X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fovs-brcompatd.c;h=594d474b1d6e831dc76a7fcb67ad0aa7c958b591;hb=250382a8fc839cb077b2636983a1c1de0d6b0e17;hp=93d9469bda82836a6bcddda132de540c0fe9a984;hpb=064af42167bf4fc9aaea2702d80ce08074b889c0;p=openvswitch
diff --git a/vswitchd/ovs-brcompatd.c b/vswitchd/ovs-brcompatd.c
index 93d9469b..594d474b 100644
--- a/vswitchd/ovs-brcompatd.c
+++ b/vswitchd/ovs-brcompatd.c
@@ -1,22 +1,21 @@
/* Copyright (c) 2008, 2009 Nicira Networks
- *
- * This program is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
*
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
*
- * You should have received a copy of the GNU General Public License
- * along with this program. If not, see .
+ * http://www.apache.org/licenses/LICENSE-2.0
*
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
*/
#include
+#include
#include
#include
#include
@@ -30,22 +29,23 @@
#include
#include
#include
+#include
#include
#include
-#include "cfg.h"
#include "command-line.h"
#include "coverage.h"
#include "daemon.h"
#include "dirs.h"
-#include "dpif.h"
+#include "dynamic-string.h"
#include "fatal-signal.h"
-#include "fault.h"
#include "leak-checker.h"
#include "netdev.h"
#include "netlink.h"
#include "ofpbuf.h"
#include "openvswitch/brcompat-netlink.h"
+#include "ovsdb-idl.h"
+#include "packets.h"
#include "poll-loop.h"
#include "process.h"
#include "signals.h"
@@ -53,6 +53,7 @@
#include "timeval.h"
#include "unixctl.h"
#include "util.h"
+#include "vswitchd/vswitch-idl.h"
#include "vlog.h"
#define THIS_MODULE VLM_brcompatd
@@ -68,25 +69,19 @@ enum bmc_action {
BMC_DEL_PORT
};
-static void parse_options(int argc, char *argv[]);
+static const char *parse_options(int argc, char *argv[]);
static void usage(void) NO_RETURN;
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 60);
-/* Maximum number of milliseconds to wait for the config file to be
- * unlocked. If set to zero, no waiting will occur. */
-static int lock_timeout = 500;
-
/* Maximum number of milliseconds to wait before pruning port entries that
* no longer exist. If set to zero, ports are never pruned. */
static int prune_timeout = 5000;
-/* Config file shared with ovs-vswitchd (usually ovs-vswitchd.conf). */
-static char *config_file;
-
-/* Command to run (via system()) to reload the ovs-vswitchd configuration
- * file. */
-static char *reload_command;
+/* Shell command to execute (via popen()) to send a control command to the
+ * running ovs-vswitchd process. The string must contain one instance of %s,
+ * which is replaced by the control command. */
+static char *appctl_command;
/* Netlink socket to listen for interface changes. */
static struct nl_sock *rtnl_sock;
@@ -171,44 +166,126 @@ static const struct nl_policy brc_dp_policy[] = {
[BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
};
-static bool
-bridge_exists(const char *name)
+static struct ovsrec_bridge *
+find_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
{
- return cfg_has_section("bridge.%s", name);
+ size_t i;
+
+ for (i = 0; i < ovs->n_bridges; i++) {
+ if (!strcmp(br_name, ovs->bridges[i]->name)) {
+ return ovs->bridges[i];
+ }
+ }
+
+ return NULL;
}
static int
-rewrite_and_reload_config(void)
-{
- if (cfg_is_dirty()) {
- int error1 = cfg_write();
- int error2 = cfg_read();
- long long int reload_start = time_msec();
- int error3 = system(reload_command);
- long long int elapsed = time_msec() - reload_start;
- COVERAGE_INC(brcompatd_reload);
- if (elapsed > 0) {
- VLOG_INFO("reload command executed in %lld ms", elapsed);
+execute_appctl_command(const char *unixctl_command, char **output)
+{
+ char *stdout_log, *stderr_log;
+ int error, status;
+ char *argv[5];
