2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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"
43 #include "meta-flow.h"
45 #include "netdev-vport.h"
47 #include "odp-execute.h"
49 #include "ofp-print.h"
52 #include "poll-loop.h"
62 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
64 /* By default, choose a priority in the middle. */
65 #define NETDEV_RULE_PRIORITY 0x8000
67 /* Configuration parameters. */
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 struct ovsthread_counter *n_hit; /* Number of flow table matches. */
105 struct ovsthread_counter *n_missed; /* Number of flow table misses. */
106 struct ovsthread_counter *n_lost; /* Number of misses not passed up. */
110 struct seq *port_seq; /* Incremented whenever a port changes. */
113 static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *,
116 /* A port in a netdev-based datapath. */
117 struct dp_netdev_port {
118 struct hmap_node node; /* Node in dp_netdev's 'ports'. */
120 struct netdev *netdev;
121 struct netdev_saved_flags *sf;
122 struct netdev_rx *rx;
123 char *type; /* Port type as requested by user. */
126 /* A flow in dp_netdev's 'flow_table'. */
127 struct dp_netdev_flow {
128 /* Packet classification. */
129 struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
131 /* Hash table index by unmasked flow.*/
132 struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
133 struct flow flow; /* The flow that created this entry. */
136 long long int used; /* Last used time, in monotonic msecs. */
137 long long int packet_count; /* Number of packets matched. */
138 long long int byte_count; /* Number of bytes matched. */
139 uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
142 struct nlattr *actions;
146 /* Interface to netdev-based datapath. */
149 struct dp_netdev *dp;
150 uint64_t last_port_seq;
153 /* All netdev-based datapaths. */
154 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
156 /* Global lock for all data. */
157 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
159 static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
160 struct dp_netdev_port **portp);
161 static int get_port_by_name(struct dp_netdev *, const char *devname,
162 struct dp_netdev_port **portp);
163 static void dp_netdev_free(struct dp_netdev *);
164 static void dp_netdev_flow_flush(struct dp_netdev *);
165 static int do_add_port(struct dp_netdev *, const char *devname,
166 const char *type, odp_port_t port_no);
167 static int do_del_port(struct dp_netdev *, odp_port_t port_no);
168 static int dpif_netdev_open(const struct dpif_class *, const char *name,
169 bool create, struct dpif **);
170 static int dp_netdev_output_userspace(struct dp_netdev *, struct ofpbuf *,
171 int queue_no, const struct flow *,
172 const struct nlattr *userdata);
173 static void dp_netdev_execute_actions(struct dp_netdev *, const struct flow *,
174 struct ofpbuf *, struct pkt_metadata *,
175 const struct nlattr *actions,
177 static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
178 struct pkt_metadata *md);
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_port_open_type(const struct dpif_class *class, const char *type)
216 return strcmp(type, "internal") ? type
217 : dpif_netdev_class_is_dummy(class) ? "dummy"
222 create_dpif_netdev(struct dp_netdev *dp)
224 uint16_t netflow_id = hash_string(dp->name, 0);
225 struct dpif_netdev *dpif;
229 dpif = xmalloc(sizeof *dpif);
230 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
232 dpif->last_port_seq = seq_read(dp->port_seq);
237 /* Choose an unused, non-zero port number and return it on success.
238 * Return ODPP_NONE on failure. */
240 choose_port(struct dp_netdev *dp, const char *name)
244 if (dp->class != &dpif_netdev_class) {
248 /* If the port name begins with "br", start the number search at
249 * 100 to make writing tests easier. */
250 if (!strncmp(name, "br", 2)) {
254 /* If the port name contains a number, try to assign that port number.
