#include "netdev.h"
#include "netdev-linux.h"
#include "netdev-vport.h"
+#include "netlink-notifier.h"
#include "netlink-socket.h"
#include "netlink.h"
#include "odp-util.h"
#include "ofpbuf.h"
+#include "openvswitch/datapath-compat.h"
#include "openvswitch/tunnel.h"
#include "packets.h"
#include "poll-loop.h"
-#include "rtnetlink.h"
-#include "rtnetlink-link.h"
#include "shash.h"
#include "sset.h"
#include "unaligned.h"
enum { LRU_MASK = LRU_MAX_PORTS - 1};
BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
+/* This ethtool flag was introduced in Linux 2.6.24, so it might be
+ * missing if we have old headers. */
+#define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */
+
struct dpif_linux_dp {
/* Generic Netlink header. */
uint8_t cmd;
/* Attributes. */
const char *name; /* OVS_DP_ATTR_NAME. */
+ uint32_t upcall_pid; /* OVS_DP_UPCALL_PID. */
struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
enum ovs_frag_handling ipv4_frags; /* OVS_DP_ATTR_IPV4_FRAGS. */
const uint32_t *sampling; /* OVS_DP_ATTR_SAMPLING. */
- uint32_t mcgroups[DPIF_N_UC_TYPES]; /* OVS_DP_ATTR_MCGROUPS. */
};
static void dpif_linux_dp_init(struct dpif_linux_dp *);
size_t key_len;
const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
size_t actions_len;
+ uint32_t upcall_pid; /* OVS_FLOW_ATTR_UPCALL_PID. */
const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
const uint64_t *used; /* OVS_FLOW_ATTR_USED. */
struct dpif dpif;
int dp_ifindex;
- /* Multicast group messages. */
- struct nl_sock *mc_sock;
- uint32_t mcgroups[DPIF_N_UC_TYPES];
+ /* Upcall messages. */
+ struct nl_sock *upcall_sock;
unsigned int listen_mask;
/* Change notification. */
struct sset changed_ports; /* Ports that have changed. */
- struct rtnetlink_notifier port_notifier;
+ struct nln_notifier *port_notifier;
bool change_error;
/* Queue of unused ports. */
/* Generic Netlink socket. */
static struct nl_sock *genl_sock;
+static struct nln *nln = NULL;
static int dpif_linux_init(void);
-static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
-static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
- void *dpif);
+static void open_dpif(const struct dpif_linux_dp *, struct dpif **);
+static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
+static void dpif_linux_port_changed(const void *vport, void *dpif);
static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
struct ofpbuf *);
if (error) {
return error;
}
- error = open_dpif(&dp, dpifp);
- ofpbuf_delete(buf);
- return error;
+ open_dpif(&dp, dpifp);
+ ofpbuf_delete(buf);
+ return 0;
}
-static int
+static void
open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
{
struct dpif_linux *dpif;
- int error;
int i;
dpif = xmalloc(sizeof *dpif);
- error = rtnetlink_link_notifier_register(&dpif->port_notifier,
- dpif_linux_port_changed, dpif);
- if (error) {
- goto error_free;
- }
+ dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
+ dpif);
dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
dp->dp_ifindex, dp->dp_ifindex);
- dpif->mc_sock = NULL;
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- dpif->mcgroups[i] = dp->mcgroups[i];
- }
+ dpif->upcall_sock = NULL;
dpif->listen_mask = 0;
dpif->dp_ifindex = dp->dp_ifindex;
sset_init(&dpif->changed_ports);
for (i = 1; i < LRU_MAX_PORTS; i++) {
dpif_linux_push_port(dpif, i);
}
- return 0;
-
-error_free:
- free(dpif);
- return error;
}
static void
dpif_linux_close(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- nl_sock_destroy(dpif->mc_sock);
- rtnetlink_link_notifier_unregister(&dpif->port_notifier);
+
+ nln_notifier_destroy(dpif->port_notifier);
+ nl_sock_destroy(dpif->upcall_sock);
sset_destroy(&dpif->changed_ports);
free(dpif->lru_bitmap);
free(dpif);
static void
dpif_linux_run(struct dpif *dpif OVS_UNUSED)
{
- rtnetlink_link_notifier_run();
+ if (nln) {
+ nln_run(nln);
+ }
}
static void
dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
{
- rtnetlink_link_notifier_wait();
+ if (nln) {
+ nln_wait(nln);
+ }
}
static int
