/*
- * 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 "dpif-linux.h"
-#include <assert.h>
#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h>
#include <net/if.h>
#include <linux/types.h>
-#include <linux/ethtool.h>
#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
+#include <poll.h>
#include <stdlib.h>
-#include <sys/ioctl.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 "svec.h"
+#include "sset.h"
+#include "timeval.h"
#include "unaligned.h"
#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
+enum { MAX_PORTS = USHRT_MAX };
+
+/* 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 {
- /* ioctl command argument. */
- int cmd;
+ /* Generic Netlink header. */
+ uint8_t cmd;
- /* struct odp_datapath header. */
- uint32_t dp_idx;
+ /* 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. */
+ 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 *);
+static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
+ const struct ofpbuf *);
+static void dpif_linux_dp_dump_start(struct nl_dump *);
static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
struct dpif_linux_dp *reply,
struct ofpbuf **bufp);
struct ofpbuf **bufp);
struct dpif_linux_flow {
- /* ioctl command argument. */
- int cmd;
+ /* Generic Netlink header. */
+ uint8_t cmd;
- /* struct odp_flow header. */
+ /* struct ovs_header. */
unsigned int nlmsg_flags;
- uint32_t dp_idx;
+ int dp_ifindex;
/* Attributes.
*
- * The 'stats', 'used', and 'state' 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. */
- const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
+ * 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 OVS_FLOW_ATTR_ACTIONS will be included in
+ * the Netlink version of the command, even if actions_len is zero. */
+ 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 uint64_t *state; /* ODP_FLOW_ATTR_STATE. */
+ 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 *);
-static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+static int dpif_linux_flow_from_ofpbuf(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(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 *);
+/* One of the dpif channels between the kernel and userspace. */
+struct dpif_channel {
+ struct nl_sock *sock; /* Netlink socket. */
+ long long int last_poll; /* Last time this channel was polled. */
+};
+
+static void report_loss(struct dpif *, struct dpif_channel *);
+
/* Datapath interface for the openvswitch Linux kernel module. */
struct dpif_linux {
struct dpif dpif;
- int fd;
+ int dp_ifindex;
- /* Used by dpif_linux_get_all_names(). */
- char *local_ifname;
- int minor;
+ /* Upcall messages. */
+ int uc_array_size; /* Size of 'channels' and 'epoll_events'. */
+ struct dpif_channel *channels;
+ struct epoll_event *epoll_events;
+ int epoll_fd; /* epoll fd that includes channel socks. */
+ int n_events; /* Num events returned by epoll_wait(). */
+ int event_offset; /* Offset into 'epoll_events'. */
/* Change notification. */
- int local_ifindex; /* Ifindex of local port. */
- struct shash changed_ports; /* Ports that have changed. */
- struct rtnetlink_notifier port_notifier;
+ struct sset changed_ports; /* Ports that have changed. */
+ struct nln_notifier *port_notifier;
bool change_error;
};
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
-static int do_ioctl(const struct dpif *, int cmd, const void *arg);
-static int open_dpif(const struct dpif_linux_vport *local_vport,
- struct dpif **);
-static int get_openvswitch_major(void);
-static int open_minor(int minor, int *fdp);
-static int make_openvswitch_device(int minor, char **fnp);
-static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
- void *dpif);
+/* 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 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_vport_from_ofpbuf(struct dpif_linux_vport *,
+ const struct ofpbuf *);
static struct dpif_linux *
dpif_linux_cast(const struct dpif *dpif)
}
static int
-dpif_linux_enumerate(struct svec *all_dps)
+dpif_linux_enumerate(struct sset *all_dps)
{
- uint32_t dp_idx;
- int major;
+ struct nl_dump dump;
+ struct ofpbuf msg;
+ int error;
- /* Check that the Open vSwitch module is loaded. */
- major = get_openvswitch_major();
- if (major < 0) {
- return -major;
+ error = dpif_linux_init();
+ if (error) {
+ return error;
}
- dp_idx = 0;
- for (;;) {
- struct dpif_linux_dp request, reply;
- struct ofpbuf *buf;
- char devname[16];
- int error;
-
- dpif_linux_dp_init(&request);
- request.dp_idx = dp_idx;
- request.cmd = ODP_DP_DUMP;
+ dpif_linux_dp_dump_start(&dump);
+ while (nl_dump_next(&dump, &msg)) {
+ struct dpif_linux_dp dp;
- error = dpif_linux_dp_transact(&request, &reply, &buf);
- if (error) {
- return error == ENODEV ? 0 : error;
+ if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
+ sset_add(all_dps, dp.name);
}
- ofpbuf_delete(buf);
-
- sprintf(devname, "dp%d", reply.dp_idx);
- svec_add(all_dps, devname);
-
- dp_idx = reply.dp_idx + 1;
}
+ return nl_dump_done(&dump);
}
static int
dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
bool create, struct dpif **dpifp)
{
- struct dpif_linux_vport request, reply;
+ struct dpif_linux_dp dp_request, dp;
struct ofpbuf *buf;
- int minor;
+ uint32_t upcall_pid;
int error;
- minor = !strncmp(name, "dp", 2)
- && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1;
- if (create) {
- struct dpif_linux_dp request, reply;
- struct ofpbuf *buf;
- int error;
-
- dpif_linux_dp_init(&request);
