#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 <stdlib.h>
-#include <sys/ioctl.h>
#include <sys/stat.h>
#include <unistd.h>
#include "dpif-provider.h"
#include "netdev.h"
#include "netdev-vport.h"
+#include "netlink-socket.h"
#include "netlink.h"
+#include "odp-util.h"
#include "ofpbuf.h"
#include "openvswitch/tunnel.h"
#include "packets.h"
#include "rtnetlink-link.h"
#include "shash.h"
#include "svec.h"
+#include "unaligned.h"
#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
+struct dpif_linux_dp {
+ /* Generic Netlink header. */
+ uint8_t cmd;
+
+ /* struct odp_header. */
+ uint32_t dp_idx;
+
+ /* Attributes. */
+ const char *name; /* ODP_DP_ATTR_NAME. */
+ struct odp_stats stats; /* ODP_DP_ATTR_STATS. */
+ enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
+ const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */
+ uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
+};
+
+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 odp_header. */
+ unsigned int nlmsg_flags;
+ uint32_t dp_idx;
+
+ /* Attributes.
+ *
+ * The 'stats' and 'used' members point to 64-bit data that might only be
+ * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
+ * access their values. */
+ const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
+ size_t key_len;
+ const struct nlattr *actions; /* ODP_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. */
+};
+
+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(const 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 *);
+
/* Datapath interface for the openvswitch Linux kernel module. */
struct dpif_linux {
struct dpif dpif;
- int fd;
+
+ /* Multicast group messages. */
+ struct nl_sock *mc_sock;
+ uint32_t mcgroups[DPIF_N_UC_TYPES];
+ unsigned int listen_mask;
/* Used by dpif_linux_get_all_names(). */
char *local_ifname;
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_internal_device(const char *name, int *dp_idx, int *port_no);
-static int lookup_minor(const char *name, int *minor);
-static int finish_open(struct dpif *, const char *local_ifname);
-static int get_openvswitch_major(void);
-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);
+/* Generic Netlink family numbers for ODP. */
+static int odp_datapath_family;
+static int odp_vport_family;
+static int odp_flow_family;
+static int odp_packet_family;
+
+/* Generic Netlink socket. */
+static struct nl_sock *genl_sock;
+
+static int dpif_linux_init(void);
+static int open_dpif(const struct dpif_linux_dp *,
+ const struct dpif_linux_vport *local_vport,
+ struct dpif **);
static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
void *dpif);
+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)
{
- int major;
+ struct nl_dump dump;
+ struct ofpbuf msg;
int error;
- int i;
- /* 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;
}
- error = 0;
- for (i = 0; i < ODP_MAX; i++) {
- struct dpif *dpif;
- char devname[16];
- int retval;
-
- sprintf(devname, "dp%d", i);
- retval = dpif_open(devname, "system", &dpif);
- if (!retval) {
- svec_add(all_dps, devname);
- dpif_uninit(dpif, true);
- } 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)) {
+ svec_add(all_dps, dp.name);
}
}
- return error;
+ 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 vport_request, vport;
+ struct dpif_linux_dp dp_request, dp;
+ struct ofpbuf *buf;
int minor;
+ int error;
+
+ error = dpif_linux_init();
+ if (error) {
+ return error;
+ }
minor = !strncmp(name, "dp", 2)
&& isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1;
- if (create) {
- if (minor >= 0) {
- return create_minor(name, minor, dpifp);
- } else {
- /* Scan for unused minor number. */
- for (minor = 0; minor < ODP_MAX; minor++) {
- int error = create_minor(name, minor, dpifp);
- if (error != EBUSY) {
- return error;
- }
- }
- /* All datapath numbers in use. */
- return ENOBUFS;
- }
+ /* Create or look up datapath. */
+ dpif_linux_dp_init(&dp_request);
+ dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET;
+ dp_request.dp_idx = minor;
+ dp_request.name = minor < 0 ? name : NULL;
+ error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
+ if (error) {
+ return error;
+ }
+ ofpbuf_delete(buf); /* Pointers inside 'dp' are now invalid! */
+
+ /* Look up local port. */
+ dpif_linux_vport_init(&vport_request);
+ vport_request.cmd = ODP_VPORT_CMD_GET;
+ vport_request.dp_idx = dp.dp_idx;
+ vport_request.port_no = ODPP_LOCAL;
+ vport_request.name = minor < 0 ? name : NULL;
