Topology rework. The mesh's job is width; depth now comes from knowledge.
Slots:
- Local/Wide collapse into ONE mesh pool: 20 desktop / 15 mobile
- Temp referral slots +4..+10 ABOVE the cap (10 desktop / 4 mobile), carry no
knowledge exchange, graduate into a freed mesh slot or expire (5 min TTL),
never evict an established mesh peer
- Fixes the Wide-never-fills bug: outbound paths hardcoded Local while the
growth gate counted Local only, so growth stopped at 71 and 20 slots were
inbound-only. Also closes register_connection paths that enforced no cap.
Knowledge — two-pool uniques announce (replaces N1/N2/N3 share lists):
- Pool 1 (forwardable): our N0-N2 uniques, deduped. Receiver stores shifted one
bounce deeper as N1-N3, reporter-tagged.
- Pool 2 (terminal): our N3, deduped against pool 1. Receiver stores as N4,
USED for search/resolution but NEVER re-announced. No N5 anywhere.
- Uniques merge mesh peers + social directs + CDN file authors + holder peers,
so receivers cannot tell how we know an ID
- Anchor entries carry addresses; every other entry is a bare ID. The pools
double as the anchor directory.
- Dedup at both store and announce; per-bounce caps; 2 MB payload ceiling
- Device-tiered depth: mobile holds 3 bounces and drops the terminal pool
Anchor convection (register loop retired — 0xC0 AnchorRegister deleted):
- Rolling in-memory caller window (no registration DB): a caller gets the 2
most recent viable prior callers, its address goes to the next 2, then
rotates out; still-connected callers preferred; RelayIntroduce coordinated
when both ends are live
- Request classes: entry (<2 connections) always served; top-up refused
cheaply under load, and the refusal feeds the adaptive weights
- Stochastic per-disconnect action {nothing | anchor intro | mesh intro} with
weights from local anchor density + refusal feedback. No thresholds, no
jitter timer. Recovery (mesh < 2) stays immediate and non-stochastic.
- Anchors mined from the uniques pools incl. retained pools of disconnected
peers; known_anchors demoted to a bootstrap cache
Security fixes from review: remote address poisoning (hearsay anchors could be
dialled — known_anchors/is_anchor now written only from completed handshakes),
N4 leaking back into announcements via the anchor mirror (infinite propagation),
missing payload size caps, unbounded candidate scoring, and a bulk SQLite write
held under the ConnectionManager mutex.
deploy.sh: version regex now captures pre-release suffixes (0.8.0-alpha would
have truncated to 0.8.0 and broken every artifact path), announce channel
derives from the suffix, and the anchor runs --publish-registry at genesis.
228 core tests; a3 integration 9/9; new c_topology_test.sh 33/33 (both
independently re-run). EDM corpse, session-relay gating, Iteration A/B intact.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LGiPD2cF75mnvneSCjdDC5
A1 — ID-class bug family (v0.6.1 persona split left network-NodeId comparisons
where posting IDs live). ~14 sites fixed with multi-persona semantics ("author is
one of MY posting identities", not == default):
- revoke_post_access / revoke_circle_access (revocation was broken for every post)
- group key create/add-member/rotate (add-member distribution now works)
- circle profile set/delete/get, encrypted-attachment CEK unwrap
- receipt/comment slot authorship, replication cycle, blob eviction own-tier
- orphaned GroupKeyRequest/Response (0xA1/0xA2) deleted (no send side existed)
A2 — Manifest privacy: AuthorManifest.author_addresses removed from struct AND
signature digest; CdnManifest tree-era fields dropped. Startup migrations re-sign
own manifests, purge unverifiable foreign copies, strip persona fields from legacy
network-id profile rows. Posting identities are now location-anonymous.
