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 *,
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_class_is_planetlab(const struct dpif_class *class)
215 return class == &dpif_planetlab_class;
219 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
221 return strcmp(type, "internal") ? type
222 : dpif_netdev_class_is_planetlab(class) ? "pltap"
223 : dpif_netdev_class_is_dummy(class) ? "dummy"
228 create_dpif_netdev(struct dp_netdev *dp)
230 uint16_t netflow_id = hash_string(dp->name, 0);
231 struct dpif_netdev *dpif;
235 dpif = xmalloc(sizeof *dpif);
236 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
238 dpif->last_port_seq = seq_read(dp->port_seq);
243 /* Choose an unused, non-zero port number and return it on success.
244 * Return ODPP_NONE on failure. */
246 choose_port(struct dp_netdev *dp, const char *name)
250 if (dp->class != &dpif_netdev_class &&
251 dp->class != &dpif_planetlab_class) {
255 /* If the port name begins with "br", start the number search at
256 * 100 to make writing tests easier. */
257 if (!strncmp(name, "br", 2)) {
261 /* If the port name contains a number, try to assign that port number.
262 * This can make writing unit tests easier because port numbers are
264 for (p = name; *p != '\0'; p++) {
265 if (isdigit((unsigned char) *p)) {
266 port_no = start_no + strtol(p, NULL, 10);
267 if (port_no > 0 && port_no < MAX_PORTS
268 && !dp->ports[port_no]) {
269 return u32_to_odp(port_no);
276 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
277 if (!dp->ports[port_no]) {
278 return u32_to_odp(port_no);
286 create_dp_netdev(const char *name, const struct dpif_class *class,
287 struct dp_netdev **dpp)
289 struct dp_netdev *dp;
293 dp = xzalloc(sizeof *dp);
295 dp->name = xstrdup(name);
297 dp->max_mtu = ETH_PAYLOAD_MAX;
298 for (i = 0; i < N_QUEUES; i++) {
299 dp->queues[i].head = dp->queues[i].tail = 0;
301 dp->queue_seq = seq_create();
302 classifier_init(&dp->cls, NULL);
303 hmap_init(&dp->flow_table);
304 list_init(&dp->port_list);
305 dp->port_seq = seq_create();
307 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
313 shash_add(&dp_netdevs, name, dp);
320 dpif_netdev_open(const struct dpif_class *class, const char *name,
321 bool create, struct dpif **dpifp)
323 struct dp_netdev *dp;
326 ovs_mutex_lock(&dp_netdev_mutex);
327 dp = shash_find_data(&dp_netdevs, name);
329 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
331 error = (dp->class != class ? EINVAL
336 *dpifp = create_dpif_netdev(dp);
338 ovs_mutex_unlock(&dp_netdev_mutex);
344 dp_netdev_purge_queues(struct dp_netdev *dp)
348 for (i = 0; i < N_QUEUES; i++) {
349 struct dp_netdev_queue *q = &dp->queues[i];
351 while (q->tail != q->head) {
352 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
353 ofpbuf_uninit(&u->upcall.packet);
354 ofpbuf_uninit(&u->buf);
360 dp_netdev_free(struct dp_netdev *dp)
362 struct dp_netdev_port *port, *next;
364 dp_netdev_flow_flush(dp);
365 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
366 do_del_port(dp, port->port_no);
368 dp_netdev_purge_queues(dp);
369 seq_destroy(dp->queue_seq);
370 classifier_destroy(&dp->cls);
371 hmap_destroy(&dp->flow_table);
372 seq_destroy(dp->port_seq);
378 dpif_netdev_close(struct dpif *dpif)
380 struct dp_netdev *dp = get_dp_netdev(dpif);
382 ovs_mutex_lock(&dp_netdev_mutex);
384 ovs_assert(dp->open_cnt > 0);
385 if (--dp->open_cnt == 0 && dp->destroyed) {
386 shash_find_and_delete(&dp_netdevs, dp->name);
391 ovs_mutex_unlock(&dp_netdev_mutex);
395 dpif_netdev_destroy(struct dpif *dpif)
397 struct dp_netdev *dp = get_dp_netdev(dpif);
399 ovs_mutex_lock(&dp_netdev_mutex);
400 dp->destroyed = true;