+
+ argv[0] = "/bin/sh";
+ argv[1] = "-c";
+ argv[2] = xasprintf(appctl_command, unixctl_command);
+ argv[3] = NULL;
+
+ /* Run process and log status. */
+ error = process_run_capture(argv, &stdout_log, &stderr_log, &status);
+ if (error) {
+ VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
+ unixctl_command, strerror(error));
+ } else if (status) {
+ char *msg = process_status_msg(status);
+ VLOG_ERR("ovs-appctl exited with error (%s)", msg);
+ free(msg);
+ error = ECHILD;
+ }
+
+ /* Deal with stdout_log. */
+ if (output) {
+ *output = stdout_log;
+ } else {
+ free(stdout_log);
+ }
+
+ /* Deal with stderr_log */
+ if (stderr_log && *stderr_log) {
+ VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
+ }
+ free(stderr_log);
+
+ free(argv[2]);
+
+ return error;
+}
+
+static void
+do_get_bridge_parts(const struct ovsrec_bridge *br, struct svec *parts,
+ int vlan, bool break_down_bonds)
+{
+ struct svec ports;
+ size_t i, j;
+
+ svec_init(&ports);
+ for (i = 0; i < br->n_ports; i++) {
+ const struct ovsrec_port *port = br->ports[i];
+
+ svec_add(&ports, port->name);
+ if (vlan >= 0) {
+ int port_vlan = port->n_tag ? *port->tag : 0;
+ if (vlan != port_vlan) {
+ continue;
+ }
}
- if (error3 == -1) {
- VLOG_ERR("failed to execute reload command: %s", strerror(errno));
- } else if (error3 != 0) {
- char *msg = process_status_msg(error3);
- VLOG_ERR("reload command exited with error (%s)", msg);
- free(msg);
+ if (break_down_bonds) {
+ for (j = 0; j < port->n_interfaces; j++) {
+ const struct ovsrec_interface *iface = port->interfaces[j];
+ svec_add(parts, iface->name);
+ }
+ } else {
+ svec_add(parts, port->name);
}
- return error1 ? error1 : error2 ? error2 : error3 ? ECHILD : 0;
}
- return 0;
+ 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 struct ovsrec_bridge *br, struct svec *ifaces,
+ int vlan)
+{
+ do_get_bridge_parts(br, 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 struct ovsrec_bridge *br, struct svec *ports,
+ int vlan)
+{
+ do_get_bridge_parts(br, ports, vlan, false);
}
+#if 0
/* 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)) {
@@ -221,32 +298,13 @@ prune_ports(void)
cfg_get_subsections(&bridges, "bridge");
for (i=0; iheader_);
+}
-/* 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. */
static bool
-netdev_exists(const char *name)
+port_is_fake_bridge(const struct ovsrec_port *port)
{
- struct stat s;
- char *filename;
- int error;
-
- filename = xasprintf("/sys/class/net/%s", name);
- error = stat(filename, &s);
- free(filename);
- return !error;
+ return (port->fake_bridge
+ && port->tag
+ && *port->tag >= 1 && *port->tag <= 4095);
}
+static void
+ovs_insert_bridge(const struct ovsrec_open_vswitch *ovs,
+ struct ovsrec_bridge *bridge)
+{
+ struct ovsrec_bridge **bridges;
+ size_t i;
+
+ bridges = xmalloc(sizeof *ovs->bridges * (ovs->n_bridges + 1));
+ for (i = 0; i < ovs->n_bridges; i++) {
+ bridges[i] = ovs->bridges[i];
+ }
+ bridges[ovs->n_bridges] = bridge;
+ ovsrec_open_vswitch_set_bridges(ovs, bridges, ovs->n_bridges + 1);
+ free(bridges);
+}
+
static int
-add_bridge(const char *br_name)
+add_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
{
- if (bridge_exists(br_name)) {
+ struct ovsrec_bridge *br;
+ struct ovsrec_port *port;
+ struct ovsrec_interface *iface;
+
+ if (find_bridge(ovs, br_name)) {
VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
return EEXIST;
} else if (netdev_exists(br_name)) {
- if (cfg_get_bool(0, "iface.%s.fake-bridge", br_name)) {
- VLOG_WARN("addbr %s: %s exists as a fake bridge",
- br_name, br_name);
- return 0;
- } else {
- VLOG_WARN("addbr %s: cannot create bridge %s because a network "
- "device named %s already exists",
