netdev_flow *netdev_flow;
942 struct flow_wildcards wc;
945 error = dpif_netdev_flow_mask_from_nlattrs(put->key, put->key_len,
946 put->mask, put->mask_len, &flow, &wc.masks);
951 ovs_mutex_lock(&dp_netdev_mutex);
952 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
954 if (put->flags & DPIF_FP_CREATE) {
955 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
957 memset(put->stats, 0, sizeof *put->stats);
959 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
968 if (put->flags & DPIF_FP_MODIFY
969 && flow_equal(&flow, &netdev_flow->flow)) {
970 error = set_flow_actions(netdev_flow, put->actions,
974 get_dpif_flow_stats(netdev_flow, put->stats);
976 if (put->flags & DPIF_FP_ZERO_STATS) {
977 clear_stats(netdev_flow);
980 } else if (put->flags & DPIF_FP_CREATE) {
983 /* Overlapping flow. */
987 ovs_mutex_unlock(&dp_netdev_mutex);
993 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
995 struct dp_netdev *dp = get_dp_netdev(dpif);
996 struct dp_netdev_flow *netdev_flow;
1000 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1005 ovs_mutex_lock(&dp_netdev_mutex);
1006 netdev_flow = dp_netdev_find_flow(dp, &key);
1009 get_dpif_flow_stats(netdev_flow, del->stats);
1011 dp_netdev_free_flow(dp, netdev_flow);
1015 ovs_mutex_unlock(&dp_netdev_mutex);
1020 struct dp_netdev_flow_state {
1023 struct nlattr *actions;
1024 struct odputil_keybuf keybuf;
1025 struct odputil_keybuf maskbuf;
1026 struct dpif_flow_stats stats;
1030 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1032 struct dp_netdev_flow_state *state;
1034 *statep = state = xmalloc(sizeof *state);
1037 state->actions = NULL;
1042 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1043 const struct nlattr **key, size_t *key_len,
1044 const struct nlattr **mask, size_t *mask_len,
1045 const struct nlattr **actions, size_t *actions_len,
1046 const struct dpif_flow_stats **stats)
1048 struct dp_netdev_flow_state *state = state_;
1049 struct dp_netdev *dp = get_dp_netdev(dpif);
1050 struct dp_netdev_flow *netdev_flow;
1051 struct hmap_node *node;
1053 ovs_mutex_lock(&dp_netdev_mutex);
1054 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1056 ovs_mutex_unlock(&dp_netdev_mutex);
1060 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1065 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1066 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1067 netdev_flow->flow.in_port.odp_port);
1070 *key_len = buf.size;
1075 struct flow_wildcards wc;
1077 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1078 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1079 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1080 odp_to_u32(wc.masks.in_port.odp_port));
1083 *mask_len = buf.size;
1087 free(state->actions);
1088 state->actions = xmemdup(netdev_flow->actions,
1089 netdev_flow->actions_len);
1091 *actions = state->actions;
1092 *actions_len = netdev_flow->actions_len;
1096 get_dpif_flow_stats(netdev_flow, &state->stats);
1097 *stats = &state->stats;
1100 ovs_mutex_unlock(&dp_netdev_mutex);
1105 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1107 struct dp_netdev_flow_state *state = state_;
1109 free(state->actions);
1115 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
1117 struct dp_netdev *dp = get_dp_netdev(dpif);
1121 if (execute->packet->size < ETH_HEADER_LEN ||
1122 execute->packet->size > UINT16_MAX) {
1126 /* Get packet metadata. */
1127 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, &md);
1129 struct ofpbuf *copy;
1132 /* Make a deep copy of 'packet', because we might modify its data. */
1133 copy = ofpbuf_clone_with_headroom(execute->packet, DP_NETDEV_HEADROOM);
1135 /* Extract flow key. */
1136 flow_extract(copy, md.skb_priority, md.pkt_mark, &md.tunnel,
1138 ovs_mutex_lock(&dp_netdev_mutex);
1139 dp_netdev_execute_actions(dp, &key, copy,
1140 execute->actions, execute->actions_len);
1141 ovs_mutex_unlock(&dp_netdev_mutex);
1142 ofpbuf_delete(copy);
1148 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1154 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1155 uint32_t queue_id, uint32_t *priority)
1157 *priority = queue_id;
1161 static struct dp_netdev_queue *
1162 find_nonempty_queue(struct dpif *dpif)
