/*
- * Copyright (c) 2008, 2009 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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 <config.h>
-#include "dpif.h"
-#include <assert.h>
+#include "dpif-linux.h"
+
#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h>
#include <net/if.h>
-#include <linux/ethtool.h>
+#include <linux/types.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 "packets.h"
#include "poll-loop.h"
-#include "svec.h"
+#include "random.h"
+#include "shash.h"
+#include "sset.h"
+#include "timeval.h"
+#include "unaligned.h"
#include "util.h"
-
#include "vlog.h"
-#define THIS_MODULE VLM_dpif_linux
+
+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 {
+ /* Generic Netlink header. */
+ uint8_t cmd;
+
+ /* struct ovs_header. */
+ int dp_ifindex;
+
+ /* Attributes. */
+ 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);
+static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
+ struct ofpbuf **bufp);
+
+struct dpif_linux_flow {
+ /* Generic Netlink header. */
+ uint8_t cmd;
+
+ /* struct ovs_header. */
+ unsigned int nlmsg_flags;
+ int dp_ifindex;
+
+ /* Attributes.
+ *
+ * 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 *mask; /* OVS_FLOW_ATTR_MASK. */
+ size_t mask_len;
+ const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
+ size_t actions_len;
+ 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_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. */
+ struct ovs_mutex upcall_lock;
+ 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 svec changed_ports; /* Ports that have changed. */
- struct rtnetlink_notifier port_notifier;
- bool change_error;
+ struct nl_sock *port_notifier; /* vport multicast group subscriber. */
};
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 lookup_minor(const char *name, int *minor);
-static int finish_open(struct dpif *, const char *local_ifname);
-static int create_minor(const char *name, int minor, struct dpif **dpifp);
-static int open_minor(int minor, struct dpif **dpifp);
-static int make_openvswitch_device(int minor, char **fnp);
-static void dpif_linux_port_changed(const struct rtnetlink_change *,
- void *dpif);
+/* Generic Netlink family numbers for OVS.
+ *
+ * Initialized by dpif_linux_init(). */
+static int ovs_datapath_family;
+static int ovs_vport_family;
+static int ovs_flow_family;
+static int ovs_packet_family;
+
+/* Generic Netlink multicast groups for OVS.
+ *
+ * Initialized by dpif_linux_init(). */
+static unsigned int ovs_vport_mcgroup;
+
+static int dpif_linux_init(void);
+static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
+static uint32_t dpif_linux_port_get_pid(const struct dpif *,
+ odp_port_t port_no);
+
+static 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)
{
+ struct nl_dump dump;
+ struct ofpbuf msg;
int error;
- int i;
- error = 0;
- for (i = 0; i < ODP_MAX; i++) {
- struct dpif *dpif;
- char devname[16];
- int retval;
+ error = dpif_linux_init();
+ if (error) {
+ return error;
+ }
- sprintf(devname, "dp%d", i);
- retval = dpif_open(devname, &dpif);
- if (!retval) {
- svec_add(all_dps, devname);
- dpif_close(dpif);
- } else if (retval != ENODEV && !error) {
- error = retval;
+ dpif_linux_dp_dump_start(&dump);
+ while (nl_dump_next(&dump, &msg)) {
+ struct dpif_linux_dp dp;
+
+ if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
+ sset_add(all_dps, dp.name);
}
}
- return error;
+ return nl_dump_done(&dump);
}
static int
-dpif_linux_open(const char *name UNUSED, char *suffix, bool create,
- struct dpif **dpifp)
+dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
+ bool create, struct dpif **dpifp)
{
- int minor;
+ struct dpif_linux_dp dp_request, dp;
+ struct ofpbuf *buf;
+ uint32_t upcall_pid;
+ int error;
- minor = !strncmp(name, "dp", 2) && isdigit(name[2]) ? atoi(name + 2) : -1;
- if (create) {
- if (minor >= 0) {
- return create_minor(suffix, minor, dpifp);
- } else {
- /* Scan for unused minor number. */
- for (minor = 0; minor < ODP_MAX; minor++) {
- int error = create_minor(suffix, minor, dpifp);
- if (error != EBUSY) {
- return error;
- }
- }
+ error = dpif_linux_init();
+ if (error) {
+ return error;
+ }
- /* All datapath numbers in use. */
- return ENOBUFS;
- }
+ /* 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 {
- struct dpif_linux *dpif;
- struct odp_port port;
- int error;
+ dp_request.cmd = OVS_DP_CMD_GET;
+ }
+ dp_request.name = name;
+ error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
+ if (error) {
+ return error;
+ }
- if (minor < 0) {
- error = lookup_minor(suffix, &minor);
- if (error) {
- return error;
- }
- }
+ error = open_dpif(&dp, dpifp);
+ ofpbuf_delete(buf);
+ return error;
+}
- error = open_minor(minor, dpifp);
- if (error) {
- return error;
- }
- dpif = dpif_linux_cast(*dpifp);
-
- /* We need the local port's ifindex for the poll function. Start by
