SV2 pool identity: persistent authority key for miner pinning (+ stats correctness fixes)

.gitignore

@@ -8,4 +8,5 @@ config.toml
 .cursor/
 .windsurf/
 .github/copilot-instructions.md
-*.sqlite
+*.sqlite
+sv2-authority.key

CHANGELOG.md

@@ -9,6 +9,56 @@ everything else bumps the **patch** version.
 
 ## [Unreleased]
 
+### Added
+- **SV2 pool identity (authority key pinning).** The Noise authority key now
+  persists across restarts (`[sv2] authority_key_file`, created on first start
+  with owner-only permissions, same pattern as bitcoind's `.cookie`), so the
+  pool keeps a stable identity that miners can pin. The base58check public key
+  (SRI `key-utils` format) is logged at startup and shown in the dashboard's
+  Connect modal with a copy button, and returned by `GET /api/info` as
+  `sv2_authority_pubkey`. Set it as the pool/authority public key on an SV2
+  miner to cryptographically verify the pool; miners that do not pin connect
+  exactly as before. `[sv2] persist_authority_key = false` opts out (fresh key
+  per process, the previous behavior).
+- `[sv2] cert_validity_secs` — validity window of the per-connection
+  certificate (default one year). Short values are useful for testing how a
+  verifying miner handles certificate expiry and clock skew.
+- Tests: full Noise handshakes against a pinning SRI initiator (correct key
+  accepted, wrong authority key rejected, expired certificate rejected, no-pin
+  still connects) plus authority-key-file round-trip/permission checks.
+
+### Changed
+- **Upgrade note (breaking for read-only deployments):** with SV2 enabled the
+  pool now creates `sv2-authority.key` (relative to its working directory) on
+  first start and **fails at boot if it cannot**. Deployments with a read-only
+  working directory, such as the shipped systemd unit with
+  `ProtectSystem=strict`, must set `[sv2] authority_key_file` to a writable
+  path (e.g. `/var/lib/solo-pool-rs/sv2-authority.key`) or set
+  `persist_authority_key = false`. Docker users who want the pool identity to
+  survive container re-creates should point it into the data volume
+  (`authority_key_file = "data/sv2-authority.key"`).
+
+### Fixed
+- Dashboard: the Connect modal **Copy buttons now actually copy** when the
+  dashboard is served over plain HTTP (the usual LAN case).
+  `navigator.clipboard` only exists in secure contexts, so the old code
+  selected the text and showed "Copied" without copying. Insecure contexts now
+  fall back to `document.execCommand('copy')`, and if even that fails the
+  button says "Copy manually" and leaves the text selected.
+- **Duplicate-share tracking** no longer misreports or permits bounded replay:
+  a share is recorded for dedup only after it validates (invalid submissions
+  previously occupied slots, so a later identical valid submit was wrongly
+  rejected as `duplicate`), and the per-session set is cleared on every
+  clean-job broadcast, scoping replay protection to live jobs instead of FIFO
+  eviction (evict-then-resubmit could inflate share/hashrate stats).
+- Pool **best-share / best-hashrate writes are monotonic end to end**: the
+  SQLite `UPDATE`s now carry a `?1 > ...` guard (matching the per-worker
+  variant) and the in-memory best-hashrate update is a CAS loop, so racing
+  writers can no longer regress a recorded best value.
+- A block accepted by the **background submit retrier** (inline attempts
+  failed, e.g. while bitcoind restarts) now updates the dashboard block count
+  and last-block panel, not just the Prometheus counters.
+
 ## [0.5.1] - 2026-06-15
 
 ### Added

Cargo.toml

@@ -77,6 +77,7 @@ charming = "0.6"
 codec_sv2 = { version = "5.0.0", features = ["noise_sv2"] }
 noise_sv2 = "1.4.2"
 secp256k1 = { version = "0.28", features = ["rand", "std"] }
+key-utils = "1.2.0"
 
 [dev-dependencies]
 tokio-test = "0.4"

