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 long long int n_hit; /* Number of flow table matches. */
106 long long int n_missed; /* Number of flow table misses. */
107 long long int n_lost; /* Number of misses not passed to client. */
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 *,
174 const struct nlattr *actions,
176 static void dp_netdev_port_input(struct dp_netdev *dp,
177 struct dp_netdev_port *port,
178 struct ofpbuf *packet, uint32_t skb_priority,
179 uint32_t pkt_mark, const struct flow_tnl *tnl);
181 static struct dpif_netdev *
182 dpif_netdev_cast(const struct dpif *dpif)
184 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
185 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
188 static struct dp_netdev *
189 get_dp_netdev(const struct dpif *dpif)
191 return dpif_netdev_cast(dpif)->dp;
195 dpif_netdev_enumerate(struct sset *all_dps)
197 struct shash_node *node;
199 ovs_mutex_lock(&dp_netdev_mutex);
200 SHASH_FOR_EACH(node, &dp_netdevs) {
201 sset_add(all_dps, node->name);
203 ovs_mutex_unlock(&dp_netdev_mutex);
209 dpif_netdev_class_is_dummy(const struct dpif_class *class)
211 return class != &dpif_netdev_class;
215 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
217 return strcmp(type, "internal") ? type
218 : dpif_netdev_class_is_dummy(class) ? "dummy"
223 create_dpif_netdev(struct dp_netdev *dp)
225 uint16_t netflow_id = hash_string(dp->name, 0);
226 struct dpif_netdev *dpif;
230 dpif = xmalloc(sizeof *dpif);
231 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
233 dpif->last_port_seq = seq_read(dp->port_seq);
238 /* Choose an unused, non-zero port number and return it on success.
239 * Return ODPP_NONE on failure. */
241 choose_port(struct dp_netdev *dp, const char *name)
245 if (dp->class != &dpif_netdev_class) {
249 /* If the port name begins with "br", start the number search at
250 * 100 to make writing tests easier. */
251 if (!strncmp(name, "br", 2)) {
255 /* If the port name contains a number, try to assign that port number.
256 * This can make writing unit tests easier because port numbers are
258 for (p = name; *p != '\0'; p++) {
259 if (isdigit((unsigned char) *p)) {
260 port_no = start_no + strtol(p, NULL, 10);
261 if (port_no > 0 && port_no < MAX_PORTS
262 && !dp->ports[port_no]) {
263 return u32_to_odp(port_no);
270 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
271 if (!dp->ports[port_no]) {
272 return u32_to_odp(port_no);
280 create_dp_netdev(const char *name, const struct dpif_class *class,
281 struct dp_netdev **dpp)
283 struct dp_netdev *dp;
287 dp = xzalloc(sizeof *dp);
289 dp->name = xstrdup(name);
291 dp->max_mtu = ETH_PAYLOAD_MAX;
292 for (i = 0; i < N_QUEUES; i++) {
293 dp->queues[i].head = dp->queues[i].tail = 0;
295 dp->queue_seq = seq_create();
296 classifier_init(&dp->cls, NULL);
297 hmap_init(&dp->flow_table);
298 list_init(&dp->port_list);
299 dp->port_seq = seq_create();
301 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
307 shash_add(&dp_netdevs, name, dp);
314 dpif_netdev_open(const struct dpif_class *class, const char *name,
315 bool create, struct dpif **dpifp)
317 struct dp_netdev *dp;
320 ovs_mutex_lock(&dp_netdev_mutex);
321 dp = shash_find_data(&dp_netdevs, name);
323 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
325 error = (dp->class != class ? EINVAL
330 *dpifp = create_dpif_netdev(dp);
332 ovs_mutex_unlock(&dp_netdev_mutex);
338 dp_netdev_purge_queues(struct dp_netdev *dp)
342 for (i = 0; i < N_QUEUES; i++) {
343 struct dp_netdev_queue *q = &dp->queues[i];