request.options_len = options->size;
}
+ if (request.type == OVS_VPORT_TYPE_NETDEV) {
+ netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
+ }
+
/* Loop until we find a port that isn't used. */
do {
request.port_no = dpif_linux_pop_port(dpif);
+ if (dpif->upcall_sock) {
+ request.upcall_pid = nl_sock_pid(dpif->upcall_sock);
+ }
error = dpif_linux_vport_transact(&request, &reply, &buf);
if (!error) {
dpif_port->name = xstrdup(reply.name);
dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
dpif_port->port_no = reply.port_no;
- if (reply.stats) {
- netdev_stats_from_rtnl_link_stats64(&dpif_port->stats,
- reply.stats);
- } else {
- memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats);
- }
ofpbuf_delete(buf);
}
return error;
dpif_port->name = (char *) vport.name;
dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
dpif_port->port_no = vport.port_no;
- if (vport.stats) {
- netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, vport.stats);
- } else {
- memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats);
- }
return 0;
}
/* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
request.actions = actions ? actions : &dummy_action;
request.actions_len = actions_len;
+ if (dpif->upcall_sock) {
+ request.upcall_pid = nl_sock_pid(dpif->upcall_sock);
+ }
if (flags & DPIF_FP_ZERO_STATS) {
request.clear = true;
}
}
static int
-dpif_linux_execute__(int dp_ifindex,
+dpif_linux_execute__(int dp_ifindex, uint32_t upcall_pid,
const struct nlattr *key, size_t key_len,
const struct nlattr *actions, size_t actions_len,
const struct ofpbuf *packet)
nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, packet->data, packet->size);
nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len);
nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len);
+ nl_msg_put_u32(buf, OVS_PACKET_ATTR_UPCALL_PID, upcall_pid);
error = nl_sock_transact(genl_sock, buf, NULL);
ofpbuf_delete(buf);
const struct ofpbuf *packet)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ uint32_t upcall_pid = 0;
- return dpif_linux_execute__(dpif->dp_ifindex, key, key_len,
+ if (dpif->upcall_sock) {
+ upcall_pid = nl_sock_pid(dpif->upcall_sock);
+ }
+
+ return dpif_linux_execute__(dpif->dp_ifindex, upcall_pid, key, key_len,
actions, actions_len, packet);
}
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int error;
- int i;
if (listen_mask == dpif->listen_mask) {
return 0;
} else if (!listen_mask) {
- nl_sock_destroy(dpif->mc_sock);
- dpif->mc_sock = NULL;
- dpif->listen_mask = 0;
- return 0;
- } else if (!dpif->mc_sock) {
- error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
+ nl_sock_destroy(dpif->upcall_sock);
+ dpif->upcall_sock = NULL;
+ } else if (!dpif->upcall_sock) {
+ struct dpif_port port;
+ struct dpif_port_dump port_dump;
+ struct dpif_flow_dump flow_dump;
+ const struct nlattr *key;
+ size_t key_len;
+
+ error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_sock);
if (error) {
return error;
}
- }
- /* Unsubscribe from old groups. */
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- if (dpif->listen_mask & (1u << i)) {
- nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
+ DPIF_PORT_FOR_EACH (&port, &port_dump, dpif_) {
+ struct dpif_linux_vport vport_request;
+
+ dpif_linux_vport_init(&vport_request);
+ vport_request.cmd = OVS_VPORT_CMD_SET;
+ vport_request.dp_ifindex = dpif->dp_ifindex;
+ vport_request.port_no = port.port_no;
+ vport_request.upcall_pid = nl_sock_pid(dpif->upcall_sock);
+ error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
+ if (error) {
+ VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on "
+ "port: %s", dpif_name(dpif_), strerror(error));
+ }
}
- }
-
- /* Update listen_mask. */
- dpif->listen_mask = listen_mask;
- /* Subscribe to new groups. */
- error = 0;
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- if (dpif->listen_mask & (1u << i)) {
- int retval;
-
- retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
- if (retval) {
- error = retval;
