#include <sys/stat.h>
#include <unistd.h>
+#include "bitmap.h"
#include "dpif-provider.h"
#include "netdev.h"
+#include "netdev-linux.h"
#include "netdev-vport.h"
#include "netlink-socket.h"
#include "netlink.h"
#include "rtnetlink.h"
#include "rtnetlink-link.h"
#include "shash.h"
-#include "svec.h"
+#include "sset.h"
#include "unaligned.h"
#include "util.h"
#include "vlog.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;
unsigned int listen_mask;
/* Change notification. */
- struct shash changed_ports; /* Ports that have changed. */
+ 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 svec *all_dps)
+dpif_linux_enumerate(struct sset *all_dps)
{
struct nl_dump dump;
struct ofpbuf msg;
struct dpif_linux_dp dp;
if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
- svec_add(all_dps, dp.name);
+ sset_add(all_dps, dp.name);
}
}
return nl_dump_done(&dump);
}
dpif->listen_mask = 0;
dpif->dp_ifindex = dp->dp_ifindex;
- shash_init(&dpif->changed_ports);
+ sset_init(&dpif->changed_ports);
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:
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
rtnetlink_link_notifier_unregister(&dpif->port_notifier);
- shash_destroy(&dpif->changed_ports);
+ 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;
}
if (dpif->change_error) {
dpif->change_error = false;
- shash_clear(&dpif->changed_ports);
+ sset_clear(&dpif->changed_ports);
return ENOBUFS;
- } else if (!shash_is_empty(&dpif->changed_ports)) {
- struct shash_node *node = shash_first(&dpif->changed_ports);
- *devnamep = shash_steal(&dpif->changed_ports, node);
+ } else if (!sset_is_empty(&dpif->changed_ports)) {
+ *devnamep = sset_pop(&dpif->changed_ports);
return 0;
} else {
return EAGAIN;
dpif_linux_port_poll_wait(const struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) {
+ if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
poll_immediate_wake();
} else {
rtnetlink_link_notifier_wait();
}
static int
-dpif_linux_flow_get(const struct dpif *dpif_,
- const struct nlattr *key, size_t key_len,
- struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
+dpif_linux_flow_get__(const struct dpif *dpif_,
+ const struct nlattr *key, size_t key_len,
+ struct dpif_linux_flow *reply, struct ofpbuf **bufp)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_flow request, reply;
- struct ofpbuf *buf;
- int error;
+ struct dpif_linux_flow request;
dpif_linux_flow_init(&request);
request.cmd = ODP_FLOW_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
request.key = key;
request.key_len = key_len;
- error = dpif_linux_flow_transact(&request, &reply, &buf);
+ return dpif_linux_flow_transact(&request, reply, bufp);
+}
+
+static int
+dpif_linux_flow_get(const struct dpif *dpif_,
+ const struct nlattr *key, size_t key_len,
+ struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
+{
+ struct dpif_linux_flow reply;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
if (!error) {
if (stats) {
dpif_linux_flow_get_stats(&reply, stats);
struct nl_dump dump;
struct dpif_linux_flow flow;
struct dpif_flow_stats stats;
+ struct ofpbuf *buf;
};
static int
nl_dump_start(&state->dump, genl_sock, buf);
ofpbuf_delete(buf);
+ state->buf = NULL;
+
return 0;
}
struct ofpbuf buf;
int error;
- if (!nl_dump_next(&state->dump, &buf)) {
- return EOF;
- }
+ do {
+ ofpbuf_delete(state->buf);
+ state->buf = NULL;
- error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
- if (!error) {
- if (key) {
- *key = state->flow.key;
- *key_len = state->flow.key_len;
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
}
- if (actions) {
- *actions = state->flow.actions;
- *actions_len = state->flow.actions_len;
+
+ error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
+ if (error) {
+ return error;
}
- if (stats) {
- dpif_linux_flow_get_stats(&state->flow, &state->stats);
- *stats = &state->stats;
+
+ if (actions && !state->flow.actions) {
+ error = dpif_linux_flow_get__(dpif_, state->flow.key,
+ state->flow.key_len,
+ &state->flow, &state->buf);
+ if (error == ENOENT) {
+ VLOG_DBG("dumped flow disappeared on get");
+ } else if (error) {
+ VLOG_WARN("error fetching dumped flow: %s", strerror(error));
+ }
}
+ } while (error);
+
+ if (actions) {
+ *actions = state->flow.actions;
+ *actions_len = state->flow.actions_len;
+ }
+ if (key) {
+ *key = state->flow.key;
+ *key_len = state->flow.key_len;
+ }
+ if (stats) {
+ dpif_linux_flow_get_stats(&state->flow, &state->stats);
+ *stats = &state->stats;
}
return error;
}
{
struct dpif_linux_flow_state *state = state_;
int error = nl_dump_done(&state->dump);
+ ofpbuf_delete(state->buf);
free(state);
return error;
}
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));
}
return reply.type == ODP_VPORT_TYPE_INTERNAL;
}
+int
+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);
+
+ 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;
+
+ nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size);
+
+ 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);
+
+ error = nl_sock_transact(genl_sock, buf, NULL);
+ ofpbuf_delete(buf);
+ return error;
+}
+
static void
dpif_linux_port_changed(const struct rtnetlink_link_change *change,
void *dpif_)
{
/* Our datapath changed, either adding a new port or deleting an
* existing one. */
- shash_add_once(&dpif->changed_ports, change->ifname, NULL);
+ sset_add(&dpif->changed_ports, change->ifname);
}
} else {
dpif->change_error = true;
}
if (a[ODP_VPORT_ATTR_MTU]) {
vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
+ } else {
+ vport->mtu = INT_MAX;
}
if (a[ODP_VPORT_ATTR_OPTIONS]) {
vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
vport->address, ETH_ADDR_LEN);
}
- if (vport->mtu) {
+ if (vport->mtu && vport->mtu != INT_MAX) {
nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
}
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)
{
git://git.onelab.eu / sliver-openvswitch.git / blobdiff