401 ovs_mutex_unlock(&dp_netdev_mutex);
407 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
409 struct dp_netdev *dp = get_dp_netdev(dpif);
411 ovs_mutex_lock(&dp_netdev_mutex);
412 stats->n_flows = hmap_count(&dp->flow_table);
413 stats->n_hit = dp->n_hit;
414 stats->n_missed = dp->n_missed;
415 stats->n_lost = dp->n_lost;
416 stats->n_masks = 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 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
646 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
648 ovs_rwlock_wrlock(&dp->cls.rwlock);
649 classifier_remove(&dp->cls, &netdev_flow->cr);
650 ovs_rwlock_unlock(&dp->cls.rwlock);
651 cls_rule_destroy(&netdev_flow->cr);
653 hmap_remove(&dp->flow_table, &netdev_flow->node);
654 free(netdev_flow->actions);
659 dp_netdev_flow_flush(struct dp_netdev *dp)
661 struct dp_netdev_flow *netdev_flow, *next;
663 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
664 dp_netdev_free_flow(dp, netdev_flow);
669 dpif_netdev_flow_flush(struct dpif *dpif)
671 struct dp_netdev *dp = get_dp_netdev(dpif);
673 ovs_mutex_lock(&dp_netdev_mutex);
674 dp_netdev_flow_flush(dp);
675 ovs_mutex_unlock(&dp_netdev_mutex);
680 struct dp_netdev_port_state {
686 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
688 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
693 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
694 struct dpif_port *dpif_port)
696 struct dp_netdev_port_state *state = state_;
697 struct dp_netdev *dp = get_dp_netdev(dpif);
700 ovs_mutex_lock(&dp_netdev_mutex);
701 for (port_idx = odp_to_u32(state->port_no);
702 port_idx < MAX_PORTS; port_idx++) {
703 struct dp_netdev_port *port = dp->ports[port_idx];
706 state->name = xstrdup(netdev_get_name(port->netdev));
707 dpif_port->name = state->name;
708 dpif_port->type = port->type;
709 dpif_port->port_no = port->port_no;
710 state->port_no = u32_to_odp(port_idx + 1);
711 ovs_mutex_unlock(&dp_netdev_mutex);
716 ovs_mutex_unlock(&dp_netdev_mutex);
722 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
724 struct dp_netdev_port_state *state = state_;
731 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
733 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
734 uint64_t new_port_seq;
737 ovs_mutex_lock(&dp_netdev_mutex);
738 new_port_seq = seq_read(dpif->dp->port_seq);
739 if (dpif->last_port_seq != new_port_seq) {
740 dpif->last_port_seq = new_port_seq;
745 ovs_mutex_unlock(&dp_netdev_mutex);
751 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
753 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
755 ovs_mutex_lock(&dp_netdev_mutex);
756 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
757 ovs_mutex_unlock(&dp_netdev_mutex);
760 static struct dp_netdev_flow *
761 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
765 ovs_rwlock_wrlock(&dp->cls.rwlock);
766 cr = classifier_lookup(&dp->cls, flow, NULL);
767 ovs_rwlock_unlock(&dp->cls.rwlock);
770 ? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
774 static struct dp_netdev_flow *
775 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
777 struct dp_netdev_flow *netdev_flow;
779 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
781 if (flow_equal(&netdev_flow->flow, flow)) {
789 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
790 struct dpif_flow_stats *stats)
792 stats->n_packets = netdev_flow->packet_count;
793 stats->n_bytes = netdev_flow->byte_count;
794 stats->used = netdev_flow->used;
795 stats->tcp_flags = netdev_flow->tcp_flags;
799 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
800 const struct nlattr *mask_key,
801 uint32_t mask_key_len, const struct flow *flow,
805 if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
