/*
- * 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 <fcntl.h>
#include <inttypes.h>
#include <net/if.h>
+#include <linux/types.h>
#include <linux/ethtool.h>
+#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
#include <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.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"
-
#include "vlog.h"
-#define THIS_MODULE VLM_dpif_linux
+
+VLOG_DEFINE_THIS_MODULE(dpif_linux);
/* Datapath interface for the openvswitch Linux kernel module. */
struct dpif_linux {
/* Change notification. */
int local_ifindex; /* Ifindex of local port. */
- struct svec changed_ports; /* Ports that have changed. */
+ 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 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);
-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 *
}
static int
-dpif_linux_open(const char *name, const char *type OVS_UNUSED, bool create,
- struct dpif **dpifp)
+dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
+ bool create, struct dpif **dpifp)
{
int minor;
* getting the local port's name. */
memset(&port, 0, sizeof port);
port.port = ODPP_LOCAL;
- if (ioctl(dpif->fd, ODP_PORT_QUERY, &port)) {
+ if (ioctl(dpif->fd, ODP_VPORT_QUERY, &port)) {
error = errno;
if (error != ENODEV) {
VLOG_WARN("%s: probe returned unexpected error: %s",
dpif_linux_close(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- rtnetlink_notifier_unregister(&dpif->port_notifier);
- svec_destroy(&dpif->changed_ports);
+ rtnetlink_link_notifier_unregister(&dpif->port_notifier);
+ shash_destroy(&dpif->changed_ports);
free(dpif->local_ifname);
close(dpif->fd);
free(dpif);
return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int);
}
+static const char *
+vport_type_to_netdev_type(const struct odp_port *odp_port)
+{
+ struct tnl_port_config tnl_config;
+
+ switch (odp_port->type) {
+ case ODP_VPORT_TYPE_UNSPEC:
+ break;
+
+ case ODP_VPORT_TYPE_NETDEV:
+ return "system";
+
+ case ODP_VPORT_TYPE_INTERNAL:
+ return "internal";
+
+ case ODP_VPORT_TYPE_PATCH:
+ return "patch";
+
+ case ODP_VPORT_TYPE_GRE:
+ memcpy(&tnl_config, odp_port->config, sizeof tnl_config);
+ return tnl_config.flags & TNL_F_IPSEC ? "ipsec_gre" : "gre";
+
+ case ODP_VPORT_TYPE_CAPWAP:
+ return "capwap";
+
+ case __ODP_VPORT_TYPE_MAX:
+ break;
+ }
+
+ VLOG_WARN_RL(&error_rl, "dp%d: port `%s' has unsupported type %"PRIu32,
+ odp_port->dp_idx, odp_port->devname, odp_port->type);
+ return "unknown";
+}
+
+static enum odp_vport_type
+netdev_type_to_vport_type(const char *type)
+{
+ return (!strcmp(type, "system") ? ODP_VPORT_TYPE_NETDEV
+ : !strcmp(type, "internal") ? ODP_VPORT_TYPE_INTERNAL
+ : !strcmp(type, "patch") ? ODP_VPORT_TYPE_PATCH
+ : (!strcmp(type, "gre")
+ || !strcmp(type, "ipsec_gre")) ? ODP_VPORT_TYPE_GRE
+ : !strcmp(type, "capwap") ? ODP_VPORT_TYPE_CAPWAP
+ : ODP_VPORT_TYPE_UNSPEC);
+}
+
static int
-dpif_linux_port_add(struct dpif *dpif_, const char *devname, uint16_t flags,
- uint16_t *port_no)
+dpif_linux_port_add(struct dpif *dpif, struct netdev *netdev,
+ uint16_t *port_nop)
{
+ const char *name = netdev_get_name(netdev);
+ const char *type = netdev_get_type(netdev);
struct odp_port port;
int error;
memset(&port, 0, sizeof port);
- strncpy(port.devname, devname, sizeof port.devname);
- port.flags = flags;
- error = do_ioctl(dpif_, ODP_PORT_ATTACH, &port);
+ strncpy(port.devname, name, sizeof port.devname);
+ netdev_vport_get_config(netdev, port.config);
+
+ port.type = netdev_type_to_vport_type(type);
+ if (port.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;
+ }
+
+ error = do_ioctl(dpif, ODP_VPORT_ATTACH, &port);
if (!error) {
- *port_no = port.port;
+ *port_nop = port.port;
}
+
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);
+}
+
+static int
+dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
+ const char *port_name, struct dpif_port *dpif_port)
{
- int tmp = port_no;
- return do_ioctl(dpif_, ODP_PORT_DETACH, &tmp);
+ struct odp_port odp_port;
+ int error;
+