- br_name, br_name, br_name);
- return EEXIST;
+ size_t i;
+
+ for (i = 0; i < ovs->n_bridges; i++) {
+ size_t j;
+ struct ovsrec_bridge *br_cfg = ovs->bridges[i];
+
+ for (j = 0; j < br_cfg->n_ports; j++) {
+ if (port_is_fake_bridge(br_cfg->ports[j])) {
+ VLOG_WARN("addbr %s: %s exists as a fake bridge",
+ br_name, br_name);
+ return 0;
+ }
+ }
}
+
+ VLOG_WARN("addbr %s: cannot create bridge %s because a network "
+ "device named %s already exists",
+ br_name, br_name, br_name);
+ return EEXIST;
}
- cfg_add_entry("bridge.%s.port=%s", br_name, br_name);
+ iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
+ ovsrec_interface_set_name(iface, br_name);
+
+ port = ovsrec_port_insert(txn_from_openvswitch(ovs));
+ ovsrec_port_set_name(port, br_name);
+ ovsrec_port_set_interfaces(port, &iface, 1);
+
+ br = ovsrec_bridge_insert(txn_from_openvswitch(ovs));
+ ovsrec_bridge_set_name(br, br_name);
+ ovsrec_bridge_set_ports(br, &port, 1);
+
+ ovs_insert_bridge(ovs, br);
+
VLOG_INFO("addbr %s: success", br_name);
return 0;
}
+static void
+add_port(const struct ovsrec_open_vswitch *ovs,
+ const struct ovsrec_bridge *br, const char *port_name)
+{
+ struct ovsrec_interface *iface;
+ struct ovsrec_port *port;
+ struct ovsrec_port **ports;
+ size_t i;
+
+ /* xxx Check conflicts? */
+ iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
+ ovsrec_interface_set_name(iface, port_name);
+
+ port = ovsrec_port_insert(txn_from_openvswitch(ovs));
+ ovsrec_port_set_name(port, port_name);
+ ovsrec_port_set_interfaces(port, &iface, 1);
+
+ ports = xmalloc(sizeof *br->ports * (br->n_ports + 1));
+ for (i = 0; i < br->n_ports; i++) {
+ ports[i] = br->ports[i];
+ }
+ ports[br->n_ports] = port;
+ ovsrec_bridge_set_ports(br, ports, br->n_ports + 1);
+ free(ports);
+}
+
+static void
+del_port(const struct ovsrec_bridge *br, const char *port_name)
+{
+ size_t i, j;
+ struct ovsrec_port *port_rec = NULL;
+
+ for (i = 0; i < br->n_ports; i++) {
+ struct ovsrec_port *port = br->ports[i];
+ if (!strcmp(port_name, port->name)) {
+ port_rec = port;
+ }
+ for (j = 0; j < port->n_interfaces; j++) {
+ struct ovsrec_interface *iface = port->interfaces[j];
+ if (!strcmp(port_name, iface->name)) {
+ ovsrec_interface_delete(iface);
+ }
+ }
+ }
+
+ /* xxx Probably can move this into the "for" loop. */
+ if (port_rec) {
+ struct ovsrec_port **ports;
+ size_t n;
+
+ ports = xmalloc(sizeof *br->ports * br->n_ports);
+ for (i = n = 0; i < br->n_ports; i++) {
+ if (br->ports[i] != port_rec) {
+ ports[n++] = br->ports[i];
+ }
+ }
+ ovsrec_bridge_set_ports(br, ports, n);
+ free(ports);
+ }
+}
+
static int
-del_bridge(const char *br_name)
+del_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
{
- if (!bridge_exists(br_name)) {
+ struct ovsrec_bridge *br = find_bridge(ovs, br_name);
+ struct ovsrec_bridge **bridges;
+ size_t i, n;
+
+ if (!br) {
VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
return ENXIO;
}
- cfg_del_section("bridge.%s", br_name);
+ del_port(br, br_name);
+
+ ovsrec_bridge_delete(br);
+
+ bridges = xmalloc(sizeof *ovs->bridges * ovs->n_bridges);
+ for (i = n = 0; i < ovs->n_bridges; i++) {
+ if (ovs->bridges[i] != br) {
+ bridges[n++] = ovs->bridges[i];
+ }
+ }
+ ovsrec_open_vswitch_set_bridges(ovs, bridges, n);
+ free(bridges);
+
VLOG_INFO("delbr %s: success", br_name);
return 0;
@@ -349,64 +514,87 @@ del_bridge(const char *br_name)
static int
parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
- const char **port_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 },
};
struct nlattr *attrs[ARRAY_SIZE(policy)];
if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
attrs, ARRAY_SIZE(policy))
- || (port_name && !attrs[BRC_GENL_A_PORT_NAME])) {