+ dpif_flow_dump_start(&flow_dump, dpif_);
+ while (dpif_flow_dump_next(&flow_dump, &key, &key_len,
+ NULL, NULL, NULL)) {
+ struct dpif_linux_flow flow_request;
+
+ dpif_linux_flow_init(&flow_request);
+ flow_request.cmd = OVS_FLOW_CMD_SET;
+ flow_request.dp_ifindex = dpif->dp_ifindex;
+ flow_request.key = key;
+ flow_request.key_len = key_len;
+ flow_request.upcall_pid = nl_sock_pid(dpif->upcall_sock);
+ error = dpif_linux_flow_transact(&flow_request, NULL, NULL);
+ if (error) {
+ VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on "
+ "flow: %s", dpif_name(dpif_), strerror(error));
}
}
+ dpif_flow_dump_done(&flow_dump);
}
- return error;
+
+ dpif->listen_mask = listen_mask;
+ return 0;
}
static int
int error;
int i;
- if (!dpif->mc_sock) {
+ if (!dpif->upcall_sock) {
return EAGAIN;
}
for (i = 0; i < 50; i++) {
int dp_ifindex;
- error = nl_sock_recv(dpif->mc_sock, &buf, false);
+ error = nl_sock_recv(dpif->upcall_sock, &buf, false);
if (error) {
return error;
}
dpif_linux_recv_wait(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (dpif->mc_sock) {
- nl_sock_wait(dpif->mc_sock, POLLIN);
+ if (dpif->upcall_sock) {
+ nl_sock_wait(dpif->upcall_sock, POLLIN);
}
}
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (dpif->mc_sock) {
- nl_sock_drain(dpif->mc_sock);
+ if (dpif->upcall_sock) {
+ nl_sock_drain(dpif->upcall_sock);
}
}
static int error = -1;
if (error < 0) {
+ unsigned int ovs_vport_mcgroup;
+
error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
&ovs_datapath_family);
if (error) {
if (!error) {
error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
}
+ if (!error) {
+ error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
+ &ovs_vport_mcgroup,
+ OVS_VPORT_MCGROUP_FALLBACK_ID);
+ }
+ if (!error) {
+ static struct dpif_linux_vport vport;
+ nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
+ dpif_linux_nln_parse, &vport);
+ }
}
return error;
ofpbuf_use_stack(&actions, &action, sizeof action);
nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
- return dpif_linux_execute__(dp_ifindex, key.data, key.size,
+ return dpif_linux_execute__(dp_ifindex, 0, key.data, key.size,
actions.data, actions.size, &packet);
}
+static bool
+dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
+{
+ struct dpif_linux_vport *vport = vport_;
+ return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
+}
+
static void
-dpif_linux_port_changed(const struct rtnetlink_link_change *change,
- void *dpif_)
+dpif_linux_port_changed(const void *vport_, void *dpif_)
{
+ const struct dpif_linux_vport *vport = vport_;
struct dpif_linux *dpif = dpif_;
- if (change) {
- if (change->master_ifindex == dpif->dp_ifindex
- && (change->nlmsg_type == RTM_NEWLINK
- || change->nlmsg_type == RTM_DELLINK))
- {
- /* Our datapath changed, either adding a new port or deleting an
- * existing one. */
- sset_add(&dpif->changed_ports, change->ifname);
+ if (vport) {
+ if (vport->dp_ifindex == dpif->dp_ifindex
+ && (vport->cmd == OVS_VPORT_CMD_NEW
+ || vport->cmd == OVS_VPORT_CMD_DEL
+ || vport->cmd == OVS_VPORT_CMD_SET)) {
+ VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
+ dpif->dpif.full_name, vport->name, vport->cmd);
+ sset_add(&dpif->changed_ports, vport->name);
}
} else {
dpif->change_error = true;
[OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
[OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
[OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
+ [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
[OVS_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct rtnl_link_stats64),
- .max_len = sizeof(struct rtnl_link_stats64),
+ .min_len = sizeof(struct ovs_vport_stats),
+ .max_len = sizeof(struct ovs_vport_stats),
.optional = true },
[OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
.min_len = ETH_ADDR_LEN,
.max_len = ETH_ADDR_LEN,
.optional = true },
- [OVS_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
[OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
[OVS_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