255 * This can make writing unit tests easier because port numbers are
257 for (p = name; *p != '\0'; p++) {
258 if (isdigit((unsigned char) *p)) {
259 port_no = start_no + strtol(p, NULL, 10);
260 if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
261 && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
262 return u32_to_odp(port_no);
269 for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
270 if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
271 return u32_to_odp(port_no);
279 create_dp_netdev(const char *name, const struct dpif_class *class,
280 struct dp_netdev **dpp)
282 struct dp_netdev *dp;
286 dp = xzalloc(sizeof *dp);
288 dp->name = xstrdup(name);
290 dp->max_mtu = ETH_PAYLOAD_MAX;
291 for (i = 0; i < N_QUEUES; i++) {
292 dp->queues[i].head = dp->queues[i].tail = 0;
294 dp->queue_seq = seq_create();
295 classifier_init(&dp->cls, NULL);
296 hmap_init(&dp->flow_table);
298 dp->n_hit = ovsthread_counter_create();
299 dp->n_missed = ovsthread_counter_create();
300 dp->n_lost = ovsthread_counter_create();
302 hmap_init(&dp->ports);
303 dp->port_seq = seq_create();
305 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
311 shash_add(&dp_netdevs, name, dp);
318 dpif_netdev_open(const struct dpif_class *class, const char *name,
319 bool create, struct dpif **dpifp)
321 struct dp_netdev *dp;
324 ovs_mutex_lock(&dp_netdev_mutex);
325 dp = shash_find_data(&dp_netdevs, name);
327 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
329 error = (dp->class != class ? EINVAL
334 *dpifp = create_dpif_netdev(dp);
336 ovs_mutex_unlock(&dp_netdev_mutex);
342 dp_netdev_purge_queues(struct dp_netdev *dp)
346 for (i = 0; i < N_QUEUES; i++) {
347 struct dp_netdev_queue *q = &dp->queues[i];
349 while (q->tail != q->head) {
350 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
351 ofpbuf_uninit(&u->upcall.packet);
352 ofpbuf_uninit(&u->buf);
358 dp_netdev_free(struct dp_netdev *dp)
360 struct dp_netdev_port *port, *next;
362 dp_netdev_flow_flush(dp);
363 HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
364 do_del_port(dp, port->port_no);
366 ovsthread_counter_destroy(dp->n_hit);
367 ovsthread_counter_destroy(dp->n_missed);
368 ovsthread_counter_destroy(dp->n_lost);
369 dp_netdev_purge_queues(dp);
370 seq_destroy(dp->queue_seq);
371 classifier_destroy(&dp->cls);
372 hmap_destroy(&dp->flow_table);
373 seq_destroy(dp->port_seq);
374 hmap_destroy(&dp->ports);
380 dpif_netdev_close(struct dpif *dpif)
382 struct dp_netdev *dp = get_dp_netdev(dpif);
384 ovs_mutex_lock(&dp_netdev_mutex);
386 ovs_assert(dp->open_cnt > 0);
387 if (--dp->open_cnt == 0 && dp->destroyed) {
388 shash_find_and_delete(&dp_netdevs, dp->name);
393 ovs_mutex_unlock(&dp_netdev_mutex);
397 dpif_netdev_destroy(struct dpif *dpif)
399 struct dp_netdev *dp = get_dp_netdev(dpif);
401 ovs_mutex_lock(&dp_netdev_mutex);
402 dp->destroyed = true;
403 ovs_mutex_unlock(&dp_netdev_mutex);
409 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
411 struct dp_netdev *dp = get_dp_netdev(dpif);
413 ovs_mutex_lock(&dp_netdev_mutex);
414 stats->n_flows = hmap_count(&dp->flow_table);
415 stats->n_hit = ovsthread_counter_read(dp->n_hit);
416 stats->n_missed = ovsthread_counter_read(dp->n_missed);
417 stats->n_lost = ovsthread_counter_read(dp->n_lost);
418 stats->n_masks = UINT32_MAX;
419 stats->n_mask_hit = UINT64_MAX;
420 ovs_mutex_unlock(&dp_netdev_mutex);
426 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
429 struct netdev_saved_flags *sf;
430 struct dp_netdev_port *port;
431 struct netdev *netdev;
432 struct netdev_rx *rx;
433 enum netdev_flags flags;
434 const char *open_type;
438 /* XXX reject devices already in some dp_netdev. */
440 /* Open and validate network device. */
441 open_type = dpif_netdev_port_open_type(dp->class, type);
