issue: 4043157 Set initial RTO to 1 second per RFC 6298

Change initial TCP retransmission timeout from 3 seconds to
1 second as recommended by RFC 6298 Section 2, with additional
safeguards for timer granularity and RFC 1122 compliance.

RFC 6298 states: "Until a round-trip time (RTT) measurement
has been made... the sender SHOULD set RTO <- 1 second"

This applies to all new TCP connections until an RTT
measurement is made.

Changes:

1. TCP RTO Calculation (src/core/lwip/tcp.c):
   - Added get_initial_rto() helper function
   - Uses round-up division
   - Prevents division trunc that could cause premature timeouts
   - Modified tcp_pcb_init() and tcp_pcb_recycle() to use helper

2. Timer Resolution Limits (src/core/util/sys_vars.cpp):
   - Added RFC 1122 validation for tcp_timer_resolution_msec
   - Enforces maximum of 500ms per RFC 1122 Section 4.2.3.2
   - Logs warning and clamps value if exceeded
   - Applied to both environment variable and config registry paths

3. Configuration Schema (xlio_config_schema.json):
   - Added "maximum": 500 constraint to timer_msec
   - Updated description to reference RFC 1122 requirement
   - Prevents invalid configurations at schema validation level

4. Documentation (README):
   - Added RFC 1122 reference to timer_msec documentation

Rationale:
Previous implementation used simple division (1000 / slow_tmr_interval)
which could result in:
- 0 ticks when interval > 1000ms (immediate timeout)
- 1 tick when interval = 1000ms (fires on next tick)
- Insufficient granularity with large timer intervals

The new implementation provides defense-in-depth:
- Schema validation prevents misconfiguration
- Runtime validation enforces RFC 1122 limits (delayed ACK ≤ 500ms)
- Round-up division ensures minimum 1 tick without truncation

This fixes incorrect SYN retransmission timing where the
first retry could occur prematurely or immediately, causing
TCP_USER_TIMEOUT tests to fail and connection establishment
issues.

Benefits:
- TCP/IP standards compliance (RFC 6298 and RFC 1122)
- Robust handling of timer granularity edge cases
- Faster connection establishment failure detection
- Prevents problematic timer configurations
- Matches Linux kernel TCP_TIMEOUT_INIT behavior

Signed-off-by: Tomer Cabouly <[email protected]>

Author: tomerdbz

README

@@ -693,6 +693,7 @@ Maps to **XLIO_TCP_TIMER_RESOLUTION_MSEC** environment variable.
 Control internal TCP timer resolution (fast timer) in milliseconds.
 Minimum value is the thread wakeup timer resolution configured in
 performance.threading.internal_handler.timer_msec.
+Maximum is 500ms per RFC 1122 Section 4.2.3.2 (delayed ACK timer must not exceed 500ms).
 Default value is 100
 
 network.protocols.tcp.timestamps

src/core/config/descriptor_providers/xlio_config_schema.json

@@ -418,9 +418,10 @@
                                 "timer_msec": {
                                     "type": "integer",
                                     "minimum": 0,
+                                    "maximum": 500,
                                     "default": 100,
                                     "title": "TCP timer interval (msec)",
-                                    "description": "Maps to XLIO_TCP_TIMER_RESOLUTION_MSEC environment variable.\nControl internal TCP timer resolution (fast timer) in milliseconds.\nMinimum value is the thread wakeup timer resolution configured in\nperformance.threading.internal_handler.timer_msec."
+                                    "description": "Maps to XLIO_TCP_TIMER_RESOLUTION_MSEC environment variable.\nControl internal TCP timer resolution (fast timer) in milliseconds.\nMinimum value is the thread wakeup timer resolution configured in\nperformance.threading.internal_handler.timer_msec.\nMaximum is 500ms per RFC 1122 Section 4.2.3.2 (delayed ACK timer must not exceed 500ms)."
                                 },
                                 "mss": {
                                     "type": "integer",

src/core/lwip/tcp.c

@@ -905,6 +905,11 @@ err_t tcp_recv_null(void *arg, struct tcp_pcb *pcb, struct pbuf *p, err_t err)
     return ERR_OK;
 }
 
