Tags

v0.7.7-beta7

mlvpn-rs 0.7.7-beta7

v0.7.7-beta6

v0.7.7-beta6: rebuilt automatic report trigger

v0.7.7-beta5

v0.7.7-beta4

0.7.7-beta4: event-based automatic loss reports (raw probe loss over a 60s window, one ticket per event, flap trigger, persisted event state, dump-file cap) on top of beta3's tx_send_mode

v0.7.7-beta3

0.7.7-beta3: tx_send_mode per-link option (default sendto; sendmmsg batching, experimental sendmmsg_gso default-off) plus doc fixes

v0.7.7-beta2

v0.7.7-beta2: delivery feedback judged from byte-counter deltas (fixes false link cuts at TCP flow start; bonded -P4 download 489/641/619 -> 642/641/656 Mbps)

v0.7.7-beta1

v0.7.7-beta1

v0.7.6-beta1

v0.7.6-beta1: benchmark build -- per-link TX workers, P+Q FEC, health telemetry, mlvpn-tui redesign, mlvpn-web, idle-backup scoring fix (b672428)

v0.7.6

v0.7.5

v0.7.5: fix automatic diagnostic reports silently 422ing against software-support (message field type mismatch)

v0.7.4

v0.7.4: full dependency update pass, including the RUSTSEC-2026-0285 rustls fix

v0.7.3-beta7

v0.7.3-beta7

Fixes a false MTU "MISMATCH" in mlvpn-tui's Session panel for
healthy, fully-negotiated but asymmetric-capability peer pairs
(found live on home-bgp/cloud-bgp after the beta6 rollout). Also
makes mlvpn-tui/mlvpnd self-test show the full Beta version string
(e.g. "0.7.3-beta7") instead of just the bare base version.

v0.7.3-beta6

v0.7.3-beta6

Adds primary-aware peer MTU negotiation and mlvpnd self-test --tunnel
--protocol tcp mode on top of v0.7.3-beta5's single-erasure FEC. First
release built with the new Stable/Beta CI naming (~betaN, not ~testingN).

v0.7.3-beta5

beta5: real TCP mode for mlvpnd self-test --tunnel

--protocol tcp|udp (tcp default), giving throughput/retransmit numbers
directly comparable to iperf3 instead of the old raw-UDP mode's
loss-blind numbers. Upload-leg retransmits are real, sender-measured;
download-leg retransmits are honestly reported as unavailable rather
than a wrong number.

docs/superpowers/specs/2026-09-16-tcp-tunnel-self-test-design.md
docs/superpowers/plans/2026-09-16-tcp-tunnel-self-test.md

v0.7.3-beta4

beta4: primary-aware, peer-negotiated tunnel MTU

Includes the full primary-aware MTU feature (effective_tunnel_mtu's
primary_idx parameter, negotiate_tunnel_mtu peer negotiation, faster
3s mtu_monitor_loop poll) plus the final-review fix for the monotonic
MTU ratchet bug (local_capability_mtu advertised to the peer instead
of the already-negotiated value) and the floor-clamping fix.

docs/superpowers/specs/2026-09-16-primary-aware-peer-mtu-design.md
docs/superpowers/plans/2026-09-16-primary-aware-peer-mtu.md

v0.7.3-beta3

v0.7.3-beta3: fixes a real production crash found during a fec_group_size sweep

fec_group_size=16 crash-looped mlvpnd on both home-bgp and cloud-bgp
(TUN device creation EINVAL) because the dual-stack (IPv6) MTU floor
was wrong (576 instead of the real, mandatory 1280 per RFC 8200).
effective_tunnel_mtu now floors correctly for any tunnel with
address6 configured, and warns when FEC's own reservation can't be
fully honored at that floor instead of silently under-reserving.

v0.7.3-beta2

v0.7.3-beta2: single-erasure FEC, fully fixed and verified end-to-end

Includes the complete single-erasure FEC feature built and hardened
this session:
- Opt-in scheduler.fec_enabled/fec_group_size XOR parity for real Data
  traffic (wire format, sender/receiver integration, diagnostics).
- MTU accounting fix so parity frames can never exceed the tunnel MTU.
- fec_enabled=false confirmed zero-cost on the receive hot path.
- Duplicate-delivery bug closed with a seen-seq floor watermark
  (formally proven + empirically verified via a reverted probe test).
- Phantom-group-member bug closed: FEC accumulation is now gated on
  confirmed real transmission, propagated from WgTunnel::encapsulate
  through wg_send through send_scheduled -- previously a silently
  dropped or WireGuard-handshake-substituted frame could poison a
  group's recovery budget or cause FEC to reconstruct and deliver a
  packet that was never actually sent by the peer.
- Real tests/veth_fec.rs integration tests, executed for real (not just
  unit-tested) on a live TUN-capable host: 3/3 passing.

Kept on the 0.7.3 version line per explicit direction -- this is a beta
iteration, not a new Cargo.toml version bump.

v0.7.3-beta1

v0.7.3-beta1: wg_substituted_total diagnostic counter

Instruments boringtun's encapsulate-substitution path (per-link and
daemon-wide lifetime counters): when a link has no current valid
WireGuard session, boringtun silently substitutes a handshake datagram
for real data with no visible error. Built after wg_rejected_total
(0.7.2) ruled out receive-side rejection as the cause of a real ~15%
TCP retransmission finding under load.

v0.7.2-beta2

v0.7.2-beta2: fixes v0.7.2-beta1's Cargo.lock/Cargo.toml mismatch

beta1 (fb59432) failed CI immediately: Cargo.toml was bumped to 0.7.2
but Cargo.lock was never committed alongside it, and CI's
--locked build correctly refused to silently regenerate the lock
file. No functional changes since beta1 -- just the missing lock file.

v0.7.2-beta1

v0.7.2-beta1: wg_rejected_total diagnostic counter

Instruments boringtun's decapsulate-rejection path (per-link and
daemon-wide lifetime counters), surfaced in mlvpn-tui and diag-dump.
Built to help root-cause a live ~15% TCP retransmission finding under
load that every pre-existing counter failed to explain.