Openvswitch原理与代码分析(7): 添加一条流表flow

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：

1. static const struct ovs_cmdl_command all_commands[] = {
2.     { "show", "switch",
3.       1, 1, ofctl_show },
4.     { "monitor", "switch [misslen] [invalid_ttl] [watch:[...]]",
5.       1, 3, ofctl_monitor },
6.     { "snoop", "switch",
7.       1, 1, ofctl_snoop },
8.     { "dump-desc", "switch",
9.       1, 1, ofctl_dump_desc },
10.     { "dump-tables", "switch",
11.       1, 1, ofctl_dump_tables },
12.     { "dump-table-features", "switch",
13.       1, 1, ofctl_dump_table_features },
14.     { "dump-table-desc", "switch",
15.       1, 1, ofctl_dump_table_desc },
16.     { "dump-flows", "switch",
17.       1, 2, ofctl_dump_flows },
18.     { "dump-aggregate", "switch",
19.       1, 2, ofctl_dump_aggregate },
20.     { "queue-stats", "switch [port [queue]]",
21.       1, 3, ofctl_queue_stats },
22.     { "queue-get-config", "switch port",
23.       2, 2, ofctl_queue_get_config },
24.     { "add-flow", "switch flow",
25.       2, 2, ofctl_add_flow },
26.     { "add-flows", "switch file",
27.       2, 2, ofctl_add_flows },
28.     { "mod-flows", "switch flow",
29.       2, 2, ofctl_mod_flows },
30.     { "del-flows", "switch [flow]",
31.       1, 2, ofctl_del_flows },
32.     { "replace-flows", "switch file",
33.       2, 2, ofctl_replace_flows },
34.     { "diff-flows", "source1 source2",
35.       2, 2, ofctl_diff_flows },
36.     { "add-meter", "switch meter",
37.       2, 2, ofctl_add_meter },
38.     { "mod-meter", "switch meter",
39.       2, 2, ofctl_mod_meter },
40.     { "del-meter", "switch meter",
41.       2, 2, ofctl_del_meters },
42.     { "del-meters", "switch",
43.       1, 1, ofctl_del_meters },
44.     { "dump-meter", "switch meter",
45.       2, 2, ofctl_dump_meters },
46.     { "dump-meters", "switch",
47.       1, 1, ofctl_dump_meters },
48.     { "meter-stats", "switch [meter]",
49.       1, 2, ofctl_meter_stats },
50.     { "meter-features", "switch",
51.       1, 1, ofctl_meter_features },
52.     { "packet-out", "switch in_port actions packet...",
53.       4, INT_MAX, ofctl_packet_out },
54.     { "dump-ports", "switch [port]",
55.       1, 2, ofctl_dump_ports },
56.     { "dump-ports-desc", "switch [port]",
57.       1, 2, ofctl_dump_ports_desc },
58.     { "mod-port", "switch iface act",
59.       3, 3, ofctl_mod_port },
60.     { "mod-table", "switch mod",
61.       3, 3, ofctl_mod_table },
62.     { "get-frags", "switch",
63.       1, 1, ofctl_get_frags },
64.     { "set-frags", "switch frag_mode",
65.       2, 2, ofctl_set_frags },
66.     { "probe", "target",
67.       1, 1, ofctl_probe },
68.     { "ping", "target [n]",
69.       1, 2, ofctl_ping },
70.     { "benchmark", "target n count",
71.       3, 3, ofctl_benchmark },
72.  
73.     { "ofp-parse", "file",
74.       1, 1, ofctl_ofp_parse },
75.     { "ofp-parse-pcap", "pcap",
76.       1, INT_MAX, ofctl_ofp_parse_pcap },
77.  
78.     { "add-group", "switch group",
79.       1, 2, ofctl_add_group },
80.     { "add-groups", "switch file",
81.       1, 2, ofctl_add_groups },
82.     { "mod-group", "switch group",
83.       1, 2, ofctl_mod_group },
84.     { "del-groups", "switch [group]",
85.       1, 2, ofctl_del_groups },
86.     { "insert-buckets", "switch [group]",
87.       1, 2, ofctl_insert_bucket },
88.     { "remove-buckets", "switch [group]",
89.       1, 2, ofctl_remove_bucket },
90.     { "dump-groups", "switch [group]",
91.       1, 2, ofctl_dump_group_desc },
92.     { "dump-group-stats", "switch [group]",
93.       1, 2, ofctl_dump_group_stats },
94.     { "dump-group-features", "switch",
95.       1, 1, ofctl_dump_group_features },
96.     { "add-tlv-map", "switch map",
97.       2, 2, ofctl_add_tlv_map },
98.     { "del-tlv-map", "switch [map]",
99.       1, 2, ofctl_del_tlv_map },
100.     { "dump-tlv-map", "switch",
101.       1, 1, ofctl_dump_tlv_map },
102.     { "help", NULL, 0, INT_MAX, ofctl_help },
103.     { "list-commands", NULL, 0, INT_MAX, ofctl_list_commands },
104.  
105.     /* Undocumented commands for testing. */
106.     { "parse-flow", NULL, 1, 1, ofctl_parse_flow },
107.     { "parse-flows", NULL, 1, 1, ofctl_parse_flows },
108.     { "parse-nx-match", NULL, 0, 0, ofctl_parse_nxm },
109.     { "parse-nxm", NULL, 0, 0, ofctl_parse_nxm },
110.     { "parse-oxm", NULL, 1, 1, ofctl_parse_oxm },
111.     { "parse-actions", NULL, 1, 1, ofctl_parse_actions },
112.     { "parse-instructions", NULL, 1, 1, ofctl_parse_instructions },
113.     { "parse-ofp10-match", NULL, 0, 0, ofctl_parse_ofp10_match },
114.     { "parse-ofp11-match", NULL, 0, 0, ofctl_parse_ofp11_match },
115.     { "parse-pcap", NULL, 1, 1, ofctl_parse_pcap },
116.     { "check-vlan", NULL, 2, 2, ofctl_check_vlan },
117.     { "print-error", NULL, 1, 1, ofctl_print_error },
118.     { "encode-error-reply", NULL, 2, 2, ofctl_encode_error_reply },
119.     { "ofp-print", NULL, 1, 2, ofctl_ofp_print },
120.     { "encode-hello", NULL, 1, 1, ofctl_encode_hello },
121.  
122.     { NULL, NULL, 0, 0, NULL },
123. };