+ || (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])) {
return EINVAL;
}
*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]);
}
+ if (count) {
+ *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
+ }
+ if (skip) {
+ *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
+ }
return 0;
}
-static void
-send_reply(uint32_t seq, int error)
+/* 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);
+ ((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
-handle_bridge_cmd(struct ofpbuf *buffer, bool add)
+handle_bridge_cmd(const struct ovsrec_open_vswitch *ovs,
+ struct ofpbuf *buffer, bool add)
{
const char *br_name;
uint32_t seq;
int error;
- error = parse_command(buffer, &seq, &br_name, NULL);
+ error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
if (!error) {
- error = add ? add_bridge(br_name) : del_bridge(br_name);
- if (!error) {
- error = rewrite_and_reload_config();
- }
- send_reply(seq, error);
+ error = add ? add_bridge(ovs, br_name) : del_bridge(ovs, br_name);
+ send_simple_reply(seq, error);
}
return error;
}
@@ -416,25 +604,20 @@ static const struct nl_policy brc_port_policy[] = {
[BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
};
-static void
-del_port(const char *br_name, const char *port_name)
-{
- cfg_del_entry("bridge.%s.port=%s", br_name, port_name);
- cfg_del_match("bonding.*.slave=%s", port_name);
- cfg_del_match("vlan.%s.*", port_name);
-}
-
static int
-handle_port_cmd(struct ofpbuf *buffer, bool add)
+handle_port_cmd(const struct ovsrec_open_vswitch *ovs,
+ struct ofpbuf *buffer, bool add)
{
const char *cmd_name = add ? "add-if" : "del-if";
const char *br_name, *port_name;
uint32_t seq;
int error;
- error = parse_command(buffer, &seq, &br_name, &port_name);
+ error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
if (!error) {
- if (!bridge_exists(br_name)) {
+ struct ovsrec_bridge *br = find_bridge(ovs, br_name);
+
+ if (!br) {
VLOG_WARN("%s %s %s: no bridge named %s",
cmd_name, br_name, port_name, br_name);
error = EINVAL;
@@ -444,21 +627,324 @@ handle_port_cmd(struct ofpbuf *buffer, bool add)
error = EINVAL;
} else {
if (add) {
- cfg_add_entry("bridge.%s.port=%s", br_name, port_name);
+ add_port(ovs, br, port_name);
} else {
- del_port(br_name, port_name);
+ del_port(br, port_name);
}
VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
- error = rewrite_and_reload_config();
}
- send_reply(seq, error);
+ send_simple_reply(seq, error);
}
return error;
}
+/* The caller is responsible for freeing '*ovs_name' if the call is
+ * successful. */
+static int
+linux_bridge_to_ovs_bridge(const struct ovsrec_open_vswitch *ovs,
+ const char *linux_name,
+ const struct ovsrec_bridge **ovs_bridge,
+ int *br_vlan)
+{
+ *ovs_bridge = find_bridge(ovs, linux_name);
+ if (*ovs_bridge) {
+ /* Bridge name is the same. We are interested in VLAN 0. */
+ *br_vlan = 0;
+ return 0;
+ } else {
+ /* No such Open vSwitch bridge 'linux_name', but there might be an
+ * internal port named 'linux_name' on some other bridge
+ * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
+ * port 'linux_name' on the bridge named 'ovs_bridge'. */
+ size_t i, j;
+
+ for (i = 0; i < ovs->n_bridges; i++) {
+ const struct ovsrec_bridge *br = ovs->bridges[i];
+
+ for (j = 0; j < br->n_ports; j++) {
+ const struct ovsrec_port *port = br->ports[j];
+
+ if (!strcmp(port->name, linux_name)) {
+ *ovs_bridge = br;
+ *br_vlan = port->n_tag ? *port->tag : -1;
+ return 0;
+ }
+ }
+
+ }
+ return ENODEV;
+ }
+}
+
+static int
+handle_fdb_query_cmd(const struct ovsrec_open_vswitch *ovs,
+ struct ofpbuf *buffer)
+{
+ /* This structure is copied directly from the Linux 2.6.30 header files.