442 error = netdev_open(devname, open_type, &netdev);
446 /* XXX reject non-Ethernet devices */
448 netdev_get_flags(netdev, &flags);
449 if (flags & NETDEV_LOOPBACK) {
450 VLOG_ERR("%s: cannot add a loopback device", devname);
451 netdev_close(netdev);
455 error = netdev_rx_open(netdev, &rx);
457 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
458 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
459 devname, ovs_strerror(errno));
460 netdev_close(netdev);
464 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
467 netdev_close(netdev);
471 port = xmalloc(sizeof *port);
472 port->port_no = port_no;
473 port->netdev = netdev;
476 port->type = xstrdup(type);
478 error = netdev_get_mtu(netdev, &mtu);
479 if (!error && mtu > dp->max_mtu) {
483 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
484 seq_change(dp->port_seq);
490 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
491 odp_port_t *port_nop)
493 struct dp_netdev *dp = get_dp_netdev(dpif);
494 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
495 const char *dpif_port;
499 ovs_mutex_lock(&dp_netdev_mutex);
500 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
501 if (*port_nop != ODPP_NONE) {
503 error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
505 port_no = choose_port(dp, dpif_port);
506 error = port_no == ODPP_NONE ? EFBIG : 0;
510 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
512 ovs_mutex_unlock(&dp_netdev_mutex);
518 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
520 struct dp_netdev *dp = get_dp_netdev(dpif);
523 ovs_mutex_lock(&dp_netdev_mutex);
524 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
525 ovs_mutex_unlock(&dp_netdev_mutex);
531 is_valid_port_number(odp_port_t port_no)
533 return port_no != ODPP_NONE;
536 static struct dp_netdev_port *
537 dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
539 struct dp_netdev_port *port;
541 HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
543 if (port->port_no == port_no) {
551 get_port_by_number(struct dp_netdev *dp,
552 odp_port_t port_no, struct dp_netdev_port **portp)
554 if (!is_valid_port_number(port_no)) {
558 *portp = dp_netdev_lookup_port(dp, port_no);
559 return *portp ? 0 : ENOENT;
564 get_port_by_name(struct dp_netdev *dp,
565 const char *devname, struct dp_netdev_port **portp)
567 struct dp_netdev_port *port;
569 HMAP_FOR_EACH (port, node, &dp->ports) {
570 if (!strcmp(netdev_get_name(port->netdev), devname)) {
579 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
581 struct dp_netdev_port *port;
584 error = get_port_by_number(dp, port_no, &port);
589 hmap_remove(&dp->ports, &port->node);
590 seq_change(dp->port_seq);
592 netdev_close(port->netdev);
593 netdev_restore_flags(port->sf);
594 netdev_rx_close(port->rx);
602 answer_port_query(const struct dp_netdev_port *port,
603 struct dpif_port *dpif_port)
605 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
606 dpif_port->type = xstrdup(port->type);
607 dpif_port->port_no = port->port_no;
611 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
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_number(dp, port_no, &port);
620 if (!error && dpif_port) {
621 answer_port_query(port, dpif_port);
623 ovs_mutex_unlock(&dp_netdev_mutex);
629 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
630 struct dpif_port *dpif_port)
632 struct dp_netdev *dp = get_dp_netdev(dpif);
633 struct dp_netdev_port *port;
636 ovs_mutex_lock(&dp_netdev_mutex);
637 error = get_port_by_name(dp, devname, &port);
638 if (!error && dpif_port) {
639 answer_port_query(port, dpif_port);
641 ovs_mutex_unlock(&dp_netdev_mutex);
647 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
649 ovs_rwlock_wrlock(&dp->cls.rwlock);
650 classifier_remove(&dp->cls, &netdev_flow->cr);
651 ovs_rwlock_unlock(&dp->cls.rwlock);
