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"
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_PORTS = 256 }; /* Maximum number of ports. */
69 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
71 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
72 * headers to be aligned on a 4-byte boundary. */
73 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
76 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
77 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
78 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
79 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
81 struct dp_netdev_upcall {
82 struct dpif_upcall upcall; /* Queued upcall information. */
83 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
86 struct dp_netdev_queue {
87 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
88 unsigned int head, tail;
91 /* Datapath based on the network device interface from netdev.h. */
93 const struct dpif_class *class;
97 int max_mtu; /* Maximum MTU of any port added so far. */
99 struct dp_netdev_queue queues[N_QUEUES];
100 struct classifier cls; /* Classifier. */
101 struct hmap flow_table; /* Flow table. */
102 struct seq *queue_seq; /* Incremented whenever a packet is queued. */
105 struct ovsthread_counter *n_hit; /* Number of flow table matches. */
106 struct ovsthread_counter *n_missed; /* Number of flow table misses. */
107 struct ovsthread_counter *n_lost; /* Number of misses not passed up. */
110 struct dp_netdev_port *ports[MAX_PORTS];
111 struct list port_list;
112 struct seq *port_seq; /* Incremented whenever a port changes. */
115 /* A port in a netdev-based datapath. */
116 struct dp_netdev_port {
117 odp_port_t port_no; /* Index into dp_netdev's 'ports'. */
118 struct list node; /* Element in dp_netdev's 'port_list'. */
119 struct netdev *netdev;
120 struct netdev_saved_flags *sf;
121 struct netdev_rx *rx;
122 char *type; /* Port type as requested by user. */
125 /* A flow in dp_netdev's 'flow_table'. */
126 struct dp_netdev_flow {
127 /* Packet classification. */
128 struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
130 /* Hash table index by unmasked flow.*/
131 struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
132 struct flow flow; /* The flow that created this entry. */
135 long long int used; /* Last used time, in monotonic msecs. */
136 long long int packet_count; /* Number of packets matched. */
137 long long int byte_count; /* Number of bytes matched. */
138 uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
141 struct nlattr *actions;
145 /* Interface to netdev-based datapath. */
148 struct dp_netdev *dp;
149 uint64_t last_port_seq;
152 /* All netdev-based datapaths. */
153 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
155 /* Global lock for all data. */
156 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
158 static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
159 struct dp_netdev_port **portp);
160 static int get_port_by_name(struct dp_netdev *, const char *devname,
161 struct dp_netdev_port **portp);
162 static void dp_netdev_free(struct dp_netdev *);
163 static void dp_netdev_flow_flush(struct dp_netdev *);
164 static int do_add_port(struct dp_netdev *, const char *devname,
165 const char *type, odp_port_t port_no);
166 static int do_del_port(struct dp_netdev *, odp_port_t port_no);
167 static int dpif_netdev_open(const struct dpif_class *, const char *name,
168 bool create, struct dpif **);
169 static int dp_netdev_output_userspace(struct dp_netdev *, struct ofpbuf *,
170 int queue_no, const struct flow *,
171 const struct nlattr *userdata);
172 static void dp_netdev_execute_actions(struct dp_netdev *, const struct flow *,
173 struct ofpbuf *, struct pkt_metadata *,
174 const struct nlattr *actions,
176 static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
177 struct pkt_metadata *md);
179 static struct dpif_netdev *
180 dpif_netdev_cast(const struct dpif *dpif)
182 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
183 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
186 static struct dp_netdev *
187 get_dp_netdev(const struct dpif *dpif)
189 return dpif_netdev_cast(dpif)->dp;
193 dpif_netdev_enumerate(struct sset *all_dps)
195 struct shash_node *node;
197 ovs_mutex_lock(&dp_netdev_mutex);
198 SHASH_FOR_EACH(node, &dp_netdevs) {
199 sset_add(all_dps, node->name);
201 ovs_mutex_unlock(&dp_netdev_mutex);
207 dpif_netdev_class_is_dummy(const struct dpif_class *class)
209 return class != &dpif_netdev_class;
213 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
215 return strcmp(type, "internal") ? type
216 : dpif_netdev_class_is_dummy(class) ? "dummy"
221 create_dpif_netdev(struct dp_netdev *dp)
223 uint16_t netflow_id = hash_string(dp->name, 0);
224 struct dpif_netdev *dpif;
228 dpif = xmalloc(sizeof *dpif);
229 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
231 dpif->last_port_seq = seq_read(dp->port_seq);
236 /* Choose an unused, non-zero port number and return it on success.
