#include <inttypes.h>
#include <net/if.h>
#include <linux/types.h>
+#include <linux/ip.h>
+#include <linux/if_tunnel.h>
#include <linux/pkt_sched.h>
+#include <linux/rtnetlink.h>
#include <poll.h>
#include <stdlib.h>
#include <strings.h>
#include "netlink-socket.h"
#include "netlink.h"
#include "odp-util.h"
-#include "ofpbuf.h"
+#include "openvswitch/ofpbuf.h"
#include "packets.h"
#include "poll-loop.h"
#include "random.h"
}
static enum ovs_vport_type
-netdev_to_ovs_vport_type(const struct netdev *netdev)
+netdev_to_ovs_vport_type(const char *type)
{
- const char *type = netdev_get_type(netdev);
-
if (!strcmp(type, "tap") || !strcmp(type, "system")) {
return OVS_VPORT_TYPE_NETDEV;
} else if (!strcmp(type, "internal")) {
}
static int
-dpif_netlink_port_add__(struct dpif_netlink *dpif, struct netdev *netdev,
+dpif_netlink_port_add__(struct dpif_netlink *dpif, const char *name,
+ enum ovs_vport_type type,
+ struct ofpbuf *options,
odp_port_t *port_nop)
OVS_REQ_WRLOCK(dpif->upcall_lock)
{
- const struct netdev_tunnel_config *tnl_cfg;
- char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
- const char *name = netdev_vport_get_dpif_port(netdev,
- namebuf, sizeof namebuf);
- const char *type = netdev_get_type(netdev);
struct dpif_netlink_vport request, reply;
struct ofpbuf *buf;
- uint64_t options_stub[64 / 8];
- struct ofpbuf options;
struct nl_sock **socksp = NULL;
uint32_t *upcall_pids;
int error = 0;
dpif_netlink_vport_init(&request);
request.cmd = OVS_VPORT_CMD_NEW;
request.dp_ifindex = dpif->dp_ifindex;
- request.type = netdev_to_ovs_vport_type(netdev);
- if (request.type == OVS_VPORT_TYPE_UNSPEC) {
+ request.type = type;
+ request.name = name;
+
+ request.port_no = *port_nop;
+ upcall_pids = vport_socksp_to_pids(socksp, dpif->n_handlers);
+ request.n_upcall_pids = socksp ? dpif->n_handlers : 1;
+ request.upcall_pids = upcall_pids;
+
+ if (options) {
+ request.options = options->data;
+ request.options_len = options->size;
+ }
+
+ error = dpif_netlink_vport_transact(&request, &reply, &buf);
+ if (!error) {
+ *port_nop = reply.port_no;
+ } else {
+ if (error == EBUSY && *port_nop != ODPP_NONE) {
+ VLOG_INFO("%s: requested port %"PRIu32" is in use",
+ dpif_name(&dpif->dpif), *port_nop);
+ }
+
+ vport_del_socksp(dpif, socksp);
+ goto exit;
+ }
+
+ if (socksp) {
+ error = vport_add_channels(dpif, *port_nop, socksp);
+ if (error) {
+ VLOG_INFO("%s: could not add channel for port %s",
+ dpif_name(&dpif->dpif), name);
+
+ /* Delete the port. */
+ dpif_netlink_vport_init(&request);
+ request.cmd = OVS_VPORT_CMD_DEL;
+ request.dp_ifindex = dpif->dp_ifindex;
+ request.port_no = *port_nop;
+ dpif_netlink_vport_transact(&request, NULL, NULL);
+ vport_del_socksp(dpif, socksp);
+ goto exit;
+ }
+ }
+ free(socksp);
+
+exit:
+ ofpbuf_delete(buf);
+ free(upcall_pids);
+
+ return error;
+}
+
+static int
+dpif_netlink_port_add_compat(struct dpif_netlink *dpif, struct netdev *netdev,
