docs(N34)+test: E13 僵尸会话根因坐实;Phase 11 判据修正+一次性报告

E13(实验方案.md 新增 + E8/E11/E12 口径修正):
- 风暴态根因=僵尸会话,非芯片残渣:PC 网卡禁用/链路断开时
  Windows 直接丢弃 socket 不发 FIN/RST → 芯片侧 ESTABLISHED 僵尸
  → 单连接监听被占(:21 拒连)→ 下一 PASV 重监听与僵尸竞态
  (accept 为垂死僵尸开枪 20ms 后 DISCONNECT)→ lftpd 5s 窗空转
  → 425;每失败轮再生僵尸 → 自续数分钟
- "残渣跨烧录存活/需断电冷启动"口径作废:板上 CH395F RSTI 接
  MCU NRST(V0.0 网表 :174),每次烧录即芯片硬件复位;"存活"是
  锤击再生错觉。正确控制=下一臂前静置 ≥90s 清僵尸窗
- 自愈链端到端验证:TIMEOUT(在途 TX ≈30s)/keepalive(空闲 60/5/3)
  → pump → relisten → 端口恢复;E11 产品路径验证 17s 判死 PASS
- 4 元组复用假设被否(pasv_probe.py:PASV 端口递增、PC 源端口随机)
- e02_probe listener=FAIL 重解读:探针挂起 socket 未关(PC 内核
  代 ACK keepalive)→ 拒连是单连接正确行为;真缺口=应用无法主动
  拆会话 → 阶段 2 sock_reset 职责

Phase 11 测试修复(net_test_task.c,实跑暴露):
- 判据 >=60000ms 是按手册公式写的,实测判死 18~31s 会在正确固件
  上假 FAIL → 改有界窗 10s~120s(下限防误杀存活空闲会话,上限
  防判死缺位)
- 报告后不释放 phase_active → 对已死连接 10ms 节奏刷 594 条假
  FAIL → phase11_run 返回完成态,主循环释放等新连接再测
- 复跑两轮验证:30137ms 单次 PASS → 自动等新连接 → 二轮 accept 正常
- Test_Guide TC-NET-1101 判据/耗时同步;ch395f_test.h Phase 14
  共存安全性注释(有整体超时不永久挡 accept)

工具:test/host/pasv_probe.py(每轮打印 PASV 端口+PC 源端口+结果)
This commit is contained in:
2026-10-13 11:23:40 +08:00
parent d346e4ee84
commit 33405a57b5
5 changed files with 228 additions and 19 deletions
+3 -1
View File
@@ -69,7 +69,9 @@ extern "C" {
//#define ENABLE_PHASE12_TESTS /* [硬] SPI 字节命令 / DMA 时延基线(L1/L2,无需 PC)—— M1 已完成,基线见说明文档 §6.1 */
/* ---- 链路时延(M2/M3;分段/阈值见 docs/测试说明/链路延迟测试说明.md)---- */
//#define ENABLE_PHASE13_TESTS /* [网] 芯片 + 队列时延(L3~L7/L10,需 PC peer)—— M2 已完成,基线见说明文档 §6.2 */
#define ENABLE_PHASE14_TESTS /* [网] 生产路径时延(L8~L11 + 模拟器 Δ)—— M3,需 peer --mode remo(:7003) */
#define ENABLE_PHASE14_TESTS /* [网] 生产路径时延(L8~L11 + 模拟器 Δ)—— M3,需 peer --mode remo(:7003);
* 与 Phase 11 共存安全:peer 缺席时有 LAT_P14_TIMEOUT_MS 整体超时
* 自动 done,不永久挡 accept */
#if defined(ENABLE_PHASE1_TESTS) || defined(ENABLE_PHASE2_TESTS) || \
defined(ENABLE_PHASE3_TESTS) || defined(ENABLE_PHASE4_TESTS) || \
+24 -9
View File
@@ -801,9 +801,9 @@ static void phase10_run(int32_t sockfd) {
* 入口条件:Socket 处于 ESTABLISHED 且对端为**死对端**——PC 连 8080 后
* 禁用网卡(存活对端会 ACK 探测、TIMEOUT 永不触发,本测试将
* 一直等待;防火墙规则对已建立连接无效,E11 实证)。
* 通过:从连接建立到 recv 返回 0 的间隔 ≥ 60s(IDLE 阈值)。
* 注:芯片定时语义与手册公式(75s)不完全一致,实测判死窗口见
* 实验方案 E11(对端失联后 ≈30s)。
* 通过:从连接建立到 recv 返回 0 的间隔落在 10s~120s 有界窗
* (芯片定时语义与手册公式 75s 不符,实测判死 18~31s——
* 见实验方案 E11 与 phase11_run 内判据注释)。
* =================================================================== */
#ifdef ENABLE_PHASE11_TESTS
@@ -811,27 +811,40 @@ static int32_t s_phase11_armed = 0;
static uint32_t s_phase11_start = 0;
static uint8_t s_phase11_buf[64]; /* 专用小缓冲:不占 32KB s_rx_buf(RAM 余量) */
static void phase11_run(int32_t sockfd) {
/* 返回值:0 = 测试进行中;1 = 已报告完成(主循环据此释放 phase_active,
* 等待**新连接**再测。勿删本机制——报告后连接已死,recv 恒返 0,
* 继续 arm 会以 10ms 节奏刷出数百条假 FAIL(2026-10-13 实跑实证) */
static int phase11_run(int32_t sockfd) {
/*
* 第一次调用:记录开始时间并 arm。
* 后续调用:非阻塞 recv,若返回 0 表示连接已断开(KeepAlive 触发)。
* 计算耗时与 60s idle 阈值比较。
* 计算耗时与有界窗判据比较。
*/
if (!s_phase11_armed) {
s_phase11_armed = 1;
s_phase11_start = HAL_GetTick();
DBG_INFO("Phase11: idle wait for KeepAlive timeout (60s)");
return;
return 0;
}
int32_t n = net_recv(sockfd, s_phase11_buf, 64, NET_MSG_DONTWAIT);
if (n == 0) {
uint32_t elapsed = HAL_GetTick() - s_phase11_start;
TEST_CHECK(elapsed >= 60000,
"Phase11: KeepAlive triggered after %lums (expect >=60000ms)", elapsed);
/* 判据(E11 实测修正版):芯片定时语义与手册公式(75s)不符——
* 生产参数实测判死窗口 18~31s(产品验证 连接+30.5s、E11 T5
* 对端失联+30s、T4 短参数 连接+~10s)。本 socket 无在途 TX、
* 对端静默且已失联 → 唯一开枪源是 keepalive,故判据为
* **有界窗**:>=10s(存活空闲会话 T2 实测 90s 不被杀,10s 下限
* 区分"误杀"与"keepalive 判死")且 <=120s(判死必须发生)。
* 勿改回 >=60000——那是对着手册公式写的,会在正确固件上假 FAIL */
TEST_CHECK(elapsed >= 10000U && elapsed <= 120000U,
"Phase11: KeepAlive triggered after %lums (expect 10000..120000ms)",
elapsed);
TEST_REPORT("Phase 11");
s_phase11_armed = 0;
return 1;
}
return 0;
}
#endif
@@ -920,7 +933,9 @@ void net_layer_test_main(void)
phase10_run(sockfd);
#endif
#ifdef ENABLE_PHASE11_TESTS
phase11_run(sockfd);
if (phase11_run(sockfd)) {
phase_active = 0; /* 已报告:释放主循环去 accept 新连接(net 层已 auto_relisten) */
}
#endif
#ifdef ENABLE_PHASE13_TESTS
/* Phase 13(链路时延 L3~L7/L10,M2):独立运行,勿与其他 Phase 混选;
+123
View File
@@ -0,0 +1,123 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
pasv_probe.py — decisive probe for the "storm" signature (N34).
Per FTP round prints: device PASV port, PC data-socket SOURCE port, round
result, and timing. Hypothesis under test: the CH395F reuses the same
ephemeral PASV port for the next data listener while the previous data
connection is still in chip-side TIME_WAIT with the same 4-tuple
(PC source port reused too) -> chip reports CONNECT then RSTs ~10ms later
(serial: "sock4 standalone accept" + "DISCONNECTED chip st=0x00/0x00").
Usage:
python pasv_probe.py --rounds 5 --delay 2 # reproduce (fast rounds)
python pasv_probe.py --rounds 3 --delay 35 # TIME_WAIT window test
"""
import argparse
import re
import socket
import sys
import time
IP_DEFAULT = "192.168.1.100"
def ftp_stor_round(ip, user, pwd, fname, payload):
"""One login -> PASV -> STOR -> verify transfer. Returns (ok, pasv_port,
pc_src_port, detail)."""
c = socket.create_connection((ip, 21), timeout=10)
c.settimeout(10)
pc_src = None
pasv_port = None
try:
def readline():
buf = b""
while not buf.endswith(b"\n"):
ch = c.recv(1)
if not ch:
break
buf += ch
return buf.decode(errors="replace").strip()
def cmd(line):
c.sendall((line + "\r\n").encode())
return readline()
greet = readline()
if not greet.startswith("220"):
return False, None, None, "no greeting: " + greet
r = cmd("USER " + user)
if not r.startswith("331"):
return False, None, None, "USER: " + r
r = cmd("PASS " + pwd)
if not r.startswith("230"):
return False, None, None, "PASS: " + r
r = cmd("PASV")
m = re.search(r"\((\d+),(\d+),(\d+),(\d+),(\d+),(\d+)\)", r)
if not r.startswith("227") or not m:
return False, None, None, "PASV: " + r
g = [int(x) for x in m.groups()]
pasv_port = g[4] * 256 + g[5]
d = socket.create_connection((ip, pasv_port), timeout=8)
pc_src = d.getsockname()[1]
r = cmd("STOR " + fname)
if not r.startswith(("125", "150")):
d.close()
return False, pasv_port, pc_src, "STOR: " + r
d.sendall(payload)
d.close()
# drain until final 2xx (multi-line or single)
final = ""
t0 = time.time()
while time.time() - t0 < 10:
r = readline()
if not r:
break
final = r
if r[:1] in ("2", "4", "5"):
break
ok = final.startswith("2")
cmd("QUIT")
return ok, pasv_port, pc_src, ("OK" if ok else "final: " + final)
finally:
c.close()
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--ip", default=IP_DEFAULT)
ap.add_argument("--rounds", type=int, default=5)
ap.add_argument("--delay", type=float, default=2.0,
help="seconds between rounds")
ap.add_argument("--user", default="admin")
ap.add_argument("--pwd", default="admin")
ap.add_argument("--size", type=int, default=1024)
args = ap.parse_args()
payload = bytes((i * 7 + 3) & 0xFF for i in range(args.size))
ports, srcs, oks = [], [], []
for i in range(args.rounds):
t = time.strftime("%H:%M:%S")
ok, port, src, detail = ftp_stor_round(args.ip, args.user, args.pwd,
"PP%d.BIN" % i, payload)
ports.append(port)
srcs.append(src)
oks.append(ok)
print("[%s] round %d: ok=%-5s pasv_port=%-6s pc_src=%-6s %s"
% (t, i + 1, ok, port, src, detail))
if i + 1 < args.rounds:
time.sleep(args.delay)
print("\nports :", ports)
print("pc_src:", srcs)
print("ok :", oks)
same_port = len(set(p for p in ports if p)) == 1
same_src = len(set(s for s in srcs if s)) == 1
print("PASV port constant: %s | PC source port constant: %s"
% (same_port, same_src))
return 0 if all(oks) else 1
if __name__ == "__main__":
sys.exit(main())