652 cls_rule_destroy(&netdev_flow->cr);
654 hmap_remove(&dp->flow_table, &netdev_flow->node);
655 free(netdev_flow->actions);
660 dp_netdev_flow_flush(struct dp_netdev *dp)
662 struct dp_netdev_flow *netdev_flow, *next;
664 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
665 dp_netdev_free_flow(dp, netdev_flow);
670 dpif_netdev_flow_flush(struct dpif *dpif)
672 struct dp_netdev *dp = get_dp_netdev(dpif);
674 ovs_mutex_lock(&dp_netdev_mutex);
675 dp_netdev_flow_flush(dp);
676 ovs_mutex_unlock(&dp_netdev_mutex);
681 struct dp_netdev_port_state {
688 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
690 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
695 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
696 struct dpif_port *dpif_port)
698 struct dp_netdev_port_state *state = state_;
699 struct dp_netdev *dp = get_dp_netdev(dpif);
700 struct hmap_node *node;
703 ovs_mutex_lock(&dp_netdev_mutex);
704 node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
706 struct dp_netdev_port *port;
708 port = CONTAINER_OF(node, struct dp_netdev_port, node);
711 state->name = xstrdup(netdev_get_name(port->netdev));
712 dpif_port->name = state->name;
713 dpif_port->type = port->type;
714 dpif_port->port_no = port->port_no;
720 ovs_mutex_unlock(&dp_netdev_mutex);
726 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
728 struct dp_netdev_port_state *state = state_;
735 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
737 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
738 uint64_t new_port_seq;
741 ovs_mutex_lock(&dp_netdev_mutex);
742 new_port_seq = seq_read(dpif->dp->port_seq);
743 if (dpif->last_port_seq != new_port_seq) {
744 dpif->last_port_seq = new_port_seq;
749 ovs_mutex_unlock(&dp_netdev_mutex);
755 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
757 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
759 ovs_mutex_lock(&dp_netdev_mutex);
760 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
761 ovs_mutex_unlock(&dp_netdev_mutex);
764 static struct dp_netdev_flow *
765 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
769 ovs_rwlock_wrlock(&dp->cls.rwlock);
770 cr = classifier_lookup(&dp->cls, flow, NULL);
771 ovs_rwlock_unlock(&dp->cls.rwlock);
774 ? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
778 static struct dp_netdev_flow *
779 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
781 struct dp_netdev_flow *netdev_flow;
783 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
785 if (flow_equal(&netdev_flow->flow, flow)) {
793 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
794 struct dpif_flow_stats *stats)
796 stats->n_packets = netdev_flow->packet_count;
797 stats->n_bytes = netdev_flow->byte_count;
798 stats->used = netdev_flow->used;
799 stats->tcp_flags = netdev_flow->tcp_flags;
803 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
804 const struct nlattr *mask_key,
805 uint32_t mask_key_len, const struct flow *flow,
809 if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
810 /* This should not happen: it indicates that
811 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
812 * disagree on the acceptable form of a mask. Log the problem
813 * as an error, with 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 mask %s", ds_cstr(&s));
828 /* Force unwildcard the in_port. */
829 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
832 /* No mask key, unwildcard everything except fields whose
833 * prerequisities are not met. */
834 memset(mask, 0x0, sizeof *mask);
836 for (id = 0; id < MFF_N_IDS; ++id) {
837 /* Skip registers and metadata. */
838 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
839 && id != MFF_METADATA) {
840 const struct mf_field *mf = mf_from_id(id);
841 if (mf_are_prereqs_ok(mf, flow)) {