+ error = dpif_linux_vport_transact(&vport_request, &vport, &buf);
+ if (error) {
+ return error;
+ } else if (vport.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 {
- struct dpif_linux *dpif;
- struct odp_port port;
- int error;
-
- if (minor < 0) {
- error = lookup_minor(name, &minor);
- if (error) {
- return error;
- }
- }
+ error = open_dpif(&dp, &vport, 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_VPORT_QUERY, &port)) {
- error = errno;
- if (error != ENODEV) {
- VLOG_WARN("%s: probe returned unexpected error: %s",
- dpif_name(*dpifp), strerror(error));
- }
- dpif_uninit(*dpifp, true);
- return error;
- }
+static int
+open_dpif(const struct dpif_linux_dp *dp,
+ const struct dpif_linux_vport *local_vport, struct dpif **dpifp)
+{
+ int dp_idx = local_vport->dp_idx;
+ struct dpif_linux *dpif;
+ char *name;
+ int error;
+ int i;
- /* Then use that to finish up opening. */
- return finish_open(&dpif->dpif, port.devname);
+ dpif = xmalloc(sizeof *dpif);
+ error = rtnetlink_link_notifier_register(&dpif->port_notifier,
+ dpif_linux_port_changed, dpif);
+ if (error) {
+ goto error_free;
+ }
+
+ name = xasprintf("dp%d", dp_idx);
+ dpif_init(&dpif->dpif, &dpif_linux_class, name, dp_idx, dp_idx);
+ free(name);
+
+ dpif->mc_sock = NULL;
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ dpif->mcgroups[i] = dp->mcgroups[i];
}
+ dpif->listen_mask = 0;
+ 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;
+
+ return 0;
+
+error_free:
+ free(dpif);
+ return error;
}
static void
rtnetlink_link_notifier_unregister(&dpif->port_notifier);
shash_destroy(&dpif->changed_ports);
free(dpif->local_ifname);
- close(dpif->fd);
free(dpif);
}
static int
dpif_linux_destroy(struct dpif *dpif_)
{
- return do_ioctl(dpif_, ODP_DP_DESTROY, NULL);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
+
+ dpif_linux_dp_init(&dp);
+ dp.cmd = ODP_DP_CMD_DEL;
+ dp.dp_idx = dpif->minor;
+ return dpif_linux_dp_transact(&dp, NULL, NULL);
}
static int
dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
{
- memset(stats, 0, sizeof *stats);
- return do_ioctl(dpif_, ODP_DP_STATS, stats);
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
+ if (!error) {
+ *stats = dp.stats;
+ ofpbuf_delete(buf);
+ }
+ return error;
}
static int
dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
{
- int drop_frags;
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
int error;
- error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags);
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
if (!error) {
- *drop_fragsp = drop_frags & 1;
+ *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP;
+ ofpbuf_delete(buf);
}
return error;
}
static int
dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
{
- int drop_frags_int = drop_frags;
- return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int);
-}
-
-static void
-translate_vport_type_to_netdev_type(struct odp_port *port)
-{
- char *type = port->type;
-
- if (!strcmp(type, "netdev")) {
- ovs_strlcpy(type, "system", sizeof port->type);
- } else if (!strcmp(type, "gre")) {
- const struct tnl_port_config *config;
-
- config = (struct tnl_port_config *)port->config;
- if (config->flags & TNL_F_IPSEC) {
- ovs_strlcpy(type, "ipsec_gre", sizeof port->type);
- }
- }
-}
-
-static void
-translate_netdev_type_to_vport_type(struct odp_port *port)
-{
- char *type = port->type;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
- if (!strcmp(type, "system")) {
- ovs_strlcpy(type, "netdev", sizeof port->type);
- } else if (!strcmp(type, "ipsec_gre")) {
- ovs_strlcpy(type, "gre", sizeof port->type);
- }
+ dpif_linux_dp_init(&dp);
+ dp.cmd = ODP_DP_CMD_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,
+dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
uint16_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 odp_port port;
+ struct dpif_linux_vport request, reply;
+ const struct ofpbuf *options;
+ struct ofpbuf *buf;
int error;
- memset(&port, 0, sizeof port);
- strncpy(port.devname, name, sizeof port.devname);
- strncpy(port.type, type, sizeof port.type);
- netdev_vport_get_config(netdev, port.config);
- translate_netdev_type_to_vport_type(&port);
+ dpif_linux_vport_init(&request);
+ request.cmd = ODP_VPORT_CMD_NEW;
+ request.dp_idx = dpif->minor;
+ request.type = netdev_vport_get_vport_type(netdev);
+ if (request.type == ODP_VPORT_TYPE_UNSPEC) {
+ VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
+ "unsupported type `%s'",
+ dpif_name(dpif_), name, type);
+ return EINVAL;
+ }