README.md

@@ -263,12 +263,14 @@ On a NerdQAxe++ (AxeOS ≥ v1.0.37):
 | Field | Value |
 |---|---|
 | Stratum | select **Stratum V2** |
-| Encryption | **on** (Noise), no authority pubkey needed; leave it unset |
+| Encryption | **on** (Noise); authority pubkey optional, see below |
 | Host / Port | `<your-server-ip>` : `3333` (same as SV1) |
 | Worker | anything (used as the SV2 `user_identity`) |
 
 The connection is secured with the SV2 **Noise** handshake (pool = responder); the device then opens an **Extended Channel** and is served `NewExtendedMiningJob` + `SetNewPrevHash` from the same `getblocktemplate` pipeline as SV1. Set `enabled = false` under `[sv2]` to refuse SV2 and serve SV1 only.
 
+**Pool identity (optional pinning).** The pool signs each connection's Noise certificate with a persistent authority key and prints the base58check public key at startup (also shown in the dashboard's Connect modal, and at `GET /api/info`). Miners that support it can pin this key to cryptographically verify they are talking to your pool; miners that leave it unset connect exactly the same, encrypted but without identity verification. The key file (`[sv2] authority_key_file`, default `sv2-authority.key`) is created on first start; `persist_authority_key = false` reverts to a fresh key per process. Both the accept and reject paths are covered by tests that run a real handshake against a pinning SRI initiator, including wrong-key and expired-certificate cases.
+
 ---
 
 ## Dashboard & metrics

TODO.md

@@ -25,20 +25,20 @@ underflow panic). Line references are as of that review and may drift.
   funnel through a dedicated writer thread, enable WAL + `synchronous=NORMAL`),
   and the dashboard `/history` + `/chart` SQLite scans (contend with the share
   path on the same connection mutex; wrap in `spawn_blocking`).
-- [ ] **Harden the duplicate-share set.** 4096-entry FIFO allows bounded replay
-  (evict-then-resubmit inflates share/hashrate stats); shares are also inserted
-  *before* validation, so invalid shares occupy slots and later identical
-  submits are misreported as `duplicate`. Scope dedup to live jobs and insert
-  only after validation passes. (`src/mining/validator.rs`)
-- [ ] **Credit background-retrier block acceptance to dashboard stats.** A block
-  accepted by the PR #6 background retrier updates Prometheus counters but not
-  the dashboard block list / pool stats (needs session context plumbing).
+- [x] **Harden the duplicate-share set** (unreleased, headed for v0.6.0):
+  shares are recorded for dedup only after validation passes, and the
+  per-session set clears on every clean-job broadcast (live-jobs scoping); the
+  4096 FIFO cap remains as a memory backstop only.
+- [x] **Credit background-retrier block acceptance to dashboard stats**
+  (unreleased, headed for v0.6.0): worker + `PoolStats` are threaded through
+  `submit_found_block` into the resubmit task; retry success now mirrors the
+  inline-success stats update.
 
 ## Low
 
-- [ ] Monotonic guard on pool best-share/best-hashrate SQLite `UPDATE`s
-  (`WHERE ?1 > ...`), matching the per-worker variant; also make the
-  best-hashrate in-memory update a CAS. (`src/stats.rs`)
+- [x] Monotonic guard on pool best-share/best-hashrate SQLite `UPDATE`s
+  (`WHERE ?1 > ...`), matching the per-worker variant; best-hashrate in-memory
+  update is now a CAS. (unreleased, headed for v0.6.0)
 - [x] Fix ghost-online accounting: repeated `mining.authorize` increments
   `active_sessions` per call but disconnect decrements once, for the last name
   only. (Fixed alongside the authorization cap: same-name re-auth is a no-op,
@@ -51,9 +51,15 @@ underflow panic). Line references are as of that review and may drift.
 ## Planned features
 