 

根据这个数据结构的定义，"add-flow"调用的函数为

1. static void
2. ofctl_add_flow(struct ovs_cmdl_context *ctx)
3. {
4.     ofctl_flow_mod(ctx->argc, ctx->argv, OFPFC_ADD);
5. }

 

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. */

 

1. static void
2. ofctl_flow_mod(int argc, char *argv[], uint16_t command)
3. {
4.     if (argc > 2 && !strcmp(argv[2], "-")) {
5.         ofctl_flow_mod_file(argc, argv, command);
6.     } else {
7.         struct ofputil_flow_mod fm;
8.         char *error;
9.         enum ofputil_protocol usable_protocols;
10.  
11.         error = parse_ofp_flow_mod_str(&fm, argc > 2 ? argv[2] : "", command,
12.                                        &usable_protocols);
13.         if (error) {
14.             ovs_fatal(0, "%s", error);
15.         }
16.         ofctl_flow_mod__(argv[1], &fm, 1, usable_protocols);
17.     }
18. }

 

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

1. static void
2. ofctl_flow_mod__(const char *remote, struct ofputil_flow_mod *fms,
3.                  size_t n_fms, enum ofputil_protocol usable_protocols)
4. {
5.     enum ofputil_protocol protocol;
6.     struct vconn *vconn;
7.     size_t i;
8.  
9.     if (bundle) {
10.         bundle_flow_mod__(remote, fms, n_fms, usable_protocols);
11.         return;
12.     }
13.  
14.     protocol = open_vconn_for_flow_mod(remote, &vconn, usable_protocols);
15.  
16.     for (i = 0; i < n_fms; i++) {
17.         struct ofputil_flow_mod *fm = &fms[i];
18.  
19.         transact_noreply(vconn, ofputil_encode_flow_mod(fm, protocol));
20.         free(CONST_CAST(struct ofpact *, fm->ofpacts));
21.     }
22.     vconn_close(vconn);
23. }

 

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。

1. static void
2. handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
3.     OVS_EXCLUDED(ofproto_mutex)
4. {
5.     enum ofperr error = handle_openflow__(ofconn, ofp_msg);
6.  
7.     if (error) {
8.         ofconn_send_error(ofconn, ofp_msg->data, error);
9.     }
10.     COVERAGE_INC(ofproto_recv_openflow);
11. }

 

handle_openflow__会做如下的调用：

1. case OFPTYPE_FLOW_MOD:
2.     return handle_flow_mod(ofconn, oh);

 

handle_flow_mod首先将openflow协议解析为fm和ofpacts

1. error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
2.                                 &ofpacts,
3.                                 u16_to_ofp(ofproto->max_ports),
4.                                 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)

1. static enum ofperr
2. ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
3.     OVS_REQUIRES(ofproto_mutex)
4. {
5.     switch (ofm->fm.command) {
6.     case OFPFC_ADD:
7.         return add_flow_start(ofproto, ofm);
8.         /* , &be->old_rules.stub[0],
9.            &be->new_rules.stub[0]); */
10.     case OFPFC_MODIFY:
11.         return modify_flows_start_loose(ofproto, ofm);
12.     case OFPFC_MODIFY_STRICT:
13.         return modify_flow_start_strict(ofproto, ofm);
14.     case OFPFC_DELETE:
15.         return delete_flows_start_loose(ofproto, ofm);
16.  
17.     case OFPFC_DELETE_STRICT:
18.         return delete_flow_start_strict(ofproto, ofm);
19.     }
20.  
21.     return OFPERR_OFPFMFC_BAD_COMMAND;
22. }

 

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。

1. static void
2. replace_rule_start(struct ofproto *ofproto, cls_version_t version,
3.                    struct rule *old_rule, struct rule *new_rule,
4.                    struct cls_conjunction *conjs, size_t n_conjs)
5. {
6.     struct oftable *table = &ofproto->tables[new_rule->table_id];
7.  
8.     /* 'old_rule' may be either an evicted rule or replaced rule. */
9.     if (old_rule) {
10.         /* Mark the old rule for removal in the next version. */
11.         cls_rule_make_invisible_in_version(&old_rule->cr, version);
12.     } else {
13.         table->n_flows++;
14.     }
15.     /* Insert flow to the classifier, so that later flow_mods may relate
16.      * to it. This is reversible, in case later errors require this to
17.      * be reverted. */
18.     ofproto_rule_insert__(ofproto, new_rule);
19.     /* Make the new rule visible for classifier lookups only from the next
20.      * version. */
21.     classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
22. }