345 while (q->tail != q->head) {
346 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
347 ofpbuf_uninit(&u->upcall.packet);
348 ofpbuf_uninit(&u->buf);
354 dp_netdev_free(struct dp_netdev *dp)
356 struct dp_netdev_port *port, *next;
358 dp_netdev_flow_flush(dp);
359 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
360 do_del_port(dp, port->port_no);
362 dp_netdev_purge_queues(dp);
363 seq_destroy(dp->queue_seq);
364 classifier_destroy(&dp->cls);
365 hmap_destroy(&dp->flow_table);
366 seq_destroy(dp->port_seq);
372 dpif_netdev_close(struct dpif *dpif)
374 struct dp_netdev *dp = get_dp_netdev(dpif);
376 ovs_mutex_lock(&dp_netdev_mutex);
378 ovs_assert(dp->open_cnt > 0);
379 if (--dp->open_cnt == 0 && dp->destroyed) {
380 shash_find_and_delete(&dp_netdevs, dp->name);
385 ovs_mutex_unlock(&dp_netdev_mutex);
389 dpif_netdev_destroy(struct dpif *dpif)
391 struct dp_netdev *dp = get_dp_netdev(dpif);
393 ovs_mutex_lock(&dp_netdev_mutex);
394 dp->destroyed = true;
395 ovs_mutex_unlock(&dp_netdev_mutex);
401 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
403 struct dp_netdev *dp = get_dp_netdev(dpif);
405 ovs_mutex_lock(&dp_netdev_mutex);
406 stats->n_flows = hmap_count(&dp->flow_table);
407 stats->n_hit = dp->n_hit;
408 stats->n_missed = dp->n_missed;
409 stats->n_lost = dp->n_lost;
410 stats->n_masks = UINT32_MAX;
411 stats->n_mask_hit = UINT64_MAX;
412 ovs_mutex_unlock(&dp_netdev_mutex);
418 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
421 struct netdev_saved_flags *sf;
422 struct dp_netdev_port *port;
423 struct netdev *netdev;
424 struct netdev_rx *rx;
425 enum netdev_flags flags;
426 const char *open_type;
430 /* XXX reject devices already in some dp_netdev. */
432 /* Open and validate network device. */
433 open_type = dpif_netdev_port_open_type(dp->class, type);
434 error = netdev_open(devname, open_type, &netdev);
438 /* XXX reject non-Ethernet devices */
440 netdev_get_flags(netdev, &flags);
441 if (flags & NETDEV_LOOPBACK) {
442 VLOG_ERR("%s: cannot add a loopback device", devname);
443 netdev_close(netdev);
447 error = netdev_rx_open(netdev, &rx);
449 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
450 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
451 devname, ovs_strerror(errno));
452 netdev_close(netdev);
456 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
459 netdev_close(netdev);
463 port = xmalloc(sizeof *port);
464 port->port_no = port_no;
465 port->netdev = netdev;
468 port->type = xstrdup(type);
470 error = netdev_get_mtu(netdev, &mtu);
471 if (!error && mtu > dp->max_mtu) {
475 list_push_back(&dp->port_list, &port->node);
476 dp->ports[odp_to_u32(port_no)] = port;
477 seq_change(dp->port_seq);
483 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
484 odp_port_t *port_nop)
486 struct dp_netdev *dp = get_dp_netdev(dpif);
487 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
488 const char *dpif_port;
492 ovs_mutex_lock(&dp_netdev_mutex);
493 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
494 if (*port_nop != ODPP_NONE) {
495 uint32_t port_idx = odp_to_u32(*port_nop);
496 if (port_idx >= MAX_PORTS) {
498 } else if (dp->ports[port_idx]) {
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 odp_to_u32(port_no) < MAX_PORTS;
537 get_port_by_number(struct dp_netdev *dp,
538 odp_port_t port_no, struct dp_netdev_port **portp)
540 if (!is_valid_port_number(port_no)) {
544 *portp = dp->ports[odp_to_u32(port_no)];
545 return *portp ? 0 : ENOENT;
550 get_port_by_name(struct dp_netdev *dp,