 - [x] **v0.4.0: non-root Docker image** (shipped in v0.4.0, 2026-06-11).
-- [ ] **SV2 identity pinning:** optional persistent Noise authority keypair via
-  config instead of the per-process ephemeral key (deferred in the
-  `protocol/sv2/noise.rs` docstring; today no miner verifies pool identity).
+- [x] **SV2 identity pinning** (unreleased, headed for v0.6.0): persistent
+  Noise authority key (`[sv2] authority_key_file`, cookie-style
+  create-on-first-start), pubkey logged at boot + shown in the dashboard
+  Connect modal + `GET /api/info`; `persist_authority_key = false` opts out,
+  `cert_validity_secs` configurable. Verified on a NerdQAxe++: pinned key
+  verifies and mines, wrong key rejected. Note: the bitaxe/nerdqaxe firmware
+  checks only the Schnorr signature, never the validity window (no wall
+  clock); upstream enforcement-toggle PRs: bitaxeorg/ESP-Miner#1796,
+  shufps/ESP-Miner-NerdQAxePlus#656.
 - [ ] **SV1-over-TLS (`stratum+ssl://`) — DEFERRED, build only on request.**
   Decision (2026-06-15): not building it. The target audience is the
   self-hosted *solo* crowd on a trusted LAN, where the value is marginal — solo

config.toml.example

@@ -77,12 +77,29 @@ found_block_dir = "found-blocks"
 # NerdQAxe++ on AxeOS >= v1.0.37, "Stratum V2" selected) at the same host:port
 # as your SV1 miners.
 #
-# The Noise authority keypair is generated automatically per process; the miner
-# does not verify pool identity, so no key configuration is required.
-#
 # Set to false to refuse SV2 and serve SV1 only.
 enabled = true
 
+# Persist the Noise authority key so the pool keeps the same identity across
+# restarts. The base58check public key is logged at startup and shown in the
+# dashboard's Connect modal; set it as the pool/authority public key on the
+# miner to cryptographically verify the pool (identity pinning). Miners that
+# do not pin connect exactly as before. Set to false for a fresh key each
+# start (pinning miners will then refuse to connect after every restart).
+persist_authority_key = true
+
+# Where the authority secret key lives (one base58check line, created with
+# owner-only permissions on first start). Relative paths resolve against the
+# service working directory, like stats_db_path. Supports ~ expansion.
+authority_key_file = "sv2-authority.key"
+
+# Validity window (seconds) of the certificate signed for each connection:
+# valid_from = now, not_valid_after = now + cert_validity_secs. Miners that
+# verify pool identity check this window against their own clock (SRI-based
+# verifiers allow 10 s of drift), so short values expose device clock skew.
+# Default: one year.
+cert_validity_secs = 31536000
+
 [bitcoin_rpc]
 # Bitcoin RPC endpoint (core/knots)
 url = "http://127.0.0.1:8332"

docker-compose.yml

@@ -35,6 +35,9 @@ services:
       #   mkdir -p data && sudo chown -R 10001:10001 data
       # Then in config.toml set stats_db_path = "data/pool_stats.sqlite" and
       # found_block_dir = "data/found-blocks" so both land in this volume.
+      # The SV2 authority key defaults to /app/sv2-authority.key, which does not
+      # survive a container re-create. If miners pin the pool identity, also set
+      # authority_key_file = "data/sv2-authority.key" so the key lives here too.
       - ./data:/app/data
 
 # Without host networking, drop `network_mode: host`, add:

packaging/systemd/solo-pool-rs.service

@@ -45,6 +45,12 @@ Group=solo-pool
 
 # systemd creates and owns /var/lib/solo-pool-rs for you. Point the SQLite
 # stats DB there:  stats_db_path = "/var/lib/solo-pool-rs/pool_stats.sqlite"
+# The SV2 authority key needs a writable path too. Its default
+# ("sv2-authority.key") resolves against the working directory, which is
+# read-only under ProtectSystem=strict below, and the pool fails at boot if it
+# cannot create the key. In config.toml set:
+#   authority_key_file = "/var/lib/solo-pool-rs/sv2-authority.key"
+# (or set persist_authority_key = false to keep an ephemeral per-process key).
 StateDirectory=solo-pool-rs
 
 # Logs go to the journal (journalctl -u solo-pool-rs). Keep log_dir empty in