+ odp_port_t *port_nop)
+ OVS_REQ_WRLOCK(dpif->upcall_lock)
+{
+ const struct netdev_tunnel_config *tnl_cfg;
+ char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
+ const char *name = netdev_vport_get_dpif_port(netdev,
+ namebuf, sizeof namebuf);
+ const char *type = netdev_get_type(netdev);
+ uint64_t options_stub[64 / 8];
+ struct ofpbuf options;
+ enum ovs_vport_type ovs_type;
+
+ ovs_type = netdev_to_ovs_vport_type(netdev_get_type(netdev));
+ if (ovs_type == OVS_VPORT_TYPE_UNSPEC) {
VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
"unsupported type `%s'",
dpif_name(&dpif->dpif), name, type);
- vport_del_socksp(dpif, socksp);
return EINVAL;
}
- request.name = name;
- if (request.type == OVS_VPORT_TYPE_NETDEV) {
+ if (ovs_type == OVS_VPORT_TYPE_NETDEV) {
#ifdef _WIN32
/* XXX : Map appropiate Windows handle */
#else
}
nl_msg_end_nested(&options, ext_ofs);
}
- request.options = options.data;
- request.options_len = options.size;
+ return dpif_netlink_port_add__(dpif, name, ovs_type, &options, port_nop);
+ } else {
+ return dpif_netlink_port_add__(dpif, name, ovs_type, NULL, port_nop);
}
- request.port_no = *port_nop;
- upcall_pids = vport_socksp_to_pids(socksp, dpif->n_handlers);
- request.n_upcall_pids = socksp ? dpif->n_handlers : 1;
- request.upcall_pids = upcall_pids;
+}
- error = dpif_netlink_vport_transact(&request, &reply, &buf);
- if (!error) {
- *port_nop = reply.port_no;
- } else {
- if (error == EBUSY && *port_nop != ODPP_NONE) {
- VLOG_INFO("%s: requested port %"PRIu32" is in use",
- dpif_name(&dpif->dpif), *port_nop);
- }
+#ifdef __linux__
- vport_del_socksp(dpif, socksp);
- goto exit;
+static int
+netdev_linux_destroy(const char *name)
+{
+ int err;
+ struct ofpbuf request, *reply;
+
+ ofpbuf_init(&request, 0);
+ nl_msg_put_nlmsghdr(&request, 0, RTM_DELLINK,
+ NLM_F_REQUEST | NLM_F_ACK);
+ ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
+ nl_msg_put_string(&request, IFLA_IFNAME, name);
+
+ err = nl_transact(NETLINK_ROUTE, &request, &reply);
+
+ if (!err) {
+ ofpbuf_uninit(reply);
}
- if (socksp) {
- error = vport_add_channels(dpif, *port_nop, socksp);
- if (error) {
- VLOG_INFO("%s: could not add channel for port %s",
- dpif_name(&dpif->dpif), name);
+ ofpbuf_uninit(&request);
+ return err;
+}
- /* Delete the port. */
- dpif_netlink_vport_init(&request);
- request.cmd = OVS_VPORT_CMD_DEL;
- request.dp_ifindex = dpif->dp_ifindex;
- request.port_no = *port_nop;
- dpif_netlink_vport_transact(&request, NULL, NULL);
- vport_del_socksp(dpif, socksp);
- goto exit;
+static int
+netdev_vxlan_destroy(const char *name)
+{
+ return netdev_linux_destroy(name);
+}
+
+static int
+netdev_gre_destroy(const char *name)
+{
+ return netdev_linux_destroy(name);
+}
+
+static int
+netdev_geneve_destroy(const char *name)
+{
+ return netdev_linux_destroy(name);
+}
+
+/*
+ * On some older systems, these enums are not defined.