- request.cmd = ODP_DP_NEW;
- request.dp_idx = minor;
- request.name = name;
- error = dpif_linux_dp_transact(&request, &reply, &buf);
- if (error) {
- return error;
- }
- minor = reply.dp_idx;
- ofpbuf_delete(buf);
+ error = dpif_linux_init();
+ if (error) {
+ return error;
}
- dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_GET;
- request.port_no = ODPP_LOCAL;
- if (minor >= 0) {
- request.dp_idx = minor;
+ /* Create or look up datapath. */
+ dpif_linux_dp_init(&dp_request);
+ if (create) {
+ dp_request.cmd = OVS_DP_CMD_NEW;
+ upcall_pid = 0;
+ dp_request.upcall_pid = &upcall_pid;
} else {
- request.name = name;
+ dp_request.cmd = OVS_DP_CMD_GET;
}
-
- error = dpif_linux_vport_transact(&request, &reply, &buf);
+ dp_request.name = name;
+ error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
if (error) {
return error;
- } else if (reply.port_no != ODPP_LOCAL) {
- /* This is an Open vSwitch device but not the local port. We
- * intentionally support only using the name of the local port as the
- * name of a datapath; otherwise, it would be too difficult to
- * enumerate all the names of a datapath. */
- error = EOPNOTSUPP;
- } else {
- error = open_dpif(&reply, dpifp);
}
+ open_dpif(&dp, dpifp);
ofpbuf_delete(buf);
- return error;
+ return 0;
}
-static int
-open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **dpifp)
+static void
+open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
{
- int dp_idx = local_vport->dp_idx;
struct dpif_linux *dpif;
- char *name;
- int error;
- int fd;
- error = open_minor(dp_idx, &fd);
- if (error) {
- goto error;
+ 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->dp_ifindex = dp->dp_ifindex;
+ sset_init(&dpif->changed_ports);
+ *dpifp = &dpif->dpif;
+}
+
+static void
+destroy_channels(struct dpif_linux *dpif)
+{
+ int i;
+
+ if (dpif->epoll_fd < 0) {
+ return;
}
- dpif = xmalloc(sizeof *dpif);
- error = rtnetlink_link_notifier_register(&dpif->port_notifier,
- dpif_linux_port_changed, dpif);
- if (error) {
- goto error_free;
+ for (i = 0; i < dpif->uc_array_size; i++ ) {
+ struct dpif_linux_vport vport_request;
+ struct dpif_channel *ch = &dpif->channels[i];
+ uint32_t upcall_pid = 0;
+
+ if (!ch->sock) {
+ continue;
+ }
+
+ /* Turn off upcalls. */
+ dpif_linux_vport_init(&vport_request);
+ vport_request.cmd = OVS_VPORT_CMD_SET;
+ vport_request.dp_ifindex = dpif->dp_ifindex;
+ vport_request.port_no = i;
+ vport_request.upcall_pid = &upcall_pid;
+ dpif_linux_vport_transact(&vport_request, NULL, NULL);
+
+ nl_sock_destroy(ch->sock);
}
- name = xasprintf("dp%d", dp_idx);
- dpif_init(&dpif->dpif, &dpif_linux_class, name, dp_idx, dp_idx);
- free(name);
+ free(dpif->channels);
+ dpif->channels = NULL;
+ dpif->uc_array_size = 0;
- dpif->fd = fd;
- dpif->local_ifname = xstrdup(local_vport->name);
- dpif->local_ifindex = local_vport->ifindex;
- dpif->minor = dp_idx;
- shash_init(&dpif->changed_ports);
- dpif->change_error = false;
- *dpifp = &dpif->dpif;
+ free(dpif->epoll_events);
+ dpif->epoll_events = NULL;
+ dpif->n_events = dpif->event_offset = 0;
- return 0;
+ close(dpif->epoll_fd);
+ dpif->epoll_fd = -1;
+}
-error_free:
- free(dpif);
- close(fd);
-error:
- return error;
+static int
+add_channel(struct dpif_linux *dpif, uint32_t port_no, struct nl_sock *sock)
+{
+ struct epoll_event event;
+
+ if (dpif->epoll_fd < 0) {
+ return 0;
+ }
+
+ /* We assume that the datapath densely chooses port numbers, which
+ * can therefore be used as an index into an array of channels. */
+ if (port_no >= dpif->uc_array_size) {
+ int new_size = port_no + 1;
+ int i;
+
+ if (new_size > 65535) {
+ VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big",
+ dpif_name(&dpif->dpif), port_no);
+ return EFBIG;
+ }
+
+ dpif->channels = xrealloc(dpif->channels,
+ new_size * sizeof *dpif->channels);
+ for (i = dpif->uc_array_size; i < new_size; i++) {
+ dpif->channels[i].sock = NULL;
+ }
+
+ dpif->epoll_events = xrealloc(dpif->epoll_events,
+ new_size * sizeof *dpif->epoll_events);
+ dpif->uc_array_size = new_size;
+ }
+
+ memset(&event, 0, sizeof event);
+ event.events = EPOLLIN;
+ event.data.u32 = port_no;
+ if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock),
+ &event) < 0) {
+ return errno;
+ }
+
+ dpif->channels[port_no].sock = sock;
+ dpif->channels[port_no].last_poll = LLONG_MIN;
+
+ return 0;
}
static void
-dpif_linux_close(struct dpif *dpif_)
+del_channel(struct dpif_linux *dpif, uint32_t port_no)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- rtnetlink_link_notifier_unregister(&dpif->port_notifier);
- shash_destroy(&dpif->changed_ports);
- free(dpif->local_ifname);
- close(dpif->fd);
- free(dpif);
+ struct dpif_channel *ch;
+
+ if (dpif->epoll_fd < 0 || port_no >= dpif->uc_array_size) {
+ return;
+ }
+
+ ch = &dpif->channels[port_no];
+ if (!ch->sock) {
+ return;
+ }
+
+ epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
+
+ nl_sock_destroy(ch->sock);
+ ch->sock = NULL;
}
-static int
-dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names)
+static void
+dpif_linux_close(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor));
- svec_add(all_names, dpif->local_ifname);
- return 0;
+ nln_notifier_destroy(dpif->port_notifier);
+ destroy_channels(dpif);
+ sset_destroy(&dpif->changed_ports);
+ free(dpif);
}
static int
struct dpif_linux_dp dp;
dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_DEL;
- dp.dp_idx = dpif->minor;
+ 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_ OVS_UNUSED)
