Openvswitch原理与代码分析(3): 添加一条流表flow

87 阅读 0 评论 58 点赞

我是靠谱客的博主靓丽抽屉，最近开发中收集的这篇文章主要介绍Openvswitch原理与代码分析(3): 添加一条流表flow，觉得挺不错的，现在分享给大家，希望可以做个参考。

概述

添加一个flow，调用的命令为

ovs-ofctl add-flow hello "hard_timeout=0 idle_timeout=0 priority=1 table=21 pkt_mark=0x55 tun_id=0x55 actions=mod_nw_dst:192.168.56.101,output:2"

这里调用的是调用ovs/utilities/ovs-ofctl.c的命令行工具

这个命令行工具支持的所有的命令及处理函数定义如下：

staticconststruct ovs_cmdl_command all_commands[] = {
{ "show", "switch",
1, 1, ofctl_show },
{ "monitor", "switch [misslen] [invalid_ttl] [watch:[...]]",
1, 3, ofctl_monitor },
{ "snoop", "switch",
1, 1, ofctl_snoop },
{ "dump-desc", "switch",
1, 1, ofctl_dump_desc },
{ "dump-tables", "switch",
1, 1, ofctl_dump_tables },
{ "dump-table-features", "switch",
1, 1, ofctl_dump_table_features },
{ "dump-table-desc", "switch",
1, 1, ofctl_dump_table_desc },
{ "dump-flows", "switch",
1, 2, ofctl_dump_flows },
{ "dump-aggregate", "switch",
1, 2, ofctl_dump_aggregate },
{ "queue-stats", "switch [port [queue]]",
1, 3, ofctl_queue_stats },
{ "queue-get-config", "switch port",
2, 2, ofctl_queue_get_config },
{ "add-flow", "switch flow",
2, 2, ofctl_add_flow },
{ "add-flows", "switch file",
2, 2, ofctl_add_flows },
{ "mod-flows", "switch flow",
2, 2, ofctl_mod_flows },
{ "del-flows", "switch [flow]",
1, 2, ofctl_del_flows },
{ "replace-flows", "switch file",
2, 2, ofctl_replace_flows },
{ "diff-flows", "source1 source2",
2, 2, ofctl_diff_flows },
{ "add-meter", "switch meter",
2, 2, ofctl_add_meter },
{ "mod-meter", "switch meter",
2, 2, ofctl_mod_meter },
{ "del-meter", "switch meter",
2, 2, ofctl_del_meters },
{ "del-meters", "switch",
1, 1, ofctl_del_meters },
{ "dump-meter", "switch meter",
2, 2, ofctl_dump_meters },
{ "dump-meters", "switch",
1, 1, ofctl_dump_meters },
{ "meter-stats", "switch [meter]",
1, 2, ofctl_meter_stats },
{ "meter-features", "switch",
1, 1, ofctl_meter_features },
{ "packet-out", "switch in_port actions packet...",
4, INT_MAX, ofctl_packet_out },
{ "dump-ports", "switch [port]",
1, 2, ofctl_dump_ports },
{ "dump-ports-desc", "switch [port]",
1, 2, ofctl_dump_ports_desc },
{ "mod-port", "switch iface act",
3, 3, ofctl_mod_port },
{ "mod-table", "switch mod",
3, 3, ofctl_mod_table },
{ "get-frags", "switch",
1, 1, ofctl_get_frags },
{ "set-frags", "switch frag_mode",
2, 2, ofctl_set_frags },
{ "probe", "target",
1, 1, ofctl_probe },
{ "ping", "target [n]",
1, 2, ofctl_ping },
{ "benchmark", "target n count",
3, 3, ofctl_benchmark },
{ "ofp-parse", "file",
1, 1, ofctl_ofp_parse },
{ "ofp-parse-pcap", "pcap",
1, INT_MAX, ofctl_ofp_parse_pcap },
{ "add-group", "switch group",
1, 2, ofctl_add_group },
{ "add-groups", "switch file",
1, 2, ofctl_add_groups },
{ "mod-group", "switch group",
1, 2, ofctl_mod_group },
{ "del-groups", "switch [group]",
1, 2, ofctl_del_groups },
{ "insert-buckets", "switch [group]",
1, 2, ofctl_insert_bucket },
{ "remove-buckets", "switch [group]",
1, 2, ofctl_remove_bucket },
{ "dump-groups", "switch [group]",
1, 2, ofctl_dump_group_desc },
{ "dump-group-stats", "switch [group]",
1, 2, ofctl_dump_group_stats },
{ "dump-group-features", "switch",
1, 1, ofctl_dump_group_features },
{ "add-tlv-map", "switch map",
2, 2, ofctl_add_tlv_map },
{ "del-tlv-map", "switch [map]",
1, 2, ofctl_del_tlv_map },
{ "dump-tlv-map", "switch",
1, 1, ofctl_dump_tlv_map },
{ "help", NULL, 0, INT_MAX, ofctl_help },
{ "list-commands", NULL, 0, INT_MAX, ofctl_list_commands },
/* Undocumented commands for testing. */
{ "parse-flow", NULL, 1, 1, ofctl_parse_flow },
{ "parse-flows", NULL, 1, 1, ofctl_parse_flows },
{ "parse-nx-match", NULL, 0, 0, ofctl_parse_nxm },
{ "parse-nxm", NULL, 0, 0, ofctl_parse_nxm },
{ "parse-oxm", NULL, 1, 1, ofctl_parse_oxm },
{ "parse-actions", NULL, 1, 1, ofctl_parse_actions },
{ "parse-instructions", NULL, 1, 1, ofctl_parse_instructions },
{ "parse-ofp10-match", NULL, 0, 0, ofctl_parse_ofp10_match },
{ "parse-ofp11-match", NULL, 0, 0, ofctl_parse_ofp11_match },
{ "parse-pcap", NULL, 1, 1, ofctl_parse_pcap },
{ "check-vlan", NULL, 2, 2, ofctl_check_vlan },
{ "print-error", NULL, 1, 1, ofctl_print_error },
{ "encode-error-reply", NULL, 2, 2, ofctl_encode_error_reply },
{ "ofp-print", NULL, 1, 2, ofctl_ofp_print },
{ "encode-hello", NULL, 1, 1, ofctl_encode_hello },
{ NULL, NULL, 0, 0, NULL },
};

