/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
+#include <poll.h>
#include <stdlib.h>
+#include <strings.h>
+#include <sys/epoll.h>
#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-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 "random.h"
#include "shash.h"
#include "sset.h"
-#include "svec.h"
+#include "timeval.h"
#include "unaligned.h"
#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
+enum { MAX_PORTS = USHRT_MAX };
+
+enum { N_CHANNELS = 17 };
+BUILD_ASSERT_DECL(IS_POW2(N_CHANNELS - 1));
+BUILD_ASSERT_DECL(N_CHANNELS > 1);
+BUILD_ASSERT_DECL(N_CHANNELS <= 32); /* We use a 32-bit word as a mask. */
+
+/* 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;
- /* struct odp_header. */
+ /* struct ovs_header. */
int dp_ifindex;
/* Attributes. */
- const char *name; /* ODP_DP_ATTR_NAME. */
- struct odp_stats stats; /* ODP_DP_ATTR_STATS. */
- enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
- const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */
- uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
+ const char *name; /* OVS_DP_ATTR_NAME. */
+ const uint32_t *upcall_pid; /* OVS_DP_UPCALL_PID. */
+ struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
};
static void dpif_linux_dp_init(struct dpif_linux_dp *);
/* Generic Netlink header. */
uint8_t cmd;
- /* struct odp_header. */
+ /* struct ovs_header. */
unsigned int nlmsg_flags;
int dp_ifindex;
/* Attributes.
*
- * The 'stats' and 'used' members point to 64-bit data that might only be
- * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
- * access their values.
+ * The 'stats' member points to 64-bit data that might only be aligned on
+ * 32-bit boundaries, so get_unaligned_u64() should be used to access its
+ * values.
*
- * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in
+ * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
* the Netlink version of the command, even if actions_len is zero. */
- const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
+ const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
size_t key_len;
- const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */
+ const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
size_t actions_len;
- const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */
- const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */
- const uint64_t *used; /* ODP_FLOW_ATTR_USED. */
- bool clear; /* ODP_FLOW_ATTR_CLEAR. */
+ const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
+ const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
+ const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */
+ bool clear; /* OVS_FLOW_ATTR_CLEAR. */
};
static void dpif_linux_flow_init(struct dpif_linux_flow *);
const struct ofpbuf *);
static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
struct ofpbuf *);
-static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
struct dpif_linux_flow *reply,
struct ofpbuf **bufp);
static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
struct dpif_flow_stats *);
+/* Packet drop monitoring.
+ *
+ * When kernel-to-user Netlink buffers overflow, the kernel notifies us that
+ * one or more packets were dropped, but it doesn't tell us anything about
+ * those packets. However, the administrator really wants to know. So we do
+ * the next best thing, and keep track of the top sources of packets received
+ * on each kernel-to-user channel, since the top sources are those that will
+ * cause the buffers to overflow.
+ *
+ * We use a variation on the "Space-Saving" algorithm in Metwally et al.,
+ * "Efficient Computation of Frequent and Top-k Elements in Data Streams", ACM
+ * Transactions on Database Systems 31:3 (2006). This algorithm yields
+ * perfectly accurate results when the data stream's unique values (in this
+ * case, port numbers) fit into our data structure, and degrades gracefully
+ * even for challenging distributions (e.g. Zipf).
+ *
+ * Our implementation is very simple, without any of the special flourishes
+ * described in the paper. It avoids the need to use a hash for lookup by
+ * keeping the constant factor (N_SKETCHES) very small. The error calculations
+ * in the paper make it sound like the results should still be satisfactory.
+ *
+ * "space-saving" and "Metwally" seem like awkward names for data structures,
+ * so we call this a "sketch" even though technically that's a different sort
+ * of summary structure.
+ */
+
+/* One of N_SKETCHES counting elements per channel in the Metwally
+ * "space-saving" algorithm. */
+enum { N_SKETCHES = 8 }; /* Number of elements per channel. */
+struct dpif_sketch {
+ uint32_t port_no; /* Port number. */
+ unsigned int hits; /* Number of hits. */
+ unsigned int error; /* Upper bound on error in 'hits'. */
+};
+
+/* One of N_CHANNELS channels per dpif between the kernel and userspace. */
+struct dpif_channel {
+ struct nl_sock *sock; /* Netlink socket. */
+ struct dpif_sketch sketches[N_SKETCHES]; /* From max to min 'hits'. */
+ long long int last_poll; /* Last time this channel was polled. */
+};
+
+static void update_sketch(struct dpif_channel *, uint32_t port_no);
+static void scale_sketches(struct dpif *);
+static void report_loss(struct dpif *, struct dpif_channel *);
+
+/* Interval, in milliseconds, at which to scale down the sketch values by a
+ * factor of 2. The Metwally algorithm doesn't do this, which makes sense in
+ * the context it assumes, but in our situation we ought to weight recent data
+ * more heavily than old data, so in my opinion this is reasonable. */
+#define SCALE_INTERVAL (60 * 1000)
+
/* Datapath interface for the openvswitch Linux kernel module. */
struct dpif_linux {
struct dpif dpif;
int dp_ifindex;
- /* Multicast group messages. */
- struct nl_sock *mc_sock;
- uint32_t mcgroups[DPIF_N_UC_TYPES];
- unsigned int listen_mask;
+ /* Upcall messages. */
+ struct dpif_channel channels[N_CHANNELS];
+ uint32_t ready_mask; /* 1-bit for each sock with unread messages. */
+ int epoll_fd; /* epoll fd that includes channel socks. */
+ long long int next_scale; /* Next time to scale down the sketches. */
/* Change notification. */
struct sset changed_ports; /* Ports that have changed. */
- struct rtnetlink_notifier port_notifier;
+ struct nln_notifier *port_notifier;
bool change_error;
+
+ /* Port number allocation. */
+ uint32_t alloc_port_no;
};
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
-/* Generic Netlink family numbers for ODP. */
-static int odp_datapath_family;
-static int odp_vport_family;
-static int odp_flow_family;
-static int odp_packet_family;
+/* Generic Netlink family numbers for OVS. */
+static int ovs_datapath_family;
+static int ovs_vport_family;
+static int ovs_flow_family;
+static int ovs_packet_family;