{
- struct dpif_linux_dp dp;
- struct ofpbuf *buf;
- int error;
+ if (nln) {
+ nln_run(nln);
+ }
+}
- error = dpif_linux_dp_get(dpif_, &dp, &buf);
- if (!error) {
- *stats = dp.stats;
- ofpbuf_delete(buf);
+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_SET;
- dp.dp_idx = dpif->minor;
- 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);
const char *type = netdev_get_type(netdev);
struct dpif_linux_vport request, reply;
const struct ofpbuf *options;
+ struct nl_sock *sock = NULL;
+ uint32_t upcall_pid;
struct ofpbuf *buf;
int error;
+ if (dpif->epoll_fd >= 0) {
+ error = nl_sock_create(NETLINK_GENERIC, &sock);
+ if (error) {
+ return error;
+ }
+ }
+
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_NEW;
- request.dp_idx = dpif->minor;
+ 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);
+ nl_sock_destroy(sock);
return EINVAL;
}
request.name = name;
request.options_len = options->size;
}
+ if (request.type == OVS_VPORT_TYPE_NETDEV) {
+ netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
+ }
+
+ request.port_no = *port_nop;
+ upcall_pid = sock ? nl_sock_pid(sock) : 0;
+ 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_), reply.port_no, upcall_pid);
+ } else {
+ if (error == EBUSY && *port_nop != UINT32_MAX) {
+ VLOG_INFO("%s: requested port %"PRIu32" is in use",
+ dpif_name(dpif_), *port_nop);
+ }
+ nl_sock_destroy(sock);
ofpbuf_delete(buf);
+ return error;
}
+ ofpbuf_delete(buf);
- return error;
+ if (sock) {
+ error = add_channel(dpif, *port_nop, sock);
+ if (error) {
+ VLOG_INFO("%s: could not add channel for port %s",
+ dpif_name(dpif_), name);
+
+ /* Delete the port. */
+ dpif_linux_vport_init(&request);
+ request.cmd = OVS_VPORT_CMD_DEL;
+ request.dp_ifindex = dpif->dp_ifindex;
+ request.port_no = *port_nop;
+ dpif_linux_vport_transact(&request, NULL, NULL);
+
+ nl_sock_destroy(sock);
+ return error;
+ }
+ }
+
+ return 0;
}
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_DEL;
- vport.dp_idx = dpif->minor;
+ 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);
+
+ del_channel(dpif, port_no);
+
+ return error;
}
static int
int error;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_GET;
- request.dp_idx = dpif_linux_cast(dpif)->minor;
+ 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 {
+ /* The UINT32_MAX "reserved" port number uses the "ovs-system"'s
+ * channel, since it is not heavily loaded. */
+ int idx = (port_no >= dpif->uc_array_size) ? 0 : port_no;
+ return nl_sock_pid(dpif->channels[idx].sock);
+ }
}
static int
dpif_linux_flow_flush(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return ioctl(dpif->fd, ODP_FLOW_FLUSH, dpif->minor) ? errno : 0;
+ struct dpif_linux_flow flow;
+
+ dpif_linux_flow_init(&flow);
+ flow.cmd = OVS_FLOW_CMD_DEL;
+ flow.dp_ifindex = dpif->dp_ifindex;
+ return dpif_linux_flow_transact(&flow, NULL, NULL);
}
struct dpif_linux_port_state {
- struct ofpbuf *buf;
- uint32_t next;
+ struct nl_dump dump;
};
static int
-dpif_linux_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
+dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
{
- *statep = xzalloc(sizeof(struct dpif_linux_port_state));
+ 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.dp_ifindex = dpif->dp_ifindex;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_vport_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, genl_sock, buf);
+ ofpbuf_delete(buf);
+
return 0;
}
static int
-dpif_linux_port_dump_next(const struct dpif *dpif, void *state_,
+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 request, reply;
- struct ofpbuf *buf;
+ struct dpif_linux_vport vport;
+ struct ofpbuf buf;
int error;
- ofpbuf_delete(state->buf);
- state->buf = NULL;
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
+ }
- dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_DUMP;
- request.dp_idx = dpif_linux_cast(dpif)->minor;
- request.port_no = state->next;
- error = dpif_linux_vport_transact(&request, &reply, &buf);
+ error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
if (error) {
- return error == ENODEV ? EOF : error;
- } else {
- dpif_port->name = (char *) reply.name;
- dpif_port->type = (char *) netdev_vport_get_netdev_type(&reply);
- dpif_port->port_no = reply.port_no;
- state->buf = buf;
- state->next = reply.port_no + 1;
- return 0;
+ return error;
}
+
+ 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_;
- ofpbuf_delete(state->buf);
+ int error = nl_dump_done(&state->dump);
+
free(state);
- return 0;
+ return error;
}
static int
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 dpif_linux_flow *reply, struct ofpbuf **bufp)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request;
+
+ dpif_linux_flow_init(&request);
+ request.cmd = OVS_FLOW_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
+ request.key = key;
+ request.key_len = key_len;
+ 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 *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_flow request, reply;
+ struct dpif_linux_flow reply;
struct ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_GET;
- request.dp_idx = dpif->minor;
- request.key = key;
- request.key_len = key_len;
- error = dpif_linux_flow_transact(&request, &reply, &buf);
+ error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
if (!error) {
if (stats) {
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 ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_NEW : ODP_FLOW_SET;
- request.dp_idx = dpif->minor;