+ memset(&odp_port, 0, sizeof odp_port);
+ odp_port.port = port_no;
+ strncpy(odp_port.devname, port_name, sizeof odp_port.devname);
+
+ 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 {
+ dpif_port->name = xstrdup(odp_port.devname);
+ dpif_port->type = xstrdup(vport_type_to_netdev_type(&odp_port));
+ dpif_port->port_no = odp_port.port;
+ return 0;
+ }
}
static int
-dpif_linux_port_query_by_number(const struct dpif *dpif_, uint16_t port_no,
- struct odp_port *port)
+dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
+ struct dpif_port *dpif_port)
{
- memset(port, 0, sizeof *port);
- port->port = port_no;
- return do_ioctl(dpif_, ODP_PORT_QUERY, port);
+ return dpif_linux_port_query__(dpif, port_no, "", dpif_port);
}
static int
-dpif_linux_port_query_by_name(const struct dpif *dpif_, const char *devname,
- struct odp_port *port)
+dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
+ struct dpif_port *dpif_port)
{
- memset(port, 0, sizeof *port);
- strncpy(port->devname, devname, sizeof port->devname);
- return do_ioctl(dpif_, ODP_PORT_QUERY, port);
+ return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
}
static int
}
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;
+ *statep = xzalloc(sizeof(struct odp_port));
+ return 0;
+}
+
+static int
+dpif_linux_port_dump_next(const struct dpif *dpif, void *state,
+ struct dpif_port *dpif_port)
+{
+ struct odp_port *odp_port = state;
+ struct odp_vport_dump dump;
int error;
- pv.ports = ports;
- pv.n_ports = n;
- error = do_ioctl(dpif_, ODP_PORT_LIST, &pv);
- return error ? -error : pv.n_ports;
+ dump.port = odp_port;
+ dump.port_no = odp_port->port;
+ error = do_ioctl(dpif, ODP_VPORT_DUMP, &dump);
+ if (error) {
+ return error;
+ } else if (odp_port->devname[0] == '\0') {
+ return EOF;
+ } else {
+ dpif_port->name = odp_port->devname;
+ dpif_port->type = (char *) vport_type_to_netdev_type(odp_port);
+ dpif_port->port_no = odp_port->port;
+ odp_port->port++;
+ return 0;
+ }
+}
+
+static int
+dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state)
+{
+ free(state);
+ return 0;
}
static int
if (dpif->change_error) {
dpif->change_error = false;
- svec_clear(&dpif->changed_ports);
+ shash_clear(&dpif->changed_ports);
return ENOBUFS;
- } else if (dpif->changed_ports.n) {
- *devnamep = dpif->changed_ports.names[--dpif->changed_ports.n];
+ } else if (!shash_is_empty(&dpif->changed_ports)) {
+ struct shash_node *node = shash_first(&dpif->changed_ports);
+ *devnamep = shash_steal(&dpif->changed_ports, node);
return 0;
} else {
return EAGAIN;
dpif_linux_port_poll_wait(const struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (dpif->changed_ports.n || dpif->change_error) {
+ 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_port_group_get(const struct dpif *dpif_, int group,
- uint16_t ports[], int n)
-{
- struct odp_port_group pg;
- int error;
-
- assert(n <= UINT16_MAX);
- pg.group = group;
- pg.ports = ports;
- pg.n_ports = n;
- error = do_ioctl(dpif_, ODP_PORT_GROUP_GET, &pg);
- return error ? -error : pg.n_ports;
-}
-
-static int
-dpif_linux_port_group_set(struct dpif *dpif_, int group,
- const uint16_t ports[], int n)
-{
- struct odp_port_group pg;
-
- assert(n <= UINT16_MAX);
- pg.group = group;
- pg.ports = (uint16_t *) ports;
- pg.n_ports = n;
- return do_ioctl(dpif_, ODP_PORT_GROUP_SET, &pg);
-}
-
static int
dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n)
{
}
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)
{
- struct odp_flowvec fv;
+ *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_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_execute(struct dpif *dpif_, uint16_t in_port,
- const union odp_action actions[], int n_actions,
+dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state)
+{
+ free(state);
+ return 0;
+}
+
+static int
+dpif_linux_execute(struct dpif *dpif_,
+ const struct nlattr *actions, size_t actions_len,
const struct ofpbuf *buf)
{
struct odp_execute execute;
memset(&execute, 0, sizeof execute);
- execute.in_port = in_port;