/* 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 uint32_t dpif_linux_port_get_pid(const struct dpif *, uint32_t port_no);
static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
struct ofpbuf *);
}
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);
{
struct dpif_linux_dp dp_request, dp;
struct ofpbuf *buf;
+ uint32_t upcall_pid;
int error;
error = dpif_linux_init();
/* Create or look up datapath. */
dpif_linux_dp_init(&dp_request);
- dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET;
+ if (create) {
+ dp_request.cmd = OVS_DP_CMD_NEW;
+ upcall_pid = 0;
+ dp_request.upcall_pid = &upcall_pid;
+ } else {
+ dp_request.cmd = OVS_DP_CMD_GET;
+ }
dp_request.name = name;
error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
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 = xzalloc(sizeof *dpif);
+ dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
+ dpif);
+ dpif->epoll_fd = -1;
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->listen_mask = 0;
+ dpif->next_scale = LLONG_MAX;
+
dpif->dp_ifindex = dp->dp_ifindex;
sset_init(&dpif->changed_ports);
- dpif->change_error = false;
*dpifp = &dpif->dpif;
+}
- return 0;
+static void
+destroy_channels(struct dpif_linux *dpif)
+{
+ struct dpif_channel *ch;
-error_free:
- free(dpif);
- return error;
+ if (dpif->epoll_fd >= 0) {
+ close(dpif->epoll_fd);
+ dpif->epoll_fd = -1;
+ }
+ for (ch = dpif->channels; ch < &dpif->channels[N_CHANNELS]; ch++) {
+ nl_sock_destroy(ch->sock);
+ ch->sock = NULL;
+ }
+ dpif->next_scale = LLONG_MAX;
}
static void
dpif_linux_close(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- rtnetlink_link_notifier_unregister(&dpif->port_notifier);
+
+ nln_notifier_destroy(dpif->port_notifier);
+ destroy_channels(dpif);
sset_destroy(&dpif->changed_ports);
free(dpif);
}
struct dpif_linux_dp dp;
dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_CMD_DEL;
+ dp.cmd = OVS_DP_CMD_DEL;
dp.dp_ifindex = dpif->dp_ifindex;
return dpif_linux_dp_transact(&dp, NULL, NULL);
}
-static int
-dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
+static void
+dpif_linux_run(struct dpif *dpif_)
{
- struct dpif_linux_dp dp;
- struct ofpbuf *buf;
- int error;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- error = dpif_linux_dp_get(dpif_, &dp, &buf);
- if (!error) {
- *stats = dp.stats;
- ofpbuf_delete(buf);
+ if (time_msec() >= dpif->next_scale) {
+ dpif->next_scale = time_msec() + SCALE_INTERVAL;
+ scale_sketches(dpif_);
+ }
+
+ if (nln) {
+ nln_run(nln);
+ }
+}
+
+static void
+dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
+{
+ if (nln) {
+ nln_wait(nln);
}
- return error;
}
static int
-dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
+dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
{
struct dpif_linux_dp dp;
struct ofpbuf *buf;
error = dpif_linux_dp_get(dpif_, &dp, &buf);
if (!error) {
- *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP;
+ stats->n_hit = dp.stats.n_hit;
+ stats->n_missed = dp.stats.n_missed;
+ stats->n_lost = dp.stats.n_lost;
+ stats->n_flows = dp.stats.n_flows;
ofpbuf_delete(buf);
}
return error;
}
-static int
-dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_dp dp;
-
- dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_CMD_SET;
- dp.dp_ifindex = dpif->dp_ifindex;
- dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO;
- return dpif_linux_dp_transact(&dp, NULL, NULL);
-}
-
static int
dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
- uint16_t *port_nop)
+ uint32_t *port_nop)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
const char *name = netdev_get_name(netdev);
struct dpif_linux_vport request, reply;
const struct ofpbuf *options;
struct ofpbuf *buf;
- int error;
+ int error, i = 0, max_ports = MAX_PORTS;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_CMD_NEW;
+ request.cmd = OVS_VPORT_CMD_NEW;
request.dp_ifindex = dpif->dp_ifindex;
request.type = netdev_vport_get_vport_type(netdev);
- if (request.type == ODP_VPORT_TYPE_UNSPEC) {
+ if (request.type == OVS_VPORT_TYPE_UNSPEC) {
VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
"unsupported type `%s'",
dpif_name(dpif_), name, type);
request.options_len = options->size;
}
- error = dpif_linux_vport_transact(&request, &reply, &buf);
- if (!error) {
- *port_nop = reply.port_no;
- ofpbuf_delete(buf);
+ if (request.type == OVS_VPORT_TYPE_NETDEV) {
+ netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
}
+ /* Unless a specific port was requested, loop until we find a port
+ * that isn't used. */
+ do {
+ uint32_t upcall_pid;
+
+ request.port_no = *port_nop != UINT32_MAX ? *port_nop
+ : ++dpif->alloc_port_no;
+ upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no);
+ request.upcall_pid = &upcall_pid;
+ error = dpif_linux_vport_transact(&request, &reply, &buf);
+
+ if (!error) {
+ *port_nop = reply.port_no;
+ VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
+ dpif_name(dpif_), request.port_no, upcall_pid);
+ } else if (error == EFBIG) {
+ /* Older datapath has lower limit. */
+ max_ports = dpif->alloc_port_no;
+ dpif->alloc_port_no = 0;
+ } else if (error == EBUSY && *port_nop != UINT32_MAX) {
+ VLOG_INFO("%s: requested port %"PRIu32" is in use",
+ dpif_name(dpif_), *port_nop);
+ }
+
+ ofpbuf_delete(buf);
+ } while ((*port_nop == UINT32_MAX) && (i++ < max_ports)
+ && (error == EBUSY || error == EFBIG));
+
return error;
}
static int
-dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
+dpif_linux_port_del(struct dpif *dpif_, uint32_t port_no)
{
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.cmd = OVS_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);
+
+ return error;
}
static int
int error;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_CMD_GET;
+ request.cmd = OVS_VPORT_CMD_GET;
request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
request.port_no = port_no;
request.name = port_name;
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.dp_ifindex != request.dp_ifindex) {
+ /* A query by name reported that 'port_name' is in some datapath
+ * other than 'dpif', but the caller wants to know about 'dpif'. */
+ error = ENODEV;
+ } else if (dpif_port) {
+ dpif_port->name = xstrdup(reply.name);
+ dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
+ dpif_port->port_no = reply.port_no;
+ }
ofpbuf_delete(buf);
}
return error;
}
static int
-dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
+dpif_linux_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
struct dpif_port *dpif_port)
{
return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
static int
dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
{
- /* If the datapath increases its range of supported ports, then it should
- * start reporting that. */
- return 1024;
+ return MAX_PORTS;
+}
+
+static uint32_t