- request.key = key;
- request.key_len = key_len;
- request.actions = actions;
- 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_DEL;
- request.dp_idx = dpif->minor;
- 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;
}
-
struct dpif_linux_flow_state {
+ struct nl_dump dump;
struct dpif_linux_flow flow;
- struct ofpbuf *buf;
struct dpif_flow_stats stats;
+ struct ofpbuf *buf;
};
static int
-dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
+dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
{
- *statep = xzalloc(sizeof(struct dpif_linux_flow_state));
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow_state *state;
+ struct dpif_linux_flow request;
+ struct ofpbuf *buf;
+
+ *statep = state = xmalloc(sizeof *state);
+
+ dpif_linux_flow_init(&request);
+ request.cmd = OVS_DP_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, genl_sock, buf);
+ ofpbuf_delete(buf);
+
+ state->buf = NULL;
+
return 0;
}
static int
-dpif_linux_flow_dump_next(const struct dpif *dpif_, void *state_,
+dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
const struct nlattr **key, size_t *key_len,
const struct nlattr **actions, size_t *actions_len,
const struct dpif_flow_stats **stats)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_flow_state *state = state_;
- struct ofpbuf *old_buf = state->buf;
- struct dpif_linux_flow request;
+ struct ofpbuf buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_DUMP;
- request.dp_idx = dpif->minor;
- request.state = state->flow.state;
- error = dpif_linux_flow_transact(&request, &state->flow, &state->buf);
- ofpbuf_delete(old_buf);
+ do {
+ ofpbuf_delete(state->buf);
+ state->buf = NULL;
- 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;
}
- return error == ENODEV ? EOF : error;
+ if (stats) {
+ dpif_linux_flow_get_stats(&state->flow, &state->stats);
+ *stats = &state->stats;
+ }
+ return error;
}
static int
dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
{
struct dpif_linux_flow_state *state = state_;
-
+ int error = nl_dump_done(&state->dump);
ofpbuf_delete(state->buf);
free(state);
- return 0;
+ return error;
+}
+
+static void
+dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
+ struct ofpbuf *buf)
+{
+ struct ovs_header *k_exec;
+
+ ofpbuf_prealloc_tailroom(buf, (64
+ + d_exec->packet->size
+ + d_exec->key_len
+ + d_exec->actions_len));
+
+ nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
+ OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
+
+ k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
+ k_exec->dp_ifindex = dp_ifindex;
+
+ 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(struct dpif *dpif_,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *packet)
+dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct odp_upcall *execute;
- struct ofpbuf *buf;
+ uint64_t request_stub[1024 / 8];
+ struct ofpbuf request;
int error;
- buf = ofpbuf_new(128 + actions_len + packet->size);
+ 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);
+
+ return error;
+}
+
+static int
+dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ return dpif_linux_execute__(dpif->dp_ifindex, execute);
+}
+
+#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_);
+
+ 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;
+
+ ovs_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();
+ }
+ }
+
+ for (i = 0; i < n_ops; i++) {
+ txnsp[i] = &auxes[i].txn;
+ }
+ nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
+
+ 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;
+
+ op->error = txn->error;
+
+ 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;
- ofpbuf_reserve(buf, sizeof *execute);
- 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);
+ 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);
+ }
+ }
- execute = ofpbuf_push_uninit(buf, sizeof *execute);
- execute->dp_idx = dpif->minor;
- execute->len = buf->size;
+ if (op->error) {
+ memset(del->stats, 0, sizeof *del->stats);
+ }
+ }
+ break;
- error = do_ioctl(dpif_, ODP_EXECUTE, buf->data);
+ case DPIF_OP_EXECUTE:
+ break;
- ofpbuf_delete(buf);
- return error;
-}
+ default:
+ NOT_REACHED();
+ }
-static int
-dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
-{
- return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask);
+ ofpbuf_uninit(&aux->request);
+ ofpbuf_uninit(&aux->reply);
+ }
}
-static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+static void
+dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
{
- return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask);
+ while (n_ops > 0) {
+ size_t chunk = MIN(n_ops, MAX_OPS);
+ dpif_linux_operate__(dpif, ops, chunk);
+ ops += chunk;
+ n_ops -= chunk;
+ }
}
static int
-dpif_linux_get_sflow_probability(const struct dpif *dpif_,
- uint32_t *probability)
+dpif_linux_recv_set(struct dpif *dpif_, bool enable)
{
- 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) {
- *probability = dp.sampling ? *dp.sampling : 0;
- ofpbuf_delete(buf);
+ if ((dpif->epoll_fd >= 0) == enable) {
+ return 0;
}
- return error;
-}
-static int
-dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_dp dp;
+ if (!enable) {
+ destroy_channels(dpif);
+ } else {
+ struct dpif_port_dump port_dump;
+ struct dpif_port port;
- dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_SET;
- dp.dp_idx = dpif->minor;
- dp.sampling = &probability;
- return dpif_linux_dp_transact(&dp, NULL, NULL);
+ dpif->epoll_fd = epoll_create(10);
+ if (dpif->epoll_fd < 0) {
+ return errno;
+ }