842 mf_mask_field(mf, mask);
852 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
857 if (odp_flow_key_to_flow(key, key_len, flow)) {
858 /* This should not happen: it indicates that odp_flow_key_from_flow()
859 * and odp_flow_key_to_flow() disagree on the acceptable form of a
860 * flow. Log the problem as an error, with enough details to enable
862 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
864 if (!VLOG_DROP_ERR(&rl)) {
868 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
869 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
876 in_port = flow->in_port.odp_port;
877 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
885 dpif_netdev_flow_get(const struct dpif *dpif,
886 const struct nlattr *nl_key, size_t nl_key_len,
887 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
889 struct dp_netdev *dp = get_dp_netdev(dpif);
890 struct dp_netdev_flow *netdev_flow;
894 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
899 ovs_mutex_lock(&dp_netdev_mutex);
900 netdev_flow = dp_netdev_find_flow(dp, &key);
903 get_dpif_flow_stats(netdev_flow, stats);
906 *actionsp = ofpbuf_clone_data(netdev_flow->actions,
907 netdev_flow->actions_len);
912 ovs_mutex_unlock(&dp_netdev_mutex);
918 set_flow_actions(struct dp_netdev_flow *netdev_flow,
919 const struct nlattr *actions, size_t actions_len)
921 netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
922 netdev_flow->actions_len = actions_len;
923 memcpy(netdev_flow->actions, actions, actions_len);
928 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
929 const struct flow_wildcards *wc,
930 const struct nlattr *actions,
933 struct dp_netdev_flow *netdev_flow;
937 netdev_flow = xzalloc(sizeof *netdev_flow);
938 netdev_flow->flow = *flow;
940 match_init(&match, flow, wc);
941 cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
942 ovs_rwlock_wrlock(&dp->cls.rwlock);
943 classifier_insert(&dp->cls, &netdev_flow->cr);
944 ovs_rwlock_unlock(&dp->cls.rwlock);
946 error = set_flow_actions(netdev_flow, actions, actions_len);
948 ovs_rwlock_wrlock(&dp->cls.rwlock);
949 classifier_remove(&dp->cls, &netdev_flow->cr);
950 ovs_rwlock_unlock(&dp->cls.rwlock);
951 cls_rule_destroy(&netdev_flow->cr);
957 hmap_insert(&dp->flow_table, &netdev_flow->node, flow_hash(flow, 0));
962 clear_stats(struct dp_netdev_flow *netdev_flow)
964 netdev_flow->used = 0;
965 netdev_flow->packet_count = 0;
966 netdev_flow->byte_count = 0;
967 netdev_flow->tcp_flags = 0;
971 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
973 struct dp_netdev *dp = get_dp_netdev(dpif);
974 struct dp_netdev_flow *netdev_flow;
976 struct flow_wildcards wc;
979 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
983 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
984 put->mask, put->mask_len,
990 ovs_mutex_lock(&dp_netdev_mutex);
991 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
993 if (put->flags & DPIF_FP_CREATE) {
994 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
996 memset(put->stats, 0, sizeof *put->stats);
998 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1007 if (put->flags & DPIF_FP_MODIFY
1008 && flow_equal(&flow, &netdev_flow->flow)) {
1009 error = set_flow_actions(netdev_flow, put->actions,
1013 get_dpif_flow_stats(netdev_flow, put->stats);
1015 if (put->flags & DPIF_FP_ZERO_STATS) {
1016 clear_stats(netdev_flow);
1019 } else if (put->flags & DPIF_FP_CREATE) {
1022 /* Overlapping flow. */
1026 ovs_mutex_unlock(&dp_netdev_mutex);
1032 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1034 struct dp_netdev *dp = get_dp_netdev(dpif);
1035 struct dp_netdev_flow *netdev_flow;
1039 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1044 ovs_mutex_lock(&dp_netdev_mutex);
1045 netdev_flow = dp_netdev_find_flow(dp, &key);
1048 get_dpif_flow_stats(netdev_flow, del->stats);