- * getting the local port's name. */
- memset(&port, 0, sizeof port);
- port.port = ODPP_LOCAL;
- if (ioctl(dpif->fd, ODP_PORT_QUERY, &port)) {
- error = errno;
- if (error != ENODEV) {
- VLOG_WARN("%s: probe returned unexpected error: %s",
- dpif_name(*dpifp), strerror(error));
- }
- dpif_close(*dpifp);
- return error;
- }
+static int
+open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
+{
+ struct dpif_linux *dpif;
- /* Then use that to finish up opening. */
- return finish_open(&dpif->dpif, port.devname);
- }
+ dpif = xzalloc(sizeof *dpif);
+ dpif->port_notifier = NULL;
+ ovs_mutex_init(&dpif->upcall_lock);
+ 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;
+ *dpifp = &dpif->dpif;
+
+ return 0;
}
static void
-dpif_linux_close(struct dpif *dpif_)
+destroy_channels(struct dpif_linux *dpif)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- rtnetlink_notifier_unregister(&dpif->port_notifier);
- svec_destroy(&dpif->changed_ports);
- free(dpif->local_ifname);
- close(dpif->fd);
- free(dpif);
+ unsigned int i;
+
+ if (dpif->epoll_fd < 0) {
+ return;
+ }
+
+ 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;
+ }
+
+ epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
+
+ /* 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 = u32_to_odp(i);
+ vport_request.upcall_pid = &upcall_pid;
+ dpif_linux_vport_transact(&vport_request, NULL, NULL);
+
+ nl_sock_destroy(ch->sock);
+ }
+
+ free(dpif->channels);
+ dpif->channels = NULL;
+ dpif->uc_array_size = 0;
+
+ free(dpif->epoll_events);
+ dpif->epoll_events = NULL;
+ dpif->n_events = dpif->event_offset = 0;
+
+ /* Don't close dpif->epoll_fd since that would cause other threads that
+ * call dpif_recv_wait(dpif) to wait on an arbitrary fd or a closed fd. */
}
static int
-dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names)
+add_channel(struct dpif_linux *dpif, odp_port_t port_no, struct nl_sock *sock)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct epoll_event event;
+ uint32_t port_idx = odp_to_u32(port_no);
+
+ 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_idx >= dpif->uc_array_size) {
+ uint32_t new_size = port_idx + 1;
+ uint32_t i;
+
+ if (new_size > MAX_PORTS) {
+ 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_idx;
+ if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock),
+ &event) < 0) {
+ return errno;
+ }
+
+ nl_sock_destroy(dpif->channels[port_idx].sock);
+ dpif->channels[port_idx].sock = sock;
+ dpif->channels[port_idx].last_poll = LLONG_MIN;
- svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor));
- svec_add(all_names, dpif->local_ifname);
return 0;
}
-static int
-dpif_linux_delete(struct dpif *dpif_)
+static void
+del_channel(struct dpif_linux *dpif, odp_port_t port_no)
{
- return do_ioctl(dpif_, ODP_DP_DESTROY, NULL);
-}
+ struct dpif_channel *ch;
+ uint32_t port_idx = odp_to_u32(port_no);
-static int
-dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
-{
- return do_ioctl(dpif_, ODP_DP_STATS, stats);
+ if (dpif->epoll_fd < 0 || port_idx >= dpif->uc_array_size) {
+ return;
+ }
+
+ ch = &dpif->channels[port_idx];
+ if (!ch->sock) {
+ return;
+ }
+
+ epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
+ dpif->event_offset = dpif->n_events = 0;
+
+ nl_sock_destroy(ch->sock);
+ ch->sock = NULL;
}
-static int
-dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
+static void
+dpif_linux_close(struct dpif *dpif_)
{
- int drop_frags;
- int error;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags);
- if (!error) {
- *drop_fragsp = drop_frags & 1;
+ nl_sock_destroy(dpif->port_notifier);
+ destroy_channels(dpif);
+ if (dpif->epoll_fd >= 0) {
+ close(dpif->epoll_fd);
}
- return error;
+ ovs_mutex_destroy(&dpif->upcall_lock);
+ free(dpif);
}
static int
-dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
+dpif_linux_destroy(struct dpif *dpif_)
{
- int drop_frags_int = drop_frags;
- return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
+
+ dpif_linux_dp_init(&dp);
+ dp.cmd = OVS_DP_CMD_DEL;
+ dp.dp_ifindex = dpif->dp_ifindex;
+ return dpif_linux_dp_transact(&dp, NULL, NULL);
}
static int
-dpif_linux_port_add(struct dpif *dpif_, const char *devname, uint16_t flags,
- uint16_t *port_no)
+dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
{
- struct odp_port port;
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
int error;
- memset(&port, 0, sizeof port);
- strncpy(port.devname, devname, sizeof port.devname);
- port.flags = flags;
- error = do_ioctl(dpif_, ODP_PORT_ADD, &port);
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
if (!error) {
- *port_no = port.port;
+ 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_port_del(struct dpif *dpif_, uint16_t port_no)
+static const char *
+get_vport_type(const struct dpif_linux_vport *vport)
{
- int tmp = port_no;