+
+ 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;
+
+ error = nl_sock_create(NETLINK_GENERIC, &sock);
+ if (error) {
+ return error;
+ }
+
+ upcall_pid = nl_sock_pid(sock);
+
+ 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));
+ nl_sock_destroy(sock);
+
+ if (error == ENODEV || error == ENOENT) {
+ /* This device isn't there, but keep trying the others. */
+ continue;
+ } else {
+ return error;
+ }
+ }
+
+ error = add_channel(dpif, port.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;
+ }
+ }
+ }
+
+ return 0;
}
static int
}
static int
-parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall)
+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_TYPE] = { .type = NL_A_U32 },
- [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 },
+ [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
- /* _ODPL_ACTION_NR only. */
- [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
-
- /* _ODPL_SFLOW_NR 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_packet *odp_packet = buf->data;
- 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;
+ int type;
+
+ ofpbuf_use_const(&b, buf->data, buf->size);
+
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
+ }
- if (!nl_policy_parse(buf, sizeof *odp_packet, odp_packet_policy,
- a, ARRAY_SIZE(odp_packet_policy))) {
+ 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 = nl_attr_get_u32(a[ODP_PACKET_ATTR_TYPE]);
+ 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 = 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 retval;
- int error;
+ int read_tries = 0;
- buf = ofpbuf_new(65536);
- retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
- if (retval < 0) {
- error = errno;
- if (error != EAGAIN) {
- VLOG_WARN_RL(&error_rl, "%s: read failed: %s",
- dpif_name(dpif_), strerror(error));
- }
- } else if (retval >= sizeof(struct odp_packet)) {
- struct odp_packet *odp_packet = buf->data;
- buf->size += retval;
+ if (dpif->epoll_fd < 0) {
+ return EAGAIN;
+ }
- if (odp_packet->len <= retval) {
- error = parse_odp_packet(buf, upcall);
- } else {
- VLOG_WARN_RL(&error_rl, "%s: discarding message truncated "
- "from %"PRIu32" bytes to %d",
- dpif_name(dpif_), odp_packet->len, retval);
- error = ERANGE;
+ if (dpif->event_offset >= dpif->n_events) {
+ int retval;
+
+ dpif->event_offset = dpif->n_events = 0;
+
+ do {
+ retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
+ dpif->uc_array_size, 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));
+ } else if (retval > 0) {
+ dpif->n_events = retval;
}
- } else if (!retval) {
- VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_));
- error = EPROTO;
- } else {
- VLOG_WARN_RL(&error_rl, "%s: discarding too-short message (%d bytes)",
- dpif_name(dpif_), retval);
- error = ERANGE;
}
- if (error) {
- ofpbuf_delete(buf);
+ while (dpif->event_offset < dpif->n_events) {
+ int idx = dpif->epoll_events[dpif->event_offset].data.u32;
+ struct dpif_channel *ch = &dpif->channels[idx];
+
+ dpif->event_offset++;
+
+ 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) {
+ return 0;
+ } else if (error) {
+ return error;
+ }
+ }
}
- return error;
+
+ return EAGAIN;
}
static void
dpif_linux_recv_wait(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- poll_fd_wait(dpif->fd, 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_);
- int i;
+ struct dpif_channel *ch;
- /* This is somewhat bogus because it assumes that the following macros have
- * fixed values, but it's going to go away later. */
-#define DP_N_QUEUES 3
-#define DP_MAX_QUEUE_LEN 100
- for (i = 0; i < DP_N_QUEUES * DP_MAX_QUEUE_LEN; i++) {
- /* Reading even 1 byte discards a whole datagram and saves time. */
- char buffer;
+ if (dpif->epoll_fd < 0) {
+ return;
+ }
- if (read(dpif->fd, &buffer, 1) != 1) {
- break;
+ for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) {
+ if (ch->sock) {
+ nl_sock_drain(ch->sock);
}
}
}
const struct dpif_class dpif_linux_class = {
"system",
- NULL,
- NULL,
dpif_linux_enumerate,
+ NULL,
dpif_linux_open,
dpif_linux_close,
- dpif_linux_get_all_names,
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,
dpif_linux_recv_purge,
};
\f
-static int get_openvswitch_major(void);
-static int get_major(const char *target);
-
static int
-do_ioctl(const struct dpif *dpif_, int cmd, const void *arg)
+dpif_linux_init(void)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return ioctl(dpif->fd, cmd, arg) ? errno : 0;
+ static int error = -1;
+
+ if (error < 0) {
+ unsigned int ovs_vport_mcgroup;
+
+ error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
+ &ovs_datapath_family);
+ if (error) {
+ VLOG_ERR("Generic Netlink family '%s' does not exist. "
+ "The Open vSwitch kernel module is probably not loaded.",
+ OVS_DATAPATH_FAMILY);
+ }
+ if (!error) {
+ error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
+ }
+ if (!error) {
+ error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
+ }
+ if (!error) {
+ 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;
}
bool
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;
-}
-
-static int
-make_openvswitch_device(int minor, char **fnp)
-{
- const char dirname[] = "/dev/net";
- int major;
- dev_t dev;
- struct stat s;
- char fn[128];
-
- *fnp = NULL;
-
- major = get_openvswitch_major();
- if (major < 0) {
- return -major;
- }
- dev = makedev(major, minor);
-
- sprintf(fn, "%s/dp%d", dirname, minor);
- if (!stat(fn, &s)) {
- if (!S_ISCHR(s.st_mode)) {
- VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing",
