Fix PDR Precedence-Based Matching and QFI-Ignore Mode for Uplink Traffic

Summary

This PR fixes critical uplink PDR (Packet Detection Rule) matching issues and adds support for QFI-ignore mode to handle gNB deployments that don't support multiple QFI per TEID.

Problems Fixed

1. Incorrect Precedence-Based PDR Selection

• Issue: UPF returned the first matching PDR instead of selecting the lowest precedence value
• Impact: Traffic with multiple PDRs per TEID+QFI (different SDF filters) was misrouted, leading to incorrect QoS treatment

2. QFI-Ignore Mode for Uplink Classification

• Issue: Some gNB deployments do not support multiple QFI per TEID, requiring UPF to classify uplink traffic using only TEID + SDF filters
• Solution: Added ignore_qfi_for_uplink mode to match PDRs based on TEID only, ignoring QFI values, and then reclassify the packets based on SDF to find matching QoS Parameters

Solutions

Fix 1: Precedence-Based Selection

Implemented proper lowest-precedence tracking in pfcp_session_s_lookup_precedence_over_n3():

• Iterate through all matching PDRs instead of returning first match
• Track PDR with lowest precedence value
• Return best match after evaluating all candidates
• Enhanced debug logging for troubleshooting

Fix 2: SDF Filter Lookup for QFI-Ignore Mode

Implemented correct SDF filter key construction to handle both standard and QFI-ignore modes:

struct session_qfi sdf_key = {
    .seid = seid,
    .qfi  = ignore_qfi_for_uplink ? packet_qfi : pdi.qfi.qfi,
};

Rationale:

• When ignore_qfi_for_uplink=false (standard mode): Use PDR's configured QFI - QFI must match exactly between packet and PDR
• When ignore_qfi_for_uplink=true (TEID-only mode): Use packet's actual QFI for SDF lookup - ensures correct SDF filter selection even when PDR matching ignores QFI, allowing proper default QoS flow handling

Testing

Test Setup

1. Deploy core network using oai-cn5g-fed/docker-compose/docker-compose-basic-nrf-qos.yaml
2. Apply patch to create QoS rules for UE with SUPI 208950000000035 multiqos.patch:
   • gbr-rule-5qi-1 = 3 Mbps (default QoS flow)
   • gbr-rule-iperf3-8080 = 20 Mbps (port 8080)
   • gbr-rule-iperf3-8081 = 10 Mbps (port 8081)

Verification Commands

# Start IPerf3 server on UE
docker exec -it ueransim-ue-5qi-1 iperf3 -s -B 12.1.1.10 -p 8080
# (also individually start it for other ports below)

# Test different ports from DN
docker exec -it oai-ext-dn iperf3 -c 12.1.1.10 -p 8080  # Should get ~20 Mbps
docker exec -it oai-ext-dn iperf3 -c 12.1.1.10 -p 8081  # Should get ~10 Mbps
docker exec -it oai-ext-dn iperf3 -c 12.1.1.10          # Should get ~3 Mbps (default)

Results

With Fix (Correct Precedence-Based QoS):

• Port 8080: 21.1 Mbps (matches 20 Mbps rule) ✅
• Port 8081: 10.6 Mbps (matches 10 Mbps rule) ✅
• Port 5201: 3.84 Mbps (matches 3 Mbps default rule) ✅

Without Fix (All Traffic Uses Default QoS):

• Port 8080: 3.84 Mbps (incorrectly uses default rule) ❌
• Port 8081: 3.48 Mbps (incorrectly uses default rule) ❌
• Port 5201: 3.66 Mbps (correctly uses default rule) ✅

Conclusion

The fix ensures that traffic is correctly classified to the appropriate QoS flow based on SDF filters and precedence values, enabling differentiated QoS treatment per application.

Files Changed

• src/upf_app/bpf/rules/pdr_and_far/pfcp_session_lookup_xdp_kernel.c

  • Precedence-based selection logic
  • SDF filter key fix for QFI-ignore mode
  • Enhanced debug logging and 3GPP compliance comments
  • eBPF rodata configuration structure

• src/upf_app/programs/pfcp_session_lookup_xdp_user.cpp - Userspace rodata configuration

• src/upf_app/include/types.h - Shared configuration struct definition

• src/upf_app/upf_config.hpp - Configuration field addition

• src/upf_app/upf_config_yaml.hpp - YAML configuration interface

• src/upf_app/upf_config_yaml.cpp - YAML parsing and mapping

• src/upf_app/SessionManager.cpp - Enhanced logging

• src/upf_app/SessionProgramManager.cpp - Fix and ensures PDRs are processed in precedence order

Configuration

QFI-ignore mode can be configured via YAML configuration file:

support_features:
  ignore_qfi_for_uplink: yes  # For gNB without multiple QFI per TEID support

Default Value: true (enabled by default)

The configuration is implemented using eBPF rodata, allowing runtime configuration without recompilation:

• Defined in struct pdr_lookup_config (shared between BPF and userspace)
• Configured from userspace before BPF program load

Backwards Compatibility

Fully backwards compatible - QFI-ignore mode is enabled by default (ignore_qfi_for_uplink: true). Operators can disable it by setting ignore_qfi_for_uplink: no in the YAML configuration if their gNB supports multiple QFI per TEID.