806 /* This should not happen: it indicates that
807 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
808 * disagree on the acceptable form of a mask. Log the problem
809 * as an error, with enough details to enable debugging. */
810 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
812 if (!VLOG_DROP_ERR(&rl)) {
816 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
818 VLOG_ERR("internal error parsing flow mask %s", ds_cstr(&s));
824 /* Force unwildcard the in_port. */
825 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
828 /* No mask key, unwildcard everything except fields whose
829 * prerequisities are not met. */
830 memset(mask, 0x0, sizeof *mask);
832 for (id = 0; id < MFF_N_IDS; ++id) {
833 /* Skip registers and metadata. */
834 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
835 && id != MFF_METADATA) {
836 const struct mf_field *mf = mf_from_id(id);
837 if (mf_are_prereqs_ok(mf, flow)) {
838 mf_mask_field(mf, mask);
848 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
853 if (odp_flow_key_to_flow(key, key_len, flow)) {
854 /* This should not happen: it indicates that odp_flow_key_from_flow()
855 * and odp_flow_key_to_flow() disagree on the acceptable form of a
856 * flow. Log the problem as an error, with enough details to enable
858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
860 if (!VLOG_DROP_ERR(&rl)) {
864 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
865 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
872 in_port = flow->in_port.odp_port;
873 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
881 dpif_netdev_flow_get(const struct dpif *dpif,
882 const struct nlattr *nl_key, size_t nl_key_len,
883 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
885 struct dp_netdev *dp = get_dp_netdev(dpif);
886 struct dp_netdev_flow *netdev_flow;
890 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
895 ovs_mutex_lock(&dp_netdev_mutex);
896 netdev_flow = dp_netdev_find_flow(dp, &key);
899 get_dpif_flow_stats(netdev_flow, stats);
902 *actionsp = ofpbuf_clone_data(netdev_flow->actions,
903 netdev_flow->actions_len);
908 ovs_mutex_unlock(&dp_netdev_mutex);
914 set_flow_actions(struct dp_netdev_flow *netdev_flow,
915 const struct nlattr *actions, size_t actions_len)
917 netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
918 netdev_flow->actions_len = actions_len;
919 memcpy(netdev_flow->actions, actions, actions_len);
924 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
925 const struct flow_wildcards *wc,
926 const struct nlattr *actions,
929 struct dp_netdev_flow *netdev_flow;
933 netdev_flow = xzalloc(sizeof *netdev_flow);
934 netdev_flow->flow = *flow;
936 match_init(&match, flow, wc);
937 cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
938 ovs_rwlock_wrlock(&dp->cls.rwlock);
939 classifier_insert(&dp->cls, &netdev_flow->cr);
940 ovs_rwlock_unlock(&dp->cls.rwlock);
942 error = set_flow_actions(netdev_flow, actions, actions_len);
944 ovs_rwlock_wrlock(&dp->cls.rwlock);
945 classifier_remove(&dp->cls, &netdev_flow->cr);
946 ovs_rwlock_unlock(&dp->cls.rwlock);
947 cls_rule_destroy(&netdev_flow->cr);
953 hmap_insert(&dp->flow_table, &netdev_flow->node, flow_hash(flow, 0));
958 clear_stats(struct dp_netdev_flow *netdev_flow)
960 netdev_flow->used = 0;
961 netdev_flow->packet_count = 0;
962 netdev_flow->byte_count = 0;
963 netdev_flow->tcp_flags = 0;
967 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
969 struct dp_netdev *dp = get_dp_netdev(dpif);
970 struct dp_netdev_flow *netdev_flow;
972 struct flow_wildcards wc;
975 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