- return do_ioctl(dpif_, ODP_PORT_DEL, &tmp);
-}
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
-static int
-dpif_linux_port_query_by_number(const struct dpif *dpif_, uint16_t port_no,
- struct odp_port *port)
-{
- memset(port, 0, sizeof *port);
- port->port = port_no;
- return do_ioctl(dpif_, ODP_PORT_QUERY, port);
-}
+ switch (vport->type) {
+ case OVS_VPORT_TYPE_NETDEV:
+ return "system";
-static int
-dpif_linux_port_query_by_name(const struct dpif *dpif_, const char *devname,
- struct odp_port *port)
-{
- memset(port, 0, sizeof *port);
- strncpy(port->devname, devname, sizeof port->devname);
- return do_ioctl(dpif_, ODP_PORT_QUERY, port);
+ case OVS_VPORT_TYPE_INTERNAL:
+ return "internal";
+
+ case OVS_VPORT_TYPE_GRE:
+ return "gre";
+
+ case OVS_VPORT_TYPE_GRE64:
+ return "gre64";
+
+ case OVS_VPORT_TYPE_VXLAN:
+ return "vxlan";
+
+ case OVS_VPORT_TYPE_LISP:
+ return "lisp";
+
+ case OVS_VPORT_TYPE_UNSPEC:
+ case __OVS_VPORT_TYPE_MAX:
+ break;
+ }
+
+ VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
+ vport->dp_ifindex, vport->name, (unsigned int) vport->type);
+ return "unknown";
}
-static int
-dpif_linux_flow_flush(struct dpif *dpif_)
+static enum ovs_vport_type
+netdev_to_ovs_vport_type(const struct netdev *netdev)
{
- return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL);
+ const char *type = netdev_get_type(netdev);
+
+ if (!strcmp(type, "tap") || !strcmp(type, "system")) {
+ return OVS_VPORT_TYPE_NETDEV;
+ } else if (!strcmp(type, "internal")) {
+ return OVS_VPORT_TYPE_INTERNAL;
+ } else if (strstr(type, "gre64")) {
+ return OVS_VPORT_TYPE_GRE64;
+ } else if (strstr(type, "gre")) {
+ return OVS_VPORT_TYPE_GRE;
+ } else if (!strcmp(type, "vxlan")) {
+ return OVS_VPORT_TYPE_VXLAN;
+ } else if (!strcmp(type, "lisp")) {
+ return OVS_VPORT_TYPE_LISP;
+ } else {
+ return OVS_VPORT_TYPE_UNSPEC;
+ }
}
static int
-dpif_linux_port_list(const struct dpif *dpif_, struct odp_port *ports, int n)
+dpif_linux_port_add__(struct dpif *dpif_, struct netdev *netdev,
+ odp_port_t *port_nop)
{
- struct odp_portvec pv;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ const struct netdev_tunnel_config *tnl_cfg;
+ char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
+ const char *name = netdev_vport_get_dpif_port(netdev,
+ namebuf, sizeof namebuf);
+ const char *type = netdev_get_type(netdev);
+ struct dpif_linux_vport request, reply;
+ struct nl_sock *sock = NULL;
+ uint32_t upcall_pid;
+ struct ofpbuf *buf;
+ uint64_t options_stub[64 / 8];
+ struct ofpbuf options;
int error;
- pv.ports = ports;
- pv.n_ports = n;
- error = do_ioctl(dpif_, ODP_PORT_LIST, &pv);
- return error ? -error : pv.n_ports;
-}
+ if (dpif->epoll_fd >= 0) {
+ error = nl_sock_create(NETLINK_GENERIC, &sock);
+ if (error) {
+ return error;
+ }
+ }
-static int
-dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ dpif_linux_vport_init(&request);
+ request.cmd = OVS_VPORT_CMD_NEW;
+ request.dp_ifindex = dpif->dp_ifindex;
+ request.type = netdev_to_ovs_vport_type(netdev);
+ 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;
- if (dpif->change_error) {
- dpif->change_error = false;
- svec_clear(&dpif->changed_ports);
- return ENOBUFS;
- } else if (dpif->changed_ports.n) {
- *devnamep = dpif->changed_ports.names[--dpif->changed_ports.n];
- return 0;
- } else {
- return EAGAIN;
+ if (request.type == OVS_VPORT_TYPE_NETDEV) {
+ netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
}
-}
-static void
-dpif_linux_port_poll_wait(const struct dpif *dpif_)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (dpif->changed_ports.n || dpif->change_error) {
- poll_immediate_wake();
+ tnl_cfg = netdev_get_tunnel_config(netdev);
+ if (tnl_cfg && tnl_cfg->dst_port != 0) {
+ ofpbuf_use_stack(&options, options_stub, sizeof options_stub);
+ nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT,
+ ntohs(tnl_cfg->dst_port));
+ request.options = options.data;
+ request.options_len = options.size;
+ }
+
+ 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 {
- rtnetlink_notifier_wait();
+ if (error == EBUSY && *port_nop != ODPP_NONE) {
+ 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);
+
+ 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_group_get(const struct dpif *dpif_, int group,
- uint16_t ports[], int n)
+dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
+ odp_port_t *port_nop)
{
- struct odp_port_group pg;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int error;
- assert(n <= UINT16_MAX);
- pg.group = group;
- pg.ports = ports;
- pg.n_ports = n;
- error = do_ioctl(dpif_, ODP_PORT_GROUP_GET, &pg);
- return error ? -error : pg.n_ports;
+ ovs_mutex_lock(&dpif->upcall_lock);
+ error = dpif_linux_port_add__(dpif_, netdev, port_nop);
+ ovs_mutex_unlock(&dpif->upcall_lock);
+
+ return error;
}
static int
-dpif_linux_port_group_set(struct dpif *dpif_, int group,