237 * Return ODPP_NONE on failure. */
239 choose_port(struct dp_netdev *dp, const char *name)
243 if (dp->class != &dpif_netdev_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 classifier_init(&dp->cls, NULL);
295 hmap_init(&dp->flow_table);
297 dp->n_hit = ovsthread_counter_create();
298 dp->n_missed = ovsthread_counter_create();
299 dp->n_lost = ovsthread_counter_create();
301 list_init(&dp->port_list);
302 dp->port_seq = seq_create();
304 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
310 shash_add(&dp_netdevs, name, dp);
317 dpif_netdev_open(const struct dpif_class *class, const char *name,
318 bool create, struct dpif **dpifp)
320 struct dp_netdev *dp;
323 ovs_mutex_lock(&dp_netdev_mutex);
324 dp = shash_find_data(&dp_netdevs, name);
326 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
328 error = (dp->class != class ? EINVAL
333 *dpifp = create_dpif_netdev(dp);
335 ovs_mutex_unlock(&dp_netdev_mutex);
341 dp_netdev_purge_queues(struct dp_netdev *dp)
345 for (i = 0; i < N_QUEUES; i++) {
346 struct dp_netdev_queue *q = &dp->queues[i];
348 while (q->tail != q->head) {
349 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
350 ofpbuf_uninit(&u->upcall.packet);
351 ofpbuf_uninit(&u->buf);
357 dp_netdev_free(struct dp_netdev *dp)
359 struct dp_netdev_port *port, *next;
361 dp_netdev_flow_flush(dp);
362 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
363 do_del_port(dp, port->port_no);
365 ovsthread_counter_destroy(dp->n_hit);
366 ovsthread_counter_destroy(dp->n_missed);
367 ovsthread_counter_destroy(dp->n_lost);
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 = ovsthread_counter_read(dp->n_hit);
414 stats->n_missed = ovsthread_counter_read(dp->n_missed);
415 stats->n_lost = ovsthread_counter_read(dp->n_lost);
416 stats->n_masks = UINT32_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 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
642 ovs_rwlock_wrlock(&dp->cls.rwlock);
643 classifier_remove(&dp->cls, &netdev_flow->cr);
644 ovs_rwlock_unlock(&dp->cls.rwlock);
645 cls_rule_destroy(&netdev_flow->cr);
647 hmap_remove(&dp->flow_table, &netdev_flow->node);
648 free(netdev_flow->actions);
653 dp_netdev_flow_flush(struct dp_netdev *dp)
655 struct dp_netdev_flow *netdev_flow, *next;
657 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
658 dp_netdev_free_flow(dp, netdev_flow);
663 dpif_netdev_flow_flush(struct dpif *dpif)
665 struct dp_netdev *dp = get_dp_netdev(dpif);
667 ovs_mutex_lock(&dp_netdev_mutex);
668 dp_netdev_flow_flush(dp);
669 ovs_mutex_unlock(&dp_netdev_mutex);
674 struct dp_netdev_port_state {
680 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
682 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
687 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
688 struct dpif_port *dpif_port)
690 struct dp_netdev_port_state *state = state_;
691 struct dp_netdev *dp = get_dp_netdev(dpif);
694 ovs_mutex_lock(&dp_netdev_mutex);
695 for (port_idx = odp_to_u32(state->port_no);
696 port_idx < MAX_PORTS; port_idx++) {
697 struct dp_netdev_port *port = dp->ports[port_idx];
700 state->name = xstrdup(netdev_get_name(port->netdev));
701 dpif_port->name = state->name;
702 dpif_port->type = port->type;
703 dpif_port->port_no = port->port_no;
704 state->port_no = u32_to_odp(port_idx + 1);
705 ovs_mutex_unlock(&dp_netdev_mutex);
710 ovs_mutex_unlock(&dp_netdev_mutex);
716 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
718 struct dp_netdev_port_state *state = state_;
725 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
727 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
728 uint64_t new_port_seq;
731 ovs_mutex_lock(&dp_netdev_mutex);
732 new_port_seq = seq_read(dpif->dp->port_seq);
733 if (dpif->last_port_seq != new_port_seq) {
734 dpif->last_port_seq = new_port_seq;
739 ovs_mutex_unlock(&dp_netdev_mutex);
745 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
747 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
749 ovs_mutex_lock(&dp_netdev_mutex);
750 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
751 ovs_mutex_unlock(&dp_netdev_mutex);
754 static struct dp_netdev_flow *
755 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
759 ovs_rwlock_wrlock(&dp->cls.rwlock);