979 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
980 put->mask, put->mask_len,
986 ovs_mutex_lock(&dp_netdev_mutex);
987 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
989 if (put->flags & DPIF_FP_CREATE) {
990 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
992 memset(put->stats, 0, sizeof *put->stats);
994 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1003 if (put->flags & DPIF_FP_MODIFY
1004 && flow_equal(&flow, &netdev_flow->flow)) {
1005 error = set_flow_actions(netdev_flow, put->actions,
1009 get_dpif_flow_stats(netdev_flow, put->stats);
1011 if (put->flags & DPIF_FP_ZERO_STATS) {
1012 clear_stats(netdev_flow);
1015 } else if (put->flags & DPIF_FP_CREATE) {
1018 /* Overlapping flow. */
1022 ovs_mutex_unlock(&dp_netdev_mutex);
1028 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1030 struct dp_netdev *dp = get_dp_netdev(dpif);
1031 struct dp_netdev_flow *netdev_flow;
1035 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1040 ovs_mutex_lock(&dp_netdev_mutex);
1041 netdev_flow = dp_netdev_find_flow(dp, &key);
1044 get_dpif_flow_stats(netdev_flow, del->stats);
1046 dp_netdev_free_flow(dp, netdev_flow);
1050 ovs_mutex_unlock(&dp_netdev_mutex);
1055 struct dp_netdev_flow_state {
1058 struct nlattr *actions;
1059 struct odputil_keybuf keybuf;
1060 struct odputil_keybuf maskbuf;
1061 struct dpif_flow_stats stats;
1065 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1067 struct dp_netdev_flow_state *state;
1069 *statep = state = xmalloc(sizeof *state);
1072 state->actions = NULL;
1077 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1078 const struct nlattr **key, size_t *key_len,
1079 const struct nlattr **mask, size_t *mask_len,
1080 const struct nlattr **actions, size_t *actions_len,
1081 const struct dpif_flow_stats **stats)
1083 struct dp_netdev_flow_state *state = state_;
1084 struct dp_netdev *dp = get_dp_netdev(dpif);
1085 struct dp_netdev_flow *netdev_flow;
1086 struct hmap_node *node;
1088 ovs_mutex_lock(&dp_netdev_mutex);
1089 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1091 ovs_mutex_unlock(&dp_netdev_mutex);
1095 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1100 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1101 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1102 netdev_flow->flow.in_port.odp_port);
1105 *key_len = buf.size;
1110 struct flow_wildcards wc;
1112 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1113 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1114 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1115 odp_to_u32(wc.masks.in_port.odp_port));
1118 *mask_len = buf.size;
1122 free(state->actions);
1123 state->actions = xmemdup(netdev_flow->actions,
1124 netdev_flow->actions_len);
1126 *actions = state->actions;
1127 *actions_len = netdev_flow->actions_len;
1131 get_dpif_flow_stats(netdev_flow, &state->stats);
1132 *stats = &state->stats;
1135 ovs_mutex_unlock(&dp_netdev_mutex);
1140 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1142 struct dp_netdev_flow_state *state = state_;
1144 free(state->actions);
1150 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1152 struct dp_netdev *dp = get_dp_netdev(dpif);
1153 struct pkt_metadata *md = &execute->md;
1156 if (execute->packet->size < ETH_HEADER_LEN ||
1157 execute->packet->size > UINT16_MAX) {
1161 /* Extract flow key. */
1162 flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1163 (union flow_in_port *)&md->in_port, &key);
1164 ovs_mutex_lock(&dp_netdev_mutex);
1165 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1166 execute->actions_len);
1167 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 ofpbuf *packet,
1263 struct pkt_metadata *md)
