135
140
/* All netdev-based datapaths. */
136
141
static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
138
/* Maximum port MTU seen so far. */
139
static int max_mtu = ETH_PAYLOAD_MAX;
143
/* Global lock for all data. */
144
static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
141
static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
146
static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
142
147
struct dp_netdev_port **portp);
143
148
static int get_port_by_name(struct dp_netdev *, const char *devname,
144
149
struct dp_netdev_port **portp);
145
150
static void dp_netdev_free(struct dp_netdev *);
146
151
static void dp_netdev_flow_flush(struct dp_netdev *);
147
152
static int do_add_port(struct dp_netdev *, const char *devname,
148
const char *type, uint32_t port_no);
149
static int do_del_port(struct dp_netdev *, uint32_t port_no);
153
const char *type, odp_port_t port_no);
154
static int do_del_port(struct dp_netdev *, odp_port_t port_no);
150
155
static int dpif_netdev_open(const struct dpif_class *, const char *name,
151
156
bool create, struct dpif **);
152
157
static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
153
158
int queue_no, const struct flow *,
159
const struct nlattr *userdata);
155
160
static void dp_netdev_execute_actions(struct dp_netdev *,
156
161
struct ofpbuf *, struct flow *,
157
162
const struct nlattr *actions,
158
163
size_t actions_len);
164
static void dp_netdev_port_input(struct dp_netdev *dp,
165
struct dp_netdev_port *port,
166
struct ofpbuf *packet, uint32_t skb_priority,
167
uint32_t skb_mark, const struct flow_tnl *tnl);
160
169
static struct dpif_netdev *
161
170
dpif_netdev_cast(const struct dpif *dpif)
427
455
dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
456
odp_port_t *port_nop)
430
458
struct dp_netdev *dp = get_dp_netdev(dpif);
459
char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
460
const char *dpif_port;
433
if (*port_nop != UINT32_MAX) {
434
if (*port_nop >= MAX_PORTS) {
436
} else if (dp->ports[*port_nop]) {
464
ovs_mutex_lock(&dp_netdev_mutex);
465
dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
466
if (*port_nop != ODPP_NONE) {
467
uint32_t port_idx = odp_to_u32(*port_nop);
468
if (port_idx >= MAX_PORTS) {
470
} else if (dp->ports[port_idx]) {
441
port_no = choose_port(dp, netdev_vport_get_dpif_port(netdev));
477
port_no = choose_port(dp, dpif_port);
478
error = port_no == ODPP_NONE ? EFBIG : 0;
444
481
*port_nop = port_no;
445
return do_add_port(dp, netdev_vport_get_dpif_port(netdev),
446
netdev_get_type(netdev), port_no);
482
error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
484
ovs_mutex_unlock(&dp_netdev_mutex);
452
dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
490
dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
454
492
struct dp_netdev *dp = get_dp_netdev(dpif);
455
return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
495
ovs_mutex_lock(&dp_netdev_mutex);
496
error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
497
ovs_mutex_unlock(&dp_netdev_mutex);
459
is_valid_port_number(uint32_t port_no)
503
is_valid_port_number(odp_port_t port_no)
461
return port_no < MAX_PORTS;
505
return odp_to_u32(port_no) < MAX_PORTS;
465
509
get_port_by_number(struct dp_netdev *dp,
466
uint32_t port_no, struct dp_netdev_port **portp)
510
odp_port_t port_no, struct dp_netdev_port **portp)
468
512
if (!is_valid_port_number(port_no)) {
472
*portp = dp->ports[port_no];
516
*portp = dp->ports[odp_to_u32(port_no)];
473
517
return *portp ? 0 : ENOENT;
1063
1173
ofpbuf_clear(&packet);
1064
1174
ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1066
error = netdev_recv(port->netdev, &packet);
1176
error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1068
dp_netdev_port_input(dp, port, &packet);
1178
dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1069
1179
} else if (error != EAGAIN && error != EOPNOTSUPP) {
1070
1180
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1071
1182
VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1072
netdev_get_name(port->netdev), strerror(error));
1183
netdev_get_name(port->netdev), ovs_strerror(error));
1075
1186
ofpbuf_uninit(&packet);
1187
ovs_mutex_unlock(&dp_netdev_mutex);
1079
1191
dpif_netdev_wait(struct dpif *dpif)
1081
struct dp_netdev *dp = get_dp_netdev(dpif);
1082
1193
struct dp_netdev_port *port;
1084
LIST_FOR_EACH (port, node, &dp->port_list) {
1085
netdev_recv_wait(port->netdev);
1195
/* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1196
* thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1197
* A makes it to poll_block().