A3 — Discovery & first contact:
- Open-slot comments: derivable slot V_x (blake3 from author + slot_binder_nonce)
so strangers pass the CDN comment-verification gate; OpenSlotDecl in FoF gating
- Greetings: HPKE-style sealed GreetingBody (return_path + fresh reply_pubkey),
throwaway outer ID, size-bucketed; messaging-first (Reply/Dismiss, no vouch)
- Registry: frozen canonical registry post + REGISTRY_POST_ID, signed registration
comments (self-certifying deletes), newest-wins per persona, --publish-registry
- Comment TTL: expires_at_ms inside signed digest v2 (context bump), rand 30-365d
ordinary / fixed 30d registrations, expiry sweep on the eviction cycle
- Single accept_incoming_comment() gate wired into all three ingest sites, closing
a pre-existing hole where both pull paths stored comments with no verification
- UI: consent panel (visible/listed/greetings, active choice at first publish),
registry search in Discover, sealed "Say hi" compose, greetings inbox
Security fixes from review: forged-tombstone injection, delete-by-sender-trust,
remote SetPolicy poisoning, stored XSS (escapeHtml quotes, 7 pre-existing sites),
greeting off-switch now revokes prior bios' slots.
No PoW (rejected: inverted cost). Session-relay gating untouched.
201 core tests pass; CLI + desktop build; 3-node integration green.
DEPLOY GATE: wire-incompatible with v0.7.3 — ALPN bump to itsgoin/4 lands in
Iteration B. Do not build/deploy/anchor-swap before that.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LGiPD2cF75mnvneSCjdDC5
Adds an optional supersedes_post_id pointer to Post for the
"re-issue with narrower access" path. Author publishes a new post
that references an earlier one; receivers can render "this is a
re-issued version of an earlier post" + offer to view the original.
Covered by PostId = BLAKE3(Post) — receivers can verify it wasn't
forged by anyone but the author.
#[serde(default)] for back-compat; existing posts deserialize with
None. All existing Post construction sites bulk-updated to set the
field to None.
Storage::get_post returns None for the field today; a follow-up can
add a dedicated column if/when receivers need to render the
pointer. The author-side re-issue helper that creates posts with
this field set comes in the Tauri/UI slice — the wire shape is in
place now.
148 tests pass (no regressions).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds the on-wire shapes for FoF Mode 2 comment-gating per
docs/fof-spec/layer-2-mode2-fof-comments.md:
- WrapSlot: per-V_x slot with 2B prefilter_tag + 48B read_ciphertext
+ 48B sign_ciphertext (sealed CEK + sealed priv_x_seed). 98 bytes
total per slot. Receiver trial-decrypts via prefilter match.
- FoFCommentGating: author-published gating block embedded in
Post.fof_gating. Carries slot_binder_nonce (32B random; replaces
spec's circular "post_id in HKDF info"), pub_post_set (1:1 with
wrap_slots, includes dummy pubkeys), wrap_slots, and revocation_list
(initially empty; revocation diffs accumulate on the BlobHeader copy).
- RevocationEntry: author-signed entry triggering retroactive comment
delete + pub_post_set removal on every file-holder that receives it.
- CommentPermission gains FriendsOfFriends variant. Existing match arm
in connection.rs handle-incoming-diff path is extended with a
"drop pending CDN four-check verification" stub (full verify in a
later slice).
- InlineComment extended with three optional fields:
pub_x_index: index into parent post's pub_post_set
group_sig: 64B ed25519 sig under priv_x
encrypted_payload: ChaCha20-Poly1305 ciphertext under CEK_comments
All #[serde(default)] for back-compat. Old comments deserialize
cleanly with None.
- Post gains optional fof_gating field for the author-signed snapshot
at publish time. PostId = BLAKE3(Post) covers it, so any tampering
is detectable. Mutations (revocation, access-grant) arrive later as
diffs against the local BlobHeader copy.
All 21 existing Post construction sites + 4 existing InlineComment
sites updated via perl -0pe sweeps to pass None for the new fields.
Full test suite: 134/134 pass (4 new slot crypto + 130 existing).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two pre-release fixes found during audit.