+dpif_linux_port_get_pid(const struct dpif *dpif_, uint32_t port_no)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ if (dpif->epoll_fd < 0) {
+ return 0;
+ } else {
+ int idx;
+
+ idx = (port_no != UINT32_MAX
+ ? 1 + (port_no & (N_CHANNELS - 2))
+ : 0);
+ return nl_sock_pid(dpif->channels[idx].sock);
+ }
}
static int
struct dpif_linux_flow flow;
dpif_linux_flow_init(&flow);
- flow.cmd = ODP_FLOW_CMD_DEL;
+ flow.cmd = OVS_FLOW_CMD_DEL;
flow.dp_ifindex = dpif->dp_ifindex;
return dpif_linux_flow_transact(&flow, NULL, NULL);
}
*statep = state = xmalloc(sizeof *state);
dpif_linux_vport_init(&request);
- request.cmd = ODP_DP_CMD_GET;
+ request.cmd = OVS_DP_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
buf = ofpbuf_new(1024);
return error;
}
- dpif_port->name = (char *) vport.name;
- dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
+ dpif_port->name = CONST_CAST(char *, vport.name);
+ dpif_port->type = CONST_CAST(char *, netdev_vport_get_netdev_type(&vport));
dpif_port->port_no = vport.port_no;
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_ OVS_UNUSED, void *state_)
{
struct dpif_linux_port_state *state = state_;
int error = nl_dump_done(&state->dump);
+
free(state);
return error;
}
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
poll_immediate_wake();
- } else {
- rtnetlink_link_notifier_wait();
}
}
struct dpif_linux_flow request;
dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_CMD_GET;
+ request.cmd = OVS_FLOW_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
request.key = key;
request.key_len = key_len;
dpif_linux_flow_get_stats(&reply, stats);
}
if (actionsp) {
- buf->data = (void *) reply.actions;
+ buf->data = CONST_CAST(struct nlattr *, reply.actions);
buf->size = reply.actions_len;
*actionsp = buf;
} else {
return error;
}
-static int
-dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
- const struct nlattr *key, size_t key_len,
- const struct nlattr *actions, size_t actions_len,
- struct dpif_flow_stats *stats)
+static void
+dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
+ struct dpif_linux_flow *request)
{
+ static struct nlattr dummy_action;
+
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ dpif_linux_flow_init(request);
+ request->cmd = (put->flags & DPIF_FP_CREATE
+ ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
+ request->dp_ifindex = dpif->dp_ifindex;
+ request->key = put->key;
+ request->key_len = put->key_len;
+ /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
+ request->actions = put->actions ? put->actions : &dummy_action;
+ request->actions_len = put->actions_len;
+ if (put->flags & DPIF_FP_ZERO_STATS) {
+ request->clear = true;
+ }
+ request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
+}
+
+static int
+dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
+{
struct dpif_linux_flow request, reply;
- struct nlattr dummy_action;
struct ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
- request.dp_ifindex = dpif->dp_ifindex;
- request.key = key;
- request.key_len = key_len;
- /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */
- request.actions = actions ? actions : &dummy_action;
- request.actions_len = actions_len;
- if (flags & DPIF_FP_ZERO_STATS) {
- request.clear = true;
- }
- request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
+ dpif_linux_init_flow_put(dpif_, put, &request);
error = dpif_linux_flow_transact(&request,
- stats ? &reply : NULL,
- stats ? &buf : NULL);
- if (!error && stats) {
- dpif_linux_flow_get_stats(&reply, stats);
+ put->stats ? &reply : NULL,
+ put->stats ? &buf : NULL);
+ if (!error && put->stats) {
+ dpif_linux_flow_get_stats(&reply, put->stats);
ofpbuf_delete(buf);
}
return error;
}
-static int
-dpif_linux_flow_del(struct dpif *dpif_,
- const struct nlattr *key, size_t key_len,
- struct dpif_flow_stats *stats)
+static void
+dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
+ struct dpif_linux_flow *request)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ dpif_linux_flow_init(request);
+ request->cmd = OVS_FLOW_CMD_DEL;
+ request->dp_ifindex = dpif->dp_ifindex;
+ request->key = del->key;
+ request->key_len = del->key_len;
+}
+
+static int
+dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
+{
struct dpif_linux_flow request, reply;
struct ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_CMD_DEL;
- request.dp_ifindex = dpif->dp_ifindex;
- request.key = key;
- request.key_len = key_len;
+ dpif_linux_init_flow_del(dpif_, del, &request);
error = dpif_linux_flow_transact(&request,
- stats ? &reply : NULL,
- stats ? &buf : NULL);
- if (!error && stats) {
- dpif_linux_flow_get_stats(&reply, stats);
+ del->stats ? &reply : NULL,
+ del->stats ? &buf : NULL);
+ if (!error && del->stats) {
+ dpif_linux_flow_get_stats(&reply, del->stats);
ofpbuf_delete(buf);
}
return error;
*statep = state = xmalloc(sizeof *state);
dpif_linux_flow_init(&request);
- request.cmd = ODP_DP_CMD_GET;
+ request.cmd = OVS_DP_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
buf = ofpbuf_new(1024);
return error;
}
-static int
-dpif_linux_execute(struct dpif *dpif_,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *packet)
+static void
+dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
+ struct ofpbuf *buf)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct odp_header *execute;
- struct ofpbuf *buf;
- int error;
+ struct ovs_header *k_exec;
- buf = ofpbuf_new(128 + actions_len + packet->size);
+ ofpbuf_prealloc_tailroom(buf, (64
+ + d_exec->packet->size
+ + d_exec->key_len
+ + d_exec->actions_len));
- nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
- ODP_PACKET_CMD_EXECUTE, 1);
+ nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
+ OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
- execute = ofpbuf_put_uninit(buf, sizeof *execute);
- execute->dp_ifindex = dpif->dp_ifindex;
+ k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
+ k_exec->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_ACTIONS, actions, actions_len);
+ nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
+ d_exec->packet->data, d_exec->packet->size);
+ nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
+ nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
+ d_exec->actions, d_exec->actions_len);
+}
+
+static int
+dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
+{
+ uint64_t request_stub[1024 / 8];
+ struct ofpbuf request;
+ int error;
+
+ ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
+ dpif_linux_encode_execute(dp_ifindex, execute, &request);
+ error = nl_sock_transact(genl_sock, &request, NULL);
+ ofpbuf_uninit(&request);
- error = nl_sock_transact(genl_sock, buf, NULL);
- ofpbuf_delete(buf);
return error;
}
static int
-dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
+dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- *listen_mask = dpif->listen_mask;