1265 struct dp_netdev_flow *netdev_flow;
1268 if (packet->size < ETH_HEADER_LEN) {
1271 flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1272 (union flow_in_port *)&md->in_port, &key);
1273 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1275 dp_netdev_flow_used(netdev_flow, packet);
1276 dp_netdev_execute_actions(dp, &key, packet, md,
1277 netdev_flow->actions,
1278 netdev_flow->actions_len);
1282 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1287 dpif_netdev_run(struct dpif *dpif)
1289 struct dp_netdev_port *port;
1290 struct dp_netdev *dp;
1291 struct ofpbuf packet;
1294 ovs_mutex_lock(&dp_netdev_mutex);
1295 dp = get_dp_netdev(dpif);
1296 ofpbuf_init(&packet, 0);
1298 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu;
1300 LIST_FOR_EACH (port, node, &dp->port_list) {
1303 /* Reset packet contents. Packet data may have been stolen. */
1304 ofpbuf_clear(&packet);
1305 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM, buf_size);
1307 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1309 struct pkt_metadata md = PKT_METADATA_INITIALIZER(port->port_no);
1310 dp_netdev_port_input(dp, &packet, &md);
1311 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1312 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1314 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1315 netdev_get_name(port->netdev), ovs_strerror(error));
1318 ofpbuf_uninit(&packet);
1319 ovs_mutex_unlock(&dp_netdev_mutex);
1323 dpif_netdev_wait(struct dpif *dpif)
1325 struct dp_netdev_port *port;
1327 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1328 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1329 * A makes it to poll_block().
1331 * But I think it doesn't matter:
1333 * - In the dpif_port_add() case, A will not wake up when a packet
1334 * arrives on the new port, but this would also happen if the
1335 * ordering were reversed.
1337 * - In the dpif_port_remove() case, A might wake up spuriously, but
1338 * that is harmless. */
1340 ovs_mutex_lock(&dp_netdev_mutex);
1341 LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1343 netdev_rx_wait(port->rx);
1346 ovs_mutex_unlock(&dp_netdev_mutex);
1350 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1351 odp_port_t out_port)
1353 struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
1355 netdev_send(p->netdev, packet);
1360 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1361 int queue_no, const struct flow *flow,
1362 const struct nlattr *userdata)
1364 struct dp_netdev_queue *q = &dp->queues[queue_no];
1365 if (q->head - q->tail < MAX_QUEUE_LEN) {
1366 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1367 struct dpif_upcall *upcall = &u->upcall;
1368 struct ofpbuf *buf = &u->buf;
1371 upcall->type = queue_no;
1373 /* Allocate buffer big enough for everything. */
1374 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1376 buf_size += NLA_ALIGN(userdata->nla_len);
1378 ofpbuf_init(buf, buf_size);
1381 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1382 upcall->key = buf->data;
1383 upcall->key_len = buf->size;
1387 upcall->userdata = ofpbuf_put(buf, userdata,
1388 NLA_ALIGN(userdata->nla_len));
1391 /* Steal packet data. */
1392 ovs_assert(packet->source == OFPBUF_MALLOC);
1393 upcall->packet = *packet;
1394 ofpbuf_use(packet, NULL, 0);
1396 seq_change(dp->queue_seq);
1405 struct dp_netdev_execute_aux {
1406 struct dp_netdev *dp;
1407 const struct flow *key;
1411 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1412 const struct pkt_metadata *md OVS_UNUSED,
1413 const struct nlattr *a, bool may_steal)
1415 struct dp_netdev_execute_aux *aux = aux_;
1416 int type = nl_attr_type(a);
1418 switch ((enum ovs_action_attr)type) {