1164 struct dp_netdev *dp = get_dp_netdev(dpif);
1167 for (i = 0; i < N_QUEUES; i++) {
1168 struct dp_netdev_queue *q = &dp->queues[i];
1169 if (q->head != q->tail) {
1177 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1180 struct dp_netdev_queue *q;
1183 ovs_mutex_lock(&dp_netdev_mutex);
1184 q = find_nonempty_queue(dpif);
1186 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1188 *upcall = u->upcall;
1189 upcall->packet = buf;
1198 ovs_mutex_unlock(&dp_netdev_mutex);
1204 dpif_netdev_recv_wait(struct dpif *dpif)
1206 struct dp_netdev *dp = get_dp_netdev(dpif);
1209 ovs_mutex_lock(&dp_netdev_mutex);
1210 seq = seq_read(dp->queue_seq);
1211 if (find_nonempty_queue(dpif)) {
1212 poll_immediate_wake();
1214 seq_wait(dp->queue_seq, seq);
1216 ovs_mutex_unlock(&dp_netdev_mutex);
1220 dpif_netdev_recv_purge(struct dpif *dpif)
1222 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1223 ovs_mutex_lock(&dp_netdev_mutex);
1224 dp_netdev_purge_queues(dpif_netdev->dp);
1225 ovs_mutex_unlock(&dp_netdev_mutex);
1229 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1230 const struct ofpbuf *packet)
1232 netdev_flow->used = time_msec();
1233 netdev_flow->packet_count++;
1234 netdev_flow->byte_count += packet->size;
1235 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1239 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1240 struct ofpbuf *packet, uint32_t skb_priority,
1241 uint32_t pkt_mark, const struct flow_tnl *tnl)
1243 struct dp_netdev_flow *netdev_flow;
1245 union flow_in_port in_port_;
1247 if (packet->size < ETH_HEADER_LEN) {
1250 in_port_.odp_port = port->port_no;
1251 flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
1252 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1254 dp_netdev_flow_used(netdev_flow, packet);
1255 dp_netdev_execute_actions(dp, &key, packet,
1256 netdev_flow->actions,
1257 netdev_flow->actions_len);
1261 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1266 dpif_netdev_run(struct dpif *dpif)
1268 struct dp_netdev_port *port;
1269 struct dp_netdev *dp;
1270 struct ofpbuf packet;
1272 ovs_mutex_lock(&dp_netdev_mutex);
1273 dp = get_dp_netdev(dpif);
1274 ofpbuf_init(&packet,
1275 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1277 LIST_FOR_EACH (port, node, &dp->port_list) {
1280 /* Reset packet contents. */
1281 ofpbuf_clear(&packet);
1282 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1284 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1286 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1287 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1288 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1290 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1291 netdev_get_name(port->netdev), ovs_strerror(error));
1294 ofpbuf_uninit(&packet);
1295 ovs_mutex_unlock(&dp_netdev_mutex);
1299 dpif_netdev_wait(struct dpif *dpif)
1301 struct dp_netdev_port *port;
1303 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1304 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1305 * A makes it to poll_block().
1307 * But I think it doesn't matter:
1309 * - In the dpif_port_add() case, A will not wake up when a packet
1310 * arrives on the new port, but this would also happen if the
1311 * ordering were reversed.
1313 * - In the dpif_port_remove() case, A might wake up spuriously, but
1314 * that is harmless. */
1316 ovs_mutex_lock(&dp_netdev_mutex);
1317 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1319 netdev_rx_wait(port->rx);
1322 ovs_mutex_unlock(&dp_netdev_mutex);
1326 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1327 int queue_no, const struct flow *flow,
1328 const struct nlattr *userdata)
1330 struct dp_netdev_queue *q = &dp->queues[queue_no];
1331 if (q->head - q->tail < MAX_QUEUE_LEN) {
1332 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1333 struct dpif_upcall *upcall = &u->upcall;
1334 struct ofpbuf *buf = &u->buf;
1337 upcall->type = queue_no;
1339 /* Allocate buffer big enough for everything. */
1340 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1342 buf_size += NLA_ALIGN(userdata->nla_len);
1344 ofpbuf_init(buf, buf_size);
1347 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1348 upcall->key = buf->data;