- return 0;
+
+ return dpif_linux_execute__(dpif->dp_ifindex, execute);
}
-static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+#define MAX_OPS 50
+
+static void
+dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
{
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);
- if (error) {
- return error;
+ struct op_auxdata {
+ struct nl_transaction txn;
+
+ struct ofpbuf request;
+ uint64_t request_stub[1024 / 8];
+
+ struct ofpbuf reply;
+ uint64_t reply_stub[1024 / 8];
+ } auxes[MAX_OPS];
+
+ struct nl_transaction *txnsp[MAX_OPS];
+ size_t i;
+
+ assert(n_ops <= MAX_OPS);
+ for (i = 0; i < n_ops; i++) {
+ struct op_auxdata *aux = &auxes[i];
+ struct dpif_op *op = ops[i];
+ struct dpif_flow_put *put;
+ struct dpif_flow_del *del;
+ struct dpif_execute *execute;
+ struct dpif_linux_flow flow;
+
+ ofpbuf_use_stub(&aux->request,
+ aux->request_stub, sizeof aux->request_stub);
+ aux->txn.request = &aux->request;
+
+ ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
+ aux->txn.reply = NULL;
+
+ switch (op->type) {
+ case DPIF_OP_FLOW_PUT:
+ put = &op->u.flow_put;
+ dpif_linux_init_flow_put(dpif_, put, &flow);
+ if (put->stats) {
+ flow.nlmsg_flags |= NLM_F_ECHO;
+ aux->txn.reply = &aux->reply;
+ }
+ dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
+ break;
+
+ case DPIF_OP_FLOW_DEL:
+ del = &op->u.flow_del;
+ dpif_linux_init_flow_del(dpif_, del, &flow);
+ if (del->stats) {
+ flow.nlmsg_flags |= NLM_F_ECHO;
+ aux->txn.reply = &aux->reply;
+ }
+ dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
+ break;
+
+ case DPIF_OP_EXECUTE:
+ execute = &op->u.execute;
+ dpif_linux_encode_execute(dpif->dp_ifindex, execute,
+ &aux->request);
+ break;
+
+ default:
+ NOT_REACHED();
}
}
- /* 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]);
- }
+ for (i = 0; i < n_ops; i++) {
+ txnsp[i] = &auxes[i].txn;
}
+ nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
- /* Update listen_mask. */
- dpif->listen_mask = listen_mask;
+ for (i = 0; i < n_ops; i++) {
+ struct op_auxdata *aux = &auxes[i];
+ struct nl_transaction *txn = &auxes[i].txn;
+ struct dpif_op *op = ops[i];
+ struct dpif_flow_put *put;
+ struct dpif_flow_del *del;
- /* Subscribe to new groups. */
- error = 0;
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- if (dpif->listen_mask & (1u << i)) {
- int retval;
+ op->error = txn->error;
- retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
- if (retval) {
- error = retval;
+ switch (op->type) {
+ case DPIF_OP_FLOW_PUT:
+ put = &op->u.flow_put;
+ if (put->stats) {
+ if (!op->error) {
+ struct dpif_linux_flow reply;
+
+ op->error = dpif_linux_flow_from_ofpbuf(&reply,
+ txn->reply);
+ if (!op->error) {
+ dpif_linux_flow_get_stats(&reply, put->stats);
+ }
+ }
+
+ if (op->error) {
+ memset(put->stats, 0, sizeof *put->stats);
+ }
+ }
+ break;
+
+ case DPIF_OP_FLOW_DEL:
+ del = &op->u.flow_del;
+ if (del->stats) {
+ if (!op->error) {
+ struct dpif_linux_flow reply;
+
+ op->error = dpif_linux_flow_from_ofpbuf(&reply,
+ txn->reply);
+ if (!op->error) {
+ dpif_linux_flow_get_stats(&reply, del->stats);
+ }
+ }
+
+ if (op->error) {
+ memset(del->stats, 0, sizeof *del->stats);
+ }
}
+ break;
+
+ case DPIF_OP_EXECUTE:
+ break;
+
+ default:
+ NOT_REACHED();
}
+
+ ofpbuf_uninit(&aux->request);
+ ofpbuf_uninit(&aux->reply);
}
- return error;
}
-static int
-dpif_linux_get_sflow_probability(const struct dpif *dpif_,
- uint32_t *probability)
+static void
+dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
{
- struct dpif_linux_dp dp;
- struct ofpbuf *buf;
+ while (n_ops > 0) {
+ size_t chunk = MIN(n_ops, MAX_OPS);
+ dpif_linux_operate__(dpif, ops, chunk);
+ ops += chunk;
+ n_ops -= chunk;
+ }
+}
+
+static void
+set_upcall_pids(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_port_dump port_dump;
+ struct dpif_port port;
int error;
- error = dpif_linux_dp_get(dpif_, &dp, &buf);
- if (!error) {
- *probability = dp.sampling ? *dp.sampling : 0;
- ofpbuf_delete(buf);
+ DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
+ uint32_t upcall_pid = dpif_linux_port_get_pid(dpif_, port.port_no);
+ 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 = &upcall_pid;
+ error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
+ if (!error) {
+ VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
+ dpif_name(&dpif->dpif), vport_request.port_no,
+ upcall_pid);
+ } else {
+ VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
+ dpif_name(&dpif->dpif), strerror(error));
+ }
}
- return error;
}
static int
-dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
+dpif_linux_recv_set(struct dpif *dpif_, bool enable)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_dp dp;
- dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_CMD_SET;
- dp.dp_ifindex = dpif->dp_ifindex;
- dp.sampling = &probability;
- return dpif_linux_dp_transact(&dp, NULL, NULL);
+ if ((dpif->epoll_fd >= 0) == enable) {
+ return 0;
+ }
+
+ if (!enable) {
+ destroy_channels(dpif);
+ } else {
+ struct dpif_channel *ch;
+ int error;
+
+ dpif->epoll_fd = epoll_create(N_CHANNELS);
+ if (dpif->epoll_fd < 0) {
+ return errno;
+ }
+
+ for (ch = dpif->channels; ch < &dpif->channels[N_CHANNELS]; ch++) {
+ int indx = ch - dpif->channels;
+ struct epoll_event event;
+
+ error = nl_sock_create(NETLINK_GENERIC, &ch->sock);
+ if (error) {
+ destroy_channels(dpif);
+ return error;
+ }
+
+ memset(&event, 0, sizeof event);
+ event.events = EPOLLIN;
+ event.data.u32 = indx;
+ if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(ch->sock),
+ &event) < 0) {
+ error = errno;
+ destroy_channels(dpif);
+ return error;
+ }
+
+ memset(ch->sketches, 0, sizeof ch->sketches);
+ ch->last_poll = LLONG_MIN;
+ }
+
+ dpif->ready_mask = 0;
+ dpif->next_scale = time_msec() + SCALE_INTERVAL;
+ }
+
+ set_upcall_pids(dpif_);
+
+ return 0;
}
static int
parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
int *dp_ifindex)
{
- static const struct nl_policy odp_packet_policy[] = {
+ static const struct nl_policy ovs_packet_policy[] = {
/* Always present. */
- [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
+ [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
.min_len = ETH_HEADER_LEN },
- [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
-
- /* ODP_PACKET_CMD_ACTION only. */
- [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
+ [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
- /* ODP_PACKET_CMD_SAMPLE only. */
- [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
- [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
+ /* OVS_PACKET_CMD_ACTION only. */
+ [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
};
- struct odp_header *odp_header;
- struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
+ struct ovs_header *ovs_header;
+ struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
struct nlmsghdr *nlmsg;
struct genlmsghdr *genl;
struct ofpbuf b;
nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
genl = ofpbuf_try_pull(&b, sizeof *genl);
- odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
- if (!nlmsg || !genl || !odp_header
- || nlmsg->nlmsg_type != odp_packet_family
- || !nl_policy_parse(&b, 0, odp_packet_policy, a,
- ARRAY_SIZE(odp_packet_policy))) {
+ ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
+ if (!nlmsg || !genl || !ovs_header
+ || nlmsg->nlmsg_type != ovs_packet_family
+ || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
+ ARRAY_SIZE(ovs_packet_policy))) {
return EINVAL;
}
- type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
- : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
- : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
+ type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
+ : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
: -1);
if (type < 0) {
return EINVAL;
memset(upcall, 0, sizeof *upcall);
upcall->type = type;
upcall->packet = buf;
- upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
- upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
- upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]);
- upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]);
- upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA]
- ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA])
+ 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]
+ ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
: 0);
- upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL]
- ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL])
- : 0);
- if (a[ODP_PACKET_ATTR_ACTIONS]) {
- upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]);
- upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
- }
-
- *dp_ifindex = odp_header->dp_ifindex;
+ *dp_ifindex = ovs_header->dp_ifindex;
return 0;
}
static int
-dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
+dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
+ struct ofpbuf *buf)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct ofpbuf *buf;
- int error;
- int i;
+ int read_tries = 0;
- if (!dpif->mc_sock) {
- return EAGAIN;
+ if (dpif->epoll_fd < 0) {
+ return EAGAIN;
}
- for (i = 0; i < 50; i++) {
- int dp_ifindex;
+ if (!dpif->ready_mask) {
+ struct epoll_event events[N_CHANNELS];
+ int retval;
+ int i;
- error = nl_sock_recv(dpif->mc_sock, &buf, false);
- if (error) {
- return error;
+ do {
+ retval = epoll_wait(dpif->epoll_fd, events, N_CHANNELS, 0);
+ } while (retval < 0 && errno == EINTR);
+ if (retval < 0) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
}
- error = parse_odp_packet(buf, upcall, &dp_ifindex);
- if (!error
- && dp_ifindex == dpif->dp_ifindex
- && dpif->listen_mask & (1u << upcall->type)) {
- return 0;
+ for (i = 0; i < retval; i++) {
+ dpif->ready_mask |= 1u << events[i].data.u32;
}
+ }
- ofpbuf_delete(buf);
- if (error) {
- return error;
+ while (dpif->ready_mask) {
+ int indx = ffs(dpif->ready_mask) - 1;
+ struct dpif_channel *ch = &dpif->channels[indx];
+
+ dpif->ready_mask &= ~(1u << indx);
+
+ for (;;) {
+ int dp_ifindex;
+ int error;
+
+ if (++read_tries > 50) {
+ return EAGAIN;
+ }
+
+ error = nl_sock_recv(ch->sock, buf, false);
+ if (error == ENOBUFS) {
+ /* ENOBUFS typically means that we've received so many
+ * packets that the buffer overflowed. Try again
+ * immediately because there's almost certainly a packet
+ * waiting for us. */
+ report_loss(dpif_, ch);
+ continue;
+ }
+
+ ch->last_poll = time_msec();
+ if (error) {
+ if (error == EAGAIN) {
+ break;
+ }
+ return error;
+ }
+
+ error = parse_odp_packet(buf, upcall, &dp_ifindex);
+ if (!error && dp_ifindex == dpif->dp_ifindex) {
+ const struct nlattr *in_port;
+
+ in_port = nl_attr_find__(upcall->key, upcall->key_len,
+ OVS_KEY_ATTR_IN_PORT);
+ if (in_port) {
+ update_sketch(ch, nl_attr_get_u32(in_port));
+ }
+ return 0;
+ }
+ 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->epoll_fd < 0) {
+ return;
}
+
+ poll_fd_wait(dpif->epoll_fd, POLLIN);
}
static void
dpif_linux_recv_purge(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_channel *ch;
+
+ if (dpif->epoll_fd < 0) {
+ return;
+ }
- if (dpif->mc_sock) {
- nl_sock_drain(dpif->mc_sock);
+ for (ch = dpif->channels; ch < &dpif->channels[N_CHANNELS]; ch++) {
+ nl_sock_drain(ch->sock);
}
}
const struct dpif_class dpif_linux_class = {
"system",
- NULL, /* run */
- NULL, /* wait */
dpif_linux_enumerate,
dpif_linux_open,
dpif_linux_close,
dpif_linux_destroy,
+ dpif_linux_run,
+ dpif_linux_wait,
dpif_linux_get_stats,
- dpif_linux_get_drop_frags,
- dpif_linux_set_drop_frags,
dpif_linux_port_add,
dpif_linux_port_del,
dpif_linux_port_query_by_number,
dpif_linux_port_query_by_name,
dpif_linux_get_max_ports,
+ dpif_linux_port_get_pid,
dpif_linux_port_dump_start,
dpif_linux_port_dump_next,
dpif_linux_port_dump_done,
dpif_linux_flow_dump_next,
dpif_linux_flow_dump_done,
dpif_linux_execute,
- dpif_linux_recv_get_mask,
- dpif_linux_recv_set_mask,
- dpif_linux_get_sflow_probability,
- dpif_linux_set_sflow_probability,
+ dpif_linux_operate,
+ dpif_linux_recv_set,
dpif_linux_queue_to_priority,
dpif_linux_recv,
dpif_linux_recv_wait,
static int error = -1;
if (error < 0) {
- error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
- &odp_datapath_family);
+ unsigned int ovs_vport_mcgroup;
+
+ 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.",
- ODP_DATAPATH_FAMILY);
+ OVS_DATAPATH_FAMILY);
}
if (!error) {
- error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family);
+ error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
}
if (!error) {
- error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family);
+ error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
}
if (!error) {
- error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
- &odp_packet_family);
+ error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
+ &ovs_packet_family);
}
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;
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;
+ return reply.type == OVS_VPORT_TYPE_INTERNAL;
+}
+
+int
+dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
+ const void *data, size_t size)
+{
+ struct ofpbuf actions, key, packet;
+ struct odputil_keybuf keybuf;
+ struct dpif_execute execute;
+ struct flow flow;
+ uint64_t action;
+
+ ofpbuf_use_const(&packet, data, size);
+ flow_extract(&packet, 0, NULL, 0, &flow);
+
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+ odp_flow_key_from_flow(&key, &flow, OVSP_NONE);
+
+ ofpbuf_use_stack(&actions, &action, sizeof action);
+ nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
+
+ execute.key = key.data;
+ execute.key_len = key.size;
+ execute.actions = actions.data;