- fn);
- } else if (s.st_rdev != dev) {
- VLOG_WARN_RL(&error_rl,
- "%s is device %u:%u but should be %u:%u, fixing",
- fn, major(s.st_rdev), minor(s.st_rdev),
- major(dev), minor(dev));
- } else {
- goto success;
- }
- if (unlink(fn)) {
- VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
- fn, strerror(errno));
- return errno;
- }
- } else if (errno == ENOENT) {
- if (stat(dirname, &s)) {
- if (errno == ENOENT) {
- if (mkdir(dirname, 0755)) {
- VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)",
- dirname, strerror(errno));
- return errno;
- }
- } else {
- VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)",
- dirname, strerror(errno));
- return errno;
- }
- }
- } else {
- VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno));
- return errno;
- }
-
- /* The device needs to be created. */
- if (mknod(fn, S_IFCHR | 0700, dev)) {
- VLOG_WARN_RL(&error_rl,
- "%s: creating character device %u:%u failed (%s)",
- fn, major(dev), minor(dev), strerror(errno));
- return errno;
- }
-
-success:
- *fnp = xstrdup(fn);
- return 0;
-}
-
-/* Return the major device number of the Open vSwitch device. If it
- * cannot be determined, a negative errno is returned. */
-static int
-get_openvswitch_major(void)
-{
- static int openvswitch_major = -1;
- if (openvswitch_major < 0) {
- openvswitch_major = get_major("openvswitch");
- }
- return openvswitch_major;
-}
-
-static int
-get_major(const char *target)
-{
- const char fn[] = "/proc/devices";
- char line[128];
- FILE *file;
- int ln;
-
- file = fopen(fn, "r");
- if (!file) {
- VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
- return -errno;
- }
-
- for (ln = 1; fgets(line, sizeof line, file); ln++) {
- char name[64];
- int major;
-
- if (!strncmp(line, "Character", 9) || line[0] == '\0') {
- /* Nothing to do. */
- } else if (!strncmp(line, "Block", 5)) {
- /* We only want character devices, so skip the rest of the file. */
- break;
- } else if (sscanf(line, "%d %63s", &major, name)) {
- if (!strcmp(name, target)) {
- fclose(file);
- return major;
- }
- } else {
- VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln);
- }
- }
-
- fclose(file);
-
- VLOG_ERR("%s: %s major not found (is the module loaded?)", fn, target);
- return -ENODEV;
+ return reply.type == OVS_VPORT_TYPE_INTERNAL;
}
-static int
-open_minor(int minor, int *fdp)
+static bool
+dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
{
- int error;
- char *fn;
-
- error = make_openvswitch_device(minor, &fn);
- if (error) {
- return error;
- }
-
- *fdp = open(fn, O_RDONLY | O_NONBLOCK);
- if (*fdp < 0) {
- error = errno;
- VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
- free(fn);
- return error;
- }
- free(fn);
- return 0;
+ 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->local_ifindex
- && (change->nlmsg_type == RTM_NEWLINK
- || change->nlmsg_type == RTM_DELLINK))
- {
- /* Our datapath changed, either adding a new port or deleting an
- * existing one. */
- shash_add_once(&dpif->changed_ports, change->ifname, NULL);
+ 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;
}
}
-
-static int
-get_dp0_fd(int *dp0_fdp)
-{
- static int dp0_fd = -1;
- if (dp0_fd < 0) {
- int error;
- int fd;
-
- error = open_minor(0, &fd);
- if (error) {
- return error;
- }
- dp0_fd = fd;
- }
- *dp0_fdp = dp0_fd;
- return 0;
-}
\f
-/* Parses the contents of 'buf', which contains a "struct odp_vport" 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 odp_vport *odp_vport;
- struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
+ struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
+ struct ovs_header *ovs_header;
+ struct nlmsghdr *nlmsg;
+ struct genlmsghdr *genl;
+ struct ofpbuf b;
dpif_linux_vport_init(vport);
- if (!nl_policy_parse(buf, sizeof *odp_vport, odp_vport_policy,
- a, ARRAY_SIZE(odp_vport_policy))) {
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
}
- odp_vport = buf->data;
- vport->dp_idx = odp_vport->dp_idx;
- 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]);
+ vport->cmd = genl->cmd;
+ 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_MTU]) {
- vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
+ if (a[OVS_VPORT_ATTR_STATS]) {
+ vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
}
- 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_ADDRESS]) {
+ vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]);
}
- if (a[ODP_VPORT_ATTR_IFINDEX]) {
- vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]);
- }
- 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_vport"
+/* 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_vport *odp_vport;
+ struct ovs_header *ovs_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
+ vport->cmd, OVS_VPORT_VERSION);
- ofpbuf_reserve(buf, sizeof odp_vport);
+ 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) {
- 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);
- }
-
- odp_vport = ofpbuf_push_uninit(buf, sizeof *odp_vport);
- odp_vport->dp_idx = vport->dp_idx;
- odp_vport->len = buf->size;
- odp_vport->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
}
/* Clears 'vport' to "empty" values. */
dpif_linux_vport_init(struct dpif_linux_vport *vport)
{
memset(vport, 0, sizeof *vport);
- vport->dp_idx = UINT32_MAX;
vport->port_no = UINT32_MAX;
}
/* 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
struct dpif_linux_vport *reply,
struct ofpbuf **bufp)
{
- struct ofpbuf *buf = NULL;
+ struct ofpbuf *request_buf;
