#include "coverage.h"
#include "daemon.h"
#include "dirs.h"
-#include "dpif.h"
#include "dynamic-string.h"
#include "fatal-signal.h"
#include "fault.h"
return 0;
}
-/* Get all the interfaces for 'bridge' as 'ifaces', breaking bonded interfaces
- * down into their constituent parts. */
static void
-get_bridge_ifaces(const char *bridge, struct svec *ifaces)
+do_get_bridge_parts(const char *bridge, struct svec *parts, int vlan,
+ bool break_down_bonds)
{
struct svec ports;
int i;
svec_init(&ports);
- svec_init(ifaces);
cfg_get_all_keys(&ports, "bridge.%s.port", bridge);
for (i = 0; i < ports.n; i++) {
const char *port_name = ports.names[i];
- if (cfg_has_section("bonding.%s", port_name)) {
+ if (vlan >= 0) {
+ int port_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
+ if (port_vlan < 0) {
+ port_vlan = 0;
+ }
+ if (vlan != port_vlan) {
+ continue;
+ }
+ }
+ if (break_down_bonds && cfg_has_section("bonding.%s", port_name)) {
struct svec slaves;
svec_init(&slaves);
cfg_get_all_keys(&slaves, "bonding.%s.slave", port_name);
- svec_append(ifaces, &slaves);
+ svec_append(parts, &slaves);
svec_destroy(&slaves);
} else {
- svec_add(ifaces, port_name);
+ svec_add(parts, port_name);
}
}
svec_destroy(&ports);
}
+/* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
+ * down into their constituent parts.
+ *
+ * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
+ * then only interfaces for trunk ports or ports with implicit VLAN 0 are
+ * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
+ * reported. */
+static void
+get_bridge_ifaces(const char *bridge, struct svec *ifaces, int vlan)
+{
+ do_get_bridge_parts(bridge, ifaces, vlan, true);
+}
+
+/* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
+ * the bond name, not broken down into their constituent interfaces.
+ *
+ * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
+ * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
+ * only port with implicit VLAN 'vlan' are reported. */
+static void
+get_bridge_ports(const char *bridge, struct svec *ports, int vlan)
+{
+ do_get_bridge_parts(bridge, ports, vlan, false);
+}
+
/* Go through the configuration file and remove any ports that no longer
* exist associated with a bridge. */
static void
prune_ports(void)
{
int i, j;
- int error;
struct svec bridges, delete;
if (cfg_lock(NULL, 0)) {
struct svec ifaces;
/* Check that each bridge interface exists. */
- get_bridge_ifaces(br_name, &ifaces);
+ svec_init(&ifaces);
+ get_bridge_ifaces(br_name, &ifaces, -1);
for (j = 0; j < ifaces.n; j++) {
const char *iface_name = ifaces.names[j];
- enum netdev_flags flags;
/* The local port and internal ports are created and destroyed by
* ovs-vswitchd itself, so don't bother checking for them at all.
continue;
}
- error = netdev_nodev_get_flags(iface_name, &flags);
- if (error == ENODEV) {
+ if (!netdev_exists(iface_name)) {
VLOG_INFO_RL(&rl, "removing dead interface %s from %s",
iface_name, br_name);
svec_add(&delete, iface_name);
- } else if (error) {
- VLOG_INFO_RL(&rl, "unknown error %d on interface %s from %s",
- error, iface_name, br_name);
}
}
svec_destroy(&ifaces);
svec_destroy(&delete);
}
-
-/* Checks whether a network device named 'name' exists and returns true if so,
- * false otherwise.
- *
- * XXX it is possible that this doesn't entirely accomplish what we want in
- * context, since ovs-vswitchd.conf may cause vswitchd to create or destroy
- * network devices based on iface.*.internal settings.
- *
- * XXX may want to move this to lib/netdev.