+ request.name = name;
- error = do_ioctl(dpif, ODP_VPORT_ATTACH, &port);
+ options = netdev_vport_get_options(netdev);
+ if (options && options->size) {
+ request.options = options->data;
+ request.options_len = options->size;
+ }
+
+ error = dpif_linux_vport_transact(&request, &reply, &buf);
if (!error) {
- *port_nop = port.port;
+ *port_nop = reply.port_no;
+ ofpbuf_delete(buf);
}
return error;
}
static int
-dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no_)
+dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
{
- int port_no = port_no_; /* Kernel expects an "int". */
- return do_ioctl(dpif_, ODP_VPORT_DETACH, &port_no);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_vport vport;
+
+ dpif_linux_vport_init(&vport);
+ vport.cmd = ODP_VPORT_CMD_DEL;
+ vport.dp_idx = dpif->minor;
+ vport.port_no = port_no;
+ return dpif_linux_vport_transact(&vport, NULL, NULL);
}
static int
dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
const char *port_name, struct dpif_port *dpif_port)
{
- struct odp_port odp_port;
+ struct dpif_linux_vport request;
+ struct dpif_linux_vport reply;
+ struct ofpbuf *buf;
int error;
- memset(&odp_port, 0, sizeof odp_port);
- odp_port.port = port_no;
- strncpy(odp_port.devname, port_name, sizeof odp_port.devname);
+ dpif_linux_vport_init(&request);
+ request.cmd = ODP_VPORT_CMD_GET;
+ request.dp_idx = dpif_linux_cast(dpif)->minor;
+ request.port_no = port_no;
+ request.name = port_name;
- error = do_ioctl(dpif, ODP_VPORT_QUERY, &odp_port);
- if (error) {
- return error;
- } else if (odp_port.dp_idx != dpif_linux_cast(dpif)->minor) {
- /* A vport named 'port_name' exists but in some other datapath. */
- return ENOENT;
- } else {
- translate_vport_type_to_netdev_type(&odp_port);
- dpif_port->name = xstrdup(odp_port.devname);
- dpif_port->type = xstrdup(odp_port.type);
- dpif_port->port_no = odp_port.port;
- return 0;
+ 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;
+ ofpbuf_delete(buf);
}
+ return error;
}
static int
dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
struct dpif_port *dpif_port)
{
- return dpif_linux_port_query__(dpif, port_no, "", dpif_port);
+ return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
}
static int
return dpif_linux_port_query__(dpif, 0, devname, 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;
+}
+
static int
dpif_linux_flow_flush(struct dpif *dpif_)
{
- return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow flow;
+
+ dpif_linux_flow_init(&flow);
+ flow.cmd = ODP_FLOW_CMD_DEL;
+ flow.dp_idx = dpif->minor;
+ return dpif_linux_flow_transact(&flow, NULL, NULL);
}
+struct dpif_linux_port_state {
+ 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 odp_port));
+ 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 = ODP_DP_CMD_GET;
+ request.dp_idx = dpif->minor;
+
+ 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 odp_port *odp_port = state;
- struct odp_vport_dump dump;
+ struct dpif_linux_port_state *state = state_;
+ struct dpif_linux_vport vport;
+ struct ofpbuf buf;
int error;
- dump.port = odp_port;
- dump.port_no = odp_port->port;
- error = do_ioctl(dpif, ODP_VPORT_DUMP, &dump);
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
+ }
+
+ error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
if (error) {
return error;
- } else if (odp_port->devname[0] == '\0') {
- return EOF;
- } else {
- translate_vport_type_to_netdev_type(odp_port);
- dpif_port->name = odp_port->devname;
- dpif_port->type = odp_port->type;
- dpif_port->port_no = odp_port->port;
- odp_port->port++;
- return 0;
}
+
+ dpif_port->name = (char *) vport.name;
+ dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
+ dpif_port->port_no = vport.port_no;
+ return 0;
}
static int
-dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state)
+dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
{
+ struct dpif_linux_port_state *state = state_;
+ int error = nl_dump_done(&state->dump);
free(state);
- return 0;
+ return error;
}
static int
}
static int
-dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n)
+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 odp_flowvec fv;
- fv.flows = flows;
- fv.n_flows = n;
- return do_ioctl(dpif_, ODP_FLOW_GET, &fv);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request, reply;
+ struct ofpbuf *buf;
+ int error;
+
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_FLOW_CMD_GET;
+ request.dp_idx = dpif->minor;
+ request.key = key;
+ request.key_len = key_len;
