/*
- * Copyright (c) 2008, 2009, 2010 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
*/
#include <config.h>
-#include "dpif.h"
+
+#include "dpif-linux.h"
#include <assert.h>
#include <ctype.h>
#include "dpif-provider.h"
#include "netdev.h"
#include "netdev-vport.h"
+#include "netlink.h"
#include "ofpbuf.h"
#include "openvswitch/tunnel.h"
+#include "packets.h"
#include "poll-loop.h"
#include "rtnetlink.h"
+#include "rtnetlink-link.h"
#include "shash.h"
#include "svec.h"
#include "util.h"
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
static int do_ioctl(const struct dpif *, int cmd, const void *arg);
-static int lookup_minor(const char *name, int *minor);
-static int finish_open(struct dpif *, const char *local_ifname);
+static int open_dpif(const struct dpif_linux_vport *local_vport,
+ struct dpif **);
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 create_minor(const char *name, int minor);
+static int open_minor(int minor, int *fdp);
static int make_openvswitch_device(int minor, char **fnp);
-static void dpif_linux_port_changed(const struct rtnetlink_change *,
+static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
void *dpif);
static struct dpif_linux *
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 ofpbuf *buf;
int minor;
+ int 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);
+ error = create_minor(name, minor);
+ if (error) {
+ return error;
+ }
} else {
/* Scan for unused minor number. */
- for (minor = 0; minor < ODP_MAX; minor++) {
- int error = create_minor(name, minor, dpifp);
- if (error != EBUSY) {
+ for (minor = 0; ; minor++) {
+ if (minor >= ODP_MAX) {
+ /* All datapath numbers in use. */
+ return ENOBUFS;
+ }
+
+ error = create_minor(name, minor);
+ if (!error) {
+ break;
+ } else if (error != EBUSY) {
return error;
}
}
-
- /* All datapath numbers in use. */
- return ENOBUFS;
}
+ }
+
+ dpif_linux_vport_init(&request);
+ request.cmd = ODP_VPORT_GET;
+ request.port_no = ODPP_LOCAL;
+ if (minor >= 0) {
+ request.dp_idx = minor;
} else {
- struct dpif_linux *dpif;
- struct odp_port port;
- int error;
+ request.name = name;
+ }
- if (minor < 0) {
- error = lookup_minor(name, &minor);
- if (error) {
- return error;
- }
- }
+ error = dpif_linux_vport_transact(&request, &reply, &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_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;
- }
+ ofpbuf_delete(buf);
+ return error;
+}
- /* Then use that to finish up opening. */
- return finish_open(&dpif->dpif, port.devname);
+static int
+open_dpif(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 fd;
+
+ error = open_minor(dp_idx, &fd);
+ if (error) {
+ goto error;
+ }
+
+ 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->fd = fd;
+ dpif->local_ifname = xstrdup(local_vport->name);
+ dpif->local_ifindex = local_vport->ifindex;
+ dpif->minor = dp_idx;
+ shash_init(&dpif->changed_ports);
+ dpif->change_error = false;
+ *dpifp = &dpif->dpif;
+
+ return 0;
+
+error_free:
+ free(dpif);
+ close(fd);
+error:
+ return error;
}
static void
dpif_linux_close(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- rtnetlink_notifier_unregister(&dpif->port_notifier);
+ rtnetlink_link_notifier_unregister(&dpif->port_notifier);
shash_destroy(&dpif->changed_ports);
free(dpif->local_ifname);
close(dpif->fd);
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;
-
- if (!strcmp(type, "system")) {
- ovs_strlcpy(type, "netdev", sizeof port->type);
- } else if (!strcmp(type, "ipsec_gre")) {
- ovs_strlcpy(type, "gre", sizeof port->type);
- }
-}
-
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_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;
+
+ options = netdev_vport_get_options(netdev);
+ if (options && options->size) {
+ request.options = options->data;
+ request.options_len = options->size;
+ }
- error = do_ioctl(dpif, ODP_VPORT_ATTACH, &port);
+ 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_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, struct odp_port *port)
+dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
+ const char *port_name, struct dpif_port *dpif_port)
{
- int error = do_ioctl(dpif, ODP_VPORT_QUERY, port);
+ struct dpif_linux_vport request;
+ struct dpif_linux_vport reply;
+ struct ofpbuf *buf;
+ int error;
+
+ dpif_linux_vport_init(&request);
+ request.cmd = ODP_VPORT_GET;
+ request.dp_idx = dpif_linux_cast(dpif)->minor;
+ request.port_no = port_no;
+ request.name = port_name;
+
+ error = dpif_linux_vport_transact(&request, &reply, &buf);
if (!error) {
- translate_vport_type_to_netdev_type(port);
+ dpif_port->name = xstrdup(reply.name);
+ dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
+ dpif_port->port_no = reply.port_no;
+ ofpbuf_delete(buf);
}
return error;
}
static int
dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
- struct odp_port *port)
+ struct dpif_port *dpif_port)
{
- memset(port, 0, sizeof *port);
- port->port = port_no;
- return dpif_linux_port_query__(dpif, port);
+ return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
}
static int
dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
- struct odp_port *port)
+ struct dpif_port *dpif_port)
{
- memset(port, 0, sizeof *port);
- strncpy(port->devname, devname, sizeof port->devname);
- return dpif_linux_port_query__(dpif, port);
+ 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
return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL);
}
+struct dpif_linux_port_state {
+ struct ofpbuf *buf;
+ uint32_t next;
+};
+
static int
-dpif_linux_port_list(const struct dpif *dpif_, struct odp_port *ports, int n)
+dpif_linux_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
{
- struct odp_portvec pv;
- unsigned int i;
+ *statep = xzalloc(sizeof(struct dpif_linux_port_state));
+ return 0;
+}
+
+static int
+dpif_linux_port_dump_next(const struct dpif *dpif, void *state_,
+ struct dpif_port *dpif_port)
+{
+ struct dpif_linux_port_state *state = state_;
+ struct dpif_linux_vport request, reply;
+ struct ofpbuf *buf;
int error;
- pv.ports = ports;
- pv.n_ports = n;
- error = do_ioctl(dpif_, ODP_VPORT_LIST, &pv);
+ ofpbuf_delete(state->buf);
+ state->buf = NULL;
+
+ 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);
if (error) {
- return -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;
}
+}
- for (i = 0; i < pv.n_ports; i++) {
- struct odp_port *port = &pv.ports[i];
-
- translate_vport_type_to_netdev_type(port);
- }
- return pv.n_ports;
+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);
+ free(state);
+ return 0;
}
static int
if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) {
poll_immediate_wake();
} else {
- rtnetlink_notifier_wait();
+ rtnetlink_link_notifier_wait();
}
}
static int
-dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n)
+dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow *flow)
{
- struct odp_flowvec fv;
- fv.flows = flows;
- fv.n_flows = n;
- return do_ioctl(dpif_, ODP_FLOW_GET, &fv);
+ return do_ioctl(dpif_, ODP_FLOW_GET, flow);
}
static int
}
static int
-dpif_linux_flow_list(const struct dpif *dpif_, struct odp_flow flows[], int n)
+dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
+{
+ *statep = xzalloc(sizeof(struct odp_flow_dump));
+ return 0;
+}
+
+static int
+dpif_linux_flow_dump_next(const struct dpif *dpif, void *state,
+ struct odp_flow *flow)
{
- struct odp_flowvec fv;
+ struct odp_flow_dump *dump = state;
int error;
- fv.flows = flows;
- fv.n_flows = n;
- error = do_ioctl(dpif_, ODP_FLOW_LIST, &fv);
- return error ? -error : fv.n_flows;
+ dump->flow = flow;
+ error = do_ioctl(dpif, ODP_FLOW_DUMP, dump);
+ return error ? error : flow->flags & ODPFF_EOF ? EOF : 0;
+}
+
+static int
+dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state)
+{
+ free(state);
+ return 0;
}
static int
}
static int
-dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp)
+parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall)
+{
+ 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_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
+