+ */
+
+#ifndef IFLA_VXLAN_MAX
+#define IFLA_VXLAN_MAX 0
+#define IFLA_VXLAN_PORT 15
+#endif
+#if IFLA_VXLAN_MAX < 20
+#define IFLA_VXLAN_UDP_ZERO_CSUM6_RX 20
+#define IFLA_VXLAN_GBP 23
+#define IFLA_VXLAN_COLLECT_METADATA 25
+#endif
+
+#if IFLA_GRE_MAX < 18
+#define IFLA_GRE_COLLECT_METADATA 18
+#endif
+
+#ifndef IFLA_GENEVE_MAX
+#define IFLA_GENEVE_MAX 0
+#define IFLA_GENEVE_PORT 5
+#endif
+
+#if IFLA_GENEVE_MAX < 6
+#define IFLA_GENEVE_COLLECT_METADATA 6
+#endif
+#if IFLA_GENEVE_MAX < 10
+#define IFLA_GENEVE_UDP_ZERO_CSUM6_RX 10
+#endif
+
+static int
+netdev_vxlan_create(struct netdev *netdev)
+{
+ int err;
+ struct ofpbuf request, *reply;
+ size_t linkinfo_off, infodata_off;
+ char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
+ const char *name = netdev_vport_get_dpif_port(netdev,
+ namebuf, sizeof namebuf);
+ struct ifinfomsg *ifinfo;
+ const struct netdev_tunnel_config *tnl_cfg;
+ tnl_cfg = netdev_get_tunnel_config(netdev);
+ if (!tnl_cfg) { /* or assert? */
+ return EINVAL;
+ }
+
+ ofpbuf_init(&request, 0);
+ nl_msg_put_nlmsghdr(&request, 0, RTM_NEWLINK,
+ NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE);
+ ifinfo = ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
+ ifinfo->ifi_change = ifinfo->ifi_flags = IFF_UP;
+ nl_msg_put_string(&request, IFLA_IFNAME, name);
+ nl_msg_put_u32(&request, IFLA_MTU, UINT16_MAX);
+ linkinfo_off = nl_msg_start_nested(&request, IFLA_LINKINFO);
+ nl_msg_put_string(&request, IFLA_INFO_KIND, "vxlan");
+ infodata_off = nl_msg_start_nested(&request, IFLA_INFO_DATA);
+ nl_msg_put_u8(&request, IFLA_VXLAN_LEARNING, 0);
+ nl_msg_put_u8(&request, IFLA_VXLAN_COLLECT_METADATA, 1);
+ nl_msg_put_u8(&request, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 1);
+ if (tnl_cfg->exts & (1 << OVS_VXLAN_EXT_GBP)) {
+ nl_msg_put_flag(&request, IFLA_VXLAN_GBP);
+ }
+ nl_msg_put_be16(&request, IFLA_VXLAN_PORT, tnl_cfg->dst_port);
+ nl_msg_end_nested(&request, infodata_off);
+ nl_msg_end_nested(&request, linkinfo_off);
+
+ err = nl_transact(NETLINK_ROUTE, &request, &reply);
+
+ if (!err) {
+ ofpbuf_uninit(reply);
+ }
+
+ /*
+ * Linux versions older than 4.3 will return EINVAL in case the VID is not
+ * set, which is sufficient to verify COLLECT_METADATA is supported.
+ */
+ if (err == EINVAL) {
+ err = EOPNOTSUPP;
+ }
+
+ ofpbuf_uninit(&request);
+ return err;
+}
+
+/*
+ * On some Linux versions, creating the device with IFLA_GRE_COLLECT_METADATA
+ * will succeed, even though that attribute is not supported. We need to verify
+ * the device has been created with that attribute. In case it has not, we
+ * destroy it and use the compat code.