+ error = dpif_linux_flow_transact(&request, &reply, &buf);
+ if (!error) {
+ if (stats) {
+ dpif_linux_flow_get_stats(&reply, stats);
+ }
+ if (actionsp) {
+ buf->data = (void *) reply.actions;
+ buf->size = reply.actions_len;
+ *actionsp = buf;
+ } else {
+ ofpbuf_delete(buf);
+ }
+ }
+ return error;
}
static int
-dpif_linux_flow_put(struct dpif *dpif_, struct odp_flow_put *put)
+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)
{
- return do_ioctl(dpif_, ODP_FLOW_PUT, put);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request, reply;
+ struct ofpbuf *buf;
+ int error;
+
+ dpif_linux_flow_init(&request);
+ request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
+ request.dp_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;
+ error = dpif_linux_flow_transact(&request,
+ stats ? &reply : NULL,
+ stats ? &buf : NULL);
+ if (!error && stats) {
+ dpif_linux_flow_get_stats(&reply, stats);
+ ofpbuf_delete(buf);
+ }
+ return error;
}
static int
-dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow)
+dpif_linux_flow_del(struct dpif *dpif_,
+ const struct nlattr *key, size_t key_len,
+ struct dpif_flow_stats *stats)
{
- return do_ioctl(dpif_, ODP_FLOW_DEL, flow);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request, reply;
+ struct ofpbuf *buf;
+ int error;
+
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_FLOW_CMD_DEL;
+ request.dp_idx = dpif->minor;
+ request.key = key;
+ request.key_len = key_len;
+ error = dpif_linux_flow_transact(&request,
+ stats ? &reply : NULL,
+ stats ? &buf : NULL);
+ if (!error && stats) {
+ dpif_linux_flow_get_stats(&reply, 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;
+};
+
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 odp_flow_dump));
+ 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 = ODP_DP_CMD_GET;
+ request.dp_idx = dpif->minor;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, genl_sock, buf);
+ ofpbuf_delete(buf);
+
return 0;
}
static int
-dpif_linux_flow_dump_next(const struct dpif *dpif, void *state,
- struct odp_flow *flow)
+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 odp_flow_dump *dump = state;
+ struct dpif_linux_flow_state *state = state_;
+ struct ofpbuf buf;
int error;
- dump->flow = flow;
- error = do_ioctl(dpif, ODP_FLOW_DUMP, dump);
- return error ? error : flow->flags & ODPFF_EOF ? EOF : 0;
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
+ }
+
+ error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
+ if (!error) {
+ if (key) {
+ *key = state->flow.key;
+ *key_len = state->flow.key_len;
+ }
+ if (actions) {
+ *actions = state->flow.actions;
+ *actions_len = state->flow.actions_len;
+ }
+ 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)
+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);
free(state);
- return 0;
+ return error;
}
static int
dpif_linux_execute(struct dpif *dpif_,
const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *buf)
+ const struct ofpbuf *packet)
{
- struct odp_execute execute;
- memset(&execute, 0, sizeof execute);
- execute.actions = (struct nlattr *) actions;
- execute.actions_len = actions_len;
- execute.data = buf->data;
- execute.length = buf->size;
- return do_ioctl(dpif_, ODP_EXECUTE, &execute);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct odp_header *execute;
+ struct ofpbuf *buf;
+ int error;
+
+ buf = ofpbuf_new(128 + actions_len + packet->size);
+
+ nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
+ ODP_PACKET_CMD_EXECUTE, 1);
+
+ execute = ofpbuf_put_uninit(buf, sizeof *execute);
+ execute->dp_idx = dpif->minor;
+
+ 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);
+
+ error = nl_sock_transact(genl_sock, buf, NULL);
+ ofpbuf_delete(buf);
+ return error;
}
static int
dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
{
- return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ *listen_mask = dpif->listen_mask;
+ return 0;
}
static int
dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
{
- return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int error;
+ int i;
+
+ if (listen_mask == dpif->listen_mask) {
+ return 0;
+ } else if (!listen_mask) {
+ nl_sock_destroy(dpif->mc_sock);
+ dpif->mc_sock = NULL;
+ dpif->listen_mask = 0;
+ return 0;
+ } else if (!dpif->mc_sock) {
+ error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
+ if (error) {
+ return error;
+ }
+ }
+
+ /* Unsubscribe from old groups. */
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ if (dpif->listen_mask & (1u << i)) {
+ nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
+ }
+ }
+
+ /* Update listen_mask. */
+ dpif->listen_mask = listen_mask;
+
+ /* Subscribe to new groups. */
+ error = 0;
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ if (dpif->listen_mask & (1u << i)) {
+ int retval;
+
+ retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
+ if (retval) {
+ error = retval;
+ }
+ }
+ }
+ return error;
}
static int
dpif_linux_get_sflow_probability(const struct dpif *dpif_,
uint32_t *probability)
{
- return do_ioctl(dpif_, ODP_GET_SFLOW_PROBABILITY, probability);
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
+ if (!error) {
+ *probability = dp.sampling ? *dp.sampling : 0;
+ ofpbuf_delete(buf);
+ }
+ return error;
}
static int
dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
{
- return do_ioctl(dpif_, ODP_SET_SFLOW_PROBABILITY, &probability);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
+
+ dpif_linux_dp_init(&dp);
+ dp.cmd = ODP_DP_CMD_SET;
+ dp.dp_idx = dpif->minor;
+ dp.sampling = &probability;
+ return dpif_linux_dp_transact(&dp, NULL, NULL);
}
static int
}
static int
-parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall)
+parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
+ uint32_t *dp_idx)
{
static const struct nl_policy odp_packet_policy[] = {
/* Always present. */
- [ODP_PACKET_ATTR_TYPE] = { .type = NL_A_U32 },
[ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
.min_len = ETH_HEADER_LEN },
[ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
- /* _ODPL_ACTION_NR only. */
+ /* ODP_PACKET_CMD_ACTION only. */
[ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
- /* _ODPL_SFLOW_NR only. */
+ /* ODP_PACKET_CMD_SAMPLE only. */
[ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
[ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
};
- struct odp_packet *odp_packet = buf->data;
+ struct odp_header *odp_header;
struct nlattr *a[ARRAY_SIZE(odp_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);
+ odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
+ if (!nlmsg || !genl || !odp_header
+ || nlmsg->nlmsg_type != odp_packet_family
+ || !nl_policy_parse(&b, 0, odp_packet_policy, a,
+ ARRAY_SIZE(odp_packet_policy))) {
+ return EINVAL;
+ }
- if (!nl_policy_parse(buf, sizeof *odp_packet, odp_packet_policy,
- a, ARRAY_SIZE(odp_packet_policy))) {
+ type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
+ : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
+ : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
+ : -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->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
}
+ *dp_idx = odp_header->dp_idx;
+
return 0;
}
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct ofpbuf *buf;
- int retval;
int error;
+ int i;
+
+ if (!dpif->mc_sock) {
+ return EAGAIN;
+ }
+
+ for (i = 0; i < 50; i++) {
+ uint32_t dp_idx;
- 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));
+ error = nl_sock_recv(dpif->mc_sock, &buf, false);
+ if (error) {
+ return error;
}
- } else if (retval >= sizeof(struct odp_packet)) {
- struct odp_packet *odp_packet = buf->data;
- buf->size += retval;
- 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;
+ error = parse_odp_packet(buf, upcall, &dp_idx);
+ if (!error
+ && dp_idx == dpif->minor
+ && dpif->listen_mask & (1u << upcall->type)) {
+ return 0;
}
- } 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);
+ 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->mc_sock) {
+ nl_sock_wait(dpif->mc_sock, POLLIN);
+ }
+}
+
+static void
+dpif_linux_recv_purge(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ if (dpif->mc_sock) {
+ nl_sock_drain(dpif->mc_sock);
+ }
}
const struct dpif_class dpif_linux_class = {
"system",
- NULL,
- NULL,
+ NULL, /* run */
+ NULL, /* wait */
dpif_linux_enumerate,
dpif_linux_open,
dpif_linux_close,
dpif_linux_port_del,
dpif_linux_port_query_by_number,
dpif_linux_port_query_by_name,
+ dpif_linux_get_max_ports,
dpif_linux_port_dump_start,
dpif_linux_port_dump_next,
dpif_linux_port_dump_done,
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) {
+ error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
+ &odp_datapath_family);
+ if (error) {
+ VLOG_ERR("Generic Netlink family '%s' does not exist. "
+ "The Open vSwitch kernel module is probably not loaded.",
+ ODP_DATAPATH_FAMILY);
+ }
+ if (!error) {
+ error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family);
+ }
+ if (!error) {
+ error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family);
+ }
+ if (!error) {
+ error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
+ &odp_packet_family);
+ }
+ if (!error) {
+ error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
+ }
+ }
+
+ return error;
}
-static int
-lookup_internal_device(const char *name, int *dp_idx, int *port_no)
+bool
+dpif_linux_is_internal_device(const char *name)
{
- struct odp_port odp_port;
- static int dp0_fd = -1;
+ struct dpif_linux_vport reply;
+ struct ofpbuf *buf;
+ int error;
- if (dp0_fd < 0) {
- int error;
- char *fn;
+ error = dpif_linux_vport_get(name, &reply, &buf);
+ if (!error) {
+ ofpbuf_delete(buf);
+ } else if (error != ENODEV) {
+ VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
+ name, strerror(error));
+ }
- error = make_openvswitch_device(0, &fn);
- if (error) {
- return error;
- }
+ return reply.type == ODP_VPORT_TYPE_INTERNAL;
+}
- dp0_fd = open(fn, O_RDONLY | O_NONBLOCK);
- if (dp0_fd < 0) {
- VLOG_WARN_RL(&error_rl, "%s: open failed (%s)",
- fn, strerror(errno));
- free(fn);
- return errno;
- }
- free(fn);
- }
+static void
+dpif_linux_port_changed(const struct rtnetlink_link_change *change,
+ void *dpif_)
+{
+ struct dpif_linux *dpif = dpif_;
- memset(&odp_port, 0, sizeof odp_port);
- strncpy(odp_port.devname, name, sizeof odp_port.devname);
- if (ioctl(dp0_fd, ODP_VPORT_QUERY, &odp_port)) {
- if (errno != ENODEV) {
- VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
- name, strerror(errno));
+ 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);
}
- return errno;
- } else if (!strcmp(odp_port.type, "internal")) {
- *dp_idx = odp_port.dp_idx;
- *port_no = odp_port.port;
- return 0;
} else {
+ dpif->change_error = true;
+ }
+}
+\f
+/* Parses the contents of 'buf', which contains a "struct odp_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
+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),
+ .optional = true },
+ [ODP_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 },
+ };
+
+ struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
+ struct odp_header *odp_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);
+ odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
+ if (!nlmsg || !genl || !odp_header
+ || nlmsg->nlmsg_type != odp_vport_family
+ || !nl_policy_parse(&b, 0, odp_vport_policy, a,
+ ARRAY_SIZE(odp_vport_policy))) {
return EINVAL;
}
+
+ vport->cmd = genl->cmd;
+ vport->dp_idx = odp_header->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]);
+ }
+ if (a[ODP_VPORT_ATTR_MTU]) {
+ vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
+ }
+ 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[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]);
+ }
+ return 0;
}
-static int
-lookup_minor(const char *name, int *minorp)
+/* Appends to 'buf' (which must initially be empty) a "struct odp_header"
+ * followed by Netlink attributes corresponding to 'vport'. */
+static void
+dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
+ struct ofpbuf *buf)
{
- int minor, port_no;
- int error;
+ struct odp_header *odp_header;
- error = lookup_internal_device(name, &minor, &port_no);
- if (error) {
- return error;
- } 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. */
- return EOPNOTSUPP;
- } else {
- *minorp = minor;
- return 0;
+ nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
+ vport->cmd, 1);
+
+ odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
+ odp_header->dp_idx = vport->dp_idx;
+
+ if (vport->port_no != UINT32_MAX) {
+ nl_msg_put_u32(buf, ODP_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->name) {
+ nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name);
+ }
+
+ if (vport->stats) {
+ nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS,
+ vport->stats, sizeof *vport->stats);
+ }
+
+ if (vport->address) {
+ nl_msg_put_unspec(buf, ODP_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,
+ 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);
}
}
-bool
-dpif_linux_is_internal_device(const char *name)
+/* Clears 'vport' to "empty" values. */
+void
+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
+ * 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)
{
- int minor, port_no;
+ struct ofpbuf *request_buf;
+ int error;
+
+ assert((reply != NULL) == (bufp != NULL));
- return !lookup_internal_device(name, &minor, &port_no);
+ 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 (reply) {
+ if (!error) {
+ error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
+ }
+ 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 = ODP_VPORT_CMD_GET;
+ request.name = name;
+
+ return dpif_linux_vport_transact(&request, reply, bufp);
+}
+\f
+/* Parses the contents of 'buf', which contains a "struct odp_header" followed
+ * 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
-make_openvswitch_device(int minor, char **fnp)
+dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
{
- const char dirname[] = "/dev/net";
- int major;
- dev_t dev;
- struct stat s;
- char fn[128];
+ 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),
+ .optional = true },
+ [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
+ [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
+ [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
+ };
- *fnp = NULL;
+ struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
+ struct odp_header *odp_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);
+ odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
+ if (!nlmsg || !genl || !odp_header
+ || nlmsg->nlmsg_type != odp_datapath_family
+ || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
+ ARRAY_SIZE(odp_datapath_policy))) {
+ return EINVAL;
+ }
- major = get_openvswitch_major();
- if (major < 0) {
- return -major;
+ dp->cmd = genl->cmd;
+ dp->dp_idx = odp_header->dp_idx;
+ dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
+ if (a[ODP_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]),
+ 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]);
}
- 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 (a[ODP_DP_ATTR_MCGROUPS]) {
+ static const struct nl_policy odp_mcgroup_policy[] = {
+ [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
+ };
+
+ struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
+
+ if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
+ mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
+ return EINVAL;
}
- if (unlink(fn)) {
- VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
- fn, strerror(errno));
- return errno;
+
+ if (mcgroups[ODP_PACKET_CMD_MISS]) {
+ dp->mcgroups[DPIF_UC_MISS]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
}
- } 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;
- }
+ if (mcgroups[ODP_PACKET_CMD_ACTION]) {
+ dp->mcgroups[DPIF_UC_ACTION]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
+ }
+ if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
+ dp->mcgroups[DPIF_UC_SAMPLE]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
}
- } 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)
+/* 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)
{
- static int openvswitch_major = -1;
- if (openvswitch_major < 0) {
- openvswitch_major = get_major("openvswitch");
+ struct odp_header *odp_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
+ NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
+
+ odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
+ odp_header->dp_idx = dp->dp_idx;
+
+ if (dp->name) {
+ nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
+ }
+
+ /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
+
+ if (dp->ipv4_frags) {
+ nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
+ }
+
+ if (dp->sampling) {
+ nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
}
- return openvswitch_major;
}
-static int
-get_major(const char *target)
+/* Clears 'dp' to "empty" values. */
+void
+dpif_linux_dp_init(struct dpif_linux_dp *dp)
{
- const char fn[] = "/proc/devices";
- char line[128];
- FILE *file;
- int ln;
+ memset(dp, 0, sizeof *dp);
+ dp->dp_idx = -1;
+}
- file = fopen(fn, "r");
- if (!file) {
- VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
- return -errno;
- }
+static void
+dpif_linux_dp_dump_start(struct nl_dump *dump)
+{
+ struct dpif_linux_dp request;
+ struct ofpbuf *buf;
- for (ln = 1; fgets(line, sizeof line, file); ln++) {
- char name[64];
- int major;
+ dpif_linux_dp_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
- 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);
+ 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 of the same form, which is decoded
+ * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
+ * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
+int
+dpif_linux_dp_transact(const struct dpif_linux_dp *request,
+ struct dpif_linux_dp *reply, struct ofpbuf **bufp)
+{
+ struct ofpbuf *request_buf;
+ int error;
+
+ assert((reply != NULL) == (bufp != NULL));
+