+ /* _ODPL_SFLOW_NR only. */
+ [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
+ };
+
+ struct odp_packet *odp_packet = buf->data;
+ struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
+
+ if (!nl_policy_parse(buf, sizeof *odp_packet, odp_packet_policy,
+ a, ARRAY_SIZE(odp_packet_policy))) {
+ return EINVAL;
+ }
+
+ memset(upcall, 0, sizeof *upcall);
+ upcall->type = nl_attr_get_u32(a[ODP_PACKET_ATTR_TYPE]);
+ upcall->packet = buf;
+ upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
+ upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
+ upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]);
+ upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]);
+ upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA]
+ ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA])
+ : 0);
+ upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL]
+ ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL])
+ : 0);
+ if (a[ODP_PACKET_ATTR_ACTIONS]) {
+ upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]);
+ upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
+ }
+
+ return 0;
+}
+
+static int
+dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct ofpbuf *buf;
int retval;
int error;
- buf = ofpbuf_new_with_headroom(65536, DPIF_RECV_MSG_PADDING);
+ buf = ofpbuf_new(65536);
retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
if (retval < 0) {
error = errno;
VLOG_WARN_RL(&error_rl, "%s: read failed: %s",
dpif_name(dpif_), strerror(error));
}
- } else if (retval >= sizeof(struct odp_msg)) {
- struct odp_msg *msg = buf->data;
- if (msg->length <= retval) {
- buf->size += retval;
- *bufp = buf;
- return 0;
+ } else 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_), msg->length, retval);
+ dpif_name(dpif_), odp_packet->len, retval);
error = ERANGE;
}
} else if (!retval) {
VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_));
error = EPROTO;
} else {
- VLOG_WARN_RL(&error_rl,
- "%s: discarding too-short message (%d bytes)",
+ VLOG_WARN_RL(&error_rl, "%s: discarding too-short message (%d bytes)",
dpif_name(dpif_), retval);
error = ERANGE;
}
- *bufp = NULL;
- ofpbuf_delete(buf);
+ if (error) {
+ ofpbuf_delete(buf);
+ }
return error;
}
poll_fd_wait(dpif->fd, 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;
+ }
+ }
+}
+
const struct dpif_class dpif_linux_class = {
"system",
NULL,
dpif_linux_port_del,
dpif_linux_port_query_by_number,
dpif_linux_port_query_by_name,
- dpif_linux_port_list,
+ dpif_linux_get_max_ports,
+ dpif_linux_port_dump_start,
+ dpif_linux_port_dump_next,
+ dpif_linux_port_dump_done,
dpif_linux_port_poll,
dpif_linux_port_poll_wait,
dpif_linux_flow_get,
dpif_linux_flow_put,
dpif_linux_flow_del,
dpif_linux_flow_flush,
- dpif_linux_flow_list,
+ dpif_linux_flow_dump_start,
+ dpif_linux_flow_dump_next,
+ dpif_linux_flow_dump_done,
dpif_linux_execute,
dpif_linux_recv_get_mask,
dpif_linux_recv_set_mask,
dpif_linux_queue_to_priority,
dpif_linux_recv,
dpif_linux_recv_wait,
+ dpif_linux_recv_purge,
};
\f
static int get_openvswitch_major(void);
return ioctl(dpif->fd, cmd, arg) ? errno : 0;
}
-static int
-lookup_minor(const char *name, int *minorp)
+bool
+dpif_linux_is_internal_device(const char *name)
{
- struct ethtool_drvinfo drvinfo;
- int minor, port_no;
- struct ifreq ifr;
+ struct dpif_linux_vport reply;
+ struct ofpbuf *buf;
int error;
- int sock;
-
- sock = socket(AF_INET, SOCK_DGRAM, 0);
- if (sock < 0) {
- VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
- error = errno;
- goto error;
- }
-
- memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
- ifr.ifr_data = (caddr_t) &drvinfo;
- memset(&drvinfo, 0, sizeof drvinfo);
- drvinfo.cmd = ETHTOOL_GDRVINFO;
- if (ioctl(sock, SIOCETHTOOL, &ifr)) {
- VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
- error = errno;
- goto error_close_sock;
- }
-
- if (strcmp(drvinfo.driver, "openvswitch")) {
- VLOG_WARN("%s is not an openvswitch device", name);
- error = EOPNOTSUPP;
- goto error_close_sock;
- }
-
- if (sscanf(drvinfo.bus_info, "%d.%d", &minor, &port_no) != 2) {
- VLOG_WARN("%s ethtool bus_info has unexpected format", name);
- error = EPROTOTYPE;
- goto error_close_sock;
- } else if (port_no != ODPP_LOCAL) {
- /* This is an Open vSwitch device but not the local port. We
- * intentionally support only using the name of the local port as the
- * name of a datapath; otherwise, it would be too difficult to
- * enumerate all the names of a datapath. */
- error = EOPNOTSUPP;
- goto error_close_sock;
+ 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));
}
- *minorp = minor;
- close(sock);
- return 0;
-
-error_close_sock:
- close(sock);
-error:
- return error;
+ return reply.type == ODP_VPORT_TYPE_INTERNAL;
}
static int
}
static int
-finish_open(struct dpif *dpif_, const char *local_ifname)
+create_minor(const char *name, int minor)
{
- 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));
+ int error;
+ int fd;
+
+ error = open_minor(minor, &fd);
+ if (error) {
return error;
}
- return 0;
-}
-static int
-create_minor(const char *name, int minor, struct dpif **dpifp)
-{
- 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);
- }
- }
+ error = ioctl(fd, ODP_DP_CREATE, name) ? errno : 0;
+ close(fd);
return error;
}
static int
-open_minor(int minor, struct dpif **dpifp)
+open_minor(int minor, int *fdp)
{
int error;
char *fn;
- int fd;
error = make_openvswitch_device(minor, &fn);
if (error) {
return error;
}
- fd = open(fn, O_RDONLY | O_NONBLOCK);
- if (fd >= 0) {
- struct dpif_linux *dpif = xmalloc(sizeof *dpif);
- error = rtnetlink_notifier_register(&dpif->port_notifier,
- dpif_linux_port_changed, dpif);
- 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);
- }
- } else {
+ *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 error;
+ return 0;
}
static void
-dpif_linux_port_changed(const struct rtnetlink_change *change, void *dpif_)
+dpif_linux_port_changed(const struct rtnetlink_link_change *change,
+ void *dpif_)
{
struct dpif_linux *dpif = dpif_;
dpif->change_error = true;
}
}
+\f
+/* Parses the contents of 'buf', which contains a "struct odp_vport" 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 odp_vport *odp_vport;
+ struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
+
+ dpif_linux_vport_init(vport);
+
+ if (!nl_policy_parse(buf, sizeof *odp_vport, odp_vport_policy,
+ a, ARRAY_SIZE(odp_vport_policy))) {
+ return EINVAL;
+ }
+ odp_vport = buf->data;
+
+ vport->dp_idx = odp_vport->dp_idx;
+ vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
+ vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
+ vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
+ if (a[ODP_VPORT_ATTR_STATS]) {
+ vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]);
+ }
+ if (a[ODP_VPORT_ATTR_ADDRESS]) {
+ vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]);
+ }
+ 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;
+}
+
+/* Appends to 'buf' (which must initially be empty) a "struct odp_vport"
+ * 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;
+
+ ofpbuf_reserve(buf, sizeof odp_vport);
+
+ 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);
+ }
+
+ 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. */
+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)
+{
+ static int dp0_fd = -1;
+ struct ofpbuf *buf = NULL;
+ int error;
+
+ assert((reply != NULL) == (bufp != NULL));
+ if (dp0_fd < 0) {
+ int fd;
+
+ error = open_minor(0, &fd);
+ if (error) {
+ goto error;
+ }
+ dp0_fd = fd;
+ }
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_vport_to_ofpbuf(request, buf);
+
+ error = ioctl(dp0_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 (error) {
+ goto error;
+ }
+ *bufp = buf;
+ } else {
+ ofpbuf_delete(buf);
+ }
+ return 0;
+
+error:
+ ofpbuf_delete(buf);
+ if (bufp) {
+ memset(reply, 0, sizeof *reply);
+ *bufp = NULL;
+ }
+ return error;
+}
+
+/* Obtains information about 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_GET;
+ request.name = name;
+
+ return dpif_linux_vport_transact(&request, reply, bufp);
+}
+