551 const char *devname, struct dp_netdev_port **portp)
553 struct dp_netdev_port *port;
555 LIST_FOR_EACH (port, node, &dp->port_list) {
556 if (!strcmp(netdev_get_name(port->netdev), devname)) {
565 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
567 struct dp_netdev_port *port;
570 error = get_port_by_number(dp, port_no, &port);
575 list_remove(&port->node);
576 dp->ports[odp_to_u32(port_no)] = NULL;
577 seq_change(dp->port_seq);
579 netdev_close(port->netdev);
580 netdev_restore_flags(port->sf);
581 netdev_rx_close(port->rx);
589 answer_port_query(const struct dp_netdev_port *port,
590 struct dpif_port *dpif_port)
592 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
593 dpif_port->type = xstrdup(port->type);
594 dpif_port->port_no = port->port_no;
598 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
599 struct dpif_port *dpif_port)
601 struct dp_netdev *dp = get_dp_netdev(dpif);
602 struct dp_netdev_port *port;
605 ovs_mutex_lock(&dp_netdev_mutex);
606 error = get_port_by_number(dp, port_no, &port);
607 if (!error && dpif_port) {
608 answer_port_query(port, dpif_port);
610 ovs_mutex_unlock(&dp_netdev_mutex);
616 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
617 struct dpif_port *dpif_port)
619 struct dp_netdev *dp = get_dp_netdev(dpif);
620 struct dp_netdev_port *port;
623 ovs_mutex_lock(&dp_netdev_mutex);
624 error = get_port_by_name(dp, devname, &port);
625 if (!error && dpif_port) {
626 answer_port_query(port, dpif_port);
628 ovs_mutex_unlock(&dp_netdev_mutex);
634 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
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, const struct dpif_execute *execute)
1146 struct dp_netdev *dp = get_dp_netdev(dpif);
1150 if (execute->packet->size < ETH_HEADER_LEN ||
1151 execute->packet->size > UINT16_MAX) {
1155 /* Get packet metadata. */
1156 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, &md);
1160 /* Extract flow key. */
1161 flow_extract(execute->packet, md.skb_priority, md.pkt_mark, &md.tunnel,
1163 ovs_mutex_lock(&dp_netdev_mutex);
1164 dp_netdev_execute_actions(dp, &key, execute->packet,
1165 execute->actions, execute->actions_len);
1166 ovs_mutex_unlock(&dp_netdev_mutex);
1172 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1178 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1179 uint32_t queue_id, uint32_t *priority)
1181 *priority = queue_id;
1185 static struct dp_netdev_queue *
1186 find_nonempty_queue(struct dpif *dpif)
1188 struct dp_netdev *dp = get_dp_netdev(dpif);
1191 for (i = 0; i < N_QUEUES; i++) {
1192 struct dp_netdev_queue *q = &dp->queues[i];
1193 if (q->head != q->tail) {
1201 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1204 struct dp_netdev_queue *q;
1207 ovs_mutex_lock(&dp_netdev_mutex);
1208 q = find_nonempty_queue(dpif);
1210 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1212 *upcall = u->upcall;
1221 ovs_mutex_unlock(&dp_netdev_mutex);
1227 dpif_netdev_recv_wait(struct dpif *dpif)
1229 struct dp_netdev *dp = get_dp_netdev(dpif);
1232 ovs_mutex_lock(&dp_netdev_mutex);
1233 seq = seq_read(dp->queue_seq);
1234 if (find_nonempty_queue(dpif)) {
1235 poll_immediate_wake();
1237 seq_wait(dp->queue_seq, seq);
1239 ovs_mutex_unlock(&dp_netdev_mutex);
1243 dpif_netdev_recv_purge(struct dpif *dpif)
1245 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1246 ovs_mutex_lock(&dp_netdev_mutex);
1247 dp_netdev_purge_queues(dpif_netdev->dp);
1248 ovs_mutex_unlock(&dp_netdev_mutex);
1252 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1253 const struct ofpbuf *packet)