int error;
- int fd;
- assert((reply != NULL) == (bufp != NULL));
+ ovs_assert((reply != NULL) == (bufp != NULL));
- error = get_dp0_fd(&fd);
+ error = dpif_linux_init();
if (error) {
- goto error;
+ if (reply) {
+ *bufp = NULL;
+ dpif_linux_vport_init(reply);
+ }
+ return error;
}
- buf = ofpbuf_new(1024);
- dpif_linux_vport_to_ofpbuf(request, buf);
-
- error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
- if (error) {
- goto error;
- }
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_vport_to_ofpbuf(request, request_buf);
+ error = nl_sock_transact(genl_sock, request_buf, bufp);
+ ofpbuf_delete(request_buf);
- if (bufp) {
- buf->size = ((struct odp_vport *) buf->data)->len;
- error = dpif_linux_vport_from_ofpbuf(reply, buf);
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
+ }
if (error) {
- goto error;
+ dpif_linux_vport_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- *bufp = buf;
- } else {
- ofpbuf_delete(buf);
- }
- return 0;
-
-error:
- ofpbuf_delete(buf);
- if (bufp) {
- memset(reply, 0, sizeof *reply);
- *bufp = NULL;
}
return error;
}
struct dpif_linux_vport request;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_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_datapath"
- * followed by Netlink attributes, into 'dp'. Returns 0 if successful,
- * otherwise a positive errno value.
+/* 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.
*
* 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
* while 'dp' is still in use. */
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 },
};
- struct odp_datapath *odp_dp;
- struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
+ struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
+ struct ovs_header *ovs_header;
+ struct nlmsghdr *nlmsg;
+ struct genlmsghdr *genl;
+ struct ofpbuf b;
dpif_linux_dp_init(dp);
- if (!nl_policy_parse(buf, sizeof *odp_dp, odp_datapath_policy,
- a, ARRAY_SIZE(odp_datapath_policy))) {
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
}
- odp_dp = buf->data;
- dp->dp_idx = odp_dp->dp_idx;
- dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
- if (a[ODP_DP_ATTR_STATS]) {
+ dp->cmd = genl->cmd;
+ 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]);
- }
+
return 0;
}
-/* Appends to 'buf' (which must initially be empty) a "struct odp_datapath"
- * followed by Netlink attributes corresponding to 'dp'. */
+/* Appends to 'buf' the Generic Netlink message described by 'dp'. */
static void
dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
{
- struct odp_datapath *odp_dp;
+ struct ovs_header *ovs_header;
- ofpbuf_reserve(buf, sizeof odp_dp);
+ nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
+ NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
+ OVS_DATAPATH_VERSION);
- if (dp->name) {
- nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
- }
+ ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
+ ovs_header->dp_ifindex = dp->dp_ifindex;
- /* 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->name) {
+ nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
}
- if (dp->sampling) {
- nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
+ if (dp->upcall_pid) {
+ nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
}
- odp_dp = ofpbuf_push_uninit(buf, sizeof *odp_dp);
- odp_dp->dp_idx = dp->dp_idx;
- odp_dp->len = buf->size;
- odp_dp->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
+ /* 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);
- dp->dp_idx = -1;
+}
+
+static void
+dpif_linux_dp_dump_start(struct nl_dump *dump)
+{
+ struct dpif_linux_dp request;
+ struct ofpbuf *buf;
+
+ dpif_linux_dp_init(&request);
+ request.cmd = OVS_DP_CMD_GET;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_dp_to_ofpbuf(&request, buf);
+ nl_dump_start(dump, genl_sock, buf);
+ ofpbuf_delete(buf);
}
/* 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_datapath 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
+ * 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'). */
+static int
dpif_linux_dp_transact(const struct dpif_linux_dp *request,
struct dpif_linux_dp *reply, struct ofpbuf **bufp)
{
- struct ofpbuf *buf = NULL;
+ struct ofpbuf *request_buf;
int error;
- int fd;
-
- assert((reply != NULL) == (bufp != NULL));
-
- error = get_dp0_fd(&fd);
- if (error) {
- goto error;
- }
- buf = ofpbuf_new(1024);
- dpif_linux_dp_to_ofpbuf(request, buf);
+ ovs_assert((reply != NULL) == (bufp != NULL));
- error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
- if (error) {
- goto error;
- }
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_dp_to_ofpbuf(request, request_buf);
+ error = nl_sock_transact(genl_sock, request_buf, bufp);
+ ofpbuf_delete(request_buf);
- if (bufp) {
- buf->size = ((struct odp_datapath *) buf->data)->len;
- error = dpif_linux_dp_from_ofpbuf(reply, buf);
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
+ }
if (error) {
- goto error;
+ dpif_linux_dp_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- *bufp = buf;
- } else {
- ofpbuf_delete(buf);
- }
- return 0;
-
-error:
- ofpbuf_delete(buf);
- if (bufp) {
- memset(reply, 0, sizeof *reply);
- *bufp = NULL;
}
return error;
}
/* 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_GET;
- request.dp_idx = dpif->minor;
+ 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_flow" followed by
- * Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
+/* 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.