- [OVS_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
};
struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
+ if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
+ vport->upcall_pid = nl_attr_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
+ }
if (a[OVS_VPORT_ATTR_STATS]) {
vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
}
if (a[OVS_VPORT_ATTR_ADDRESS]) {
vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]);
}
- if (a[OVS_VPORT_ATTR_MTU]) {
- vport->mtu = nl_attr_get_u32(a[OVS_VPORT_ATTR_MTU]);
- } else {
- vport->mtu = INT_MAX;
- }
if (a[OVS_VPORT_ATTR_OPTIONS]) {
vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
if (a[OVS_VPORT_ATTR_IFINDEX]) {
vport->ifindex = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFINDEX]);
}
- if (a[OVS_VPORT_ATTR_IFLINK]) {
- vport->iflink = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFLINK]);
- }
return 0;
}
nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
}
+ nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
+
if (vport->stats) {
nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
vport->stats, sizeof *vport->stats);
vport->address, ETH_ADDR_LEN);
}
- if (vport->mtu && vport->mtu != INT_MAX) {
- nl_msg_put_u32(buf, OVS_VPORT_ATTR_MTU, vport->mtu);
- }
-
if (vport->options) {
nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
vport->options, vport->options_len);
if (vport->ifindex) {
nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFINDEX, vport->ifindex);
}
-
- if (vport->iflink) {
- nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFLINK, vport->iflink);
- }
}
/* Clears 'vport' to "empty" values. */
.optional = true },
[OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
[OVS_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
- [OVS_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
};
struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
dp->sampling = nl_attr_get(a[OVS_DP_ATTR_SAMPLING]);
}
- if (a[OVS_DP_ATTR_MCGROUPS]) {
- static const struct nl_policy ovs_mcgroup_policy[] = {
- [OVS_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
- [OVS_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
- [OVS_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
- };
-
- struct nlattr *mcgroups[ARRAY_SIZE(ovs_mcgroup_policy)];
-
- if (!nl_parse_nested(a[OVS_DP_ATTR_MCGROUPS], ovs_mcgroup_policy,
- mcgroups, ARRAY_SIZE(ovs_mcgroup_policy))) {
- return EINVAL;
- }
-
- if (mcgroups[OVS_PACKET_CMD_MISS]) {
- dp->mcgroups[DPIF_UC_MISS]
- = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_MISS]);
- }
- if (mcgroups[OVS_PACKET_CMD_ACTION]) {
- dp->mcgroups[DPIF_UC_ACTION]
- = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_ACTION]);
- }
- if (mcgroups[OVS_PACKET_CMD_SAMPLE]) {
- dp->mcgroups[DPIF_UC_SAMPLE]
- = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_SAMPLE]);
- }
- }
-
return 0;
}
nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
}
+ nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, dp->upcall_pid);
+
/* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
if (dp->ipv4_frags) {
static const struct nl_policy ovs_flow_policy[] = {
[OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
[OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
+ [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
[OVS_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
.min_len = sizeof(struct ovs_flow_stats),
.max_len = sizeof(struct ovs_flow_stats),
flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
}
+ if (a[OVS_FLOW_ATTR_UPCALL_PID]) {
+ flow->upcall_pid = nl_attr_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
+ }
if (a[OVS_FLOW_ATTR_STATS]) {
flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
}
flow->actions, flow->actions_len);
}
+ nl_msg_put_u32(buf, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
+
/* We never need to send these to the kernel. */
assert(!flow->stats);
assert(!flow->tcp_flags);
git://git.onelab.eu / sliver-openvswitch.git / blobdiff