1255 netdev_flow->used = time_msec();
1256 netdev_flow->packet_count++;
1257 netdev_flow->byte_count += packet->size;
1258 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1262 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1263 struct ofpbuf *packet, uint32_t skb_priority,
1264 uint32_t pkt_mark, const struct flow_tnl *tnl)
1266 struct dp_netdev_flow *netdev_flow;
1268 union flow_in_port in_port_;
1270 if (packet->size < ETH_HEADER_LEN) {
1273 in_port_.odp_port = port->port_no;
1274 flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
1275 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1277 dp_netdev_flow_used(netdev_flow, packet);
1278 dp_netdev_execute_actions(dp, &key, packet,
1279 netdev_flow->actions,
1280 netdev_flow->actions_len);
1284 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1289 dpif_netdev_run(struct dpif *dpif)
1291 struct dp_netdev_port *port;
1292 struct dp_netdev *dp;
1293 struct ofpbuf packet;
1296 ovs_mutex_lock(&dp_netdev_mutex);
1297 dp = get_dp_netdev(dpif);
1298 ofpbuf_init(&packet, 0);
1300 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu;
1302 LIST_FOR_EACH (port, node, &dp->port_list) {
1305 /* Reset packet contents. Packet data may have been stolen. */
1306 ofpbuf_clear(&packet);
1307 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM, buf_size);
1309 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1311 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1312 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1315 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1316 netdev_get_name(port->netdev), ovs_strerror(error));
1319 ofpbuf_uninit(&packet);
1320 ovs_mutex_unlock(&dp_netdev_mutex);
1324 dpif_netdev_wait(struct dpif *dpif)
1326 struct dp_netdev_port *port;
1328 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1329 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1330 * A makes it to poll_block().
1332 * But I think it doesn't matter:
1334 * - In the dpif_port_add() case, A will not wake up when a packet
1335 * arrives on the new port, but this would also happen if the
1336 * ordering were reversed.
1338 * - In the dpif_port_remove() case, A might wake up spuriously, but
1339 * that is harmless. */
1341 ovs_mutex_lock(&dp_netdev_mutex);
1342 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1344 netdev_rx_wait(port->rx);
1347 ovs_mutex_unlock(&dp_netdev_mutex);
1351 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1352 int queue_no, const struct flow *flow,
1353 const struct nlattr *userdata)
1355 struct dp_netdev_queue *q = &dp->queues[queue_no];
1356 if (q->head - q->tail < MAX_QUEUE_LEN) {
1357 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1358 struct dpif_upcall *upcall = &u->upcall;
1359 struct ofpbuf *buf = &u->buf;
1362 upcall->type = queue_no;
1364 /* Allocate buffer big enough for everything. */
1365 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1367 buf_size += NLA_ALIGN(userdata->nla_len);
1369 ofpbuf_init(buf, buf_size);
1372 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1373 upcall->key = buf->data;
1374 upcall->key_len = buf->size;
1378 upcall->userdata = ofpbuf_put(buf, userdata,
1379 NLA_ALIGN(userdata->nla_len));
1382 /* Steal packet data. */
1383 ovs_assert(packet->source == OFPBUF_MALLOC);
1384 upcall->packet = *packet;
1385 ofpbuf_use(packet, NULL, 0);
1387 seq_change(dp->queue_seq);
1396 struct dp_netdev_execute_aux {
1397 struct dp_netdev *dp;
1398 const struct flow *key;