- const uint16_t ports[], int n)
+dpif_linux_port_del__(struct dpif *dpif_, odp_port_t port_no)
{
- struct odp_port_group pg;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_vport vport;
+ int error;
- assert(n <= UINT16_MAX);
- pg.group = group;
- pg.ports = (uint16_t *) ports;
- pg.n_ports = n;
- return do_ioctl(dpif_, ODP_PORT_GROUP_SET, &pg);
-}
+ dpif_linux_vport_init(&vport);
+ vport.cmd = OVS_VPORT_CMD_DEL;
+ vport.dp_ifindex = dpif->dp_ifindex;
+ vport.port_no = port_no;
+ error = dpif_linux_vport_transact(&vport, NULL, NULL);
-static int
-dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n)
-{
- struct odp_flowvec fv;
- fv.flows = flows;
- fv.n_flows = n;
- return do_ioctl(dpif_, ODP_FLOW_GET, &fv);
-}
+ del_channel(dpif, port_no);
-static int
-dpif_linux_flow_put(struct dpif *dpif_, struct odp_flow_put *put)
-{
- return do_ioctl(dpif_, ODP_FLOW_PUT, put);
+ return error;
}
static int
-dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow)
+dpif_linux_port_del(struct dpif *dpif_, odp_port_t port_no)
{
- return do_ioctl(dpif_, ODP_FLOW_DEL, flow);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int error;
+
+ ovs_mutex_lock(&dpif->upcall_lock);
+ error = dpif_linux_port_del__(dpif_, port_no);
+ ovs_mutex_unlock(&dpif->upcall_lock);
+
+ return error;
}
static int
-dpif_linux_flow_list(const struct dpif *dpif_, struct odp_flow flows[], int n)
+dpif_linux_port_query__(const struct dpif *dpif, odp_port_t port_no,
+ const char *port_name, struct dpif_port *dpif_port)
{
- struct odp_flowvec fv;
+ struct dpif_linux_vport request;
+ struct dpif_linux_vport reply;
+ struct ofpbuf *buf;
int error;
- fv.flows = flows;
- fv.n_flows = n;
- error = do_ioctl(dpif_, ODP_FLOW_LIST, &fv);
- return error ? -error : fv.n_flows;
+ dpif_linux_vport_init(&request);
+ 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) {
+ 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(get_vport_type(&reply));
+ dpif_port->port_no = reply.port_no;
+ }
+ ofpbuf_delete(buf);
+ }
+ return error;
}
static int
-dpif_linux_execute(struct dpif *dpif_, uint16_t in_port,
- const union odp_action actions[], int n_actions,
- const struct ofpbuf *buf)
+dpif_linux_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
+ struct dpif_port *dpif_port)
{
- struct odp_execute execute;
- memset(&execute, 0, sizeof execute);
- execute.in_port = in_port;
- execute.actions = (union odp_action *) actions;
- execute.n_actions = n_actions;
- execute.data = buf->data;
- execute.length = buf->size;
- return do_ioctl(dpif_, ODP_EXECUTE, &execute);
+ return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
}
static int
-dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
+dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
+ struct dpif_port *dpif_port)
{
- return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask);
+ return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
}
-static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+static uint32_t
+dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
{
- return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask);
+ return MAX_PORTS;
}
-static int
-dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp)
+static uint32_t
+dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct ofpbuf *buf;
- int retval;
- int error;
-
- 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_msg)) {
- struct odp_msg *msg = buf->data;
- if (msg->length <= retval) {
- buf->size += retval;
- *bufp = buf;
- return 0;
- } else {
- VLOG_WARN_RL(&error_rl, "%s: discarding message truncated "
- "from %zu bytes to %d",
- dpif_name(dpif_), msg->length, retval);
- error = ERANGE;
- }
- } 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;
+ uint32_t port_idx = odp_to_u32(port_no);
+ uint32_t pid = 0;
+
+ ovs_mutex_lock(&dpif->upcall_lock);
+ if (dpif->epoll_fd >= 0) {
+ /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s
+ * channel, since it is not heavily loaded. */
+ uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx;
+ pid = nl_sock_pid(dpif->channels[idx].sock);
}
+ ovs_mutex_unlock(&dpif->upcall_lock);
- *bufp = NULL;
- ofpbuf_delete(buf);
- return error;
+ return pid;
}
-static void
-dpif_linux_recv_wait(struct dpif *dpif_)
+static int
+dpif_linux_flow_flush(struct dpif *dpif_)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- poll_fd_wait(dpif->fd, POLLIN);
+ const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ 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);
}
-const struct dpif_class dpif_linux_class = {
- "", /* This is the default class. */
- "linux",
- NULL,
- NULL,
- dpif_linux_enumerate,
- dpif_linux_open,
- dpif_linux_close,
- dpif_linux_get_all_names,
- dpif_linux_delete,
- 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_port_list,