1050 dp_netdev_free_flow(dp, netdev_flow);
1054 ovs_mutex_unlock(&dp_netdev_mutex);
1059 struct dp_netdev_flow_state {
1062 struct nlattr *actions;
1063 struct odputil_keybuf keybuf;
1064 struct odputil_keybuf maskbuf;
1065 struct dpif_flow_stats stats;
1069 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1071 struct dp_netdev_flow_state *state;
1073 *statep = state = xmalloc(sizeof *state);
1076 state->actions = NULL;
1081 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1082 const struct nlattr **key, size_t *key_len,
1083 const struct nlattr **mask, size_t *mask_len,
1084 const struct nlattr **actions, size_t *actions_len,
1085 const struct dpif_flow_stats **stats)
1087 struct dp_netdev_flow_state *state = state_;
1088 struct dp_netdev *dp = get_dp_netdev(dpif);
1089 struct dp_netdev_flow *netdev_flow;
1090 struct hmap_node *node;
1092 ovs_mutex_lock(&dp_netdev_mutex);
1093 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1095 ovs_mutex_unlock(&dp_netdev_mutex);
1099 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1104 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1105 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1106 netdev_flow->flow.in_port.odp_port);
1109 *key_len = buf.size;
1114 struct flow_wildcards wc;
1116 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1117 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1118 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1119 odp_to_u32(wc.masks.in_port.odp_port));
1122 *mask_len = buf.size;
1126 free(state->actions);
1127 state->actions = xmemdup(netdev_flow->actions,
1128 netdev_flow->actions_len);
1130 *actions = state->actions;
1131 *actions_len = netdev_flow->actions_len;
1135 get_dpif_flow_stats(netdev_flow, &state->stats);
1136 *stats = &state->stats;
1139 ovs_mutex_unlock(&dp_netdev_mutex);
1144 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1146 struct dp_netdev_flow_state *state = state_;
1148 free(state->actions);
1154 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1156 struct dp_netdev *dp = get_dp_netdev(dpif);
1157 struct pkt_metadata *md = &execute->md;
1160 if (execute->packet->size < ETH_HEADER_LEN ||
1161 execute->packet->size > UINT16_MAX) {
1165 /* Extract flow key. */
1166 flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1167 (union flow_in_port *)&md->in_port, &key);
1168 ovs_mutex_lock(&dp_netdev_mutex);
1169 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1170 execute->actions_len);
1171 ovs_mutex_unlock(&dp_netdev_mutex);
1176 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1182 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1183 uint32_t queue_id, uint32_t *priority)
1185 *priority = queue_id;
1189 static struct dp_netdev_queue *
1190 find_nonempty_queue(struct dpif *dpif)
1192 struct dp_netdev *dp = get_dp_netdev(dpif);
1195 for (i = 0; i < N_QUEUES; i++) {
1196 struct dp_netdev_queue *q = &dp->queues[i];
1197 if (q->head != q->tail) {
1205 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1208 struct dp_netdev_queue *q;
1211 ovs_mutex_lock(&dp_netdev_mutex);
1212 q = find_nonempty_queue(dpif);
1214 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1216 *upcall = u->upcall;
1225 ovs_mutex_unlock(&dp_netdev_mutex);
1231 dpif_netdev_recv_wait(struct dpif *dpif)
1233 struct dp_netdev *dp = get_dp_netdev(dpif);
1236 ovs_mutex_lock(&dp_netdev_mutex);
1237 seq = seq_read(dp->queue_seq);
1238 if (find_nonempty_queue(dpif)) {
1239 poll_immediate_wake();
1241 seq_wait(dp->queue_seq, seq);
1243 ovs_mutex_unlock(&dp_netdev_mutex);
1247 dpif_netdev_recv_purge(struct dpif *dpif)
1249 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1250 ovs_mutex_lock(&dp_netdev_mutex);
1251 dp_netdev_purge_queues(dpif_netdev->dp);
