#include "netlink.h"
#include "odp-util.h"
#include "ofpbuf.h"
-#include "openvswitch/datapath-compat.h"
#include "packets.h"
#include "poll-loop.h"
#include "random.h"
/* Attributes. */
const char *name; /* OVS_DP_ATTR_NAME. */
- const uint32_t *upcall_pid; /* OVS_DP_UPCALL_PID. */
+ const uint32_t *upcall_pid; /* OVS_DP_ATTR_UPCALL_PID. */
struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
+ struct ovs_dp_megaflow_stats megaflow_stats;
+ /* OVS_DP_ATTR_MEGAFLOW_STATS.*/
};
static void dpif_linux_dp_init(struct dpif_linux_dp *);
int dp_ifindex;
/* Upcall messages. */
- pthread_mutex_t upcall_lock;
+ struct ovs_mutex upcall_lock;
int uc_array_size; /* Size of 'channels' and 'epoll_events'. */
struct dpif_channel *channels;
struct epoll_event *epoll_events;
/* Change notification. */
struct nl_sock *port_notifier; /* vport multicast group subscriber. */
+ bool refresh_channels;
};
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
static uint32_t dpif_linux_port_get_pid(const struct dpif *,
odp_port_t port_no);
+static int dpif_linux_refresh_channels(struct dpif *);
static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
struct ofpbuf *);
dpif = xzalloc(sizeof *dpif);
dpif->port_notifier = NULL;
- xpthread_mutex_init(&dpif->upcall_lock, NULL);
+ ovs_mutex_init(&dpif->upcall_lock);
dpif->epoll_fd = -1;
dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
if (dpif->epoll_fd >= 0) {
close(dpif->epoll_fd);
}
- xpthread_mutex_destroy(&dpif->upcall_lock);
+ ovs_mutex_destroy(&dpif->upcall_lock);
free(dpif);
}
return dpif_linux_dp_transact(&dp, NULL, NULL);
}
+static void
+dpif_linux_run(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ if (dpif->refresh_channels) {
+ dpif->refresh_channels = false;
+ dpif_linux_refresh_channels(dpif_);
+ }
+}
+
static int
dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
{
stats->n_missed = dp.stats.n_missed;
stats->n_lost = dp.stats.n_lost;
stats->n_flows = dp.stats.n_flows;
+ stats->n_masks = dp.megaflow_stats.n_masks;
+ stats->n_mask_hit = dp.megaflow_stats.n_mask_hit;
ofpbuf_delete(buf);
}
return error;
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int error;
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
error = dpif_linux_port_add__(dpif_, netdev, port_nop);
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
return error;
}
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int error;
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
error = dpif_linux_port_del__(dpif_, port_no);
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
return error;
}
return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
}
-static odp_port_t
+static uint32_t
dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
{
- return u32_to_odp(MAX_PORTS);
+ return MAX_PORTS;
}
static uint32_t
uint32_t port_idx = odp_to_u32(port_no);
uint32_t pid = 0;
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
if (dpif->epoll_fd >= 0) {
/* The ODPP_NONE "reserved" port number uses the "ovs-system"'s
* channel, since it is not heavily loaded. */
uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx;
pid = nl_sock_pid(dpif->channels[idx].sock);
}
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
return pid;
}
struct nl_dump dump;
};
-static int
-dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
+static void
+dpif_linux_port_dump_start__(const struct dpif *dpif_, struct nl_dump *dump)
{
const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_port_state *state;
struct dpif_linux_vport request;
struct ofpbuf *buf;
- *statep = state = xmalloc(sizeof *state);
-
dpif_linux_vport_init(&request);
- request.cmd = OVS_DP_CMD_GET;
+ request.cmd = OVS_VPORT_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
buf = ofpbuf_new(1024);
dpif_linux_vport_to_ofpbuf(&request, buf);
- nl_dump_start(&state->dump, NETLINK_GENERIC, buf);
+ nl_dump_start(dump, NETLINK_GENERIC, buf);
ofpbuf_delete(buf);
+}
+
+static int
+dpif_linux_port_dump_start(const struct dpif *dpif, void **statep)
+{
+ struct dpif_linux_port_state *state;
+
+ *statep = state = xmalloc(sizeof *state);
+ dpif_linux_port_dump_start__(dpif, &state->dump);
return 0;
}
+static bool
+dpif_linux_port_dump_next__(const struct dpif *dpif_, struct nl_dump *dump,
+ struct dpif_linux_vport *vport)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct ofpbuf buf;
+ int error;
+
+ if (!nl_dump_next(dump, &buf)) {
+ return EOF;
+ }
+
+ error = dpif_linux_vport_from_ofpbuf(vport, &buf);
+ if (error) {
+ VLOG_WARN_RL(&error_rl, "%s: failed to parse vport record (%s)",
+ dpif_name(&dpif->dpif), ovs_strerror(error));
+ }
+ return error;
+}
+
static int
dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
struct dpif_port *dpif_port)
{
struct dpif_linux_port_state *state = state_;
struct dpif_linux_vport vport;
- struct ofpbuf buf;
int error;
- if (!nl_dump_next(&state->dump, &buf)) {
- return EOF;
- }
-
- error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
+ error = dpif_linux_port_dump_next__(dpif, &state->dump, &vport);
if (error) {
return error;
}
-
dpif_port->name = CONST_CAST(char *, vport.name);
dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
dpif_port->port_no = vport.port_no;
|| vport.cmd == OVS_VPORT_CMD_SET)) {
VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
dpif->dpif.full_name, vport.name, vport.cmd);
+ if (vport.cmd == OVS_VPORT_CMD_DEL) {
+ dpif->refresh_channels = true;
+ }
*devnamep = xstrdup(vport.name);
+ ofpbuf_uninit(&buf);
return 0;
- } else {
- continue;
}
}
- } else if (error == EAGAIN) {
- return EAGAIN;
+ } else if (error != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)",
+ ovs_strerror(error));
+ nl_sock_drain(dpif->port_notifier);
+ error = ENOBUFS;
}
- VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)",
- ovs_strerror(error));
- nl_sock_drain(dpif->port_notifier);
- return ENOBUFS;
+ ofpbuf_uninit(&buf);
+ if (error) {
+ return error;
+ }
}
}
*statep = state = xmalloc(sizeof *state);
dpif_linux_flow_init(&request);
- request.cmd = OVS_DP_CMD_GET;
+ request.cmd = OVS_FLOW_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
buf = ofpbuf_new(1024);
}
}
+/* Synchronizes 'dpif->channels' with the set of vports currently in 'dpif' in
+ * the kernel, by adding a new channel for any kernel vport that lacks one and
+ * deleting any channels that have no backing kernel vports. */
static int
-dpif_linux_recv_set__(struct dpif *dpif_, bool enable)
+dpif_linux_refresh_channels(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ unsigned long int *keep_channels;
+ struct dpif_linux_vport vport;
+ size_t keep_channels_nbits;
+ struct nl_dump dump;
+ int retval = 0;
+ size_t i;
- if ((dpif->epoll_fd >= 0) == enable) {
- return 0;
+ /* To start with, we need an epoll fd. */
+ if (dpif->epoll_fd < 0) {
+ dpif->epoll_fd = epoll_create(10);
+ if (dpif->epoll_fd < 0) {
+ return errno;
+ }
}
- if (!enable) {
- destroy_channels(dpif);
- } else {
- struct dpif_port_dump port_dump;
- struct dpif_port port;
+ keep_channels_nbits = dpif->uc_array_size;
+ keep_channels = bitmap_allocate(keep_channels_nbits);
- if (dpif->epoll_fd < 0) {
- dpif->epoll_fd = epoll_create(10);
- if (dpif->epoll_fd < 0) {
- return errno;
- }
- }
+ dpif->n_events = dpif->event_offset = 0;
- DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
- struct dpif_linux_vport vport_request;
- struct nl_sock *sock;
- uint32_t upcall_pid;
- int error;
+ dpif_linux_port_dump_start__(dpif_, &dump);
+ while (dpif_linux_port_dump_next__(dpif_, &dump, &vport)) {
+ uint32_t port_no = odp_to_u32(vport.port_no);
+ struct nl_sock *sock = (port_no < dpif->uc_array_size
+ ? dpif->channels[port_no].sock
+ : NULL);
+ bool new_sock = !sock;
+ int error;
+ if (new_sock) {
error = nl_sock_create(NETLINK_GENERIC, &sock);
if (error) {
- return error;
+ retval = error;
+ goto error;
}
+ }
- upcall_pid = nl_sock_pid(sock);
+ /* Configure the vport to deliver misses to 'sock'. */
+ if (!vport.upcall_pid || *vport.upcall_pid != nl_sock_pid(sock)) {
+ uint32_t upcall_pid = nl_sock_pid(sock);
+ 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.port_no = vport.port_no;
vport_request.upcall_pid = &upcall_pid;
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->dpif), ovs_strerror(error));
- nl_sock_destroy(sock);
- if (error == ENODEV || error == ENOENT) {
- /* This device isn't there, but keep trying the others. */
- continue;
+ if (error != ENODEV && error != ENOENT) {
+ retval = error;
} else {
- return error;
+ /* The vport isn't really there, even though the dump says
+ * it is. Probably we just hit a race after a port
+ * disappeared. */
}
+ goto error;
}
+ }
- error = add_channel(dpif, port.port_no, sock);
+ if (new_sock) {
+ error = add_channel(dpif, vport.port_no, sock);
if (error) {
VLOG_INFO("%s: could not add channel for port %s",
- dpif_name(dpif_), port.name);
- nl_sock_destroy(sock);
- return error;
+ dpif_name(dpif_), vport.name);
+ retval = error;
+ goto error;
}
}
+
+ if (port_no < keep_channels_nbits) {
+ bitmap_set1(keep_channels, port_no);
+ }
+ continue;
+
+ error:
+ nl_sock_destroy(sock);
+ }
+ nl_dump_done(&dump);
+
+ /* Discard any saved channels that we didn't reuse. */
+ for (i = 0; i < keep_channels_nbits; i++) {
+ if (!bitmap_is_set(keep_channels, i)) {
+ nl_sock_destroy(dpif->channels[i].sock);
+ dpif->channels[i].sock = NULL;
+ }
}
+ free(keep_channels);
- return 0;
+ return retval;
+}
+
+static int
+dpif_linux_recv_set__(struct dpif *dpif_, bool enable)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ if ((dpif->epoll_fd >= 0) == enable) {
+ return 0;
+ } else if (!enable) {
+ destroy_channels(dpif);
+ return 0;
+ } else {
+ return dpif_linux_refresh_channels(dpif_);
+ }
}
static int
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int error;
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
error = dpif_linux_recv_set__(dpif_, enable);
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
return error;
}
memset(upcall, 0, sizeof *upcall);
upcall->type = type;
- upcall->packet = buf;
- upcall->packet->data = CONST_CAST(struct nlattr *,
- nl_attr_get(a[OVS_PACKET_ATTR_PACKET]));
- upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
upcall->key = CONST_CAST(struct nlattr *,
nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
upcall->userdata = a[OVS_PACKET_ATTR_USERDATA];
+
+ /* Allow overwriting the netlink attribute header without reallocating. */
+ ofpbuf_use_stub(&upcall->packet,
+ CONST_CAST(struct nlattr *,
+ nl_attr_get(a[OVS_PACKET_ATTR_PACKET])) - 1,
+ nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]) +
+ sizeof(struct nlattr));
+ upcall->packet.data = (char *)upcall->packet.data + sizeof(struct nlattr);
+ upcall->packet.size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
+
*dp_ifindex = ovs_header->dp_ifindex;
return 0;
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int error;
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
error = dpif_linux_recv__(dpif_, upcall, buf);
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
return error;
}
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
if (dpif->epoll_fd >= 0) {
poll_fd_wait(dpif->epoll_fd, POLLIN);
}
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
}
static void
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- xpthread_mutex_lock(&dpif->upcall_lock);
+ ovs_mutex_lock(&dpif->upcall_lock);
if (dpif->epoll_fd >= 0) {
struct dpif_channel *ch;
}
}
}
- xpthread_mutex_unlock(&dpif->upcall_lock);
+ ovs_mutex_unlock(&dpif->upcall_lock);
}
const struct dpif_class dpif_linux_class = {
dpif_linux_open,
dpif_linux_close,
dpif_linux_destroy,
- NULL, /* run */
+ dpif_linux_run,
NULL, /* wait */
dpif_linux_get_stats,
dpif_linux_port_add,
}
if (!error) {
error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
- &ovs_vport_mcgroup,
- OVS_VPORT_MCGROUP_FALLBACK_ID);
+ &ovs_vport_mcgroup);
}
ovsthread_once_done(&once);
[OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
[OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
.optional = true },
+ [OVS_DP_ATTR_MEGAFLOW_STATS] = {
+ NL_POLICY_FOR(struct ovs_dp_megaflow_stats),
+ .optional = true },
};
struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
sizeof dp->stats);
}
+ if (a[OVS_DP_ATTR_MEGAFLOW_STATS]) {
+ /* Can't use structure assignment because Netlink doesn't ensure
+ * sufficient alignment for 64-bit members. */
+ memcpy(&dp->megaflow_stats, nl_attr_get(a[OVS_DP_ATTR_MEGAFLOW_STATS]),
+ sizeof dp->megaflow_stats);
+ }
+
return 0;
}
dpif_linux_dp_init(struct dpif_linux_dp *dp)
{
memset(dp, 0, sizeof *dp);
+ dp->megaflow_stats.n_masks = UINT32_MAX;
+ dp->megaflow_stats.n_mask_hit = UINT64_MAX;
}
static void
ofpbuf_delete(request_buf);
if (reply) {
+ dpif_linux_dp_init(reply);
if (!error) {
error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
}
if (error) {
- dpif_linux_dp_init(reply);
ofpbuf_delete(*bufp);
*bufp = NULL;
}
time_msec() - ch->last_poll);
}
- VLOG_WARN("%s: lost packet on channel %td%s",
+ VLOG_WARN("%s: lost packet on channel %"PRIdPTR"%s",
dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));
ds_destroy(&s);
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff