#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"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
struct dpif_linux_dp {
- /* ioctl command argument. */
- int cmd;
+ /* Generic Netlink header. */
+ uint8_t cmd;
- /* struct odp_datapath header. */
- uint32_t dp_idx;
+ /* struct odp_header. */
+ int dp_ifindex;
/* Attributes. */
const char *name; /* ODP_DP_ATTR_NAME. */
struct odp_stats stats; /* ODP_DP_ATTR_STATS. */
enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */
+ uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
};
static void dpif_linux_dp_init(struct dpif_linux_dp *);
+static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
+ const struct ofpbuf *);
+static void dpif_linux_dp_dump_start(struct nl_dump *);
static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
struct dpif_linux_dp *reply,
struct ofpbuf **bufp);
struct ofpbuf **bufp);
struct dpif_linux_flow {
- /* ioctl command argument. */
- int cmd;
+ /* Generic Netlink header. */
+ uint8_t cmd;
- /* struct odp_flow header. */
+ /* struct odp_header. */
unsigned int nlmsg_flags;
- uint32_t dp_idx;
+ int dp_ifindex;
/* Attributes.
*
- * The 'stats', 'used', and 'state' members point to 64-bit data that might
- * only be aligned on 32-bit boundaries, so get_unaligned_u64() should be
- * used to access their values. */
+ * 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.
+ *
+ * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in
+ * the Netlink version of the command, even if actions_len is zero. */
const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
size_t key_len;
const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */
const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */
const uint64_t *used; /* ODP_FLOW_ATTR_USED. */
bool clear; /* ODP_FLOW_ATTR_CLEAR. */
- const uint64_t *state; /* ODP_FLOW_ATTR_STATE. */
};
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);
/* 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;
+ /* Multicast group messages. */
+ struct nl_sock *mc_sock;
+ uint32_t mcgroups[DPIF_N_UC_TYPES];
+ unsigned int listen_mask;
/* Change notification. */
- int local_ifindex; /* Ifindex of local port. */
struct shash changed_ports; /* Ports that have changed. */
struct rtnetlink_notifier port_notifier;
bool change_error;
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
-static int do_ioctl(const struct dpif *, int cmd, const void *arg);
-static int open_dpif(const struct dpif_linux_vport *local_vport,
- struct dpif **);
-static int get_openvswitch_major(void);
-static int open_minor(int minor, int *fdp);
-static int make_openvswitch_device(int minor, char **fnp);
+/* 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 *, 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)
{
- uint32_t dp_idx;
- int major;
+ struct nl_dump dump;
+ struct ofpbuf msg;
+ int error;
- /* Check that the Open vSwitch module is loaded. */
- major = get_openvswitch_major();
- if (major < 0) {
- return -major;
+ error = dpif_linux_init();
+ if (error) {
+ return error;
}
- dp_idx = 0;
- for (;;) {
- struct dpif_linux_dp request, reply;
- struct ofpbuf *buf;
- char devname[16];
- int error;
-
- dpif_linux_dp_init(&request);
- request.dp_idx = dp_idx;
- request.cmd = ODP_DP_DUMP;
+ dpif_linux_dp_dump_start(&dump);
+ while (nl_dump_next(&dump, &msg)) {
+ struct dpif_linux_dp dp;
- error = dpif_linux_dp_transact(&request, &reply, &buf);
- if (error) {
- return error == ENODEV ? 0 : error;
+ if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
+ svec_add(all_dps, dp.name);
}
- ofpbuf_delete(buf);
-
- sprintf(devname, "dp%d", reply.dp_idx);
- svec_add(all_dps, devname);
-
- dp_idx = reply.dp_idx + 1;
}
+ return nl_dump_done(&dump);
}
static int
dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
bool create, struct dpif **dpifp)
{
- struct dpif_linux_vport request, reply;
+ struct dpif_linux_dp dp_request, dp;
struct ofpbuf *buf;
- int minor;
int error;
- minor = !strncmp(name, "dp", 2)
- && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1;
- if (create) {
- struct dpif_linux_dp request, reply;
- struct ofpbuf *buf;
- int error;
-
- dpif_linux_dp_init(&request);
- request.cmd = ODP_DP_NEW;
- request.dp_idx = minor;
- request.name = name;
- error = dpif_linux_dp_transact(&request, &reply, &buf);
- if (error) {
- return error;
- }
- minor = reply.dp_idx;
- ofpbuf_delete(buf);
- }
-
- dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_GET;
- request.port_no = ODPP_LOCAL;
- if (minor >= 0) {
- request.dp_idx = minor;
- } else {
- request.name = name;
+ error = dpif_linux_init();
+ if (error) {
+ return error;
}
- error = dpif_linux_vport_transact(&request, &reply, &buf);
+ /* 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.name = name;
+ error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
if (error) {
return error;
- } else if (reply.port_no != ODPP_LOCAL) {
- /* This is an Open vSwitch device but not the local port. We
- * intentionally support only using the name of the local port as the
- * name of a datapath; otherwise, it would be too difficult to
- * enumerate all the names of a datapath. */
- error = EOPNOTSUPP;
- } else {
- error = open_dpif(&reply, dpifp);
}
-
+ error = open_dpif(&dp, dpifp);
ofpbuf_delete(buf);
+
return error;
}
static int
-open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **dpifp)
+open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
{
- int dp_idx = local_vport->dp_idx;
struct dpif_linux *dpif;
- char *name;
int error;
- int fd;
-
- error = open_minor(dp_idx, &fd);
- if (error) {
- goto error;
- }
+ int i;
dpif = xmalloc(sizeof *dpif);
error = rtnetlink_link_notifier_register(&dpif->port_notifier,
goto error_free;
}
- name = xasprintf("dp%d", dp_idx);
- dpif_init(&dpif->dpif, &dpif_linux_class, name, dp_idx, dp_idx);
- free(name);
+ dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
+ dp->dp_ifindex, dp->dp_ifindex);
- dpif->fd = fd;
- dpif->local_ifname = xstrdup(local_vport->name);
- dpif->local_ifindex = local_vport->ifindex;
- dpif->minor = dp_idx;
+ dpif->mc_sock = NULL;
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ dpif->mcgroups[i] = dp->mcgroups[i];
+ }
+ dpif->listen_mask = 0;
+ dpif->dp_ifindex = dp->dp_ifindex;
shash_init(&dpif->changed_ports);
dpif->change_error = false;
*dpifp = &dpif->dpif;
error_free:
free(dpif);
- close(fd);
-error:
return error;
}
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
rtnetlink_link_notifier_unregister(&dpif->port_notifier);
shash_destroy(&dpif->changed_ports);
- free(dpif->local_ifname);
- close(dpif->fd);
free(dpif);
}
-static int
-dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
-
- svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor));
- svec_add(all_names, dpif->local_ifname);
- return 0;
-}
-
static int
dpif_linux_destroy(struct dpif *dpif_)
{
struct dpif_linux_dp dp;
dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_DEL;
- dp.dp_idx = dpif->minor;
+ dp.cmd = ODP_DP_CMD_DEL;
+ dp.dp_ifindex = dpif->dp_ifindex;
return dpif_linux_dp_transact(&dp, NULL, NULL);
}
struct dpif_linux_dp dp;
dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_SET;
- dp.dp_idx = dpif->minor;
+ dp.cmd = ODP_DP_CMD_SET;
+ dp.dp_ifindex = dpif->dp_ifindex;
dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO;
return dpif_linux_dp_transact(&dp, NULL, NULL);
}
int error;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_NEW;
- request.dp_idx = dpif->minor;
+ request.cmd = ODP_VPORT_CMD_NEW;
+ request.dp_ifindex = dpif->dp_ifindex;
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 "
struct dpif_linux_vport vport;
dpif_linux_vport_init(&vport);
- vport.cmd = ODP_VPORT_DEL;
- vport.dp_idx = dpif->minor;
+ vport.cmd = ODP_VPORT_CMD_DEL;
+ vport.dp_ifindex = dpif->dp_ifindex;
vport.port_no = port_no;
return dpif_linux_vport_transact(&vport, NULL, NULL);
}
int error;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_GET;
- request.dp_idx = dpif_linux_cast(dpif)->minor;
+ request.cmd = ODP_VPORT_CMD_GET;
+ request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
request.port_no = port_no;
request.name = port_name;
dpif_linux_flow_flush(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return ioctl(dpif->fd, ODP_FLOW_FLUSH, dpif->minor) ? errno : 0;
+ struct dpif_linux_flow flow;
+
+ dpif_linux_flow_init(&flow);
+ flow.cmd = ODP_FLOW_CMD_DEL;
+ flow.dp_ifindex = dpif->dp_ifindex;
+ return dpif_linux_flow_transact(&flow, NULL, NULL);
}
struct dpif_linux_port_state {
- struct ofpbuf *buf;
- uint32_t next;
+ struct nl_dump dump;
};
static int
-dpif_linux_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
+dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
{
- *statep = xzalloc(sizeof(struct dpif_linux_port_state));
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_port_state *state;
+ struct dpif_linux_vport request;
+ struct ofpbuf *buf;
+
+ *statep = state = xmalloc(sizeof *state);
+
+ dpif_linux_vport_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_vport_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, genl_sock, buf);
+ ofpbuf_delete(buf);
+
return 0;
}
static int
-dpif_linux_port_dump_next(const struct dpif *dpif, void *state_,
+dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
struct dpif_port *dpif_port)
{
struct dpif_linux_port_state *state = state_;
- struct dpif_linux_vport request, reply;
- struct ofpbuf *buf;
+ struct dpif_linux_vport vport;
+ struct ofpbuf buf;
int error;
- ofpbuf_delete(state->buf);
- state->buf = NULL;
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
+ }
- dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_DUMP;
- request.dp_idx = dpif_linux_cast(dpif)->minor;
- request.port_no = state->next;
- error = dpif_linux_vport_transact(&request, &reply, &buf);
+ error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
if (error) {
- return error == ENODEV ? EOF : error;
- } else {
- dpif_port->name = (char *) reply.name;
- dpif_port->type = (char *) netdev_vport_get_netdev_type(&reply);
- dpif_port->port_no = reply.port_no;
- state->buf = buf;
- state->next = reply.port_no + 1;
- return 0;
+ return error;
}
+
+ dpif_port->name = (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_)
{
struct dpif_linux_port_state *state = state_;
- ofpbuf_delete(state->buf);
+ int error = nl_dump_done(&state->dump);
free(state);
- return 0;
+ return error;
}
static int
}
}
+static int
+dpif_linux_flow_get__(const struct dpif *dpif_,
+ const struct nlattr *key, size_t key_len,
+ struct dpif_linux_flow *reply, struct ofpbuf **bufp)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request;
+
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_FLOW_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
+ request.key = key;
+ request.key_len = key_len;
+ return dpif_linux_flow_transact(&request, reply, bufp);
+}
+
static int
dpif_linux_flow_get(const struct dpif *dpif_,
const struct nlattr *key, size_t key_len,
struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_flow request, reply;
+ struct dpif_linux_flow reply;
struct ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_GET;
- request.dp_idx = dpif->minor;
- request.key = key;
- request.key_len = key_len;
- error = dpif_linux_flow_transact(&request, &reply, &buf);
+ error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
if (!error) {
if (stats) {
dpif_linux_flow_get_stats(&reply, stats);
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_flow request, reply;
+ struct nlattr dummy_action;
struct ofpbuf *buf;
int error;
dpif_linux_flow_init(&request);
- request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_NEW : ODP_FLOW_SET;
- request.dp_idx = dpif->minor;
+ request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
+ request.dp_ifindex = dpif->dp_ifindex;
request.key = key;
request.key_len = key_len;
- request.actions = actions;
+ /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */
+ request.actions = actions ? actions : &dummy_action;
request.actions_len = actions_len;
if (flags & DPIF_FP_ZERO_STATS) {
request.clear = true;
int error;
dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_DEL;
- request.dp_idx = dpif->minor;
+ request.cmd = ODP_FLOW_CMD_DEL;
+ request.dp_ifindex = dpif->dp_ifindex;
request.key = key;
request.key_len = key_len;
error = dpif_linux_flow_transact(&request,
return error;
}
-
struct dpif_linux_flow_state {
+ struct nl_dump dump;
struct dpif_linux_flow flow;
- struct ofpbuf *buf;
struct dpif_flow_stats stats;
+ struct ofpbuf *buf;
};
static int
-dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
+dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
{
- *statep = xzalloc(sizeof(struct dpif_linux_flow_state));
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow_state *state;
+ struct dpif_linux_flow request;
+ struct ofpbuf *buf;
+
+ *statep = state = xmalloc(sizeof *state);
+
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(&request, buf);
+ nl_dump_start(&state->dump, genl_sock, buf);
+ ofpbuf_delete(buf);
+
+ state->buf = NULL;
+
return 0;
}
static int
-dpif_linux_flow_dump_next(const struct dpif *dpif_, void *state_,
+dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
const struct nlattr **key, size_t *key_len,
const struct nlattr **actions, size_t *actions_len,
const struct dpif_flow_stats **stats)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_flow_state *state = state_;
- struct ofpbuf *old_buf = state->buf;
- struct dpif_linux_flow request;
+ struct ofpbuf buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = ODP_FLOW_DUMP;
- request.dp_idx = dpif->minor;
- request.state = state->flow.state;
- error = dpif_linux_flow_transact(&request, &state->flow, &state->buf);
- ofpbuf_delete(old_buf);
+ do {
+ ofpbuf_delete(state->buf);
+ state->buf = NULL;
- if (!error) {
- if (key) {
- *key = state->flow.key;
- *key_len = state->flow.key_len;
+ if (!nl_dump_next(&state->dump, &buf)) {
+ return EOF;
}
- if (actions) {
- *actions = state->flow.actions;
- *actions_len = state->flow.actions_len;
+
+ error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
+ if (error) {
+ return error;
}
- if (stats) {
- dpif_linux_flow_get_stats(&state->flow, &state->stats);
- *stats = &state->stats;
+
+ if (actions && !state->flow.actions) {
+ error = dpif_linux_flow_get__(dpif_, state->flow.key,
+ state->flow.key_len,
+ &state->flow, &state->buf);
+ if (error == ENOENT) {
+ VLOG_DBG("dumped flow disappeared on get");
+ } else if (error) {
+ VLOG_WARN("error fetching dumped flow: %s", strerror(error));
+ }
}
+ } while (error);
+
+ if (actions) {
+ *actions = state->flow.actions;
+ *actions_len = state->flow.actions_len;
+ }
+ if (key) {
+ *key = state->flow.key;
+ *key_len = state->flow.key_len;
+ }
+ if (stats) {
+ dpif_linux_flow_get_stats(&state->flow, &state->stats);
+ *stats = &state->stats;
}
- return error == ENODEV ? EOF : error;
+ return error;
}
static int
dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
{
struct dpif_linux_flow_state *state = state_;
-
+ int error = nl_dump_done(&state->dump);
ofpbuf_delete(state->buf);
free(state);
- return 0;
+ return error;
}
static int
const struct ofpbuf *packet)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct odp_upcall *execute;
+ struct odp_header *execute;
struct ofpbuf *buf;
int error;
buf = ofpbuf_new(128 + actions_len + packet->size);
- ofpbuf_reserve(buf, sizeof *execute);
- nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
- nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
+ nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
+ ODP_PACKET_CMD_EXECUTE, 1);
- execute = ofpbuf_push_uninit(buf, sizeof *execute);
- execute->dp_idx = dpif->minor;
- execute->len = buf->size;
+ execute = ofpbuf_put_uninit(buf, sizeof *execute);
+ execute->dp_ifindex = dpif->dp_ifindex;
- error = do_ioctl(dpif_, ODP_EXECUTE, buf->data);
+ 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
struct dpif_linux_dp dp;
dpif_linux_dp_init(&dp);
- dp.cmd = ODP_DP_SET;
- dp.dp_idx = dpif->minor;
+ dp.cmd = ODP_DP_CMD_SET;
+ dp.dp_ifindex = dpif->dp_ifindex;
dp.sampling = &probability;
return dpif_linux_dp_transact(&dp, NULL, NULL);
}
}
static int
-parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall)
+parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
+ int *dp_ifindex)
{
static const struct nl_policy odp_packet_policy[] = {
/* 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_ifindex = odp_header->dp_ifindex;
+
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++) {
+ int dp_ifindex;
- 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_ifindex);
+ if (!error
+ && dp_ifindex == dpif->dp_ifindex
+ && 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_);
- int i;
-
- /* This is somewhat bogus because it assumes that the following macros have
- * fixed values, but it's going to go away later. */
-#define DP_N_QUEUES 3
-#define DP_MAX_QUEUE_LEN 100
- for (i = 0; i < DP_N_QUEUES * DP_MAX_QUEUE_LEN; i++) {
- /* Reading even 1 byte discards a whole datagram and saves time. */
- char buffer;
- if (read(dpif->fd, &buffer, 1) != 1) {
- break;
- }
+ 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_get_all_names,
dpif_linux_destroy,
dpif_linux_get_stats,
dpif_linux_get_drop_frags,
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;
}
bool
return reply.type == ODP_VPORT_TYPE_INTERNAL;
}
-static int
-make_openvswitch_device(int minor, char **fnp)
-{
- const char dirname[] = "/dev/net";
- int major;
- dev_t dev;
- struct stat s;
- char fn[128];
-
- *fnp = NULL;
-
- major = get_openvswitch_major();
- if (major < 0) {
- return -major;
- }
- dev = makedev(major, minor);
-
- sprintf(fn, "%s/dp%d", dirname, minor);
- if (!stat(fn, &s)) {
- if (!S_ISCHR(s.st_mode)) {
- VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing",
- fn);
- } else if (s.st_rdev != dev) {
- VLOG_WARN_RL(&error_rl,
- "%s is device %u:%u but should be %u:%u, fixing",
- fn, major(s.st_rdev), minor(s.st_rdev),
- major(dev), minor(dev));
- } else {
- goto success;
- }
- if (unlink(fn)) {
- VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
- fn, strerror(errno));
- return errno;
- }
- } else if (errno == ENOENT) {
- if (stat(dirname, &s)) {
- if (errno == ENOENT) {
- if (mkdir(dirname, 0755)) {
- VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)",
- dirname, strerror(errno));
- return errno;
- }
- } else {
- VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)",
- dirname, strerror(errno));
- return errno;
- }
- }
- } else {
- VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno));
- return errno;
- }
-
- /* The device needs to be created. */
- if (mknod(fn, S_IFCHR | 0700, dev)) {
- VLOG_WARN_RL(&error_rl,
- "%s: creating character device %u:%u failed (%s)",
- fn, major(dev), minor(dev), strerror(errno));
- return errno;
- }
-
-success:
- *fnp = xstrdup(fn);
- return 0;
-}
-
-/* Return the major device number of the Open vSwitch device. If it
- * cannot be determined, a negative errno is returned. */
-static int
-get_openvswitch_major(void)
-{
- static int openvswitch_major = -1;
- if (openvswitch_major < 0) {
- openvswitch_major = get_major("openvswitch");
- }
- return openvswitch_major;
-}
-
-static int
-get_major(const char *target)
-{
- const char fn[] = "/proc/devices";
- char line[128];
- FILE *file;
- int ln;
-
- file = fopen(fn, "r");
- if (!file) {
- VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
- return -errno;
- }
-
- for (ln = 1; fgets(line, sizeof line, file); ln++) {
- char name[64];
- int major;
-
- if (!strncmp(line, "Character", 9) || line[0] == '\0') {
- /* Nothing to do. */
- } else if (!strncmp(line, "Block", 5)) {
- /* We only want character devices, so skip the rest of the file. */
- break;
- } else if (sscanf(line, "%d %63s", &major, name)) {
- if (!strcmp(name, target)) {
- fclose(file);
- return major;
- }
- } else {
- VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln);
- }
- }
-
- fclose(file);
-
- VLOG_ERR("%s: %s major not found (is the module loaded?)", fn, target);
- return -ENODEV;
-}
-
-static int
-open_minor(int minor, int *fdp)
-{
- int error;
- char *fn;
-
- error = make_openvswitch_device(minor, &fn);
- if (error) {
- return error;
- }
-
- *fdp = open(fn, O_RDONLY | O_NONBLOCK);
- if (*fdp < 0) {
- error = errno;
- VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
- free(fn);
- return error;
- }
- free(fn);
- return 0;
-}
-
static void
dpif_linux_port_changed(const struct rtnetlink_link_change *change,
void *dpif_)
struct dpif_linux *dpif = dpif_;
if (change) {
- if (change->master_ifindex == dpif->local_ifindex
+ if (change->master_ifindex == dpif->dp_ifindex
&& (change->nlmsg_type == RTM_NEWLINK
|| change->nlmsg_type == RTM_DELLINK))
{
dpif->change_error = true;
}
}
-
-static int
-get_dp0_fd(int *dp0_fdp)
-{
- static int dp0_fd = -1;
- if (dp0_fd < 0) {
- int error;
- int fd;
-
- error = open_minor(0, &fd);
- if (error) {
- return error;
- }
- dp0_fd = fd;
- }
- *dp0_fdp = dp0_fd;
- return 0;
-}
\f
-/* Parses the contents of 'buf', which contains a "struct odp_vport" followed
+/* Parses the contents of 'buf', which contains a "struct odp_header" followed
* by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
* positive errno value.
*
[ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
};
- struct odp_vport *odp_vport;
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);
- if (!nl_policy_parse(buf, sizeof *odp_vport, odp_vport_policy,
- a, ARRAY_SIZE(odp_vport_policy))) {
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
}
- odp_vport = buf->data;
- vport->dp_idx = odp_vport->dp_idx;
+ vport->cmd = genl->cmd;
+ vport->dp_ifindex = odp_header->dp_ifindex;
vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
}
if (a[ODP_VPORT_ATTR_MTU]) {
vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
+ } else {
+ vport->mtu = INT_MAX;
}
if (a[ODP_VPORT_ATTR_OPTIONS]) {
vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
return 0;
}
-/* Appends to 'buf' (which must initially be empty) a "struct odp_vport"
+/* Appends to 'buf' (which must initially be empty) a "struct 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)
{
- struct odp_vport *odp_vport;
+ struct odp_header *odp_header;
- ofpbuf_reserve(buf, sizeof odp_vport);
+ 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_ifindex = vport->dp_ifindex;
if (vport->port_no != UINT32_MAX) {
nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
vport->address, ETH_ADDR_LEN);
}
- if (vport->mtu) {
+ if (vport->mtu && vport->mtu != INT_MAX) {
nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
}
if (vport->iflink) {
nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink);
}
-
- odp_vport = ofpbuf_push_uninit(buf, sizeof *odp_vport);
- odp_vport->dp_idx = vport->dp_idx;
- odp_vport->len = buf->size;
- odp_vport->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
}
/* Clears 'vport' to "empty" values. */
dpif_linux_vport_init(struct dpif_linux_vport *vport)
{
memset(vport, 0, sizeof *vport);
- vport->dp_idx = UINT32_MAX;
vport->port_no = UINT32_MAX;
}
struct dpif_linux_vport *reply,
struct ofpbuf **bufp)
{
- struct ofpbuf *buf = NULL;
+ struct ofpbuf *request_buf;
int error;
- int fd;
assert((reply != NULL) == (bufp != NULL));
- error = get_dp0_fd(&fd);
- if (error) {
- goto error;
- }
-
- buf = ofpbuf_new(1024);
- dpif_linux_vport_to_ofpbuf(request, buf);
+ 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);
- error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
- if (error) {
- goto error;
- }
-
- if (bufp) {
- buf->size = ((struct odp_vport *) buf->data)->len;
- error = dpif_linux_vport_from_ofpbuf(reply, buf);
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
+ }
if (error) {
- goto error;
+ dpif_linux_vport_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- *bufp = buf;
- } else {
- ofpbuf_delete(buf);
- }
- return 0;
-
-error:
- ofpbuf_delete(buf);
- if (bufp) {
- memset(reply, 0, sizeof *reply);
- *bufp = NULL;
}
return error;
}
struct dpif_linux_vport request;
dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_GET;
+ request.cmd = 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_datapath"
- * followed by Netlink attributes, into 'dp'. Returns 0 if successful,
- * otherwise a positive errno value.
+/* Parses the contents of 'buf', which contains a "struct 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. */
.optional = true },
[ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
[ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
+ [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
};
- struct odp_datapath *odp_dp;
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);
- if (!nl_policy_parse(buf, sizeof *odp_dp, odp_datapath_policy,
- a, ARRAY_SIZE(odp_datapath_policy))) {
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
}
- odp_dp = buf->data;
- dp->dp_idx = odp_dp->dp_idx;
+ dp->cmd = genl->cmd;
+ dp->dp_ifindex = odp_header->dp_ifindex;
dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
if (a[ODP_DP_ATTR_STATS]) {
/* Can't use structure assignment because Netlink doesn't ensure
if (a[ODP_DP_ATTR_SAMPLING]) {
dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
}
+
+ if (a[ODP_DP_ATTR_MCGROUPS]) {
+ static const struct nl_policy odp_mcgroup_policy[] = {
+ [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
+ };
+
+ struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
+
+ if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
+ mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
+ return EINVAL;
+ }
+
+ if (mcgroups[ODP_PACKET_CMD_MISS]) {
+ dp->mcgroups[DPIF_UC_MISS]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
+ }
+ if (mcgroups[ODP_PACKET_CMD_ACTION]) {
+ dp->mcgroups[DPIF_UC_ACTION]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
+ }
+ if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
+ dp->mcgroups[DPIF_UC_SAMPLE]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
+ }
+ }
+
return 0;
}
-/* Appends to 'buf' (which must initially be empty) a "struct odp_datapath"
- * followed by Netlink attributes corresponding to 'dp'. */
+/* Appends to 'buf' the Generic Netlink message described by 'dp'. */
static void
dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
{
- struct odp_datapath *odp_dp;
+ struct odp_header *odp_header;
- ofpbuf_reserve(buf, sizeof odp_dp);
+ 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_ifindex = dp->dp_ifindex;
if (dp->name) {
nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
if (dp->sampling) {
nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
}
-
- odp_dp = ofpbuf_push_uninit(buf, sizeof *odp_dp);
- odp_dp->dp_idx = dp->dp_idx;
- odp_dp->len = buf->size;
- odp_dp->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
}
/* Clears 'dp' to "empty" values. */
dpif_linux_dp_init(struct dpif_linux_dp *dp)
{
memset(dp, 0, sizeof *dp);
- dp->dp_idx = -1;
+}
+
+static void
+dpif_linux_dp_dump_start(struct nl_dump *dump)
+{
+ struct dpif_linux_dp request;
+ struct ofpbuf *buf;
+
+ dpif_linux_dp_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_dp_to_ofpbuf(&request, buf);
+ nl_dump_start(dump, genl_sock, buf);
+ ofpbuf_delete(buf);
}
/* Executes 'request' in the kernel datapath. If the command fails, returns a
* positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
* without doing anything else. If 'reply' and 'bufp' are nonnull, then the
- * result of the command is expected to be an odp_datapath also, which is
- * decoded and stored in '*reply' and '*bufp'. The caller must free '*bufp'
- * when the reply is no longer needed ('reply' will contain pointers into
- * '*bufp'). */
+ * 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 *buf = NULL;
+ struct ofpbuf *request_buf;
int error;
- int fd;
assert((reply != NULL) == (bufp != NULL));
- error = get_dp0_fd(&fd);
- if (error) {
- goto error;
- }
-
- buf = ofpbuf_new(1024);
- dpif_linux_dp_to_ofpbuf(request, buf);
+ 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);
- error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
- if (error) {
- goto error;
- }
-
- if (bufp) {
- buf->size = ((struct odp_datapath *) buf->data)->len;
- error = dpif_linux_dp_from_ofpbuf(reply, buf);
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
+ }
if (error) {
- goto error;
+ dpif_linux_dp_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- *bufp = buf;
- } else {
- ofpbuf_delete(buf);
- }
- return 0;
-
-error:
- ofpbuf_delete(buf);
- if (bufp) {
- memset(reply, 0, sizeof *reply);
- *bufp = NULL;
}
return error;
}
struct dpif_linux_dp request;
dpif_linux_dp_init(&request);
- request.cmd = ODP_DP_GET;
- request.dp_idx = dpif->minor;
+ request.cmd = ODP_DP_CMD_GET;
+ request.dp_ifindex = dpif->dp_ifindex;
return dpif_linux_dp_transact(&request, reply, bufp);
}
\f
-/* Parses the contents of 'buf', which contains a "struct odp_flow" followed by
- * Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
+/* Parses the contents of 'buf', which contains a "struct 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'
[ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
[ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
/* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
- [ODP_FLOW_ATTR_STATE] = { .type = NL_A_U64, .optional = true },
};
- struct odp_flow *odp_flow;
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);
- if (!nl_policy_parse(buf, sizeof *odp_flow, odp_flow_policy,
- a, ARRAY_SIZE(odp_flow_policy))) {
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ 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;
}
- odp_flow = buf->data;
- flow->nlmsg_flags = odp_flow->nlmsg_flags;
- flow->dp_idx = odp_flow->dp_idx;
+ flow->nlmsg_flags = nlmsg->nlmsg_flags;
+ flow->dp_ifindex = odp_header->dp_ifindex;
flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
if (a[ODP_FLOW_ATTR_ACTIONS]) {
if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
}
- if (a[ODP_FLOW_ATTR_STATE]) {
- flow->state = nl_attr_get(a[ODP_FLOW_ATTR_STATE]);
+ if (a[ODP_FLOW_ATTR_USED]) {
+ flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]);
}
return 0;
}
-/* Appends to 'buf' (which must initially be empty) a "struct odp_flow"
+/* Appends to 'buf' (which must initially be empty) a "struct 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)
{
- struct odp_flow *odp_flow;
+ struct odp_header *odp_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
+ NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
+ flow->cmd, 1);
- ofpbuf_reserve(buf, sizeof odp_flow);
+ odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
+ odp_header->dp_ifindex = flow->dp_ifindex;
if (flow->key_len) {
nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
}
- if (flow->actions_len) {
+ if (flow->actions || flow->actions_len) {
nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
flow->actions, flow->actions_len);
}
if (flow->clear) {
nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
}
-
- if (flow->state) {
- nl_msg_put_u64(buf, ODP_FLOW_ATTR_STATE,
- get_unaligned_u64(flow->state));
- }
-
- odp_flow = ofpbuf_push_uninit(buf, sizeof *odp_flow);
- odp_flow->nlmsg_flags = flow->nlmsg_flags;
- odp_flow->dp_idx = flow->dp_idx;
- odp_flow->len = buf->size;
- odp_flow->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
}
/* Clears 'flow' to "empty" values. */
/* Executes 'request' in the kernel datapath. If the command fails, returns a
* positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
* without doing anything else. If 'reply' and 'bufp' are nonnull, then the
- * result of the command is expected to be an odp_flow also, which is decoded
- * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
- * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
+ * 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 *buf = NULL;
+ struct ofpbuf *request_buf;
int error;
- int fd;
assert((reply != NULL) == (bufp != NULL));
- error = get_dp0_fd(&fd);
- if (error) {
- goto error;
- }
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(request, request_buf);
+ error = nl_sock_transact(genl_sock, request_buf, bufp);
+ ofpbuf_delete(request_buf);
- buf = ofpbuf_new(1024);
- dpif_linux_flow_to_ofpbuf(request, buf);
-
- error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
- if (error) {
- goto error;
- }
-
- if (bufp) {
- buf->size = ((struct odp_flow *) buf->data)->len;
- error = dpif_linux_flow_from_ofpbuf(reply, buf);
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
+ }
if (error) {
- goto error;
+ dpif_linux_flow_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
}
- *bufp = buf;
- } else {
- ofpbuf_delete(buf);
- }
- return 0;
-
-error:
- ofpbuf_delete(buf);
- if (bufp) {
- memset(reply, 0, sizeof *reply);
- *bufp = NULL;
}
return error;
}