- execute.actions = (union odp_action *) actions;
- execute.n_actions = n_actions;
+ execute.actions = (struct nlattr *) actions;
+ execute.actions_len = actions_len;
execute.data = buf->data;
execute.length = buf->size;
return do_ioctl(dpif_, ODP_EXECUTE, &execute);
}
static int
-dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp)
+dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
+ uint32_t queue_id, uint32_t *priority)
+{
+ if (queue_id < 0xf000) {
+ *priority = TC_H_MAKE(1 << 16, queue_id + 1);
+ return 0;
+ } else {
+ return EINVAL;
+ }
+}
+
+static int
+parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall)
+{
+ 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;
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;
}
dpif_linux_port_del,
dpif_linux_port_query_by_number,
dpif_linux_port_query_by_name,
- dpif_linux_port_list,
+ 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_port_group_get,
- dpif_linux_port_group_set,
dpif_linux_flow_get,
dpif_linux_flow_put,
dpif_linux_flow_del,
dpif_linux_flow_flush,
- dpif_linux_flow_list,
+ dpif_linux_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_get_sflow_probability,
dpif_linux_set_sflow_probability,
+ dpif_linux_queue_to_priority,
dpif_linux_recv,
dpif_linux_recv_wait,
};
}
static int
-lookup_minor(const char *name, int *minorp)
+lookup_internal_device(const char *name, int *dp_idx, int *port_no)
{
- struct ethtool_drvinfo drvinfo;
- int minor, port_no;
- struct ifreq ifr;
- int error;
- int sock;
+ struct odp_port odp_port;
+ static int dp0_fd = -1;
- sock = socket(AF_INET, SOCK_DGRAM, 0);
- if (sock < 0) {
- VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
- error = errno;
- goto error;
- }
+ if (dp0_fd < 0) {
+ int error;
+ char *fn;
- memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
- ifr.ifr_data = (caddr_t) &drvinfo;
+ error = make_openvswitch_device(0, &fn);
+ if (error) {
+ return error;
+ }
- 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;
+ 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);
}
- if (strcmp(drvinfo.driver, "openvswitch")) {
- VLOG_WARN("%s is not an openvswitch device", name);
- error = EOPNOTSUPP;
- goto error_close_sock;
+ 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));
+ }
+ return errno;
+ } else if (odp_port.type == ODP_VPORT_TYPE_INTERNAL) {
+ *dp_idx = odp_port.dp_idx;
+ *port_no = odp_port.port;
+ return 0;
+ } else {
+ return EINVAL;
}
+}
+
+static int
+lookup_minor(const char *name, int *minorp)
+{
+ int minor, port_no;
+ int error;
- 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;
+ 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. */
- error = EOPNOTSUPP;
- goto error_close_sock;
+ return EOPNOTSUPP;
+ } else {
+ *minorp = minor;
+ return 0;
}
+}
- *minorp = minor;
- close(sock);
- return 0;
+bool
+dpif_linux_is_internal_device(const char *name)
+{
+ int minor, port_no;
-error_close_sock:
- close(sock);
-error:
- return error;
+ return !lookup_internal_device(name, &minor, &port_no);
}
static int
return major;
}
} else {
- static bool warned;
- if (!warned) {
- VLOG_WARN("%s:%d: syntax error", fn, ln);
- }
- warned = true;
+ VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln);
}
}
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);
+ error = rtnetlink_link_notifier_register(&dpif->port_notifier,
+ dpif_linux_port_changed,
+ dpif);
if (!error) {
char *name;
dpif->local_ifname = NULL;
dpif->minor = minor;
dpif->local_ifindex = 0;
- svec_init(&dpif->changed_ports);
+ shash_init(&dpif->changed_ports);
dpif->change_error = false;
*dpifp = &dpif->dpif;
} else {
}
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_;
{
/* Our datapath changed, either adding a new port or deleting an
* existing one. */
- if (!svec_contains(&dpif->changed_ports, change->ifname)) {
- svec_add(&dpif->changed_ports, change->ifname);
- svec_sort(&dpif->changed_ports);
- }
+ shash_add_once(&dpif->changed_ports, change->ifname, NULL);
}
} else {
dpif->change_error = true;