#include <sys/stat.h>
#include <unistd.h>
+#include "bitmap.h"
#include "dpif-provider.h"
+#include "dynamic-string.h"
+#include "flow.h"
#include "netdev.h"
+#include "netdev-linux.h"
#include "netdev-vport.h"
#include "netlink-socket.h"
#include "netlink.h"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
+enum { LRU_MAX_PORTS = 1024 };
+enum { LRU_MASK = LRU_MAX_PORTS - 1};
+BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
+
struct dpif_linux_dp {
/* Generic Netlink header. */
uint8_t cmd;
struct sset changed_ports; /* Ports that have changed. */
struct rtnetlink_notifier port_notifier;
bool change_error;
+
+ /* Queue of unused ports. */
+ unsigned long *lru_bitmap;
+ uint16_t lru_ports[LRU_MAX_PORTS];
+ size_t lru_head;
+ size_t lru_tail;
};
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
return CONTAINER_OF(dpif, struct dpif_linux, dpif);
}
+static void
+dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
+{
+ if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
+ bitmap_set1(dp->lru_bitmap, port);
+ dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
+ }
+}
+
+static uint32_t
+dpif_linux_pop_port(struct dpif_linux *dp)
+{
+ uint16_t port;
+
+ if (dp->lru_head == dp->lru_tail) {
+ return UINT32_MAX;
+ }
+
+ port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
+ bitmap_set0(dp->lru_bitmap, port);
+ return port;
+}
+
static int
dpif_linux_enumerate(struct sset *all_dps)
{
dpif->change_error = false;
*dpifp = &dpif->dpif;
+ dpif->lru_head = dpif->lru_tail = 0;
+ dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
+ bitmap_set1(dpif->lru_bitmap, ODPP_LOCAL);
+ for (i = 1; i < LRU_MAX_PORTS; i++) {
+ dpif_linux_push_port(dpif, i);
+ }
return 0;
error_free:
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);
sset_destroy(&dpif->changed_ports);
+ free(dpif->lru_bitmap);
free(dpif);
}
request.options_len = options->size;
}
- error = dpif_linux_vport_transact(&request, &reply, &buf);
- if (!error) {
- *port_nop = reply.port_no;
+ /* Loop until we find a port that isn't used. */
+ do {
+ request.port_no = dpif_linux_pop_port(dpif);
+ error = dpif_linux_vport_transact(&request, &reply, &buf);
+
+ if (!error) {
+ *port_nop = reply.port_no;
+ }
ofpbuf_delete(buf);
- }
+ } while (request.port_no != UINT32_MAX
+ && (error == EBUSY || error == EFBIG));
return error;
}
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_vport vport;
+ int error;
dpif_linux_vport_init(&vport);
vport.cmd = ODP_VPORT_CMD_DEL;
vport.dp_ifindex = dpif->dp_ifindex;
vport.port_no = port_no;
- return dpif_linux_vport_transact(&vport, NULL, NULL);
+ error = dpif_linux_vport_transact(&vport, NULL, NULL);
+
+ if (!error) {
+ dpif_linux_push_port(dpif, port_no);
+ }
+ return error;
}
static int
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;
struct dpif_linux_port_state {
struct nl_dump dump;
+ unsigned long *port_bitmap; /* Ports in the datapath. */
+ bool complete; /* Dump completed without error. */
};
static int
struct ofpbuf *buf;
*statep = state = xmalloc(sizeof *state);
+ state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
+ state->complete = false;
dpif_linux_vport_init(&request);
request.cmd = ODP_DP_CMD_GET;
int error;
if (!nl_dump_next(&state->dump, &buf)) {
+ state->complete = true;
return EOF;
}
return error;
}
+ if (vport.port_no < LRU_MAX_PORTS) {
+ bitmap_set1(state->port_bitmap, vport.port_no);
+ }
+
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;
}
static int
-dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
+dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_port_state *state = state_;
int error = nl_dump_done(&state->dump);
+
+ if (state->complete) {
+ uint16_t i;
+
+ for (i = 0; i < LRU_MAX_PORTS; i++) {
+ if (!bitmap_is_set(state->port_bitmap, i)) {
+ dpif_linux_push_port(dpif, i);
+ }
+ }
+ }
+
+ free(state->port_bitmap);
free(state);
return error;
}
}
static int
-dpif_linux_execute(struct dpif *dpif_,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *packet)
+dpif_linux_execute__(int dp_ifindex,
+ const struct nlattr *key, size_t key_len,
+ const struct nlattr *actions, size_t actions_len,