*
* 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
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. */
- [ODP_FLOW_ATTR_STATE] = { .type = NL_A_U64, .optional = true },
+ [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 odp_flow *odp_flow;
- struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
+ struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
+ struct ovs_header *ovs_header;
+ struct nlmsghdr *nlmsg;
+ struct genlmsghdr *genl;
+ struct ofpbuf b;
dpif_linux_flow_init(flow);
- if (!nl_policy_parse(buf, sizeof *odp_flow, odp_flow_policy,
- a, ARRAY_SIZE(odp_flow_policy))) {
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
}
- odp_flow = buf->data;
- flow->nlmsg_flags = odp_flow->nlmsg_flags;
- flow->dp_idx = odp_flow->dp_idx;
- 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->nlmsg_flags = nlmsg->nlmsg_flags;
+ 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_STATE]) {
- flow->state = nl_attr_get(a[ODP_FLOW_ATTR_STATE]);
+ 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_flow"
+/* 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_flow *odp_flow;
+ struct ovs_header *ovs_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
+ NLM_F_REQUEST | flow->nlmsg_flags,
+ flow->cmd, OVS_FLOW_VERSION);
- ofpbuf_reserve(buf, sizeof odp_flow);
+ 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_len) {
- nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
+ if (flow->actions || flow->actions_len) {
+ nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
flow->actions, flow->actions_len);
}
/* We never need to send these to the kernel. */
- assert(!flow->stats);
- assert(!flow->tcp_flags);
- assert(!flow->used);
+ ovs_assert(!flow->stats);
+ ovs_assert(!flow->tcp_flags);
+ ovs_assert(!flow->used);
if (flow->clear) {
- nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
- }
-
- if (flow->state) {
- nl_msg_put_u64(buf, ODP_FLOW_ATTR_STATE,
- get_unaligned_u64(flow->state));
+ nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
}
-
- odp_flow = ofpbuf_push_uninit(buf, sizeof *odp_flow);
- odp_flow->nlmsg_flags = flow->nlmsg_flags;
- odp_flow->dp_idx = flow->dp_idx;
- odp_flow->len = buf->size;
- odp_flow->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
}
/* Clears 'flow' to "empty" values. */
-void
+static void
dpif_linux_flow_init(struct dpif_linux_flow *flow)
{
memset(flow, 0, sizeof *flow);
/* 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_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,
+ * 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'). */
+static int
+dpif_linux_flow_transact(struct dpif_linux_flow *request,
struct dpif_linux_flow *reply, struct ofpbuf **bufp)
{
- struct ofpbuf *buf = NULL;
+ struct ofpbuf *request_buf;
int error;
- int fd;
- assert((reply != NULL) == (bufp != NULL));
+ ovs_assert((reply != NULL) == (bufp != NULL));
- error = get_dp0_fd(&fd);
- if (error) {
- goto error;
+ if (reply) {
+ request->nlmsg_flags |= NLM_F_ECHO;
}
- buf = ofpbuf_new(1024);
- dpif_linux_flow_to_ofpbuf(request, buf);
-
- error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
- if (error) {
- goto error;
- }
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(request, request_buf);
+ error = nl_sock_transact(genl_sock, request_buf, bufp);
+ ofpbuf_delete(request_buf);
- if (bufp) {
- buf->size = ((struct odp_flow *) buf->data)->len;
- error = dpif_linux_flow_from_ofpbuf(reply, buf);
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
+ }
if (error) {
- goto error;
+ dpif_linux_flow_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- *bufp = buf;
- } else {
- ofpbuf_delete(buf);
- }
- return 0;
-
-error:
- ofpbuf_delete(buf);
- if (bufp) {
- memset(reply, 0, sizeof *reply);
- *bufp = NULL;
}
return error;
}
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
+/* 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 ds s;
+
+ if (VLOG_DROP_WARN(&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);
+ }
+ VLOG_WARN("%s: lost packet on channel %td%s",
+ dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));
+ ds_destroy(&s);
+}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff