Add server_id_fallback mechanism to prevent false mismatch reports during server ID transitions

Summary:
During server ID transitions, existing connections continue using old server IDs in TCP header options while new connections use new server IDs. This causes BPF comparison logic to report false mismatches for correctly routed existing connections.

Confirmed the issue within a shiv test backend.
          <idle>-0       [008] ..s.1 563303.012647: bpf_trace_printk: [op=PASSIVE_ESTABLISHED_CB rport=34390 remote_ip4=100663423] passive estab received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.012755: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.013949: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.014361: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.014435: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.015225: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.015266: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451
          <idle>-0       [008] ..s.1 563303.015268: bpf_trace_printk: [op=PARSE_HDR_OPT_CB rport=34390 remote_ip4=100663423] passive received wrong server id: option=7274496, server=46451

This diff introduces a server_id_fallback mechanism to handle server ID transitions gracefully:

**Core Implementation:**
- Add `server_id_fallback` field to `server_info` struct in BPF and C++
- Add `fallbackV6Id_` field to `TcpPktRouter` class for server ID v2 format
- Route server IDs based on format: IDs ≥131072 (2<<16) → fallbackV6Id_, IDs <131072 → v6Id_
- Update BPF packet validation to accept packets matching either primary OR fallback server ID
- Use fallback server ID in TCP headers when primary server ID is 0

**API Enhancements:**
- Add `getFallbackServerIdV6()` method to access fallback server ID
- Update `getServerIdV6()` to return primary ID when available, fallback otherwise
- Enhanced `updateServerInfo()` to populate both fields in BPF map

**BPF Logic Updates:**
- `handle_passive_write_hdr_opt()`: Use server_id_fallback when server_id is 0
- `handle_passive_parse_hdr()` and `handle_passive_estab()`: Accept packets with either server ID
- Enhanced logging shows both primary and fallback server ID values

**Comprehensive Testing:**
- 3 unit tests validating server ID routing, BPF map population, and field preservation
- 3 integration tests covering server ID v2 functionality, seamless transitions, and format validation
- Extended test infrastructure with flexible server ID management

This eliminates false mismatch reports during server ID transitions while maintaining full backward compatibility.
 ---
> Generated by [Confucius Code Assist (CCA)](https://www.internalfb.com/wiki/Confucius/Analect/Shared_Analects/Confucius_Code_Assist_(CCA)/)
[Session](https://www.internalfb.com/confucius?session_id=14e196ac-8842-11f0-a62e-b9c865999367&tab=Chat), [Trace](https://www.internalfb.com/confucius?session_id=14e196ac-8842-11f0-a62e-b9c865999367&tab=Trace)

Reviewed By: avasylev

Differential Revision: D81549160

fbshipit-source-id: 7e69d3e3b0c1f834c62a25c5ed2834af21a3091c

Author: Fei Chen

katran/tpr/TPRTypes.h

@@ -33,6 +33,9 @@ struct server_info {
   uint8_t kde_zones;
   // only ipv6 is supported
   uint32_t server_id;
+  // when default server id is not available, backend uses server id v2 as
+  // fallback
+  uint32_t server_id_fallback;
 };
 
 // Tentative stats for different transport events

katran/tpr/TcpPktRouter.cpp

@@ -12,7 +12,6 @@ namespace katran_tpr {
 // max 24-bit value
 constexpr uint32_t kMaxServerId = (1 << 24) - 1;
 constexpr uint32_t kServerInfoIndex = 0;
-const std::string kServerInfoMap = "server_infos";
 
 TcpPktRouter::TcpPktRouter(
     RunningMode mode,
@@ -112,11 +111,23 @@ bool TcpPktRouter::setServerIdV6(uint32_t id) {
   if (id > kMaxServerId) {
     return false;
   }
-  if (v6Id_ == id) {
-    LOG(WARNING) << "Server id is already set to " << id;
-    return true;
+
+  // Route ID to appropriate field based on server ID format
+  if (id >= (2 << 16)) {
+    // Server ID v2 format - set to fallback field
+    if (fallbackV6Id_ == id) {
+      LOG(WARNING) << "Server fallback id is already set to " << id;
+      return true;
+    }
+    fallbackV6Id_ = id;
+  } else {
+    // Regular server ID format - set to primary field
+    if (v6Id_ == id) {
+      LOG(WARNING) << "Server id is already set to " << id;
+      return true;
+    }
+    v6Id_ = id;
   }
-  v6Id_ = id;
   if (isInitialized_) {
     auto updateRes = updateServerInfo();
     if (updateRes.hasError()) {
@@ -151,21 +162,28 @@ TcpPktRouter::updateServerInfo() noexcept {
     info.running_mode = RunningMode::SERVER;
     info.kde_enabled = kdeEnabled_;
     info.kde_zones = kdeZones_;
+
+    // Use the appropriately routed server IDs
     info.server_id = v6Id_;
-    if (info.server_id == 0) {
-      LOG(WARNING) << "TCP Pkt router is set but server_id is 0. Please check "
-                      "if the id has been set properly.";
+    info.server_id_fallback = fallbackV6Id_;
+
+    if (info.server_id == 0 && info.server_id_fallback == 0) {
+      LOG(WARNING)
+          << "TCP Pkt router is set but both server_id and fallback are 0. Please check "
+             "if the id has been set properly.";
     }
   } else {
     info.running_mode = RunningMode::CLIENT;
     info.server_id = 0;
+    info.server_id_fallback = 0;
   }
   int mapFd = bpf_map__fd(skel_->maps.server_infos);
   auto updateRes = BpfUtil::updateMapElement(mapFd, kServerInfoIndex, info);
   if (updateRes.hasError()) {
     return makeError(updateRes.error(), __func__);
   }
-  LOG(INFO) << "set server_id=" << info.server_id;
+  LOG(INFO) << "set server_id=" << info.server_id
+            << " server_id_fallback=" << info.server_id_fallback;
   return folly::Unit();
 }
 

katran/tpr/TcpPktRouter.h

@@ -68,7 +68,12 @@ class TcpPktRouter {
    * Note: only IPv6 is supported.
    */
   uint32_t getServerIdV6() const noexcept {
-    return v6Id_;
+    // in case v6Id is not available, use fallbackV6Id_ instead.
+    return v6Id_ ? v6Id_ : fallbackV6Id_;
+  }
+
+  uint32_t getFallbackServerIdV6() const noexcept {
+    return fallbackV6Id_;
   }
 
   folly::Expected<folly::Unit, std::system_error> setServerKDEZone(
@@ -103,7 +108,8 @@ class TcpPktRouter {
   folly::Expected<folly::Unit, std::system_error> updateServerInfo() noexcept;
 
   bool isInitialized_{false};
-  uint32_t v6Id_;
+  uint32_t v6Id_{0};
+  uint32_t fallbackV6Id_{0}; // Server ID v2 fallback when >= 2<<16
   bool kdeEnabled_;
   uint8_t kdeZones_{0};
   std::optional<uint32_t> serverPort_;

katran/tpr/bpf/tcp_pkt_router_passive_hdlr.h

@@ -60,14 +60,17 @@ static inline int handle_passive_parse_hdr(
     TPR_PRINT(skops, "passive received 0 server id");
     return PASS;
   }
-  if (s_info->server_id != hdr_opt.server_id) {
+  // Check if incoming server_id matches either primary or fallback server_id
+  if (s_info->server_id != hdr_opt.server_id &&
+      s_info->server_id_fallback != hdr_opt.server_id) {
     // read the server_id. But not itself. Packet is misrouted.
     stat->error_bad_id++;
     TPR_PRINT(
         skops,
-        "passive received wrong server id: option=%d, server=%d",
+        "passive received wrong server id: option=%d, server=%d, fallback=%d",
         hdr_opt.server_id,
-        s_info->server_id);
+        s_info->server_id,
+        s_info->server_id_fallback);
     return PASS;
   } else {
     stat->server_id_read++;
@@ -94,7 +97,12 @@ static inline int handle_passive_write_hdr_opt(
 
   hdr_opt.kind = TCP_SRV_HDR_OPT_KIND;
   hdr_opt.len = TCP_HDR_OPT_LEN;
-  hdr_opt.server_id = s_info->server_id;
+  // If server_id is non-zero, use it; otherwise use server_id_fallback
+  if (s_info->server_id != 0) {
+    hdr_opt.server_id = s_info->server_id;
+  } else {
+    hdr_opt.server_id = s_info->server_id_fallback;
+  }
   err = bpf_store_hdr_opt(skops, &hdr_opt, sizeof(hdr_opt), NO_FLAGS);
   if (err) {
     stat->error_write_opt++;
@@ -123,15 +131,18 @@ static inline int handle_passive_estab(
     unset_parse_hdr_cb_flags(skops, stat);
     return err;
   }
-  if (s_info->server_id != hdr_opt.server_id) {
+  // Check if incoming server_id matches either primary or fallback server_id
+  if (s_info->server_id != hdr_opt.server_id &&
+      s_info->server_id_fallback != hdr_opt.server_id) {
     stat->error_bad_id++;
     // the peer sent the server_id but it is wrong.
     // keep on sending the server_id and reading peer's tcp-hdr
     TPR_PRINT(
         skops,
-        "passive estab received wrong server id: option=%d, server=%d",
+        "passive estab received wrong server id: option=%d, server=%d, fallback=%d",
         hdr_opt.server_id,
-        s_info->server_id);
+        s_info->server_id,
+        s_info->server_id_fallback);
     return set_parse_hdr_cb_flags(skops, stat);
   } else {
     stat->server_id_read++;

katran/tpr/bpf/tcp_pkt_router_structs.h

@@ -10,6 +10,9 @@ struct server_info {
   // zones supported by kde
   __u8 kde_zones;
   __u32 server_id;
+  // when default server id is not available, backend users server id v2 as
+  // fallback
+  __u32 server_id_fallback;
 };
 
 // struct that represents a header options in TCP packet