1252 ovs_mutex_unlock(&dp_netdev_mutex);
1256 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1257 const struct ofpbuf *packet)
1259 netdev_flow->used = time_msec();
1260 netdev_flow->packet_count++;
1261 netdev_flow->byte_count += packet->size;
1262 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1266 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1267 struct pkt_metadata *md)
1269 struct dp_netdev_flow *netdev_flow;
1272 if (packet->size < ETH_HEADER_LEN) {
1275 flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1276 (union flow_in_port *)&md->in_port, &key);
1277 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1279 dp_netdev_flow_used(netdev_flow, packet);
1280 dp_netdev_execute_actions(dp, &key, packet, md,
1281 netdev_flow->actions,
1282 netdev_flow->actions_len);
1283 ovsthread_counter_inc(dp->n_hit, 1);
1285 ovsthread_counter_inc(dp->n_missed, 1);
1286 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1291 dpif_netdev_run(struct dpif *dpif)
1293 struct dp_netdev_port *port;
1294 struct dp_netdev *dp;
1295 struct ofpbuf packet;
1298 ovs_mutex_lock(&dp_netdev_mutex);
1299 dp = get_dp_netdev(dpif);
1300 ofpbuf_init(&packet, 0);
1302 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu;
1304 HMAP_FOR_EACH (port, node, &dp->ports) {
1307 /* Reset packet contents. Packet data may have been stolen. */
1308 ofpbuf_clear(&packet);
1309 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM, buf_size);
1311 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1313 struct pkt_metadata md = PKT_METADATA_INITIALIZER(port->port_no);
1314 dp_netdev_port_input(dp, &packet, &md);
1315 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1316 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1318 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1319 netdev_get_name(port->netdev), ovs_strerror(error));
1322 ofpbuf_uninit(&packet);
1323 ovs_mutex_unlock(&dp_netdev_mutex);
1327 dpif_netdev_wait(struct dpif *dpif)
1329 struct dp_netdev_port *port;
1331 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1332 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1333 * A makes it to poll_block().
1335 * But I think it doesn't matter:
1337 * - In the dpif_port_add() case, A will not wake up when a packet
1338 * arrives on the new port, but this would also happen if the
1339 * ordering were reversed.
1341 * - In the dpif_port_remove() case, A might wake up spuriously, but
1342 * that is harmless. */
1344 ovs_mutex_lock(&dp_netdev_mutex);
1345 HMAP_FOR_EACH (port, node, &get_dp_netdev(dpif)->ports) {
1347 netdev_rx_wait(port->rx);
1350 ovs_mutex_unlock(&dp_netdev_mutex);
1354 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1355 odp_port_t out_port)
1357 struct dp_netdev_port *p = dp_netdev_lookup_port(dp, out_port);
1359 netdev_send(p->netdev, packet);
1364 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1365 int queue_no, const struct flow *flow,
1366 const struct nlattr *userdata)
1368 struct dp_netdev_queue *q = &dp->queues[queue_no];
1369 if (q->head - q->tail < MAX_QUEUE_LEN) {
1370 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1371 struct dpif_upcall *upcall = &u->upcall;
1372 struct ofpbuf *buf = &u->buf;
1375 upcall->type = queue_no;
1377 /* Allocate buffer big enough for everything. */
1378 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1380 buf_size += NLA_ALIGN(userdata->nla_len);
1382 ofpbuf_init(buf, buf_size);
1385 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1386 upcall->key = buf->data;
1387 upcall->key_len = buf->size;
1391 upcall->userdata = ofpbuf_put(buf, userdata,
1392 NLA_ALIGN(userdata->nla_len));
1395 /* Steal packet data. */
1396 ovs_assert(packet->source == OFPBUF_MALLOC);
1397 upcall->packet = *packet;
1398 ofpbuf_use(packet, NULL, 0);
1400 seq_change(dp->queue_seq);