+ 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 (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'). */
+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;
- fclose(file);
+ dpif_linux_dp_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
+ request.dp_idx = dpif->minor;
- VLOG_ERR("%s: %s major not found (is the module loaded?)", fn, target);
- return -ENODEV;
+ return dpif_linux_dp_transact(&request, reply, bufp);
}
-
+\f
+/* Parses the contents of 'buf', which contains a "struct odp_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
-finish_open(struct dpif *dpif_, const char *local_ifname)
+dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
+ const struct ofpbuf *buf)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- dpif->local_ifname = xstrdup(local_ifname);
- dpif->local_ifindex = if_nametoindex(local_ifname);
- if (!dpif->local_ifindex) {
- int error = errno;
- dpif_uninit(dpif_, true);
- VLOG_WARN("could not get ifindex of %s device: %s",
- local_ifname, strerror(errno));
- return error;
+ 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),
+ .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. */
+ };
+
+ struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
+ struct odp_header *odp_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);
+ odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
+ if (!nlmsg || !genl || !odp_header
+ || nlmsg->nlmsg_type != odp_flow_family
+ || !nl_policy_parse(&b, 0, odp_flow_policy, a,
+ ARRAY_SIZE(odp_flow_policy))) {
+ return EINVAL;
+ }
+
+ flow->nlmsg_flags = nlmsg->nlmsg_flags;
+ flow->dp_idx = odp_header->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]);
+ }
+ if (a[ODP_FLOW_ATTR_STATS]) {
+ flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
+ }
+ if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
+ flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
}
return 0;
}
-static int
-create_minor(const char *name, int minor, struct dpif **dpifp)
+/* Appends to 'buf' (which must initially be empty) a "struct odp_header"
+ * followed by Netlink attributes corresponding to 'flow'. */
+static void
+dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
+ struct ofpbuf *buf)
{
- int error = open_minor(minor, dpifp);
- if (!error) {
- error = do_ioctl(*dpifp, ODP_DP_CREATE, name);
- if (!error) {
- error = finish_open(*dpifp, name);
- } else {
- dpif_uninit(*dpifp, true);
- }
+ struct odp_header *odp_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
+ NLM_F_REQUEST | flow->nlmsg_flags, flow->cmd, 1);
+
+ odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
+ odp_header->dp_idx = flow->dp_idx;
+
+ if (flow->key_len) {
+ nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
+ }
+
+ if (flow->actions_len) {
+ nl_msg_put_unspec(buf, ODP_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);
+
+ if (flow->clear) {
+ nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
}
- return error;
}
-static int
-open_minor(int minor, struct dpif **dpifp)
+/* Clears 'flow' to "empty" values. */
+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'). */
+int
+dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+ struct dpif_linux_flow *reply, struct ofpbuf **bufp)
{
+ struct ofpbuf *request_buf;
int error;
- char *fn;
- int fd;
- error = make_openvswitch_device(minor, &fn);
- if (error) {
- return error;
- }
+ assert((reply != NULL) == (bufp != NULL));
- fd = open(fn, O_RDONLY | O_NONBLOCK);
- if (fd >= 0) {
- struct dpif_linux *dpif = xmalloc(sizeof *dpif);
- error = rtnetlink_link_notifier_register(&dpif->port_notifier,
- dpif_linux_port_changed,
- dpif);
+ 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 (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;
- shash_init(&dpif->changed_ports);
- dpif->change_error = false;
- *dpifp = &dpif->dpif;
- } else {
- free(dpif);
+ error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
+ }
+ if (error) {
+ dpif_linux_flow_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- } else {
- error = errno;
- VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
}
- free(fn);
-
return error;
}
static void
-dpif_linux_port_changed(const struct rtnetlink_link_change *change,
- void *dpif_)
+dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
+ struct dpif_flow_stats *stats)
{
- 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 (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_unaligned_u64(flow->used) : 0;
+ stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
}
+