+ */
+static int
+netdev_gre_verify(const char *name)
+{
+ int err;
+ struct ofpbuf request, *reply;
+ struct ifinfomsg *ifmsg;
+
+ static const struct nl_policy rtlink_policy[] = {
+ [IFLA_LINKINFO] = { .type = NL_A_NESTED },
+ };
+ static const struct nl_policy linkinfo_policy[] = {
+ [IFLA_INFO_KIND] = { .type = NL_A_STRING },
+ [IFLA_INFO_DATA] = { .type = NL_A_NESTED },
+ };
+ static const struct nl_policy gre_policy[] = {
+ [IFLA_GRE_COLLECT_METADATA] = { .type = NL_A_FLAG },
+ };
+
+ ofpbuf_init(&request, 0);
+ nl_msg_put_nlmsghdr(&request, 0, RTM_GETLINK,
+ NLM_F_REQUEST);
+ ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
+ nl_msg_put_string(&request, IFLA_IFNAME, name);
+
+ err = nl_transact(NETLINK_ROUTE, &request, &reply);
+ if (!err) {
+ struct nlattr *rtlink[ARRAY_SIZE(rtlink_policy)];
+ struct nlattr *linkinfo[ARRAY_SIZE(linkinfo_policy)];
+ struct nlattr *gre[ARRAY_SIZE(gre_policy)];
+
+ err = EINVAL;
+ ifmsg = ofpbuf_at(reply, NLMSG_HDRLEN, sizeof *ifmsg);
+ if (nl_policy_parse(reply, NLMSG_HDRLEN + sizeof *ifmsg,
+ rtlink_policy, rtlink, ARRAY_SIZE(rtlink_policy))) {
+ if (nl_parse_nested(rtlink[IFLA_LINKINFO], linkinfo_policy,
+ linkinfo, ARRAY_SIZE(linkinfo_policy)) &&
+ !strcmp(nl_attr_get_string(linkinfo[IFLA_INFO_KIND]),
+ "gretap")) {
+ if (nl_parse_nested(linkinfo[IFLA_INFO_DATA], gre_policy, gre,
+ ARRAY_SIZE(gre_policy)) &&
+ nl_attr_get_flag(gre[IFLA_GRE_COLLECT_METADATA])) {
+ err = 0;
+ }
+ }
}
+ ofpbuf_uninit(reply);
}
- free(socksp);
+ ofpbuf_uninit(&request);
+ return err;
+}
-exit:
- ofpbuf_delete(buf);
- free(upcall_pids);
+static int
+netdev_gre_create(struct netdev *netdev)
+{
+ int err;
+ struct ofpbuf request, *reply;
+ size_t linkinfo_off, infodata_off;
+ char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
+ const char *name = netdev_vport_get_dpif_port(netdev,
+ namebuf, sizeof namebuf);
+ struct ifinfomsg *ifinfo;
+ const struct netdev_tunnel_config *tnl_cfg;
+ tnl_cfg = netdev_get_tunnel_config(netdev);
+ if (!tnl_cfg) { /* or assert? */
+ return EINVAL;
+ }
+
+ ofpbuf_init(&request, 0);
+ nl_msg_put_nlmsghdr(&request, 0, RTM_NEWLINK,
+ NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE);
+ ifinfo = ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
+ ifinfo->ifi_change = ifinfo->ifi_flags = IFF_UP;
+ nl_msg_put_string(&request, IFLA_IFNAME, name);
+ nl_msg_put_u32(&request, IFLA_MTU, UINT16_MAX);
+ linkinfo_off = nl_msg_start_nested(&request, IFLA_LINKINFO);
+ nl_msg_put_string(&request, IFLA_INFO_KIND, "gretap");
+ infodata_off = nl_msg_start_nested(&request, IFLA_INFO_DATA);
+ nl_msg_put_flag(&request, IFLA_GRE_COLLECT_METADATA);
+ nl_msg_end_nested(&request, infodata_off);
+ nl_msg_end_nested(&request, linkinfo_off);
+
+ err = nl_transact(NETLINK_ROUTE, &request, &reply);
+ if (!err) {
+ ofpbuf_uninit(reply);
+ }
+
+ if (!err && (err = netdev_gre_verify(name))) {
+ netdev_gre_destroy(name);
+ }
+
+ /*
+ * If tunnel metadata is not supported, EEXIST will be returned for zero
+ * addresses tunnel. We still need to verify metadata has been set as above.