- dpif_linux_port_poll,
- dpif_linux_port_poll_wait,
- dpif_linux_port_group_get,
- dpif_linux_port_group_set,
- dpif_linux_flow_get,
- dpif_linux_flow_put,
- dpif_linux_flow_del,
- dpif_linux_flow_flush,
- dpif_linux_flow_list,
- dpif_linux_execute,
- dpif_linux_recv_get_mask,
- dpif_linux_recv_set_mask,
- dpif_linux_recv,
- dpif_linux_recv_wait,
+struct dpif_linux_port_state {
+ struct nl_dump dump;
};
-\f
-static int get_openvswitch_major(void);
-static int get_major(const char *target, int default_major);
-
-static int
-do_ioctl(const struct dpif *dpif_, int cmd, const void *arg)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return ioctl(dpif->fd, cmd, arg) ? errno : 0;
-}
static int
-lookup_minor(const char *name, int *minorp)
+dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
{
- struct ethtool_drvinfo drvinfo;
- int minor, port_no;
- struct ifreq ifr;
- int error;
- int sock;
-
- sock = socket(AF_INET, SOCK_DGRAM, 0);
- if (sock < 0) {
- VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
- error = errno;
- goto error;
- }
-
- memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
- ifr.ifr_data = (caddr_t) &drvinfo;
+ const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_port_state *state;
+ struct dpif_linux_vport request;
+ struct ofpbuf *buf;
- memset(&drvinfo, 0, sizeof drvinfo);
- drvinfo.cmd = ETHTOOL_GDRVINFO;
- if (ioctl(sock, SIOCETHTOOL, &ifr)) {
- VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
- error = errno;
- goto error_close_sock;
- }
+ *statep = state = xmalloc(sizeof *state);
- if (strcmp(drvinfo.driver, "openvswitch")) {
- VLOG_WARN("%s is not an openvswitch device", name);
- error = EOPNOTSUPP;
- goto error_close_sock;
- }
+ dpif_linux_vport_init(&request);
+ request.cmd = OVS_VPORT_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
- if (sscanf(drvinfo.bus_info, "%d.%d", &minor, &port_no) != 2) {
- VLOG_WARN("%s ethtool bus_info has unexpected format", name);
- error = EPROTOTYPE;
- goto error_close_sock;
- } else if (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;
- goto error_close_sock;
- }
+ buf = ofpbuf_new(1024);
+ dpif_linux_vport_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, NETLINK_GENERIC, buf);
+ ofpbuf_delete(buf);
- *minorp = minor;
- close(sock);
return 0;
-
-error_close_sock:
- close(sock);
-error:
- return error;
}
static int
-make_openvswitch_device(int minor, char **fnp)
+dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
+ struct dpif_port *dpif_port)
{
- dev_t dev = makedev(get_openvswitch_major(), minor);
- const char dirname[] = "/dev/net";
- struct stat s;
- char fn[128];
+ struct dpif_linux_port_state *state = state_;
+ struct dpif_linux_vport vport;
+ struct ofpbuf buf;
+ int error;
- *fnp = NULL;
- 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 instead of %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;
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
}
- /* 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;
+ error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
+ if (error) {
+ return error;
}
-success:
- *fnp = xstrdup(fn);
+ dpif_port->name = CONST_CAST(char *, vport.name);
+ dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
+ dpif_port->port_no = vport.port_no;
return 0;
}
-
static int
-get_openvswitch_major(void)
+dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
{
- static unsigned int openvswitch_major;
- if (!openvswitch_major) {
- enum { DEFAULT_MAJOR = 248 };
- openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR);
- }
- return openvswitch_major;
+ struct dpif_linux_port_state *state = state_;
+ int error = nl_dump_done(&state->dump);
+
+ free(state);
+ return error;
}
static int
-get_major(const char *target, int default_major)
+dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
{
- const char fn[] = "/proc/devices";
- char line[128];
- FILE *file;
- int ln;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ /* Lazily create the Netlink socket to listen for notifications. */
+ if (!dpif->port_notifier) {
+ struct nl_sock *sock;
+ int error;
+
+ error = nl_sock_create(NETLINK_GENERIC, &sock);
+ if (error) {
+ return error;
+ }
+
+ error = nl_sock_join_mcgroup(sock, ovs_vport_mcgroup);
+ if (error) {
+ nl_sock_destroy(sock);
+ return error;
+ }
+ dpif->port_notifier = sock;
- file = fopen(fn, "r");
- if (!file) {
- VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
- goto error;
+ /* We have no idea of the current state so report that everything
+ * changed. */
+ return ENOBUFS;
}
- for (ln = 1; fgets(line, sizeof line, file); ln++) {
- char name[64];
- int major;
+ for (;;) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ uint64_t buf_stub[4096 / 8];
+ struct ofpbuf buf;
+ int error;
- 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;
+ ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
+ error = nl_sock_recv(dpif->port_notifier, &buf, false);
+ if (!error) {
+ struct dpif_linux_vport vport;
+
+ error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
+ if (!error) {
+ 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);
+ *devnamep = xstrdup(vport.name);
+ ofpbuf_uninit(&buf);
+ return 0;
+ }
}
+ } else if (error != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)",
+ ovs_strerror(error));
+ nl_sock_drain(dpif->port_notifier);
+ error = ENOBUFS;
+ }
+
+ ofpbuf_uninit(&buf);
+ if (error) {
+ return error;
+ }
+ }
+}
+
+static void
+dpif_linux_port_poll_wait(const struct dpif *dpif_)
+{
+ const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ if (dpif->port_notifier) {
+ nl_sock_wait(dpif->port_notifier, POLLIN);
+ } else {
+ poll_immediate_wake();
+ }
+}
+
+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)
+{
+ const 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_flow reply;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
+ if (!error) {
+ if (stats) {
+ dpif_linux_flow_get_stats(&reply, stats);
+ }
+ if (actionsp) {
+ buf->data = CONST_CAST(struct nlattr *, reply.actions);
+ buf->size = reply.actions_len;
+ *actionsp = buf;
} else {
- static bool warned;
- if (!warned) {
- VLOG_WARN("%s:%d: syntax error", fn, ln);
+ ofpbuf_delete(buf);
+ }
+ }
+ return error;
+}
+
+static void
+dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
+ struct dpif_linux_flow *request)
+{
+ static const struct nlattr dummy_action;
+
+ const 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;
+ request->mask = put->mask;
+ request->mask_len = put->mask_len;
+ /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
+ request->actions = (put->actions
+ ? put->actions
+ : CONST_CAST(struct nlattr *, &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_init_flow_put(dpif_, put, &request);
+ error = dpif_linux_flow_transact(&request,
+ 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 void
+dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
+ struct dpif_linux_flow *request)
+{
+ const 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_init_flow_del(dpif_, del, &request);
+ error = dpif_linux_flow_transact(&request,
+ 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 dpif_flow_stats stats;
+ struct ofpbuf *buf;
+};
+
+static int
+dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
+{
+ const 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_FLOW_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, NETLINK_GENERIC, buf);
+ ofpbuf_delete(buf);
+
+ state->buf = NULL;
+
+ return 0;
+}
+
+static int
+dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
+ const struct nlattr **key, size_t *key_len,
+ const struct nlattr **mask, size_t *mask_len,
+ const struct nlattr **actions, size_t *actions_len,
+ const struct dpif_flow_stats **stats)
+{
+ struct dpif_linux_flow_state *state = state_;
+ struct ofpbuf buf;
+ int error;
+
+ do {
+ ofpbuf_delete(state->buf);
+ state->buf = NULL;
+
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
+ }
+
+ error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
+ if (error) {
+ return error;
+ }
+
+ 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",
+ ovs_strerror(error));
+ }
+ }
+ } while (error);
+
+ if (actions) {
+ *actions = state->flow.actions;
+ *actions_len = state->flow.actions_len;
+ }
+ if (key) {
+ *key = state->flow.key;
+ *key_len = state->flow.key_len;
+ }
+ if (mask) {
+ *mask = state->flow.mask;
+ *mask_len = state->flow.mask ? state->flow.mask_len : 0;
+ }
+ 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 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__(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_transact(NETLINK_GENERIC, &request, NULL);
+ ofpbuf_uninit(&request);
+
+ return error;
+}
+
+static int
+dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
+{
+ const 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)
+{
+ const 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;
}
- warned = true;
+ 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();
}
}
- VLOG_ERR("%s: %s major not found (is the module loaded?), using "
- "default major %d", fn, target, default_major);
-error:
- VLOG_INFO("using default major %d for %s", default_major, target);
- return default_major;
+ for (i = 0; i < n_ops; i++) {
+ txnsp[i] = &auxes[i].txn;
+ }
+ nl_transact_multiple(NETLINK_GENERIC, 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;
+
+ 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);
+ }
+}
+
+static void
+dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
+{
+ 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
-finish_open(struct dpif *dpif_, const char *local_ifname)