- *
- * XXX why not just use netdev_nodev_get_flags() or similar function? */
-static bool
-netdev_exists(const char *name)
-{
- struct stat s;
- char *filename;
- int error;
-
- filename = xasprintf("/sys/class/net/%s", name);
- error = stat(filename, &s);
- free(filename);
- return !error;
-}
-
static int
add_bridge(const char *br_name)
{
const char **port_name, uint64_t *count, uint64_t *skip)
{
static const struct nl_policy policy[] = {
- [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
+ [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
[BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
[BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
[BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
attrs, ARRAY_SIZE(policy))
+ || (br_name && !attrs[BRC_GENL_A_DP_NAME])
|| (port_name && !attrs[BRC_GENL_A_PORT_NAME])
|| (count && !attrs[BRC_GENL_A_FDB_COUNT])
|| (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
}
*seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
- *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
+ if (br_name) {
+ *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
+ }
if (port_name) {
*port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
}
return 0;
}
-static void
-send_reply(uint32_t seq, int error, struct ofpbuf *fdb_query_data)
+/* Composes and returns a reply to a request made by the datapath with Netlink
+ * sequence number 'seq' and error code 'error'. The caller may add additional
+ * attributes to the message, then it may send it with send_reply(). */
+static struct ofpbuf *
+compose_reply(uint32_t seq, int error)
{
- struct ofpbuf msg;
- int retval;
-
- /* Compose reply. */
- ofpbuf_init(&msg, 0);
- nl_msg_put_genlmsghdr(&msg, brc_sock, 32, brc_family, NLM_F_REQUEST,
+ struct ofpbuf *reply = ofpbuf_new(4096);
+ nl_msg_put_genlmsghdr(reply, brc_sock, 32, brc_family, NLM_F_REQUEST,
BRC_GENL_C_DP_RESULT, 1);
- ((struct nlmsghdr *) msg.data)->nlmsg_seq = seq;
- nl_msg_put_u32(&msg, BRC_GENL_A_ERR_CODE, error);
- if (fdb_query_data) {
- nl_msg_put_unspec(&msg, BRC_GENL_A_FDB_DATA,
- fdb_query_data->data, fdb_query_data->size);
- }
+ ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
+ nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
+ return reply;
+}
- /* Send reply. */
- retval = nl_sock_send(brc_sock, &msg, false);
+/* Sends 'reply' to the datapath and frees it. */
+static void
+send_reply(struct ofpbuf *reply)
+{
+ int retval = nl_sock_send(brc_sock, reply, false);
if (retval) {
VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
strerror(retval));
}
- ofpbuf_uninit(&msg);
+ ofpbuf_delete(reply);
+}
+
+/* Composes and sends a reply to a request made by the datapath with Netlink
+ * sequence number 'seq' and error code 'error'. */
+static void
+send_simple_reply(uint32_t seq, int error)
+{
+ send_reply(compose_reply(seq, error));
}
static int
if (!error) {
error = rewrite_and_reload_config();
}
- send_reply(seq, error, NULL);
+ send_simple_reply(seq, error);
}
return error;
}
VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
error = rewrite_and_reload_config();
}
- send_reply(seq, error, NULL);
+ send_simple_reply(seq, error);
}
return error;
}
+/* Returns the name of the bridge that contains a port named 'port_name', as a
+ * malloc'd string that the caller must free, or a null pointer if no bridge
+ * contains a port named 'port_name'. */
+static char *
+get_bridge_containing_port(const char *port_name)
+{
+ struct svec matches;
+ const char *start, *end;
+
+ svec_init(&matches);
+ cfg_get_matches(&matches, "bridge.*.port=%s", port_name);
+ if (!matches.n) {
+ return 0;
+ }
+
+ start = matches.names[0] + strlen("bridge.");
+ end = strstr(start, ".port=");
+ assert(end);
+ return xmemdup0(start, end - start);
+}
+
+static int
+linux_bridge_to_ovs_bridge(const char *linux_bridge,
+ char **ovs_bridge, int *br_vlan)
+{
+ if (bridge_exists(linux_bridge)) {
+ /* Bridge name is the same. We are interested in VLAN 0. */
+ *ovs_bridge = xstrdup(linux_bridge);
+ *br_vlan = 0;
+ return 0;
+ } else {
+ /* No such Open vSwitch bridge 'linux_bridge', but there might be an