1404 ovsthread_counter_inc(dp->n_lost, 1);
1409 struct dp_netdev_execute_aux {
1410 struct dp_netdev *dp;
1411 const struct flow *key;
1415 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1416 const struct pkt_metadata *md OVS_UNUSED,
1417 const struct nlattr *a, bool may_steal)
1419 struct dp_netdev_execute_aux *aux = aux_;
1420 int type = nl_attr_type(a);
1422 switch ((enum ovs_action_attr)type) {
1423 case OVS_ACTION_ATTR_OUTPUT:
1424 dp_netdev_output_port(aux->dp, packet, u32_to_odp(nl_attr_get_u32(a)));
1427 case OVS_ACTION_ATTR_USERSPACE: {
1428 const struct nlattr *userdata;
1430 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1432 /* Make a copy if we are not allowed to steal the packet's data. */
1434 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1436 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1439 ofpbuf_uninit(packet);
1443 case OVS_ACTION_ATTR_PUSH_VLAN:
1444 case OVS_ACTION_ATTR_POP_VLAN:
1445 case OVS_ACTION_ATTR_PUSH_MPLS:
1446 case OVS_ACTION_ATTR_POP_MPLS:
1447 case OVS_ACTION_ATTR_SET:
1448 case OVS_ACTION_ATTR_SAMPLE:
1449 case OVS_ACTION_ATTR_UNSPEC:
1450 case __OVS_ACTION_ATTR_MAX:
1456 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1457 struct ofpbuf *packet, struct pkt_metadata *md,
1458 const struct nlattr *actions, size_t actions_len)
1460 struct dp_netdev_execute_aux aux = {dp, key};
1462 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1465 const struct dpif_class dpif_netdev_class = {
1467 dpif_netdev_enumerate,
1468 dpif_netdev_port_open_type,
1471 dpif_netdev_destroy,
1474 dpif_netdev_get_stats,
1475 dpif_netdev_port_add,
1476 dpif_netdev_port_del,
1477 dpif_netdev_port_query_by_number,
1478 dpif_netdev_port_query_by_name,
1479 NULL, /* port_get_pid */
1480 dpif_netdev_port_dump_start,
1481 dpif_netdev_port_dump_next,
1482 dpif_netdev_port_dump_done,
1483 dpif_netdev_port_poll,
1484 dpif_netdev_port_poll_wait,
1485 dpif_netdev_flow_get,
1486 dpif_netdev_flow_put,
1487 dpif_netdev_flow_del,
1488 dpif_netdev_flow_flush,
1489 dpif_netdev_flow_dump_start,
1490 dpif_netdev_flow_dump_next,
1491 dpif_netdev_flow_dump_done,
1492 dpif_netdev_execute,
1494 dpif_netdev_recv_set,
1495 dpif_netdev_queue_to_priority,
1497 dpif_netdev_recv_wait,
1498 dpif_netdev_recv_purge,
1502 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1503 const char *argv[], void *aux OVS_UNUSED)
1505 struct dp_netdev_port *port;
1506 struct dp_netdev *dp;
1509 dp = shash_find_data(&dp_netdevs, argv[1]);
1510 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1511 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1515 if (get_port_by_name(dp, argv[2], &port)) {
1516 unixctl_command_reply_error(conn, "unknown port");
1520 port_no = u32_to_odp(atoi(argv[3]));
1521 if (!port_no || port_no == ODPP_NONE) {
1522 unixctl_command_reply_error(conn, "bad port number");
1525 if (dp_netdev_lookup_port(dp, port_no)) {
1526 unixctl_command_reply_error(conn, "port number already in use");
1529 hmap_remove(&dp->ports, &port->node);
1530 port->port_no = port_no;
1531 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
1532 seq_change(dp->port_seq);
1533 unixctl_command_reply(conn, NULL);
1537 dpif_dummy_register__(const char *type)
1539 struct dpif_class *class;
1541 class = xmalloc(sizeof *class);
1542 *class = dpif_netdev_class;
1543 class->type = xstrdup(type);
1544 dp_register_provider(class);
1548 dpif_dummy_register(bool override)
1555 dp_enumerate_types(&types);
1556 SSET_FOR_EACH (type, &types) {
1557 if (!dp_unregister_provider(type)) {
1558 dpif_dummy_register__(type);
1561 sset_destroy(&types);
1564 dpif_dummy_register__("dummy");
1566 unixctl_command_register("dpif-dummy/change-port-number",
1567 "DP PORT NEW-NUMBER",
1568 3, 3, dpif_dummy_change_port_number, NULL);