+ const struct ofpbuf *packet)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct odp_header *execute;
struct ofpbuf *buf;
int error;
ODP_PACKET_CMD_EXECUTE, 1);
execute = ofpbuf_put_uninit(buf, sizeof *execute);
- execute->dp_ifindex = dpif->dp_ifindex;
+ execute->dp_ifindex = dp_ifindex;
nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
+ nl_msg_put_unspec(buf, ODP_PACKET_ATTR_KEY, key, key_len);
nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
error = nl_sock_transact(genl_sock, buf, NULL);
return error;
}
+static int
+dpif_linux_execute(struct dpif *dpif_,
+ const struct nlattr *key, size_t key_len,
+ const struct nlattr *actions, size_t actions_len,
+ const struct ofpbuf *packet)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ return dpif_linux_execute__(dpif->dp_ifindex, key, key_len,
+ actions, actions_len, packet);
+}
+
static int
dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
{
const struct dpif_class dpif_linux_class = {
"system",
- NULL, /* run */
- NULL, /* wait */
dpif_linux_enumerate,
dpif_linux_open,
dpif_linux_close,
dpif_linux_destroy,
+ NULL, /* run */
+ NULL, /* wait */
dpif_linux_get_stats,
dpif_linux_get_drop_frags,
dpif_linux_set_drop_frags,
error = dpif_linux_vport_get(name, &reply, &buf);
if (!error) {
ofpbuf_delete(buf);
- } else if (error != ENODEV) {
+ } else if (error != ENODEV && error != ENOENT) {
VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
name, strerror(error));
}
dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
const void *data, size_t size)
{
- struct odp_header *execute;
- struct ofpbuf *buf;
- size_t actions_ofs;
- int error;
-
- buf = ofpbuf_new(128 + size);
+ struct ofpbuf actions, key, packet;
+ struct odputil_keybuf keybuf;
+ struct flow flow;
+ uint64_t action;
- nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
- ODP_PACKET_CMD_EXECUTE, 1);
-
- execute = ofpbuf_put_uninit(buf, sizeof *execute);
- execute->dp_ifindex = dp_ifindex;
+ ofpbuf_use_const(&packet, data, size);
+ flow_extract(&packet, htonll(0), 0, &flow);
- nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size);
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+ odp_flow_key_from_flow(&key, &flow);
- actions_ofs = nl_msg_start_nested(buf, ODP_PACKET_ATTR_ACTIONS);
- nl_msg_put_u32(buf, ODP_ACTION_ATTR_OUTPUT, port_no);
- nl_msg_end_nested(buf, actions_ofs);
+ ofpbuf_use_stack(&actions, &action, sizeof action);
+ nl_msg_put_u32(&actions, ODP_ACTION_ATTR_OUTPUT, port_no);
- error = nl_sock_transact(genl_sock, buf, NULL);
- ofpbuf_delete(buf);
- return error;
+ return dpif_linux_execute__(dp_ifindex, key.data, key.size,
+ actions.data, actions.size, &packet);
}
static void
assert((reply != NULL) == (bufp != NULL));
+ error = dpif_linux_init();
+ if (error) {
+ if (reply) {
+ *bufp = NULL;
+ dpif_linux_vport_init(reply);
+ }
+ return error;
+ }
+
request_buf = ofpbuf_new(1024);
dpif_linux_vport_to_ofpbuf(request, request_buf);
error = nl_sock_transact(genl_sock, request_buf, bufp);
}
/* Clears 'dp' to "empty" values. */
-void
+static void
dpif_linux_dp_init(struct dpif_linux_dp *dp)
{
memset(dp, 0, sizeof *dp);
* result of the command is expected to be of the same form, which is decoded
* and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
* reply is no longer needed ('reply' will contain pointers into '*bufp'). */
-int
+static int
dpif_linux_dp_transact(const struct dpif_linux_dp *request,
struct dpif_linux_dp *reply, struct ofpbuf **bufp)
{
/* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
* The caller must free '*bufp' when the reply is no longer needed ('reply'
* will contain pointers into '*bufp'). */
-int
+static int
dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
struct ofpbuf **bufp)
{
}
/* Clears 'flow' to "empty" values. */
-void
+static void
dpif_linux_flow_init(struct dpif_linux_flow *flow)
{
memset(flow, 0, sizeof *flow);
* result of the command is expected to be a flow also, which is decoded and
* stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
* is no longer needed ('reply' will contain pointers into '*bufp'). */
-int
+static int
dpif_linux_flow_transact(const struct dpif_linux_flow *request,
struct dpif_linux_flow *reply, struct ofpbuf **bufp)
{