+ * internal port named 'linux_bridge' on some other bridge
+ * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
+ * port 'linux_bridge' on the bridge named 'ovs_bridge'. */
+ const char *port_name = linux_bridge;
+
+ *ovs_bridge = get_bridge_containing_port(port_name);
+ *br_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
+ if (*ovs_bridge && *br_vlan >= 0) {
+ return 0;
+ } else {
+ free(*ovs_bridge);
+ return ENODEV;
+ }
+ }
+}
+
static int
handle_fdb_query_cmd(struct ofpbuf *buffer)
{
int n_local_macs;
int i;
+ /* Impedance matching between the vswitchd and Linux kernel notions of what
+ * a bridge is. The kernel only handles a single VLAN per bridge, but
+ * vswitchd can deal with all the VLANs on a single bridge. We have to
+ * pretend that the former is the case even though the latter is the
+ * implementation. */
+ const char *linux_bridge; /* Name used by brctl. */
+ char *ovs_bridge; /* Name used by ovs-vswitchd. */
+ int br_vlan; /* VLAN tag. */
+ struct svec ifaces;
+
struct ofpbuf query_data;
+ struct ofpbuf *reply;
char *unixctl_command;
uint64_t count, skip;
- const char *br_name;
- struct svec ifaces;
char *output;
char *save_ptr;
uint32_t seq;
int error;
/* Parse the command received from brcompat_mod. */
- error = parse_command(buffer, &seq, &br_name, NULL, &count, &skip);
+ error = parse_command(buffer, &seq, &linux_bridge, NULL, &count, &skip);
if (error) {
return error;
}
+ /* Figure out vswitchd bridge and VLAN. */
+ cfg_read();
+ error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
+ if (error) {
+ send_simple_reply(seq, error);
+ return error;
+ }
+
/* Fetch the forwarding database using ovs-appctl. */
- unixctl_command = xasprintf("fdb/show %s", br_name);
+ unixctl_command = xasprintf("fdb/show %s", ovs_bridge);
error = execute_appctl_command(unixctl_command, &output);
free(unixctl_command);
if (error) {
- send_reply(seq, error, NULL);
+ free(ovs_bridge);
+ send_simple_reply(seq, error);
return error;
}
/* Fetch the MAC address for each interface on the bridge, so that we can
* fill in the is_local field in the response. */
- cfg_read();
- get_bridge_ifaces(br_name, &ifaces);
+ svec_init(&ifaces);
+ get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
local_macs = xmalloc(ifaces.n * sizeof *local_macs);
n_local_macs = 0;
for (i = 0; i < ifaces.n; i++) {
const char *iface_name = ifaces.names[i];
struct mac *mac = &local_macs[n_local_macs];
- if (!netdev_nodev_get_etheraddr(iface_name, mac->addr)) {
- n_local_macs++;
+ struct netdev *netdev;
+
+ error = netdev_open(iface_name, NETDEV_ETH_TYPE_NONE, &netdev);
+ if (netdev) {
+ if (!netdev_get_etheraddr(netdev, mac->addr)) {
+ n_local_macs++;
+ }
+ netdev_close(netdev);
}
}
svec_destroy(&ifaces);
continue;
}
+ if (vlan != br_vlan) {
+ continue;
+ }
+
if (skip > 0) {
skip--;
continue;
}
free(output);
- send_reply(seq, 0, &query_data);
+ /* Compose and send reply to datapath. */
+ reply = compose_reply(seq, 0);
+ nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
+ query_data.data, query_data.size);
+ send_reply(reply);
+
+ /* Free memory. */
ofpbuf_uninit(&query_data);
+ free(ovs_bridge);
+
+ return 0;
+}
+
+static void
+send_ifindex_reply(uint32_t seq, struct svec *ifaces)
+{
+ struct ofpbuf *reply;
+ const char *iface;
+ size_t n_indices;
+ int *indices;
+ size_t i;
+
+ /* Make sure that any given interface only occurs once. This shouldn't
+ * happen, but who knows what people put into their configuration files. */
+ svec_sort_unique(ifaces);
+
+ /* Convert 'ifaces' into ifindexes. */
+ n_indices = 0;
+ indices = xmalloc(ifaces->n * sizeof *indices);
+ SVEC_FOR_EACH (i, iface, ifaces) {
+ int ifindex = if_nametoindex(iface);
+ if (ifindex) {
+ indices[n_indices++] = ifindex;
+ }
+ }
+
+ /* Compose and send reply. */
+ reply = compose_reply(seq, 0);
+ nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
+ indices, n_indices * sizeof *indices);
+ send_reply(reply);
+
+ /* Free memory. */
+ free(indices);
+}
+
+static int
+handle_get_bridges_cmd(struct ofpbuf *buffer)
+{
+ struct svec bridges;
+ const char *br_name;
+ size_t i;
+
+ uint32_t seq;
+
+ int error;
+
+ /* Parse Netlink command.