1419 case OVS_ACTION_ATTR_OUTPUT:
1420 dp_netdev_output_port(aux->dp, packet, u32_to_odp(nl_attr_get_u32(a)));
1423 case OVS_ACTION_ATTR_USERSPACE: {
1424 const struct nlattr *userdata;
1426 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1428 /* Make a copy if we are not allowed to steal the packet's data. */
1430 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1432 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1435 ofpbuf_uninit(packet);
1439 case OVS_ACTION_ATTR_PUSH_VLAN:
1440 case OVS_ACTION_ATTR_POP_VLAN:
1441 case OVS_ACTION_ATTR_PUSH_MPLS:
1442 case OVS_ACTION_ATTR_POP_MPLS:
1443 case OVS_ACTION_ATTR_SET:
1444 case OVS_ACTION_ATTR_SAMPLE:
1445 case OVS_ACTION_ATTR_UNSPEC:
1446 case __OVS_ACTION_ATTR_MAX:
1452 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1453 struct ofpbuf *packet, struct pkt_metadata *md,
1454 const struct nlattr *actions, size_t actions_len)
1456 struct dp_netdev_execute_aux aux = {dp, key};
1458 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1461 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1462 dpif_netdev_enumerate, \
1463 dpif_netdev_port_open_type, \
1465 dpif_netdev_close, \
1466 dpif_netdev_destroy, \
1469 dpif_netdev_get_stats, \
1470 dpif_netdev_port_add, \
1471 dpif_netdev_port_del, \
1472 dpif_netdev_port_query_by_number, \
1473 dpif_netdev_port_query_by_name, \
1474 dpif_netdev_get_max_ports, \
1475 NULL, /* port_get_pid */ \
1476 dpif_netdev_port_dump_start, \
1477 dpif_netdev_port_dump_next, \
1478 dpif_netdev_port_dump_done, \
1479 dpif_netdev_port_poll, \
1480 dpif_netdev_port_poll_wait, \
1481 dpif_netdev_flow_get, \
1482 dpif_netdev_flow_put, \
1483 dpif_netdev_flow_del, \
1484 dpif_netdev_flow_flush, \
1485 dpif_netdev_flow_dump_start, \
1486 dpif_netdev_flow_dump_next, \
1487 dpif_netdev_flow_dump_done, \
1488 dpif_netdev_execute, \
1489 NULL, /* operate */ \
1490 dpif_netdev_recv_set, \
1491 dpif_netdev_queue_to_priority, \
1493 dpif_netdev_recv_wait, \
1494 dpif_netdev_recv_purge, \
1496 const struct dpif_class dpif_netdev_class = {
1498 DPIF_NETDEV_CLASS_FUNCTIONS
1501 const struct dpif_class dpif_planetlab_class = {
1503 DPIF_NETDEV_CLASS_FUNCTIONS
1507 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1508 const char *argv[], void *aux OVS_UNUSED)
1510 struct dp_netdev_port *port;
1511 struct dp_netdev *dp;
1514 dp = shash_find_data(&dp_netdevs, argv[1]);
1515 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1516 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1520 if (get_port_by_name(dp, argv[2], &port)) {
1521 unixctl_command_reply_error(conn, "unknown port");
1525 port_no = atoi(argv[3]);
1526 if (port_no <= 0 || port_no >= MAX_PORTS) {
1527 unixctl_command_reply_error(conn, "bad port number");
1530 if (dp->ports[port_no]) {
1531 unixctl_command_reply_error(conn, "port number already in use");
1534 dp->ports[odp_to_u32(port->port_no)] = NULL;
1535 dp->ports[port_no] = port;
1536 port->port_no = u32_to_odp(port_no);
1537 seq_change(dp->port_seq);
1538 unixctl_command_reply(conn, NULL);
1542 dpif_dummy_register__(const char *type)
1544 struct dpif_class *class;
1546 class = xmalloc(sizeof *class);
1547 *class = dpif_netdev_class;
1548 class->type = xstrdup(type);
1549 dp_register_provider(class);
1553 dpif_dummy_register(bool override)
1560 dp_enumerate_types(&types);
1561 SSET_FOR_EACH (type, &types) {
1562 if (!dp_unregister_provider(type)) {
1563 dpif_dummy_register__(type);
1566 sset_destroy(&types);
1569 dpif_dummy_register__("dummy");
1571 unixctl_command_register("dpif-dummy/change-port-number",
1572 "DP PORT NEW-NUMBER",
1573 3, 3, dpif_dummy_change_port_number, NULL);