+ * It would be more straightforward to #include , but
+ * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
+ * with old header files won't have it. */
+ struct __fdb_entry {
+ __u8 mac_addr[6];
+ __u8 port_no;
+ __u8 is_local;
+ __u32 ageing_timer_value;
+ __u8 port_hi;
+ __u8 pad0;
+ __u16 unused;
+ };
+
+ struct mac {
+ uint8_t addr[6];
+ };
+ struct mac *local_macs;
+ 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_name; /* Name used by brctl. */
+ const struct ovsrec_bridge *ovs_bridge; /* Bridge 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;
+ char *output;
+ char *save_ptr;
+ uint32_t seq;
+ int error;
+
+ /* Parse the command received from brcompat_mod. */
+ error = parse_command(buffer, &seq, &linux_name, NULL, &count, &skip);
+ if (error) {
+ return error;
+ }
+
+ /* Figure out vswitchd bridge and VLAN. */
+ error = linux_bridge_to_ovs_bridge(ovs, linux_name,
+ &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", ovs_bridge->name);
+ error = execute_appctl_command(unixctl_command, &output);
+ free(unixctl_command);
+ if (error) {
+ 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. */
+ 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];
+ 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);
+
+ /* Parse the response from ovs-appctl and convert it to binary format to
+ * pass back to the kernel. */
+ ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
+ save_ptr = NULL;
+ strtok_r(output, "\n", &save_ptr); /* Skip header line. */
+ while (count > 0) {
+ struct __fdb_entry *entry;
+ int port, vlan, age;
+ uint8_t mac[ETH_ADDR_LEN];
+ char *line;
+ bool is_local;
+
+ line = strtok_r(NULL, "\n", &save_ptr);
+ if (!line) {
+ break;
+ }
+
+ if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
+ &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
+ != 2 + ETH_ADDR_SCAN_COUNT + 1) {
+ struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
+ continue;
+ }
+
+ if (vlan != br_vlan) {
+ continue;
+ }
+
+ if (skip > 0) {
+ skip--;
+ continue;
+ }
+
+ /* Is this the MAC address of an interface on the bridge? */
+ is_local = false;
+ for (i = 0; i < n_local_macs; i++) {
+ if (eth_addr_equals(local_macs[i].addr, mac)) {
+ is_local = true;
+ break;
+ }
+ }
+
+ entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
+ memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
+ entry->port_no = port & 0xff;
+ entry->is_local = is_local;
+ entry->ageing_timer_value = age * HZ;
+ entry->port_hi = (port & 0xff00) >> 8;
+ entry->pad0 = 0;
+ entry->unused = 0;
+ count--;
+ }
+ free(output);
+
+ /* 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);
+
+ 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(const struct ovsrec_open_vswitch *ovs,
+ struct ofpbuf *buffer)
+{
+ struct svec bridges;
+ size_t i, j;
+
+ 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. */
+ svec_init(&bridges);
+ for (i = 0; i < ovs->n_bridges; i++) {
+ const struct ovsrec_bridge *br = ovs->bridges[i];
+
+ svec_add(&bridges, br->name);
+ for (j = 0; j < br->n_ports; j++) {
+ const struct ovsrec_port *port = br->ports[j];
+
+ if (port->fake_bridge) {
+ svec_add(&bridges, port->name);
+ }
+ }
+ }
+
+ send_ifindex_reply(seq, &bridges);
+ svec_destroy(&bridges);
+
+ return 0;
+}
+
static int
-brc_recv_update(void)
+handle_get_ports_cmd(const struct ovsrec_open_vswitch *ovs,
+ struct ofpbuf *buffer)
+{
+ uint32_t seq;
+