1) GroupKeyDistribute admin forgery (critical)
`group_key_distribution::try_apply_distribution_post` trusted the
`admin` field inside the decrypted payload without verifying it
matched the post's author. Exploit: any peer who learns a victim's
posting NodeId (public — appears as a recipient on any DM/group
post) and observes a target group_id in the wild could craft an
encrypted distribution post claiming to be from the legitimate
admin. The victim's storage uses INSERT OR REPLACE on group_keys,
so a successful forgery would overwrite the victim's legitimate
group key record and stored seed, breaking future rotations / key
distributions from the real admin.
Fix: reject the distribution post when `content.admin != post.author`.
Added test `forged_admin_is_rejected` that seeds a legitimate
record, attempts a forgery, and asserts the legitimate record is
untouched.
2) Cap concurrent port-scan hole punches at 1 (bandwidth)
`hole_punch_with_scanning` fires ~100 QUIC ClientHellos/sec for up
to SCAN_MAX_DURATION_SECS (300s), ~1 Mbps per active scanner. With
no cap, the growth loop / anchor referrals / replication paths
could spawn several scanners at once and drive sustained multi-Mbps
upload — particularly pathological on obfuscated VPNs where every
probe stalls at a proxy timeout, explaining the reported 10 Mbps
sustained upload after anchor connect.
Fix: module-level `tokio::sync::Semaphore(1)` guarding entry to the
scanning loop. Second-and-beyond callers fall back to the cheaper
`hole_punch_parallel` (standard punching, no 100/sec port walk)
instead of spawning another scanner. Permit is held for the scanner
lifetime and released on return. Added unit test
`scanner_semaphore_caps_concurrent_scans_at_one`.
Both changes leave the successful-call path untouched (single scanner
still runs; legitimate key distributions still apply). 120 / 120 core
tests pass.
Removes the last persona-signed direct push on the wire. Group/circle
seeds no longer travel via the 0xA0 `GroupKeyDistribute` uni-stream from
admin to member. Instead the admin publishes an encrypted post containing
the seed + metadata; each member is a recipient; the post propagates via
the normal CDN. Members decrypt with their posting secret to recover the
seed.
Eliminates the wire-level coordination signal between an admin endpoint
and each member endpoint when a group is created, a member is added, or
a key is rotated.
Core pieces:
- New `VisibilityIntent::GroupKeyDistribute` variant.
- New `types::GroupKeyDistributionContent` — JSON payload inside the
encrypted post: group_id, circle_name, epoch, group_public_key, admin,
canonical_root_post_id, group_seed.
- New `group_key_distribution` module:
- `build_distribution_post(admin, admin_secret, record, group_seed, members)`
returns `(PostId, Post, PostVisibility::Encrypted)` — wraps the CEK
per member using standard `crypto::encrypt_post`.
- `try_apply_distribution_post(s, post, visibility, our_personas)`
iterates every posting identity's secret trying to decrypt; on
success stores `group_key` + `group_seed` and returns true.
- `process_pending(s, our_personas)` scans stored
GroupKeyDistribute-intent posts and applies any we can decrypt.
Node API:
- `add_to_circle`: builds a distribution post wrapping the current seed
to just the new member, stores with intent=GroupKeyDistribute, and
propagates via `update_neighbor_manifests_as` (no direct push).
- `create_group_key_inner`: at group creation, after wrapping keys for
every non-self member, builds one distribution post addressed to all
of them and propagates through the CDN.
- `rotate_group_key`: same pattern at epoch rotation.
- New `Node::process_group_key_distributions` — scans and applies.
`sync_all` now calls it automatically so seeds take effect right after
a pull cycle.
Removals (wire-breaking; v0.6.2 already forked):
- MessageType 0xA0 (`GroupKeyDistribute`), its payload struct, the
handler in connection.rs, and `Network::push_group_key` all deleted.
ConnectionManager's `secret_seed` (network secret) is no longer used for
group-key unwrapping — that shifted to posting secrets in the apply
pass, matching the v0.6.1+ identity split where group keys are wrapped
to posting NodeIds.
Tests: new `member_decrypts_and_applies` covers a recipient decrypting +
storing the seed and a non-recipient failing to apply. Workspace
compiles clean; 118 / 118 core tests pass on a stable run (pre-existing
flaky `relay_cooldown` test with a 1ms timing window is unrelated).
Scott Reimers