+ *
+ * The command doesn't actually have any arguments, but we need the
+ * sequence number to send the reply. */
+ error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
+ if (error) {
+ return error;
+ }
+
+ /* Get all the real bridges and all the fake ones. */
+ cfg_read();
+ svec_init(&bridges);
+ cfg_get_subsections(&bridges, "bridge");
+ SVEC_FOR_EACH (i, br_name, &bridges) {
+ const char *iface_name;
+ struct svec ifaces;
+ size_t j;
+
+ svec_init(&ifaces);
+ get_bridge_ifaces(br_name, &ifaces, -1);
+ SVEC_FOR_EACH (j, iface_name, &ifaces) {
+ if (cfg_get_bool(0, "iface.%s.fake-bridge", iface_name)) {
+ svec_add(&bridges, iface_name);
+ }
+ }
+ svec_destroy(&ifaces);
+ }
+
+ send_ifindex_reply(seq, &bridges);
+ svec_destroy(&bridges);
+
+ return 0;
+}
+
+static int
+handle_get_ports_cmd(struct ofpbuf *buffer)
+{
+ uint32_t seq;
+
+ const char *linux_bridge;
+ char *ovs_bridge;
+ int br_vlan;
+
+ struct svec ports;
+
+ int error;
+
+ /* Parse Netlink command. */
+ error = parse_command(buffer, &seq, &linux_bridge, NULL, NULL, NULL);
+ if (error) {
+ return error;
+ }
+
+ cfg_read();
+ error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
+ if (error) {
+ send_simple_reply(seq, error);
+ return error;
+ }
+
+ svec_init(&ports);
+ get_bridge_ports(ovs_bridge, &ports, br_vlan);
+ svec_sort(&ports);
+ svec_del(&ports, linux_bridge);
+ send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
+ svec_destroy(&ports);
+
+ free(ovs_bridge);
return 0;
}
retval = handle_fdb_query_cmd(buffer);
break;
+ case BRC_GENL_C_GET_BRIDGES:
+ retval = handle_get_bridges_cmd(buffer);
+ break;
+
+ case BRC_GENL_C_GET_PORTS:
+ retval = handle_get_ports_cmd(buffer);
+ break;
+
default:
retval = EPROTO;
}
const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
char br_name[IFNAMSIZ];
uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
- struct svec ports;
- enum netdev_flags flags;
if (!if_indextoname(br_idx, br_name)) {
ofpbuf_delete(buf);
return;
}
- if (netdev_nodev_get_flags(port_name, &flags) == ENODEV) {
+ if (!netdev_exists(port_name)) {
/* Network device is really gone. */
+ struct svec ports;
+
VLOG_INFO("network device %s destroyed, "
"removing from bridge %s", port_name, br_name);
+
svec_init(&ports);
cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
svec_sort(&ports);
del_port(br_name, port_name);
rewrite_and_reload_config();
}
+ svec_destroy(&ports);
} else {
/* A network device by that name exists even though the kernel
* told us it had disappeared. Probably, what happened was
}
}
- cfg_read();
+ retval = cfg_read();
+ if (retval) {
+ ovs_fatal(retval, "could not read config file");
+ }
for (;;) {
unixctl_server_run(unixctl);
brc_recv_update();
+ netdev_run();
/* If 'prune_timeout' is non-zero, we actively prune from the
* config file any 'bridge.<br_name>.port' entries that are no
nl_sock_wait(brc_sock, POLLIN);
unixctl_server_wait(unixctl);
+ netdev_wait();
poll_block();
}
"use --help for usage");
}
+ cfg_init();
config_file = argv[0];
error = cfg_set_file(config_file);
if (error) {