根据这个数据结构的定义，"add-flow"调用的函数为

staticvoid
ofctl_add_flow(struct ovs_cmdl_context *ctx)
{
ofctl_flow_mod(ctx->argc, ctx->argv, OFPFC_ADD);
}

调用ofctl_flow_mod，parse_ofp_flow_mod_str将字符串解析为ofputil_flow_mod fm

ofputil_flow_mod包含两个最重要的成员变量：

struct match match，所谓match就是一个key。

struct ofpact *ofpacts; /* Series of "struct ofpact"s. */

staticvoid
ofctl_flow_mod(int argc,char *argv[], uint16_t command)
{
if (argc > 2 && !strcmp(argv[2], "-")) {
ofctl_flow_mod_file(argc, argv, command);
} else {
struct ofputil_flow_mod fm;
char *error;
enum ofputil_protocol usable_protocols;
error = parse_ofp_flow_mod_str(&fm, argc > 2 ? argv[2] : "", command,
&usable_protocols);
if (error) {
ovs_fatal(0, "%s", error);
}
ofctl_flow_mod__(argv[1], &fm, 1, usable_protocols);
}
}

ofctl_flow_mod__会打开一个指向ovs-vswitchd的socket，将flow match变成openflow的协议，发出去transact_noreply

staticvoid
ofctl_flow_mod__(constchar *remote,struct ofputil_flow_mod *fms,
size_t n_fms,enum ofputil_protocol usable_protocols)
{
enum ofputil_protocol protocol;
struct vconn *vconn;
size_t i;
if (bundle) {
bundle_flow_mod__(remote, fms, n_fms, usable_protocols);
return;
}
protocol = open_vconn_for_flow_mod(remote, &vconn, usable_protocols);
for (i = 0; i < n_fms; i++) {
struct ofputil_flow_mod *fm = &fms[i];
transact_noreply(vconn, ofputil_encode_flow_mod(fm, protocol));
free(CONST_CAST(struct ofpact *, fm->ofpacts));
}
vconn_close(vconn);
}

Ovs-vswitchd会监听socket，在ovs-vswitchd.c中bridge_run每个bridge会监听消息，ofproto_run监听openflow的调用，connmgr_run网络连接管理，ofconn_run管理socket连接。

connmgr_run(p->connmgr, handle_openflow);会设置当有openflow调用的时候，handle_openflow会被调用。

staticvoid
handle_openflow(struct ofconn *ofconn,conststruct ofpbuf *ofp_msg)
OVS_EXCLUDED(ofproto_mutex)
{
enum ofperr error = handle_openflow__(ofconn, ofp_msg);
if (error) {
ofconn_send_error(ofconn, ofp_msg->data, error);
}
COVERAGE_INC(ofproto_recv_openflow);
}

handle_openflow__会做如下的调用：

case OFPTYPE_FLOW_MOD:
return handle_flow_mod(ofconn, oh);

handle_flow_mod首先将openflow协议解析为fm和ofpacts

error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
&ofpacts,
u16_to_ofp(ofproto->max_ports),
ofproto->n_tables);

然后调用static enum ofperr handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm, const struct flow_mod_requester *req)

会调用static enum ofperr ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) OVS_REQUIRES(ofproto_mutex)

staticenum ofperr
ofproto_flow_mod_start(struct ofproto *ofproto,struct ofproto_flow_mod *ofm)
OVS_REQUIRES(ofproto_mutex)
{
switch (ofm->fm.command) {
case OFPFC_ADD:
return add_flow_start(ofproto, ofm);
/* , &be->old_rules.stub[0],
&be->new_rules.stub[0]); */
case OFPFC_MODIFY:
return modify_flows_start_loose(ofproto, ofm);
case OFPFC_MODIFY_STRICT:
return modify_flow_start_strict(ofproto, ofm);
case OFPFC_DELETE:
return delete_flows_start_loose(ofproto, ofm);
case OFPFC_DELETE_STRICT:
return delete_flow_start_strict(ofproto, ofm);
}
return OFPERR_OFPFMFC_BAD_COMMAND;
}

在函数add_flow_start中，首先cls_rule_init(&cr, &fm->match, fm->priority); 将match也即key变成一个cls_rule，cls_rule是一个压缩版本的match，match是一个整个数据结构保存整个package，从L1一直到L4全都有，比较大，如果保存在内存太浪费，cls_rule中有一个minimatch，是用压缩的方式保存match，也即如果match中为0的部分不保存，采取稀疏矩阵的方式。

接下来创建一个新的rule，error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables, rule, new_rule);

最后replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs); 将rule替换现在的rule，有则替换，没有则插入。