1199
* But I think it doesn't matter:
1201
* - In the dpif_port_add() case, A will not wake up when a packet
1202
* arrives on the new port, but this would also happen if the
1203
* ordering were reversed.
1205
* - In the dpif_port_remove() case, A might wake up spuriously, but
1206
* that is harmless. */
1208
ovs_mutex_lock(&dp_netdev_mutex);
1209
LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
1211
netdev_rx_wait(port->rx);
1090
dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1092
struct eth_header *eh = packet->l2;
1094
memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1095
memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1099
dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1214
ovs_mutex_unlock(&dp_netdev_mutex);
1218
dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port)
1220
struct dp_netdev *dp = dp_;
1102
1221
struct dp_netdev_port *p = dp->ports[out_port];
1104
1223
netdev_send(p->netdev, packet);
1109
1228
dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1110
int queue_no, const struct flow *flow, uint64_t arg)
1229
int queue_no, const struct flow *flow,
1230
const struct nlattr *userdata)
1112
1232
struct dp_netdev_queue *q = &dp->queues[queue_no];
1113
struct dp_netdev_upcall *u;
1114
struct dpif_upcall *upcall;
1118
if (q->head - q->tail >= MAX_QUEUE_LEN) {
1233
if (q->head - q->tail < MAX_QUEUE_LEN) {
1234
struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1235
struct dpif_upcall *upcall = &u->upcall;
1236
struct ofpbuf *buf = &u->buf;
1239
upcall->type = queue_no;
1241
/* Allocate buffer big enough for everything. */
1242
buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1244
buf_size += NLA_ALIGN(userdata->nla_len);
1246
ofpbuf_init(buf, buf_size);
1249
odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1250
upcall->key = buf->data;
1251
upcall->key_len = buf->size;
1255
upcall->userdata = ofpbuf_put(buf, userdata,
1256
NLA_ALIGN(userdata->nla_len));
1261
* We adjust 'data' and 'size' in 'buf' so that only the packet itself
1262
* is visible in 'upcall->packet'. The ODP flow and (if present)
1263
* userdata become part of the headroom. */
1264
ofpbuf_put_zeros(buf, 2);
1265
buf->data = ofpbuf_put(buf, packet->data, packet->size);
1266
buf->size = packet->size;
1267
upcall->packet = buf;
1120
1272
return ENOBUFS;
1123
u = &q->upcalls[q->head++ & QUEUE_MASK];
1126
ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1127
odp_flow_key_from_flow(buf, flow, flow->in_port);
1128
key_len = buf->size;
1129
ofpbuf_pull(buf, key_len);
1130
ofpbuf_reserve(buf, 2);
1131
ofpbuf_put(buf, packet->data, packet->size);
1133
upcall = &u->upcall;
1134
upcall->type = queue_no;
1135
upcall->packet = buf;
1136
upcall->key = buf->base;
1137
upcall->key_len = key_len;
1138
upcall->userdata = arg;
1144
dp_netdev_sample(struct dp_netdev *dp,
1145
struct ofpbuf *packet, struct flow *key,
1146
const struct nlattr *action)
1148
const struct nlattr *subactions = NULL;
1149
const struct nlattr *a;
1152
NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1153
int type = nl_attr_type(a);
1155
switch ((enum ovs_sample_attr) type) {
1156
case OVS_SAMPLE_ATTR_PROBABILITY:
1157
if (random_uint32() >= nl_attr_get_u32(a)) {
1162
case OVS_SAMPLE_ATTR_ACTIONS:
1166
case OVS_SAMPLE_ATTR_UNSPEC:
1167
case __OVS_SAMPLE_ATTR_MAX:
1173
dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1174
nl_attr_get_size(subactions));
1178
dp_netdev_action_userspace(struct dp_netdev *dp,
1179
struct ofpbuf *packet, struct flow *key,
1180
const struct nlattr *a)
1182
const struct nlattr *userdata_attr;
1185
userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1186
userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1277
dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1278
const struct flow *key,
1279
const struct nlattr *userdata)
1187
1281
dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1191
execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1193
enum ovs_key_attr type = nl_attr_type(a);
1194
const struct ovs_key_ipv4 *ipv4_key;
1195
const struct ovs_key_ipv6 *ipv6_key;
1196
const struct ovs_key_tcp *tcp_key;
1197
const struct ovs_key_udp *udp_key;
1200
case OVS_KEY_ATTR_PRIORITY:
1201
case OVS_KEY_ATTR_SKB_MARK:
1202
case OVS_KEY_ATTR_TUNNEL:
1203
/* not implemented */
1206
case OVS_KEY_ATTR_ETHERNET:
1207
dp_netdev_set_dl(packet,
1208
nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1211
case OVS_KEY_ATTR_IPV4:
1212
ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1213
packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1214
ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1217
case OVS_KEY_ATTR_IPV6:
1218
ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1219
packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1220
ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1221
ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1224
case OVS_KEY_ATTR_TCP:
1225
tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1226
packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1229
case OVS_KEY_ATTR_UDP:
1230
udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1231
packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1234
case OVS_KEY_ATTR_UNSPEC:
1235
case OVS_KEY_ATTR_ENCAP:
1236
case OVS_KEY_ATTR_ETHERTYPE:
1237
case OVS_KEY_ATTR_IN_PORT:
1238
case OVS_KEY_ATTR_VLAN:
1239
case OVS_KEY_ATTR_ICMP:
1240
case OVS_KEY_ATTR_ICMPV6:
1241
case OVS_KEY_ATTR_ARP:
1242
case OVS_KEY_ATTR_ND:
1243
case __OVS_KEY_ATTR_MAX:
1250
1285
dp_netdev_execute_actions(struct dp_netdev *dp,
1251
1286
struct ofpbuf *packet, struct flow *key,
1252
1287
const struct nlattr *actions,
1253
1288
size_t actions_len)
1255
const struct nlattr *a;
1258
NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1259
int type = nl_attr_type(a);
1261
switch ((enum ovs_action_attr) type) {
1262
case OVS_ACTION_ATTR_OUTPUT:
1263
dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1266
case OVS_ACTION_ATTR_USERSPACE:
1267
dp_netdev_action_userspace(dp, packet, key, a);
1270
case OVS_ACTION_ATTR_PUSH_VLAN: {
1271
const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
1272
eth_push_vlan(packet, vlan->vlan_tci);
1276
case OVS_ACTION_ATTR_POP_VLAN:
1277
eth_pop_vlan(packet);
1280
case OVS_ACTION_ATTR_SET:
1281
execute_set_action(packet, nl_attr_get(a));
1284
case OVS_ACTION_ATTR_SAMPLE:
1285
dp_netdev_sample(dp, packet, key, a);
1288
case OVS_ACTION_ATTR_UNSPEC:
1289
case __OVS_ACTION_ATTR_MAX:
1290
odp_execute_actions(dp, packet, key, actions, actions_len,
1291
dp_netdev_output_port, dp_netdev_action_userspace);
1295
1294
const struct dpif_class dpif_netdev_class = {