1349 upcall->key_len = buf->size;
1353 upcall->userdata = ofpbuf_put(buf, userdata,
1354 NLA_ALIGN(userdata->nla_len));
1359 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1360 * is visible in 'upcall->packet'. The ODP flow and (if present)
1361 * userdata become part of the headroom. */
1362 ofpbuf_put_zeros(buf, 2);
1363 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1364 buf->size = packet->size;
1365 upcall->packet = buf;
1367 seq_change(dp->queue_seq);
1376 struct dp_netdev_execute_aux {
1377 struct dp_netdev *dp;
1378 const struct flow *key;
1382 dp_netdev_action_output(void *aux_, struct ofpbuf *packet,
1383 const struct flow *flow OVS_UNUSED,
1384 odp_port_t out_port)
1386 struct dp_netdev_execute_aux *aux = aux_;
1387 struct dp_netdev_port *p = aux->dp->ports[odp_to_u32(out_port)];
1389 netdev_send(p->netdev, packet);
1394 dp_netdev_action_userspace(void *aux_, struct ofpbuf *packet,
1395 const struct flow *flow OVS_UNUSED,
1396 const struct nlattr *a)
1398 struct dp_netdev_execute_aux *aux = aux_;
1399 const struct nlattr *userdata;
1401 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1402 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1407 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1408 struct ofpbuf *packet,
1409 const struct nlattr *actions, size_t actions_len)
1411 struct dp_netdev_execute_aux aux = {dp, key};
1412 struct flow md = *key; /* Packet metadata, may be modified by actions. */
1414 odp_execute_actions(&aux, packet, &md, actions, actions_len,
1415 dp_netdev_action_output, dp_netdev_action_userspace);
1418 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1419 dpif_netdev_enumerate, \
1420 dpif_netdev_port_open_type, \
1422 dpif_netdev_close, \
1423 dpif_netdev_destroy, \
1426 dpif_netdev_get_stats, \
1427 dpif_netdev_port_add, \
1428 dpif_netdev_port_del, \
1429 dpif_netdev_port_query_by_number, \
1430 dpif_netdev_port_query_by_name, \
1431 dpif_netdev_get_max_ports, \
1432 NULL, /* port_get_pid */ \
1433 dpif_netdev_port_dump_start, \
1434 dpif_netdev_port_dump_next, \
1435 dpif_netdev_port_dump_done, \
1436 dpif_netdev_port_poll, \
1437 dpif_netdev_port_poll_wait, \
1438 dpif_netdev_flow_get, \
1439 dpif_netdev_flow_put, \
1440 dpif_netdev_flow_del, \
1441 dpif_netdev_flow_flush, \
1442 dpif_netdev_flow_dump_start, \
1443 dpif_netdev_flow_dump_next, \
1444 dpif_netdev_flow_dump_done, \
1445 dpif_netdev_execute, \
1446 NULL, /* operate */ \
1447 dpif_netdev_recv_set, \
1448 dpif_netdev_queue_to_priority, \
1450 dpif_netdev_recv_wait, \
1451 dpif_netdev_recv_purge, \
1453 const struct dpif_class dpif_netdev_class = {
1455 DPIF_NETDEV_CLASS_FUNCTIONS
1458 const struct dpif_class dpif_planetlab_class = {
1460 DPIF_NETDEV_CLASS_FUNCTIONS
1464 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1465 const char *argv[], void *aux OVS_UNUSED)
1467 struct dp_netdev_port *port;
1468 struct dp_netdev *dp;
1471 dp = shash_find_data(&dp_netdevs, argv[1]);
1472 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1473 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1477 if (get_port_by_name(dp, argv[2], &port)) {
1478 unixctl_command_reply_error(conn, "unknown port");
1482 port_no = atoi(argv[3]);
1483 if (port_no <= 0 || port_no >= MAX_PORTS) {
1484 unixctl_command_reply_error(conn, "bad port number");
1487 if (dp->ports[port_no]) {
1488 unixctl_command_reply_error(conn, "port number already in use");
1491 dp->ports[odp_to_u32(port->port_no)] = NULL;
1492 dp->ports[port_no] = port;
1493 port->port_no = u32_to_odp(port_no);
1494 seq_change(dp->port_seq);
1495 unixctl_command_reply(conn, NULL);
1499 dpif_dummy_register__(const char *type)
1501 struct dpif_class *class;
1503 class = xmalloc(sizeof *class);
1504 *class = dpif_netdev_class;
1505 class->type = xstrdup(type);
1506 dp_register_provider(class);
1510 dpif_dummy_register(bool override)
1517 dp_enumerate_types(&types);
1518 SSET_FOR_EACH (type, &types) {
1519 if (!dp_unregister_provider(type)) {
1520 dpif_dummy_register__(type);
1523 sset_destroy(&types);
1526 dpif_dummy_register__("dummy");
1528 unixctl_command_register("dpif-dummy/change-port-number",
1529 "DP PORT NEW-NUMBER",
1530 3, 3, dpif_dummy_change_port_number, NULL);