+static inline u32_t get_initial_rto(void)
+{
+    return (1000 + slow_tmr_interval - 1) / slow_tmr_interval;
+}
+
 void tcp_pcb_init(struct tcp_pcb *pcb, u8_t prio, void *container)
 {
     u32_t iss;
@@ -927,9 +932,10 @@ void tcp_pcb_init(struct tcp_pcb *pcb, u8_t prio, void *container)
     pcb->mss = pcb->advtsd_mss;
     pcb->user_timeout_ms = 0;
     pcb->ticks_since_data_sent = -1;
-    pcb->rto = 3000 / slow_tmr_interval;
+    // Set initial RTO to 1 second as per RFC 6298
+    pcb->rto = get_initial_rto();
     pcb->sa = 0;
-    pcb->sv = 3000 / slow_tmr_interval;
+    pcb->sv = get_initial_rto();
     pcb->rtime = -1;
 #if TCP_CC_ALGO_MOD
     switch (lwip_cc_algo_module) {
@@ -985,9 +991,10 @@ void tcp_pcb_recycle(struct tcp_pcb *pcb)
     pcb->flags = 0;
     pcb->user_timeout_ms = 0;
     pcb->ticks_since_data_sent = -1;
-    pcb->rto = 3000 / slow_tmr_interval;
+    // Set initial RTO to 1 second as per RFC 6298
+    pcb->rto = get_initial_rto();
     pcb->sa = 0;
-    pcb->sv = 3000 / slow_tmr_interval;
+    pcb->sv = get_initial_rto();
     pcb->nrtx = 0;
     pcb->dupacks = 0;
     pcb->rtime = -1;

src/core/util/sys_vars.cpp

@@ -1621,6 +1621,16 @@ void mce_sys_var::get_env_params()
         tcp_timer_resolution_msec = timer_resolution_msec;
     }
 
+    // RFC 1122 Section 4.2.3.2: Delayed ACK timer must not exceed 500ms
+    // This limits TCP timer resolution to ensure protocol compliance and proper RTO calculations
+    if (tcp_timer_resolution_msec > 500) {
+        vlog_printf(VLOG_WARNING,
+                    "TCP timer resolution [%s=%d] exceeds RFC 1122 maximum of 500ms. "
+                    "Clamping to 500ms to ensure protocol compliance.\n",
+                    SYS_VAR_TCP_TIMER_RESOLUTION_MSEC, tcp_timer_resolution_msec);
+        tcp_timer_resolution_msec = 500;
+    }
+
     if ((env_ptr = getenv(SYS_VAR_INTERNAL_THREAD_CPUSET))) {
         snprintf(internal_thread_cpuset, FILENAME_MAX, "%s", env_ptr);
     }
@@ -2744,6 +2754,16 @@ void mce_sys_var::configure_memory_limits(const config_registry &registry)
         tcp_timer_resolution_msec = timer_resolution_msec;
     }
 
+    // RFC 1122 Section 4.2.3.2: Delayed ACK timer must not exceed 500ms
+    // This limits TCP timer resolution to ensure protocol compliance and proper RTO calculations
+    if (tcp_timer_resolution_msec > 500) {
+        vlog_printf(VLOG_WARNING,
+                    "TCP timer resolution [%s=%d] exceeds RFC 1122 maximum of 500ms. "
+                    "Clamping to 500ms to ensure protocol compliance.\n",
+                    SYS_VAR_TCP_TIMER_RESOLUTION_MSEC, tcp_timer_resolution_msec);
+        tcp_timer_resolution_msec = 500;
+    }
+
     if (registry.value_exists("performance.threading.cpuset")) {
         snprintf(internal_thread_cpuset, FILENAME_MAX, "%s",
                  registry.get_value<std::string>("performance.threading.cpuset").c_str());