760 cr = classifier_lookup(&dp->cls, flow, NULL);
761 ovs_rwlock_unlock(&dp->cls.rwlock);
764 ? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
768 static struct dp_netdev_flow *
769 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
771 struct dp_netdev_flow *netdev_flow;
773 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
775 if (flow_equal(&netdev_flow->flow, flow)) {
783 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
784 struct dpif_flow_stats *stats)
786 stats->n_packets = netdev_flow->packet_count;
787 stats->n_bytes = netdev_flow->byte_count;
788 stats->used = netdev_flow->used;
789 stats->tcp_flags = netdev_flow->tcp_flags;
793 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
794 const struct nlattr *mask_key,
795 uint32_t mask_key_len, const struct flow *flow,
799 if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
800 /* This should not happen: it indicates that
801 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
802 * disagree on the acceptable form of a mask. Log the problem
803 * as an error, with enough details to enable debugging. */
804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
806 if (!VLOG_DROP_ERR(&rl)) {
810 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
812 VLOG_ERR("internal error parsing flow mask %s", ds_cstr(&s));
818 /* Force unwildcard the in_port. */
819 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
822 /* No mask key, unwildcard everything except fields whose
823 * prerequisities are not met. */
824 memset(mask, 0x0, sizeof *mask);
826 for (id = 0; id < MFF_N_IDS; ++id) {
827 /* Skip registers and metadata. */
828 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
829 && id != MFF_METADATA) {
830 const struct mf_field *mf = mf_from_id(id);
831 if (mf_are_prereqs_ok(mf, flow)) {
832 mf_mask_field(mf, mask);
842 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
847 if (odp_flow_key_to_flow(key, key_len, flow)) {
848 /* This should not happen: it indicates that odp_flow_key_from_flow()
849 * and odp_flow_key_to_flow() disagree on the acceptable form of a
850 * flow. Log the problem as an error, with enough details to enable
852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
854 if (!VLOG_DROP_ERR(&rl)) {
858 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
859 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
866 in_port = flow->in_port.odp_port;
867 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
875 dpif_netdev_flow_get(const struct dpif *dpif,
876 const struct nlattr *nl_key, size_t nl_key_len,
877 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
879 struct dp_netdev *dp = get_dp_netdev(dpif);
880 struct dp_netdev_flow *netdev_flow;
884 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
889 ovs_mutex_lock(&dp_netdev_mutex);
890 netdev_flow = dp_netdev_find_flow(dp, &key);
893 get_dpif_flow_stats(netdev_flow, stats);
896 *actionsp = ofpbuf_clone_data(netdev_flow->actions,
897 netdev_flow->actions_len);
902 ovs_mutex_unlock(&dp_netdev_mutex);
908 set_flow_actions(struct dp_netdev_flow *netdev_flow,
909 const struct nlattr *actions, size_t actions_len)
911 netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
912 netdev_flow->actions_len = actions_len;
913 memcpy(netdev_flow->actions, actions, actions_len);
918 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
919 const struct flow_wildcards *wc,
920 const struct nlattr *actions,
923 struct dp_netdev_flow *netdev_flow;
927 netdev_flow = xzalloc(sizeof *netdev_flow);
928 netdev_flow->flow = *flow;
930 match_init(&match, flow, wc);
931 cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
932 ovs_rwlock_wrlock(&dp->cls.rwlock);
933 classifier_insert(&dp->cls, &netdev_flow->cr);
934 ovs_rwlock_unlock(&dp->cls.rwlock);
936 error = set_flow_actions(netdev_flow, actions, actions_len);
938 ovs_rwlock_wrlock(&dp->cls.rwlock);
939 classifier_remove(&dp->cls, &netdev_flow->cr);
940 ovs_rwlock_unlock(&dp->cls.rwlock);
941 cls_rule_destroy(&netdev_flow->cr);
947 hmap_insert(&dp->flow_table, &netdev_flow->node, flow_hash(flow, 0));
952 clear_stats(struct dp_netdev_flow *netdev_flow)
954 netdev_flow->used = 0;
955 netdev_flow->packet_count = 0;
956 netdev_flow->byte_count = 0;