+ execute.actions_len = actions.size;
+ execute.packet = &packet;
+ return dpif_linux_execute__(dp_ifindex, &execute);
+}
+
+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;
}
}
\f
-/* Parses the contents of 'buf', which contains a "struct odp_header" followed
+/* Parses the contents of 'buf', which contains a "struct ovs_header" followed
* by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
* positive errno value.
*
dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
const struct ofpbuf *buf)
{
- static const struct nl_policy odp_vport_policy[] = {
- [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
- [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
- [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
- [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct rtnl_link_stats64),
- .max_len = sizeof(struct rtnl_link_stats64),
+ static const struct nl_policy ovs_vport_policy[] = {
+ [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] = { NL_POLICY_FOR(struct ovs_vport_stats),
.optional = true },
- [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
+ [OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
.min_len = ETH_ADDR_LEN,
.max_len = ETH_ADDR_LEN,
.optional = true },
- [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
- [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
- [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
- [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
+ [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
};
- struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
- struct odp_header *odp_header;
+ struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
+ struct ovs_header *ovs_header;
struct nlmsghdr *nlmsg;
struct genlmsghdr *genl;
struct ofpbuf b;
ofpbuf_use_const(&b, buf->data, buf->size);
nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
genl = ofpbuf_try_pull(&b, sizeof *genl);
- odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
- if (!nlmsg || !genl || !odp_header
- || nlmsg->nlmsg_type != odp_vport_family
- || !nl_policy_parse(&b, 0, odp_vport_policy, a,
- ARRAY_SIZE(odp_vport_policy))) {
+ ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
+ if (!nlmsg || !genl || !ovs_header
+ || nlmsg->nlmsg_type != ovs_vport_family
+ || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
+ ARRAY_SIZE(ovs_vport_policy))) {
return EINVAL;
}
vport->cmd = genl->cmd;
- vport->dp_ifindex = odp_header->dp_ifindex;
- vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
- vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
- vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
- if (a[ODP_VPORT_ATTR_STATS]) {
- vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]);
- }
- if (a[ODP_VPORT_ATTR_ADDRESS]) {
- vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]);
- }
- if (a[ODP_VPORT_ATTR_MTU]) {
- vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
- } else {
- vport->mtu = INT_MAX;
+ vport->dp_ifindex = ovs_header->dp_ifindex;
+ 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(a[OVS_VPORT_ATTR_UPCALL_PID]);
}
- if (a[ODP_VPORT_ATTR_OPTIONS]) {
- vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
- vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]);
+ if (a[OVS_VPORT_ATTR_STATS]) {
+ vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
}
- if (a[ODP_VPORT_ATTR_IFINDEX]) {
- vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]);
+ if (a[OVS_VPORT_ATTR_ADDRESS]) {
+ vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]);
}
- if (a[ODP_VPORT_ATTR_IFLINK]) {
- vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]);
+ 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]);
}
return 0;
}
-/* Appends to 'buf' (which must initially be empty) a "struct odp_header"
+/* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
* followed by Netlink attributes corresponding to 'vport'. */
static void
dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
struct ofpbuf *buf)
{
- struct odp_header *odp_header;
+ struct ovs_header *ovs_header;
- nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
- vport->cmd, 1);
+ nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
+ vport->cmd, OVS_VPORT_VERSION);
- odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
- odp_header->dp_ifindex = vport->dp_ifindex;
+ ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
+ ovs_header->dp_ifindex = vport->dp_ifindex;
if (vport->port_no != UINT32_MAX) {
- nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
+ nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
}
- if (vport->type != ODP_VPORT_TYPE_UNSPEC) {
- nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type);
+ if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
+ nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
}
if (vport->name) {
- nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name);
+ nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
+ }
+
+ if (vport->upcall_pid) {
+ nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
}
if (vport->stats) {
- nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS,
+ nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
vport->stats, sizeof *vport->stats);
}
if (vport->address) {
- nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS,
+ nl_msg_put_unspec(buf, OVS_VPORT_ATTR_ADDRESS,
vport->address, ETH_ADDR_LEN);
}
- if (vport->mtu && vport->mtu != INT_MAX) {
- nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
- }
-
if (vport->options) {
- nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS,
+ nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
vport->options, vport->options_len);
}
-
- if (vport->ifindex) {
- nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex);
- }
-
- if (vport->iflink) {
- nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink);
- }
}
/* Clears 'vport' to "empty" values. */
/* Executes 'request' in the kernel datapath. If the command fails, returns a
* positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
* without doing anything else. If 'reply' and 'bufp' are nonnull, then the
- * result of the command is expected to be an odp_vport also, which is decoded
+ * result of the command is expected to be an ovs_vport 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
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);
struct dpif_linux_vport request;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_CMD_GET;
+ request.cmd = OVS_VPORT_CMD_GET;
request.name = name;
return dpif_linux_vport_transact(&request, reply, bufp);
}
\f
-/* Parses the contents of 'buf', which contains a "struct odp_header" followed
+/* Parses the contents of 'buf', which contains a "struct ovs_header" followed
* by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
* positive errno value.