+dpif_linux_recv_set__(struct dpif *dpif_, bool enable)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- dpif->local_ifname = strdup(local_ifname);
- dpif->local_ifindex = if_nametoindex(local_ifname);
- if (!dpif->local_ifindex) {
- int error = errno;
- dpif_close(dpif_);
- VLOG_WARN("could not get ifindex of %s device: %s",
- local_ifname, strerror(errno));
- return error;
+
+ if ((dpif->epoll_fd >= 0) == enable) {
+ return 0;
}
+
+ if (!enable) {
+ destroy_channels(dpif);
+ } else {
+ struct dpif_port_dump port_dump;
+ struct dpif_port port;
+
+ if (dpif->epoll_fd < 0) {
+ 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), ovs_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
-create_minor(const char *name, int minor, struct dpif **dpifp)
+dpif_linux_recv_set(struct dpif *dpif_, bool enable)
{
- int error = open_minor(minor, dpifp);
- if (!error) {
- error = do_ioctl(*dpifp, ODP_DP_CREATE, name);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int error;
+
+ ovs_mutex_lock(&dpif->upcall_lock);
+ error = dpif_linux_recv_set__(dpif_, enable);
+ ovs_mutex_unlock(&dpif->upcall_lock);
+
+ return error;
+}
+
+static int
+dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
+ uint32_t queue_id, uint32_t *priority)
+{
+ if (queue_id < 0xf000) {
+ *priority = TC_H_MAKE(1 << 16, queue_id + 1);
+ return 0;
+ } else {
+ return EINVAL;
+ }
+}
+
+static int
+parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
+ int *dp_ifindex)
+{
+ static const struct nl_policy ovs_packet_policy[] = {
+ /* Always present. */
+ [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
+ .min_len = ETH_HEADER_LEN },
+ [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
+
+ /* OVS_PACKET_CMD_ACTION only. */
+ [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true },
+ };
+
+ 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;
+ }
+
+ 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 = 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];
+ *dp_ifindex = ovs_header->dp_ifindex;
+
+ return 0;
+}
+
+static int
+dpif_linux_recv__(struct dpif *dpif_, struct dpif_upcall *upcall,
+ struct ofpbuf *buf)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int read_tries = 0;
+
+ if (dpif->epoll_fd < 0) {
+ return EAGAIN;
+ }
+
+ 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)", ovs_strerror(errno));
+ } else if (retval > 0) {
+ dpif->n_events = retval;
+ }
+ }
+
+ 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 EAGAIN;
+}
+
+static int
+dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
+ struct ofpbuf *buf)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int error;
+
+ ovs_mutex_lock(&dpif->upcall_lock);
+ error = dpif_linux_recv__(dpif_, upcall, buf);
+ ovs_mutex_unlock(&dpif->upcall_lock);
+
+ return error;
+}
+
+static void
+dpif_linux_recv_wait(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ ovs_mutex_lock(&dpif->upcall_lock);
+ if (dpif->epoll_fd >= 0) {
+ poll_fd_wait(dpif->epoll_fd, POLLIN);
+ }
+ ovs_mutex_unlock(&dpif->upcall_lock);
+}
+
+static void
+dpif_linux_recv_purge(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ ovs_mutex_lock(&dpif->upcall_lock);
+ if (dpif->epoll_fd >= 0) {
+ struct dpif_channel *ch;
+
+ for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size];
+ ch++) {
+ if (ch->sock) {
+ nl_sock_drain(ch->sock);
+ }
+ }
+ }
+ ovs_mutex_unlock(&dpif->upcall_lock);
+}
+
+const struct dpif_class dpif_linux_class = {
+ "system",
+ dpif_linux_enumerate,
+ NULL,
+ dpif_linux_open,
+ dpif_linux_close,
+ dpif_linux_destroy,
+ NULL, /* run */
+ NULL, /* wait */
+ dpif_linux_get_stats,
+ 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_port_poll,
+ dpif_linux_port_poll_wait,
+ dpif_linux_flow_get,
+ dpif_linux_flow_put,
+ dpif_linux_flow_del,
+ dpif_linux_flow_flush,
+ dpif_linux_flow_dump_start,
+ dpif_linux_flow_dump_next,
+ dpif_linux_flow_dump_done,
+ dpif_linux_execute,
+ 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
+dpif_linux_init(void)
+{
+ static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
+ static int error;
+
+ if (ovsthread_once_start(&once)) {
+ 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 = finish_open(*dpifp, name);
- } else {
- dpif_close(*dpifp);
+ 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_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
+ &ovs_vport_mcgroup);
+ }
+
+ ovsthread_once_done(&once);
}
+
return error;
}
+bool
+dpif_linux_is_internal_device(const char *name)
+{
+ struct dpif_linux_vport reply;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_vport_get(name, &reply, &buf);
+ if (!error) {
+ ofpbuf_delete(buf);
+ } else if (error != ENODEV && error != ENOENT) {
+ VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
+ name, ovs_strerror(error));
+ }
+
+ return reply.type == OVS_VPORT_TYPE_INTERNAL;
+}
+\f
+/* 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.