957 netdev_flow->tcp_flags = 0;
961 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
963 struct dp_netdev *dp = get_dp_netdev(dpif);
964 struct dp_netdev_flow *netdev_flow;
966 struct flow_wildcards wc;
969 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
973 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
974 put->mask, put->mask_len,
980 ovs_mutex_lock(&dp_netdev_mutex);
981 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
983 if (put->flags & DPIF_FP_CREATE) {
984 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
986 memset(put->stats, 0, sizeof *put->stats);
988 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
997 if (put->flags & DPIF_FP_MODIFY
998 && flow_equal(&flow, &netdev_flow->flow)) {
999 error = set_flow_actions(netdev_flow, put->actions,
1003 get_dpif_flow_stats(netdev_flow, put->stats);
1005 if (put->flags & DPIF_FP_ZERO_STATS) {
1006 clear_stats(netdev_flow);
1009 } else if (put->flags & DPIF_FP_CREATE) {
1012 /* Overlapping flow. */
1016 ovs_mutex_unlock(&dp_netdev_mutex);
1022 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1024 struct dp_netdev *dp = get_dp_netdev(dpif);
1025 struct dp_netdev_flow *netdev_flow;
1029 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1034 ovs_mutex_lock(&dp_netdev_mutex);
1035 netdev_flow = dp_netdev_find_flow(dp, &key);
1038 get_dpif_flow_stats(netdev_flow, del->stats);
1040 dp_netdev_free_flow(dp, netdev_flow);
1044 ovs_mutex_unlock(&dp_netdev_mutex);
1049 struct dp_netdev_flow_state {
1052 struct nlattr *actions;
1053 struct odputil_keybuf keybuf;
1054 struct odputil_keybuf maskbuf;
1055 struct dpif_flow_stats stats;
1059 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1061 struct dp_netdev_flow_state *state;
1063 *statep = state = xmalloc(sizeof *state);
1066 state->actions = NULL;
1071 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1072 const struct nlattr **key, size_t *key_len,
1073 const struct nlattr **mask, size_t *mask_len,
1074 const struct nlattr **actions, size_t *actions_len,
1075 const struct dpif_flow_stats **stats)
1077 struct dp_netdev_flow_state *state = state_;
1078 struct dp_netdev *dp = get_dp_netdev(dpif);
1079 struct dp_netdev_flow *netdev_flow;
1080 struct hmap_node *node;
1082 ovs_mutex_lock(&dp_netdev_mutex);
1083 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1085 ovs_mutex_unlock(&dp_netdev_mutex);
1089 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1094 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1095 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1096 netdev_flow->flow.in_port.odp_port);
1099 *key_len = buf.size;
1104 struct flow_wildcards wc;
1106 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1107 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1108 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1109 odp_to_u32(wc.masks.in_port.odp_port));
1112 *mask_len = buf.size;
1116 free(state->actions);
1117 state->actions = xmemdup(netdev_flow->actions,
1118 netdev_flow->actions_len);
1120 *actions = state->actions;
1121 *actions_len = netdev_flow->actions_len;
1125 get_dpif_flow_stats(netdev_flow, &state->stats);
1126 *stats = &state->stats;
1129 ovs_mutex_unlock(&dp_netdev_mutex);
1134 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1136 struct dp_netdev_flow_state *state = state_;
1138 free(state->actions);
1144 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1146 struct dp_netdev *dp = get_dp_netdev(dpif);
1147 struct pkt_metadata *md = &execute->md;
1150 if (execute->packet->size < ETH_HEADER_LEN ||
1151 execute->packet->size > UINT16_MAX) {
1155 /* Extract flow key. */
1156 flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1157 (union flow_in_port *)&md->in_port, &key);
1158 ovs_mutex_lock(&dp_netdev_mutex);
1159 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1160 execute->actions_len);
1161 ovs_mutex_unlock(&dp_netdev_mutex);
1166 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1172 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1173 uint32_t queue_id, uint32_t *priority)
1175 *priority = queue_id;
1179 static struct dp_netdev_queue *
1180 find_nonempty_queue(struct dpif *dpif)
1182 struct dp_netdev *dp = get_dp_netdev(dpif);