+ const char *linux_name;
+ const struct ovsrec_bridge *ovs_bridge;
+ int br_vlan;
+
+ struct svec ports;
+
+ int error;
+
+ /* Parse Netlink command. */
+ error = parse_command(buffer, &seq, &linux_name, NULL, NULL, NULL);
+ if (error) {
+ return error;
+ }
+
+ error = linux_bridge_to_ovs_bridge(ovs, linux_name,
+ &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_name);
+ send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
+ svec_destroy(&ports);
+
+ return 0;
+}
+
+static void
+brc_recv_update(const struct ovsrec_open_vswitch *ovs)
{
int retval;
struct ofpbuf *buffer;
@@ -477,7 +963,7 @@ brc_recv_update(void)
if (retval != EAGAIN) {
VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
}
- return retval;
+ return;
}
genlmsghdr = nl_msg_genlmsghdr(buffer);
@@ -492,40 +978,47 @@ brc_recv_update(void)
goto error;
}
- if (cfg_lock(NULL, lock_timeout)) {
- /* Couldn't lock config file. */
- retval = EAGAIN;
- goto error;
- }
-
switch (genlmsghdr->cmd) {
case BRC_GENL_C_DP_ADD:
- retval = handle_bridge_cmd(buffer, true);
+ handle_bridge_cmd(ovs, buffer, true);
break;
case BRC_GENL_C_DP_DEL:
- retval = handle_bridge_cmd(buffer, false);
+ handle_bridge_cmd(ovs, buffer, false);
break;
case BRC_GENL_C_PORT_ADD:
- retval = handle_port_cmd(buffer, true);
+ handle_port_cmd(ovs, buffer, true);
break;
case BRC_GENL_C_PORT_DEL:
- retval = handle_port_cmd(buffer, false);
+ handle_port_cmd(ovs, buffer, false);
+ break;
+
+ case BRC_GENL_C_FDB_QUERY:
+ handle_fdb_query_cmd(ovs, buffer);
+ break;
+
+ case BRC_GENL_C_GET_BRIDGES:
+ handle_get_bridges_cmd(ovs, buffer);
+ break;
+
+ case BRC_GENL_C_GET_PORTS:
+ handle_get_ports_cmd(ovs, buffer);
break;
default:
- retval = EPROTO;
+ VLOG_WARN_RL(&rl, "received unknown brc netlink command: %d\n",
+ genlmsghdr->cmd);
+ break;
}
- cfg_unlock();
-
error:
ofpbuf_delete(buffer);
- return retval;
+ return;
}
+#if 0
/* Check for interface configuration changes announced through RTNL. */
static void
rtnl_recv_update(void)
@@ -564,7 +1057,6 @@ rtnl_recv_update(void)
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;
if (!if_indextoname(br_idx, br_name)) {
ofpbuf_delete(buf);
@@ -578,35 +1070,93 @@ rtnl_recv_update(void)
return;
}
- svec_init(&ports);
- cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
- svec_sort(&ports);
- if (svec_contains(&ports, port_name)) {
- del_port(br_name, port_name);
- rewrite_and_reload_config();
+ 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);
+ if (svec_contains(&ports, port_name)) {
+ del_port(br_name, port_name);
+ }
+ svec_destroy(&ports);
+ } else {
+ /* A network device by that name exists even though the kernel
+ * told us it had disappeared. Probably, what happened was
+ * this:
+ *
+ * 1. Device destroyed.
+ * 2. Notification sent to us.
+ * 3. New device created with same name as old one.
+ * 4. ovs-brcompatd notified, removes device from bridge.
+ *
+ * There's no a priori reason that in this situation that the
+ * new device with the same name should remain in the bridge;
+ * on the contrary, that would be unexpected. *But* there is
+ * one important situation where, if we do this, bad things
+ * happen. This is the case of XenServer Tools version 5.0.0,
+ * which on boot of a Windows VM cause something like this to
+ * happen on the Xen host:
+ *
+ * i. Create tap1.0 and vif1.0.
+ * ii. Delete tap1.0.
+ * iii. Delete vif1.0.
+ * iv. Re-create vif1.0.