+ */
+ if (err == EINVAL || err == EEXIST) {
+ err = EOPNOTSUPP;
+ }
+
+ ofpbuf_uninit(&request);
+ return err;
+}
+
+static int
+netdev_geneve_create(struct netdev *netdev)
+{
+ int err;
+ struct ofpbuf request, *reply;
+ size_t linkinfo_off, infodata_off;
+ char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
+ const char *name = netdev_vport_get_dpif_port(netdev,
+ namebuf, sizeof namebuf);
+ struct ifinfomsg *ifinfo;
+ const struct netdev_tunnel_config *tnl_cfg;
+ tnl_cfg = netdev_get_tunnel_config(netdev);
+ if (!tnl_cfg) { /* or assert? */
+ return EINVAL;
+ }
+
+ ofpbuf_init(&request, 0);
+ nl_msg_put_nlmsghdr(&request, 0, RTM_NEWLINK,
+ NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE);
+ ifinfo = ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
+ ifinfo->ifi_change = ifinfo->ifi_flags = IFF_UP;
+ nl_msg_put_string(&request, IFLA_IFNAME, name);
+ nl_msg_put_u32(&request, IFLA_MTU, UINT16_MAX);
+ linkinfo_off = nl_msg_start_nested(&request, IFLA_LINKINFO);
+ nl_msg_put_string(&request, IFLA_INFO_KIND, "geneve");
+ infodata_off = nl_msg_start_nested(&request, IFLA_INFO_DATA);
+ nl_msg_put_flag(&request, IFLA_GENEVE_COLLECT_METADATA);
+ nl_msg_put_u8(&request, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1);
+ nl_msg_put_be16(&request, IFLA_GENEVE_PORT, tnl_cfg->dst_port);
+ nl_msg_end_nested(&request, infodata_off);
+ nl_msg_end_nested(&request, linkinfo_off);
+
+ err = nl_transact(NETLINK_ROUTE, &request, &reply);
+
+ if (!err) {
+ ofpbuf_uninit(reply);
+ }
+
+ /*
+ * Linux versions older than 4.3 will return EINVAL in case the GENEVE_ID is
+ * not set, which is sufficient to verify COLLECT_METADATA is supported.
+ */
+ if (err == EINVAL) {
+ err = EOPNOTSUPP;
+ }
+
+ ofpbuf_uninit(&request);
+ return err;
+}
+
+#else
+
+static int
+netdev_vxlan_create(struct netdev *netdev OVS_UNUSED)
+{
+ return EOPNOTSUPP;
+}
+
+static int
+netdev_gre_create(struct netdev *netdev OVS_UNUSED)
+{
+ return EOPNOTSUPP;
+}
+
+static int
+netdev_geneve_create(struct netdev *netdev OVS_UNUSED)
+{
+ return EOPNOTSUPP;
+}
+
+static int
+netdev_vxlan_destroy(const char *name OVS_UNUSED)
+{
+ return EOPNOTSUPP;
+}
+
+static int
+netdev_gre_destroy(const char *name OVS_UNUSED)
+{
+ return EOPNOTSUPP;
+}
+
+static int
+netdev_geneve_destroy(const char *name OVS_UNUSED)
+{
+ return EOPNOTSUPP;
+}
+
+#endif
+
+static int
+dpif_netlink_port_query__(const struct dpif_netlink *dpif, odp_port_t port_no,
+ const char *port_name, struct dpif_port *dpif_port);
+
+static int
+dpif_netlink_port_create(struct netdev *netdev)
+{
+ switch (netdev_to_ovs_vport_type(netdev_get_type(netdev))) {
+ case OVS_VPORT_TYPE_VXLAN:
+ return netdev_vxlan_create(netdev);
+ case OVS_VPORT_TYPE_GRE:
+ return netdev_gre_create(netdev);
+ case OVS_VPORT_TYPE_GENEVE:
+ return netdev_geneve_create(netdev);
+ case OVS_VPORT_TYPE_NETDEV:
+ case OVS_VPORT_TYPE_INTERNAL:
+ case OVS_VPORT_TYPE_LISP:
+ case OVS_VPORT_TYPE_STT:
+ case OVS_VPORT_TYPE_UNSPEC:
+ case __OVS_VPORT_TYPE_MAX:
+ default:
+ return EOPNOTSUPP;
+ }
+ return 0;
+}
+
+static int
+dpif_netlink_port_destroy(const char *name, const char *type)
+{
+ switch (netdev_to_ovs_vport_type(type)) {
+ case OVS_VPORT_TYPE_VXLAN:
+ return netdev_vxlan_destroy(name);
+ case OVS_VPORT_TYPE_GRE:
+ return netdev_gre_destroy(name);
+ case OVS_VPORT_TYPE_GENEVE:
+ return netdev_geneve_destroy(name);
+ case OVS_VPORT_TYPE_NETDEV:
+ case OVS_VPORT_TYPE_INTERNAL:
+ case OVS_VPORT_TYPE_LISP:
+ case OVS_VPORT_TYPE_STT:
+ case OVS_VPORT_TYPE_UNSPEC:
+ case __OVS_VPORT_TYPE_MAX:
+ default:
+ return EOPNOTSUPP;
+ }
+ return 0;
+}
+
+static int
+dpif_netlink_port_create_and_add(struct dpif_netlink *dpif, struct netdev *netdev,
+ odp_port_t *port_nop)
+ OVS_REQ_WRLOCK(dpif->upcall_lock)
+{
+ int error;
+ char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
+ const char *name = netdev_vport_get_dpif_port(netdev,
+ namebuf, sizeof namebuf);
+
+ error = dpif_netlink_port_create(netdev);
+ if (error) {
+ return error;
+ }
+
+ error = dpif_netlink_port_add__(dpif, name, OVS_VPORT_TYPE_NETDEV, NULL, port_nop);
+ if (error) {
+ VLOG_DBG("failed to add port, destroying: %d", error);
+ dpif_netlink_port_destroy(name, netdev_get_type(netdev));
+ }
return error;
}
int error;
fat_rwlock_wrlock(&dpif->upcall_lock);
- error = dpif_netlink_port_add__(dpif, netdev, port_nop);
+ error = dpif_netlink_port_create_and_add(dpif, netdev, port_nop);
+ if (error == EOPNOTSUPP) {
+ error = dpif_netlink_port_add_compat(dpif, netdev, port_nop);
+ }
fat_rwlock_unlock(&dpif->upcall_lock);
return error;
{
struct dpif_netlink_vport vport;
int error;
+ struct dpif_port dpif_port;
+
+ error = dpif_netlink_port_query__(dpif, port_no, NULL, &dpif_port);
+ if (error) {
+ return error;
+ }
dpif_netlink_vport_init(&vport);
vport.cmd = OVS_VPORT_CMD_DEL;
vport_del_channels(dpif, port_no);
+ dpif_netlink_port_destroy(dpif_port.name, dpif_port.type);
+ dpif_port_destroy(&dpif_port);
+
return error;
}
error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
&ovs_datapath_family);
if (error) {
- VLOG_ERR("Generic Netlink family '%s' does not exist. "
- "The Open vSwitch kernel module is probably not loaded.",
- OVS_DATAPATH_FAMILY);
+ VLOG_WARN("Generic Netlink family '%s' does not exist. "
+ "The Open vSwitch kernel module is probably not loaded.",
+ OVS_DATAPATH_FAMILY);
}
if (!error) {
error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
projects / cascardo / ovs.git / blobdiff