1185 for (i = 0; i < N_QUEUES; i++) {
1186 struct dp_netdev_queue *q = &dp->queues[i];
1187 if (q->head != q->tail) {
1195 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1198 struct dp_netdev_queue *q;
1201 ovs_mutex_lock(&dp_netdev_mutex);
1202 q = find_nonempty_queue(dpif);
1204 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1206 *upcall = u->upcall;
1215 ovs_mutex_unlock(&dp_netdev_mutex);
1221 dpif_netdev_recv_wait(struct dpif *dpif)
1223 struct dp_netdev *dp = get_dp_netdev(dpif);
1226 ovs_mutex_lock(&dp_netdev_mutex);
1227 seq = seq_read(dp->queue_seq);
1228 if (find_nonempty_queue(dpif)) {
1229 poll_immediate_wake();
1231 seq_wait(dp->queue_seq, seq);
1233 ovs_mutex_unlock(&dp_netdev_mutex);
1237 dpif_netdev_recv_purge(struct dpif *dpif)
1239 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1240 ovs_mutex_lock(&dp_netdev_mutex);
1241 dp_netdev_purge_queues(dpif_netdev->dp);
1242 ovs_mutex_unlock(&dp_netdev_mutex);
1246 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1247 const struct ofpbuf *packet)
1249 netdev_flow->used = time_msec();
1250 netdev_flow->packet_count++;
1251 netdev_flow->byte_count += packet->size;
1252 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1256 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1257 struct pkt_metadata *md)
1259 struct dp_netdev_flow *netdev_flow;
1262 if (packet->size < ETH_HEADER_LEN) {
1265 flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1266 (union flow_in_port *)&md->in_port, &key);
1267 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1269 dp_netdev_flow_used(netdev_flow, packet);
1270 dp_netdev_execute_actions(dp, &key, packet, md,
1271 netdev_flow->actions,
1272 netdev_flow->actions_len);
1273 ovsthread_counter_inc(dp->n_hit, 1);
1275 ovsthread_counter_inc(dp->n_missed, 1);
1276 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1281 dpif_netdev_run(struct dpif *dpif)
1283 struct dp_netdev_port *port;
1284 struct dp_netdev *dp;
1285 struct ofpbuf packet;
1288 ovs_mutex_lock(&dp_netdev_mutex);
1289 dp = get_dp_netdev(dpif);
1290 ofpbuf_init(&packet, 0);
1292 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu;
1294 LIST_FOR_EACH (port, node, &dp->port_list) {
1297 /* Reset packet contents. Packet data may have been stolen. */
1298 ofpbuf_clear(&packet);
1299 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM, buf_size);
1301 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1303 struct pkt_metadata md = PKT_METADATA_INITIALIZER(port->port_no);
1304 dp_netdev_port_input(dp, &packet, &md);
1305 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1306 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1308 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1309 netdev_get_name(port->netdev), ovs_strerror(error));
1312 ofpbuf_uninit(&packet);
1313 ovs_mutex_unlock(&dp_netdev_mutex);
1317 dpif_netdev_wait(struct dpif *dpif)
1319 struct dp_netdev_port *port;
1321 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1322 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1323 * A makes it to poll_block().
1325 * But I think it doesn't matter:
1327 * - In the dpif_port_add() case, A will not wake up when a packet
1328 * arrives on the new port, but this would also happen if the
1329 * ordering were reversed.
1331 * - In the dpif_port_remove() case, A might wake up spuriously, but
1332 * that is harmless. */
1334 ovs_mutex_lock(&dp_netdev_mutex);
1335 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1337 netdev_rx_wait(port->rx);
1340 ovs_mutex_unlock(&dp_netdev_mutex);
1344 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1345 odp_port_t out_port)
1347 struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
1349 netdev_send(p->netdev, packet);
1354 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1355 int queue_no, const struct flow *flow,
1356 const struct nlattr *userdata)
1358 struct dp_netdev_queue *q = &dp->queues[queue_no];
1359 if (q->head - q->tail < MAX_QUEUE_LEN) {
1360 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1361 struct dpif_upcall *upcall = &u->upcall;
1362 struct ofpbuf *buf = &u->buf;
1365 upcall->type = queue_no;
1367 /* Allocate buffer big enough for everything. */