+ *
+ * (XenServer Tools 5.5.0 does not exhibit this behavior, and
+ * neither does a VM without Tools installed at all.@.)
+ *
+ * Steps iii and iv happen within a few seconds of each other.
+ * Step iv causes /etc/xensource/scripts/vif to run, which in
+ * turn calls ovs-cfg-mod to add the new device to the bridge.
+ * If step iv happens after step 4 (in our first list of
+ * steps), then all is well, but if it happens between 3 and 4
+ * (which can easily happen if ovs-brcompatd has to wait to
+ * lock the configuration file), then we will remove the new
+ * incarnation from the bridge instead of the old one!
+ *
+ * So, to avoid this problem, we do nothing here. This is
+ * strictly incorrect except for this one particular case, and
+ * perhaps that will bite us someday. If that happens, then we
+ * will have to somehow track network devices by ifindex, since
+ * a new device will have a new ifindex even if it has the same
+ * name as an old device.
+ */
+ VLOG_INFO("kernel reported network device %s removed but "
+ "a device by that name exists (XS Tools 5.0.0?)",
+ port_name);
}
cfg_unlock();
}
ofpbuf_delete(buf);
}
}
+#endif
int
main(int argc, char *argv[])
{
struct unixctl_server *unixctl;
+ const char *remote;
+ struct ovsdb_idl *idl;
int retval;
set_program_name(argv[0]);
- register_fault_handlers();
time_init();
vlog_init();
- parse_options(argc, argv);
+ vlog_set_levels(VLM_ANY_MODULE, VLF_CONSOLE, VLL_WARN);
+ vlog_set_levels(VLM_reconnect, VLF_ANY_FACILITY, VLL_WARN);
+
+ remote = parse_options(argc, argv);
signal(SIGPIPE, SIG_IGN);
process_init();
die_if_already_running();
- daemonize();
+ daemonize_start();
retval = unixctl_server_create(NULL, &unixctl);
if (retval) {
@@ -624,12 +1174,33 @@ main(int argc, char *argv[])
}
}
- cfg_read();
+ daemonize_complete();
+
+ idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
for (;;) {
+ const struct ovsrec_open_vswitch *ovs;
+ struct ovsdb_idl_txn *txn;
+ enum ovsdb_idl_txn_status status;
+
+ ovsdb_idl_run(idl);
+
+ txn = ovsdb_idl_txn_create(idl);
+
unixctl_server_run(unixctl);
- brc_recv_update();
+ ovs = ovsrec_open_vswitch_first(idl);
+ if (ovs) {
+ brc_recv_update(ovs);
+ } else if (ovsdb_idl_get_seqno(idl)) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ VLOG_WARN_RL(&rl, "%s: database does not contain any Open vSwitch "
+ "configuration", remote);
+ } else {
+ /* Haven't yet received initial database contents. */
+ }
+ netdev_run();
+#if 0
/* If 'prune_timeout' is non-zero, we actively prune from the
* config file any 'bridge..port' entries that are no
* longer valid. We use two methods:
@@ -647,44 +1218,96 @@ main(int argc, char *argv[])
nl_sock_wait(rtnl_sock, POLLIN);
poll_timer_wait(prune_timeout);
}
+#endif
+
+ while ((status = ovsdb_idl_txn_commit(txn)) == TXN_INCOMPLETE) {
+ ovsdb_idl_run(idl);
+ ovsdb_idl_wait(idl);
+ ovsdb_idl_txn_wait(txn);
+ poll_block();
+ }
+ ovsdb_idl_txn_destroy(txn);
+
+ switch (status) {
+ case TXN_INCOMPLETE:
+ NOT_REACHED();
+
+ case TXN_ABORTED:
+ /* Should not happen--we never call ovsdb_idl_txn_abort(). */
+ ovs_fatal(0, "transaction aborted");
+
+ case TXN_SUCCESS:
+ case TXN_UNCHANGED:
+ break;
+
+ case TXN_TRY_AGAIN:
+ /* xxx Handle this better! */
+ printf("xxx We need to try again!\n");
+ break;
+
+ case TXN_ERROR:
+ /* xxx Is this what we want to do? */
+ ovs_fatal(0, "transaction error");
+
+ default:
+ NOT_REACHED();
+ }
nl_sock_wait(brc_sock, POLLIN);
+ ovsdb_idl_wait(idl);
unixctl_server_wait(unixctl);
+ netdev_wait();
poll_block();
}
+ ovsdb_idl_destroy(idl);
+
return 0;
}
static void
+validate_appctl_command(void)
+{
+ const char *p;
+ int n;
+
+ n = 0;
+ for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
+ if (p[1] == '%') {
+ /* Nothing to do. */
+ } else if (p[1] == 's') {
+ n++;
+ } else {
+ ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
+ }
+ }
+ if (n != 1) {
+ ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
+ }
+}
+
+static const char *
parse_options(int argc, char *argv[])
{
enum {
- OPT_LOCK_TIMEOUT = UCHAR_MAX + 1,
OPT_PRUNE_TIMEOUT,
- OPT_RELOAD_COMMAND,
+ OPT_APPCTL_COMMAND,
VLOG_OPTION_ENUMS,
LEAK_CHECKER_OPTION_ENUMS
};
static struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"version", no_argument, 0, 'V'},
- {"lock-timeout", required_argument, 0, OPT_LOCK_TIMEOUT},
{"prune-timeout", required_argument, 0, OPT_PRUNE_TIMEOUT},
- {"reload-command", required_argument, 0, OPT_RELOAD_COMMAND},
+ {"appctl-command", required_argument, 0, OPT_APPCTL_COMMAND},
DAEMON_LONG_OPTIONS,
VLOG_LONG_OPTIONS,
LEAK_CHECKER_LONG_OPTIONS,
{0, 0, 0, 0},
};
char *short_options = long_options_to_short_options(long_options);
- int error;
- reload_command = xasprintf("%s/ovs-appctl -t "
- "%s/ovs-vswitchd.`cat %s/ovs-vswitchd.pid`.ctl "
- "-e vswitchd/reload 2>&1 "
- "| /usr/bin/logger -t brcompatd-reload",
- ovs_bindir, ovs_rundir, ovs_rundir);
+ appctl_command = xasprintf("%s/ovs-appctl %%s", ovs_bindir);
for (;;) {
int c;
@@ -702,16 +1325,12 @@ parse_options(int argc, char *argv[])
OVS_PRINT_VERSION(0, 0);
exit(EXIT_SUCCESS);
- case OPT_LOCK_TIMEOUT:
- lock_timeout = atoi(optarg);
- break;
-
case OPT_PRUNE_TIMEOUT:
prune_timeout = atoi(optarg) * 1000;
break;
- case OPT_RELOAD_COMMAND:
- reload_command = optarg;
+ case OPT_APPCTL_COMMAND:
+ appctl_command = optarg;
break;
VLOG_OPTION_HANDLERS
@@ -727,20 +1346,17 @@ parse_options(int argc, char *argv[])
}
free(short_options);
+ validate_appctl_command();
+
argc -= optind;
argv += optind;
if (argc != 1) {
- ovs_fatal(0, "exactly one non-option argument required; "
+ ovs_fatal(0, "database socket is non-option argument; "
"use --help for usage");
}
- config_file = argv[0];
- error = cfg_set_file(config_file);
- if (error) {
- ovs_fatal(error, "failed to add configuration file \"%s\"",
- config_file);
- }
+ return argv[0];
}
static void
@@ -751,9 +1367,8 @@ usage(void)
"CONFIG is the configuration file used by ovs-vswitchd.\n",
program_name, program_name);
printf("\nConfiguration options:\n"
- " --reload-command=COMMAND shell command to reload ovs-vswitchd\n"
+ " --appctl-command=COMMAND shell command to run ovs-appctl\n"
" --prune-timeout=SECS wait at most SECS before pruning ports\n"
- " --lock-timeout=MSECS wait at most MSECS for CONFIG to unlock\n"
);
daemon_usage();
vlog_usage();
@@ -761,6 +1376,6 @@ usage(void)
" -h, --help display this help message\n"
" -V, --version display version information\n");
leak_checker_usage();
- printf("\nThe default reload command is:\n%s\n", reload_command);
+ printf("\nThe default appctl command is:\n%s\n", appctl_command);
exit(EXIT_SUCCESS);
}