1402 dp_netdev_action_output(void *aux_, struct ofpbuf *packet,
1403 const struct flow *flow OVS_UNUSED,
1404 odp_port_t out_port)
1406 struct dp_netdev_execute_aux *aux = aux_;
1407 struct dp_netdev_port *p = aux->dp->ports[odp_to_u32(out_port)];
1409 netdev_send(p->netdev, packet);
1414 dp_netdev_action_userspace(void *aux_, struct ofpbuf *packet,
1415 const struct flow *flow OVS_UNUSED,
1416 const struct nlattr *a, bool may_steal)
1418 struct dp_netdev_execute_aux *aux = aux_;
1419 const struct nlattr *userdata;
1421 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1423 /* Make a copy if we are not allowed to steal the packet's data. */
1425 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1427 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1430 ofpbuf_uninit(packet);
1435 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1436 struct ofpbuf *packet,
1437 const struct nlattr *actions, size_t actions_len)
1439 struct dp_netdev_execute_aux aux = {dp, key};
1440 struct flow md = *key; /* Packet metadata, may be modified by actions. */
1442 odp_execute_actions(&aux, packet, &md, actions, actions_len,
1443 dp_netdev_action_output, dp_netdev_action_userspace);
1446 const struct dpif_class dpif_netdev_class = {
1448 dpif_netdev_enumerate,
1449 dpif_netdev_port_open_type,
1452 dpif_netdev_destroy,
1455 dpif_netdev_get_stats,
1456 dpif_netdev_port_add,
1457 dpif_netdev_port_del,
1458 dpif_netdev_port_query_by_number,
1459 dpif_netdev_port_query_by_name,
1460 dpif_netdev_get_max_ports,
1461 NULL, /* port_get_pid */
1462 dpif_netdev_port_dump_start,
1463 dpif_netdev_port_dump_next,
1464 dpif_netdev_port_dump_done,
1465 dpif_netdev_port_poll,
1466 dpif_netdev_port_poll_wait,
1467 dpif_netdev_flow_get,
1468 dpif_netdev_flow_put,
1469 dpif_netdev_flow_del,
1470 dpif_netdev_flow_flush,
1471 dpif_netdev_flow_dump_start,
1472 dpif_netdev_flow_dump_next,
1473 dpif_netdev_flow_dump_done,
1474 dpif_netdev_execute,
1476 dpif_netdev_recv_set,
1477 dpif_netdev_queue_to_priority,
1479 dpif_netdev_recv_wait,
1480 dpif_netdev_recv_purge,
1484 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1485 const char *argv[], void *aux OVS_UNUSED)
1487 struct dp_netdev_port *port;
1488 struct dp_netdev *dp;
1491 dp = shash_find_data(&dp_netdevs, argv[1]);
1492 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1493 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1497 if (get_port_by_name(dp, argv[2], &port)) {
1498 unixctl_command_reply_error(conn, "unknown port");
1502 port_no = atoi(argv[3]);
1503 if (port_no <= 0 || port_no >= MAX_PORTS) {
1504 unixctl_command_reply_error(conn, "bad port number");
1507 if (dp->ports[port_no]) {
1508 unixctl_command_reply_error(conn, "port number already in use");
1511 dp->ports[odp_to_u32(port->port_no)] = NULL;
1512 dp->ports[port_no] = port;
1513 port->port_no = u32_to_odp(port_no);
1514 seq_change(dp->port_seq);
1515 unixctl_command_reply(conn, NULL);
1519 dpif_dummy_register__(const char *type)
1521 struct dpif_class *class;
1523 class = xmalloc(sizeof *class);
1524 *class = dpif_netdev_class;
1525 class->type = xstrdup(type);
1526 dp_register_provider(class);
1530 dpif_dummy_register(bool override)
1537 dp_enumerate_types(&types);
1538 SSET_FOR_EACH (type, &types) {
1539 if (!dp_unregister_provider(type)) {
1540 dpif_dummy_register__(type);
1543 sset_destroy(&types);
1546 dpif_dummy_register__("dummy");
1548 unixctl_command_register("dpif-dummy/change-port-number",
1549 "DP PORT NEW-NUMBER",
1550 3, 3, dpif_dummy_change_port_number, NULL);