1368 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1370 buf_size += NLA_ALIGN(userdata->nla_len);
1372 ofpbuf_init(buf, buf_size);
1375 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1376 upcall->key = buf->data;
1377 upcall->key_len = buf->size;
1381 upcall->userdata = ofpbuf_put(buf, userdata,
1382 NLA_ALIGN(userdata->nla_len));
1385 /* Steal packet data. */
1386 ovs_assert(packet->source == OFPBUF_MALLOC);
1387 upcall->packet = *packet;
1388 ofpbuf_use(packet, NULL, 0);
1390 seq_change(dp->queue_seq);
1394 ovsthread_counter_inc(dp->n_lost, 1);
1399 struct dp_netdev_execute_aux {
1400 struct dp_netdev *dp;
1401 const struct flow *key;
1405 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1406 const struct pkt_metadata *md OVS_UNUSED,
1407 const struct nlattr *a, bool may_steal)
1409 struct dp_netdev_execute_aux *aux = aux_;
1410 int type = nl_attr_type(a);
1412 switch ((enum ovs_action_attr)type) {
1413 case OVS_ACTION_ATTR_OUTPUT:
1414 dp_netdev_output_port(aux->dp, packet, u32_to_odp(nl_attr_get_u32(a)));
1417 case OVS_ACTION_ATTR_USERSPACE: {
1418 const struct nlattr *userdata;
1420 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1422 /* Make a copy if we are not allowed to steal the packet's data. */
1424 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1426 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1429 ofpbuf_uninit(packet);
1433 case OVS_ACTION_ATTR_PUSH_VLAN:
1434 case OVS_ACTION_ATTR_POP_VLAN:
1435 case OVS_ACTION_ATTR_PUSH_MPLS:
1436 case OVS_ACTION_ATTR_POP_MPLS:
1437 case OVS_ACTION_ATTR_SET:
1438 case OVS_ACTION_ATTR_SAMPLE:
1439 case OVS_ACTION_ATTR_UNSPEC:
1440 case __OVS_ACTION_ATTR_MAX:
1446 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1447 struct ofpbuf *packet, struct pkt_metadata *md,
1448 const struct nlattr *actions, size_t actions_len)
1450 struct dp_netdev_execute_aux aux = {dp, key};
1452 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1455 const struct dpif_class dpif_netdev_class = {
1457 dpif_netdev_enumerate,
1458 dpif_netdev_port_open_type,
1461 dpif_netdev_destroy,
1464 dpif_netdev_get_stats,
1465 dpif_netdev_port_add,
1466 dpif_netdev_port_del,
1467 dpif_netdev_port_query_by_number,
1468 dpif_netdev_port_query_by_name,
1469 NULL, /* port_get_pid */
1470 dpif_netdev_port_dump_start,
1471 dpif_netdev_port_dump_next,
1472 dpif_netdev_port_dump_done,
1473 dpif_netdev_port_poll,
1474 dpif_netdev_port_poll_wait,
1475 dpif_netdev_flow_get,
1476 dpif_netdev_flow_put,
1477 dpif_netdev_flow_del,
1478 dpif_netdev_flow_flush,
1479 dpif_netdev_flow_dump_start,
1480 dpif_netdev_flow_dump_next,
1481 dpif_netdev_flow_dump_done,
1482 dpif_netdev_execute,
1484 dpif_netdev_recv_set,
1485 dpif_netdev_queue_to_priority,
1487 dpif_netdev_recv_wait,
1488 dpif_netdev_recv_purge,
1492 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1493 const char *argv[], void *aux OVS_UNUSED)
1495 struct dp_netdev_port *port;
1496 struct dp_netdev *dp;
1499 dp = shash_find_data(&dp_netdevs, argv[1]);
1500 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1501 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1505 if (get_port_by_name(dp, argv[2], &port)) {
1506 unixctl_command_reply_error(conn, "unknown port");
1510 port_no = atoi(argv[3]);
1511 if (port_no <= 0 || port_no >= MAX_PORTS) {
1512 unixctl_command_reply_error(conn, "bad port number");
1515 if (dp->ports[port_no]) {
1516 unixctl_command_reply_error(conn, "port number already in use");
1519 dp->ports[odp_to_u32(port->port_no)] = NULL;
1520 dp->ports[port_no] = port;
1521 port->port_no = u32_to_odp(port_no);
1522 seq_change(dp->port_seq);
1523 unixctl_command_reply(conn, NULL);
1527 dpif_dummy_register__(const char *type)
1529 struct dpif_class *class;
1531 class = xmalloc(sizeof *class);
1532 *class = dpif_netdev_class;
1533 class->type = xstrdup(type);
1534 dp_register_provider(class);
1538 dpif_dummy_register(bool override)
1545 dp_enumerate_types(&types);
1546 SSET_FOR_EACH (type, &types) {
1547 if (!dp_unregister_provider(type)) {
1548 dpif_dummy_register__(type);
1551 sset_destroy(&types);
1554 dpif_dummy_register__("dummy");
1556 unixctl_command_register("dpif-dummy/change-port-number",
1557 "DP PORT NEW-NUMBER",
1558 3, 3, dpif_dummy_change_port_number, NULL);