+ *
+ * 'vport' will contain pointers into 'buf', so the caller should not free
+ * 'buf' while 'vport' is still in use. */
static int
-open_minor(int minor, struct dpif **dpifp)
+dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
+ const struct ofpbuf *buf)
+{
+ 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 },
+ [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
+ };
+
+ 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);
+
+ 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;
+ }
+
+ vport->cmd = genl->cmd;
+ vport->dp_ifindex = ovs_header->dp_ifindex;
+ vport->port_no = nl_attr_get_odp_port(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[OVS_VPORT_ATTR_STATS]) {
+ vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
+ }
+ 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 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 ovs_header *ovs_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
+ vport->cmd, OVS_VPORT_VERSION);
+
+ ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
+ ovs_header->dp_ifindex = vport->dp_ifindex;
+
+ if (vport->port_no != ODPP_NONE) {
+ nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
+ }
+
+ 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, 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, OVS_VPORT_ATTR_STATS,
+ vport->stats, sizeof *vport->stats);
+ }
+
+ if (vport->options) {
+ nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
+ vport->options, vport->options_len);
+ }
+}
+
+/* Clears 'vport' to "empty" values. */
+void
+dpif_linux_vport_init(struct dpif_linux_vport *vport)
{
+ memset(vport, 0, sizeof *vport);
+ vport->port_no = ODPP_NONE;
+}
+
+/* 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 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
+dpif_linux_vport_transact(const struct dpif_linux_vport *request,
+ struct dpif_linux_vport *reply,
+ struct ofpbuf **bufp)
+{
+ struct ofpbuf *request_buf;
int error;
- char *fn;
- int fd;
- error = make_openvswitch_device(minor, &fn);
+ ovs_assert((reply != NULL) == (bufp != NULL));
+
+ error = dpif_linux_init();
if (error) {
+ if (reply) {
+ *bufp = NULL;
+ dpif_linux_vport_init(reply);
+ }
return error;
}
- fd = open(fn, O_RDONLY | O_NONBLOCK);
- if (fd >= 0) {
- struct dpif_linux *dpif = xmalloc(sizeof *dpif);
- error = rtnetlink_notifier_register(&dpif->port_notifier,
- dpif_linux_port_changed, dpif);
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_vport_to_ofpbuf(request, request_buf);
+ error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
+ ofpbuf_delete(request_buf);
+
+ if (reply) {
if (!error) {
- char *name;
-
- name = xasprintf("dp%d", minor);
- dpif_init(&dpif->dpif, &dpif_linux_class, name, minor, minor);
- free(name);
-
- dpif->fd = fd;
- dpif->local_ifname = NULL;
- dpif->minor = minor;
- dpif->local_ifindex = 0;
- svec_init(&dpif->changed_ports);
- dpif->change_error = false;
- *dpifp = &dpif->dpif;
- } else {
- free(dpif);
+ error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
}
- } else {
- error = errno;
- VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
+ if (error) {
+ dpif_linux_vport_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
+ }
+ }
+ return error;
+}
+
+/* Obtains information about the kernel vport named 'name' 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
+dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
+ struct ofpbuf **bufp)
+{
+ struct dpif_linux_vport request;
+
+ dpif_linux_vport_init(&request);
+ 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 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 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 },
+ };
+
+ 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);
+
+ 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;
+ }
+
+ 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[OVS_DP_ATTR_STATS]),
+ sizeof dp->stats);
}
- free(fn);
+ return 0;
+}
+
+/* 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 ovs_header *ovs_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
+ NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
+ OVS_DATAPATH_VERSION);
+
+ ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
+ ovs_header->dp_ifindex = dp->dp_ifindex;
+
+ if (dp->name) {
+ nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
+ }
+
+ if (dp->upcall_pid) {
+ nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
+ }
+
+ /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
+}
+
+/* Clears 'dp' to "empty" values. */
+static void
+dpif_linux_dp_init(struct dpif_linux_dp *dp)
+{
+ memset(dp, 0, sizeof *dp);
+}
+
+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, NETLINK_GENERIC, 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 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 *request_buf;
+ int error;
+
+ ovs_assert((reply != NULL) == (bufp != NULL));
+
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_dp_to_ofpbuf(request, request_buf);
+ error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
+ ofpbuf_delete(request_buf);
+
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
+ }
+ if (error) {
+ dpif_linux_dp_init(reply);
+ ofpbuf_delete(*bufp);
+ *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'). */
+static int
+dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
+ struct ofpbuf **bufp)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp request;
+
+ dpif_linux_dp_init(&request);
+ 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 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'
+ * while 'flow' is still in use. */
+static int
+dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
+ const struct ofpbuf *buf)
+{
+ static const struct nl_policy ovs_flow_policy[] = {
+ [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
+ [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true },
+ [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
+ [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
+ .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 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);
+
+ 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;
+ }
+
+ 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_MASK]) {
+ flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]);
+ flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]);
+ }
+ 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[OVS_FLOW_ATTR_STATS]) {
+ flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
+ }
+ if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
+ flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
+ }
+ 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 ovs_header"
+ * followed by Netlink attributes corresponding to 'flow'. */
static void
-dpif_linux_port_changed(const struct rtnetlink_change *change, void *dpif_)
-{
- 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. */
- if (!svec_contains(&dpif->changed_ports, change->ifname)) {
- svec_add(&dpif->changed_ports, change->ifname);
- svec_sort(&dpif->changed_ports);
- }
+dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
+ struct ofpbuf *buf)
+{
+ 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);
+
+ 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, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
+ }
+
+ if (flow->mask_len) {
+ nl_msg_put_unspec(buf, OVS_FLOW_ATTR_MASK, flow->mask, flow->mask_len);
+ }
+
+ 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. */
+ ovs_assert(!flow->stats);
+ ovs_assert(!flow->tcp_flags);
+ ovs_assert(!flow->used);
+
+ if (flow->clear) {
+ nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
+ }
+}
+
+/* Clears 'flow' to "empty" values. */
+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 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 *request_buf;
+ int error;
+
+ ovs_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_transact(NETLINK_GENERIC, request_buf, bufp);
+ ofpbuf_delete(request_buf);
+
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
}
+ if (error) {
+ dpif_linux_flow_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
+ }
+ }
+ return error;
+}
+
+static void
+dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
+ struct dpif_flow_stats *stats)
+{
+ if (flow->stats) {
+ stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
+ stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
} else {
- dpif->change_error = true;
+ stats->n_packets = 0;
+ stats->n_bytes = 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);
}