*
static int
dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
{
- static const struct nl_policy odp_datapath_policy[] = {
- [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
- [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct odp_stats),
- .max_len = sizeof(struct odp_stats),
+ static const struct nl_policy ovs_datapath_policy[] = {
+ [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
+ [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
.optional = true },
- [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
- [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
- [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
};
- struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
- struct odp_header *odp_header;
+ struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
+ struct ovs_header *ovs_header;
struct nlmsghdr *nlmsg;
struct genlmsghdr *genl;
struct ofpbuf b;
ofpbuf_use_const(&b, buf->data, buf->size);
nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
genl = ofpbuf_try_pull(&b, sizeof *genl);
- odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
- if (!nlmsg || !genl || !odp_header
- || nlmsg->nlmsg_type != odp_datapath_family
- || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
- ARRAY_SIZE(odp_datapath_policy))) {
+ ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
+ if (!nlmsg || !genl || !ovs_header
+ || nlmsg->nlmsg_type != ovs_datapath_family
+ || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
+ ARRAY_SIZE(ovs_datapath_policy))) {
return EINVAL;
}
dp->cmd = genl->cmd;
- dp->dp_ifindex = odp_header->dp_ifindex;
- dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
- if (a[ODP_DP_ATTR_STATS]) {
+ dp->dp_ifindex = ovs_header->dp_ifindex;
+ dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
+ if (a[OVS_DP_ATTR_STATS]) {
/* Can't use structure assignment because Netlink doesn't ensure
* sufficient alignment for 64-bit members. */
- memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]),
+ memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
sizeof dp->stats);
}
- if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
- dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
- }
- if (a[ODP_DP_ATTR_SAMPLING]) {
- dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
- }
-
- if (a[ODP_DP_ATTR_MCGROUPS]) {
- static const struct nl_policy odp_mcgroup_policy[] = {
- [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
- [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
- [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
- };
-
- struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
-
- if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
- mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
- return EINVAL;
- }
-
- if (mcgroups[ODP_PACKET_CMD_MISS]) {
- dp->mcgroups[DPIF_UC_MISS]
- = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
- }
- if (mcgroups[ODP_PACKET_CMD_ACTION]) {
- dp->mcgroups[DPIF_UC_ACTION]
- = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
- }
- if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
- dp->mcgroups[DPIF_UC_SAMPLE]
- = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
- }
- }
return 0;
}
static void
dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
{
- struct odp_header *odp_header;
+ struct ovs_header *ovs_header;
- nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
- NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
+ nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
+ NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
+ OVS_DATAPATH_VERSION);
- odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
- odp_header->dp_ifindex = dp->dp_ifindex;
+ ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
+ ovs_header->dp_ifindex = dp->dp_ifindex;
if (dp->name) {
- nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
+ nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
}
- /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
-
- if (dp->ipv4_frags) {
- nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
+ if (dp->upcall_pid) {
+ nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
}
- if (dp->sampling) {
- nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
- }
+ /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
}
/* Clears 'dp' to "empty" values. */
-void
+static void
dpif_linux_dp_init(struct dpif_linux_dp *dp)
{
memset(dp, 0, sizeof *dp);
struct ofpbuf *buf;
dpif_linux_dp_init(&request);
- request.cmd = ODP_DP_CMD_GET;
+ request.cmd = OVS_DP_CMD_GET;
buf = ofpbuf_new(1024);
dpif_linux_dp_to_ofpbuf(&request, buf);
* 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)
{
struct dpif_linux_dp request;
dpif_linux_dp_init(&request);
- request.cmd = ODP_DP_CMD_GET;
+ request.cmd = OVS_DP_CMD_GET;
request.dp_ifindex = dpif->dp_ifindex;
return dpif_linux_dp_transact(&request, reply, bufp);
}
\f
-/* Parses the contents of 'buf', which contains a "struct odp_header" followed
+/* Parses the contents of 'buf', which contains a "struct ovs_header" followed
* by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
* positive errno value.
*
dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
const struct ofpbuf *buf)
{
- static const struct nl_policy odp_flow_policy[] = {
- [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
- [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
- [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct odp_flow_stats),
- .max_len = sizeof(struct odp_flow_stats),
+ 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_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
.optional = true },
- [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
- [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
- /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
+ [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
+ [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
+ /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
};
- struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
- struct odp_header *odp_header;
+ struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
+ struct ovs_header *ovs_header;
struct nlmsghdr *nlmsg;
struct genlmsghdr *genl;
struct ofpbuf b;
ofpbuf_use_const(&b, buf->data, buf->size);
nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
genl = ofpbuf_try_pull(&b, sizeof *genl);
- odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
- if (!nlmsg || !genl || !odp_header
- || nlmsg->nlmsg_type != odp_flow_family
- || !nl_policy_parse(&b, 0, odp_flow_policy, a,
- ARRAY_SIZE(odp_flow_policy))) {
+ ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
+ if (!nlmsg || !genl || !ovs_header
+ || nlmsg->nlmsg_type != ovs_flow_family
+ || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
+ ARRAY_SIZE(ovs_flow_policy))) {
return EINVAL;
}
flow->nlmsg_flags = nlmsg->nlmsg_flags;
- flow->dp_ifindex = odp_header->dp_ifindex;
- flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
- flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
- if (a[ODP_FLOW_ATTR_ACTIONS]) {
- flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]);
- flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]);
+ flow->dp_ifindex = ovs_header->dp_ifindex;
+ flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
+ flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
+ if (a[OVS_FLOW_ATTR_ACTIONS]) {
+ flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
+ flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
}
- if (a[ODP_FLOW_ATTR_STATS]) {
- flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
+ if (a[OVS_FLOW_ATTR_STATS]) {
+ flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
}
- if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
- flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
+ if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
+ flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
}
- if (a[ODP_FLOW_ATTR_USED]) {
- flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]);
+ if (a[OVS_FLOW_ATTR_USED]) {
+ flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
}
return 0;
}
-/* Appends to 'buf' (which must initially be empty) a "struct odp_header"
+/* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
* followed by Netlink attributes corresponding to 'flow'. */
static void
dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
struct ofpbuf *buf)
{
- struct odp_header *odp_header;
+ struct ovs_header *ovs_header;
- nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
- NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
- flow->cmd, 1);
+ nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
+ NLM_F_REQUEST | flow->nlmsg_flags,
+ flow->cmd, OVS_FLOW_VERSION);
- odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
- odp_header->dp_ifindex = flow->dp_ifindex;
+ ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
+ ovs_header->dp_ifindex = flow->dp_ifindex;
if (flow->key_len) {
- nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
+ nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
}
if (flow->actions || flow->actions_len) {
- nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
+ nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
flow->actions, flow->actions_len);
}
assert(!flow->used);
if (flow->clear) {
- nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
+ nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
}
}
/* 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
-dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+static int
+dpif_linux_flow_transact(struct dpif_linux_flow *request,
struct dpif_linux_flow *reply, struct ofpbuf **bufp)
{
struct ofpbuf *request_buf;
assert((reply != NULL) == (bufp != NULL));
+ if (reply) {
+ request->nlmsg_flags |= NLM_F_ECHO;
+ }
+
request_buf = ofpbuf_new(1024);
dpif_linux_flow_to_ofpbuf(request, request_buf);
error = nl_sock_transact(genl_sock, request_buf, bufp);
stats->n_packets = 0;
stats->n_bytes = 0;
}
- stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
+ stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
}
+\f
+/* Metwally "space-saving" algorithm implementation. */
+
+/* Updates 'ch' to record that a packet was received on 'port_no'. */
+static void
+update_sketch(struct dpif_channel *ch, uint32_t port_no)
+{
+ struct dpif_sketch *sk;
+
+ /* Find an existing counting element for 'port_no' or, if none, replace the
+ * counting element with the fewest hits by 'port_no'. */
+ for (sk = ch->sketches; ; sk++) {
+ if (port_no == sk->port_no) {
+ break;
+ } else if (sk == &ch->sketches[N_SKETCHES - 1]) {
+ sk->port_no = port_no;
+ sk->error = sk->hits;
+ break;
+ }
+ }
+
+ /* Increment the hit count, then re-sort the counting elements (usually
+ * nothing needs to be done). */
+ sk->hits++;
+ while (sk > ch->sketches && sk[-1].hits > sk->hits) {
+ struct dpif_sketch tmp = sk[-1];
+ sk[-1] = *sk;
+ *sk = tmp;
+ sk--;
+ }
+}
+
+/* Divide the counts of all the the counting elements in 'dpif' by 2. See the
+ * comment on SCALE_INTERVAL. */
+static void
+scale_sketches(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_channel *ch;
+
+ for (ch = dpif->channels; ch < &dpif->channels[N_CHANNELS]; ch++) {
+ struct dpif_sketch *sk;
+
+ for (sk = ch->sketches; sk < &ch->sketches[N_SKETCHES]; sk++) {
+ sk->hits /= 2;
+ sk->error /= 2;
+ }
+ }
+}
+/* Logs information about a packet that was recently lost in 'ch' (in
+ * 'dpif_'). */
+static void
+report_loss(struct dpif *dpif_, struct dpif_channel *ch)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
+ struct dpif_sketch *sk;
+ struct ds s;
+
+ if (VLOG_DROP_ERR(&rl)) {
+ return;
+ }
+
+ ds_init(&s);
+ if (ch->last_poll != LLONG_MIN) {
+ ds_put_format(&s, " (last polled %lld ms ago)",
+ time_msec() - ch->last_poll);
+ }
+ ds_put_cstr(&s, ", most frequent sources are");
+ for (sk = ch->sketches; sk < &ch->sketches[N_SKETCHES]; sk++) {
+ if (sk->hits) {
+ struct dpif_port port;
+
+ ds_put_format(&s, " %"PRIu32, sk->port_no);
+ if (!dpif_port_query_by_number(dpif_, sk->port_no, &port)) {
+ ds_put_format(&s, "(%s)", port.name);
+ dpif_port_destroy(&port);
+ }
+ if (sk->error) {
+ ds_put_format(&s, ": %u to %u,",
+ sk->hits - sk->error, sk->hits);
+ } else {
+ ds_put_format(&s, ": %u,", sk->hits);
+ }
+ }
+ }
+ ds_chomp(&s, ',');
+
+ VLOG_WARN("%s: lost packet on channel %td%s",
+ dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));
+ ds_destroy(&s);
+}