• v1.17.3 a20f86f25c

    agorise released this 2026-09-12 05:10:51 +00:00 | 78 commits to main since this release

    Morphit v1.17.3

    Follow-up to v1.17.2, driven entirely by what two real upgrades revealed. The
    headline item is that nobody could set up a warrant canary — the setup script
    was being killed mid-prompt by a shell error, and the same defect sat in the
    weekly refresh that keeps an existing canary alive. No protocol or consensus
    change.

    Fixed

    • Warrant-canary setup no longer dies mid-prompt. Answering the questions
      would get as far as "Operator name" and then stop with "Canary setup didn't
      finish", with no explanation. The script was not finishing early — it was being
      terminated. It looks for your instance URL to offer as a default, and the
      search used a pipeline that reports "not found" as an error; combined with the
      script's own abort-on-any-error setting, the first configuration file that
      lacked that setting killed it outright. Since almost no existing instance has
      that setting recorded, this affected essentially everyone. The lookup now
      tolerates a miss, tries more places, and on a Tor-only box falls back to your
      .onion as the default.
    • The weekly canary refresh had the same defect. Identical construct in the
      refresh job that the timer runs. A node whose configuration lacked one
      particular setting — or that simply had no .onion among its chain endpoints —
      would have its refresh killed silently, letting a published canary go stale on
      its own schedule with no error any operator would ever see. That is the worse
      version of the bug, because nothing surfaces it until readers notice the date.
    • The seeding self-check can find your addresses again. Every instance was
      reporting "no hidden address configured" and skipping the frontend check, even
      with a live .onion and a working domain. Two causes: the check read one
      setting from one file, and it runs as an unprivileged user that cannot read
      Tor's address file at all. The upgrade now looks these up itself — where it has
      the permission to — and hands them down, with more fallbacks and a message
      naming exactly where it looked when it still finds nothing.
    • The snapshot mirror installs itself on hand-built servers. Its weekly timer
      was only installed by the automated server setup, which never runs on an
      instance that was installed by hand — including the canonical one. The upgrade
      now installs and enables the timer directly, so the feature is not silently
      inert on exactly the servers that matter most.
    • The stray "New major version of npm available!" banner is gone. It was
      printing after the upgrade's own success message, which read like part of the
      result. Suppressed at the last step that could emit it, so it holds however
      that step is run.

    Changed

    • The "Zero use of clearnet internet" badge now shows a medal (🏅) instead of a
      shield (🛡️)
      in the instance directory. It marks an achievement a node has
      earned by proving all seven outbound paths are hidden, not a protective
      feature — the medal reads that way.

    Security

    • The address lookup deliberately does not read hidden addresses from the
      chain-endpoint configuration. That file lists other operators' .onion
      addresses, and reading it would have made this instance probe a stranger's node
      and report it as its own working seeder.

    Notes

    • No protocol/consensus change. Everything here is operator-facing: setup,
      upgrade, seeding and reporting accuracy.
    • If your canary setup failed on v1.17.2, re-run it after upgrading —
      sudo morphit-ops → "Harden this server", or
      sudo bash /opt/morphit/scripts/canary/setup.sh.
    • OPERATIONS.md §52 covers the snapshot mirror pipeline and how to check it.
    Downloads
  • v1.17.2 71d9696995

    agorise released this 2026-09-12 02:07:57 +00:00 | 79 commits to main since this release

    Morphit v1.17.2

    Hidden-seeding and fast-sync release. Two things were quietly broken in the same
    place: a Tor/I2P-only instance could not upgrade from the federation for weeks
    because of a single missing firewall rule, and a Tor/I2P-only instance could not
    fast-sync at all because the snapshot it needs was only ever reachable over
    clearnet. Both are fixed, along with the checks that reported success while the
    first one was broken. No protocol or consensus change.

    Added

    • Every instance now mirrors the federation indexer snapshot. Fast-sync is
      what gets a brand-new node from an empty database to a live orderbook in
      minutes instead of days, and it depends on one small (~600 kB) file that
      @morphit anchors on-chain. Until now nothing kept that file reachable from
      anywhere except the canonical box and a few public clearnet IPFS gateways —
      which made one box a single point of failure for every new instance, and made
      fast-sync impossible for a zero-clearnet node. Your instance now pins that file
      and re-serves it over your clearnet origin, your .onion and your .b32.i2p,
      refreshing weekly and after every upgrade. Mirroring is not a trust claim: a
      newcomer proves every byte against the signed on-chain SHA-256, so a bad mirror
      is caught by arithmetic rather than by reputation. Nothing to configure.
    • Fast-sync works over Tor and I2P. A new node now prefers federation peers
      on the transport it already speaks, ahead of any clearnet source. A hidden-only
      node omits clearnet sources entirely and falls back to a full replay rather than
      reaching for a public gateway — finishing faster is never worth deanonymising
      the box. Peers are discovered from the same chain history the node already reads
      to find the snapshot, so a brand-new box with no indexer yet still finds a
      private mirror, and no address is baked into the source tree to go stale.
    • The frontend-to-gateway firewall path now repairs itself. Every upgrade
      probes the connection from inside the frontend container and, if it is being
      dropped, fixes it in place — firewall rule, then a fallback rule, then a
      container restart, verifying after each and reporting which one worked. An
      instance admin is never asked to paste a firewall command.

    Fixed

    • Hidden-only instances can upgrade from the federation again. The frontend
      proxies /ipfs/ and /ipns/ to the IPFS gateway on the host, but the firewall
      had no rule for that port — so the container-to-host connection was dropped, nginx
      timed out, and every .onion / .b32.i2p release fetch returned a stock 404. Any
      instance running the web firewall was therefore unable to serve releases over
      Tor or I2P
      , which is why hidden-only nodes fell back to stale copies from
      elsewhere and failed their integrity check. Hardening now opens that port to the
      web-firewall network only (never the public), when this box hosts IPFS — and the
      self-heal above delivers the same fix to instances the installer never manages.
    • The seeding self-check actually runs now. The check introduced for this
      release used a Bash-only construct in a script that runs under /bin/sh, which
      on Ubuntu aborts the moment it is called. In practice the entire per-transport
      verification — and the step after it — never executed on a single real box. It
      is now plain POSIX shell, and a probe that cannot connect reports that instead
      of killing the script, which matters because an unreachable gateway is exactly
      what it exists to detect.
    • No more false "your firewall is broken" warning. The same check asked the
      frontend for the release using the loopback address, which the web firewall
      rejects on every path even on a perfectly healthy box — so it would have told
      every operator at once that their firewall was dropping connections. It now asks
      using the instance's real hostname, and treats a refusal from its own probe as
      inconclusive rather than as a fault.
    • Reachability checks no longer time out on healthy nodes. The Tor and I2P
      checks downloaded the full ~33 MB release tarball, which routinely outruns any
      sane timeout over a hidden transport and reported working nodes as broken. They
      now fetch a small file over the identical path.
    • "Working seeder" is verified, not assumed. The previous self-check asked
      the local gateway whether it served the release and then announced the box was a
      working seeder over every transport. That passed happily while no peer could fetch
      anything. It now checks the path a peer actually uses — through the frontend, then
      end-to-end over each configured .onion and .b32.i2p — and reports each
      transport separately, with the exact fix when one fails.
    • No more false "unsigned tarball will be refused" warning. A release fetched
      over Tor/I2P carries no detached signature by design: it is verified against the
      SHA-256 anchored on-chain, read from the node's own indexer. The warning now
      describes what actually happens — the on-chain hash is used, and the upgrade is
      refused only if neither a signature nor a matching hash can be established.

    Changed

    • The health report no longer overstates IPFS seeding. Its "ok" verdict is based
      on local timer runs (the release is pinned, the IPNS record was rebroadcast), which
      does not prove a peer can fetch it. The wording now says exactly that, and points at
      the upgrade's seed step for the real per-transport reachability check.
    • Snapshot publishing is a single-signer job, and is off by default. Exactly one
      instance in the federation exports and anchors the snapshot; everyone else mirrors
      it. An ordinary instance will never begin anchoring snapshots under its own account
      just because it was installed with defaults.

    Security

    • The new firewall rule is scoped to the pinned web-firewall network and opens only
      the read-only release gateway, which serves pinned releases and nothing else. It is
      gated on the box actually hosting IPFS, so a node without IPFS opens no new port,
      and a Tor/I2P-only box gains no clearnet exposure.
    • A mirroring instance verifies the snapshot against the on-chain SHA-256 before it
      will serve it, and removes the pin if the hash does not match — so a box can never
      become a mirror for content the signer did not publish.
    • The self-heal only ever adds a rule permitting the pinned container network to reach
      the local read-only gateway, and verifies by observing the connection rather than
      trusting a command's exit code.

    Notes

    • No protocol/consensus change. All changes are operator-facing seeding, upgrade,
      fast-sync and reporting accuracy.
    • Mirroring rides on IPFS release hosting: a box without Kubo skips it silently.
      OPERATIONS.md §52 covers the whole pipeline, including how to opt out.
    Downloads
  • v1.17.1 02b32c1731

    agorise released this 2026-09-11 04:57:46 +00:00 | 80 commits to main since this release

    Morphit v1.17.1

    Smooth-install / self-heal release. A run of real operator pain — a fresh OS
    reinstall, a Tor-only upgrade, and a censored-region node — surfaced a handful of
    places where an install or upgrade needed a human to notice and fix something by
    hand. Every one of them now handles itself, or fails loud with the exact fix. No
    protocol or consensus change.

    Fixed

    • An upgrade no longer leaves a containerized frontend serving the OLD build.
      The BunkerWeb/containerized frontend bind-mounts the built site, and a rebuilt
      image is byte-identical, so docker compose up left the running container pinned
      to the pre-upgrade build — the upgrade reported success while /verify.json still
      showed the old version. The frontend is now force-recreated so it re-binds the
      fresh build, and the post-upgrade check self-heals (restart + re-verify) before
      it will warn.
    • operator_tag now shows on containerized instances. The tag was stamped into
      the served verify.json only on the bare-metal deploy path, so every
      container-served instance read operator_tag: null despite a correct config and
      on-chain registration. The stamp now runs before both deploy paths. (Cosmetic —
      fee attribution always used the runtime config and was never affected.)
    • A reinstalled IPFS origin host no longer goes publicly dark on IPFS. Stock
      firewall opened only 22/80/443, so a fresh install left the IPFS swarm port closed
      and public gateways couldn't fetch the seeded release. Hardening now opens 4001
      (tcp + udp)
      automatically when the box hosts IPFS, and the post-harden self-test
      flags it if it's ever not open.
    • The warrant-canary upload survives a reinstall. A fresh OS re-keys only root,
      so the canary's morphit@ upload failed with "Permission denied". Install now
      propagates the operator's login keys to the morphit service user.
    • Canary autorenew no longer fails silently under nvm. The weekly timer runs with
      a minimal PATH and couldn't find a version-manager node; the generated service now
      pins node's directory into its PATH.

    Added

    • Turnkey canary setup. Remote (laptop-signed) setup now offers to generate a
      dedicated passphrase-less upload key, wires it behind an SSH alias, and points the
      refresh at it — so the weekly timer never prompts and the canary can't go falsely
      stale. Same-box setup is drivable with near-zero data entry.
    • Upgrade offers a canary when the footer link would 404. If the site footer's
      canary link has no canary behind it (bad for visitors and SEO), the upgrade offers a
      turnkey same-box setup — signing key, first canary, and weekly schedule, automatic —
      and is a single keystroke to decline. It stays silent for operators who already run
      a remote canary.
    • Every instance self-verifies it's a working release seeder. After seeding a
      release to IPFS, the instance now confirms its own gateway actually serves it (the
      path a peer uses) and says so plainly — or warns loudly if it doesn't — so the
      federation never silently loses a seeder.
    • Hidden-only nodes get a clear, verified offline path. When no federation peer
      yet serves a release over Tor/I2P, the fail-closed message now points to
      morphit-ops upgrade --from-file, which verifies the tarball against the SHA-256
      @morphit anchored on-chain (read from the node's own indexer — zero clearnet).

    Changed

    • morphit-ops edit already offered to re-publish your on-chain registration after a
      name/tag/origin/contact change; this is a reminder that the operator tag attributes
      fee earnings, so a config that drifts to a wrong/reserved tag is now easy to catch
      and correct.

    Security

    • Every change is hidden-only-safe: the 4001 rule is gated on the box actually hosting
      IPFS; the canary offer never fights a remote-signed canary; and the offline upgrade
      path verifies against the on-chain anchor with no clearnet dependency. No new network
      exposure on a Tor/I2P-only box.

    Notes

    • No protocol/consensus change. All changes are operator-facing install, upgrade,
      recovery, and seeding robustness.
    Downloads
  • v1.17.0 69ca9a2d76

    agorise released this 2026-09-10 23:37:58 +00:00 | 82 commits to main since this release

    Morphit v1.17.0

    Operator-safety release. Hardening can no longer strand an operator or take a site
    dark; fast-sync works on a stock box and recovers a stranded flagship from any
    synced peer; Matrix alerts set themselves up with zero hand-editing; and hidden-only
    instances are audited clean of any clearnet reliance.

    Fixed

    • Hardening can no longer lock you out of SSH. harden set PermitRootLogin no
      whenever any key was authorized — including a root-only key — locking out an
      operator whose only login was root. It now disables root login only when a
      non-root keyed user exists, and otherwise keeps key-only root
      (prohibit-password), so it can never strand the sole operator.
    • Hardening can no longer take the public site dark. Enabling UFW flushes
      nftables, wiping Docker's forwarding chains — external traffic reached the box
      but the DNAT-forward to the frontend container was dropped (the box looked
      healthy to itself while the world timed out). harden now re-asserts Docker's
      chains
      after the firewall change, verifies the NAT rules returned, and
      hard-fails if they didn't.
    • Fast-sync works on a stock box. The snapshot bootstrap validated against a
      half-loaded environment and bailed (MORPHIT_INDEXER_CHAIN_ID: Required) because
      the fast-sync wrapper only inherited part of the operator env. It now sources
      the same env files the indexer unit does
      , so a correctly-configured box can
      fast-sync.
    • matrix-bot no longer crash-loops on an E2EE key collision. Reusing an access
      token after a reinstall causes a "one time key … already exists" collision; the
      bot now self-clears the dead crypto store, prints a plain-language fix, and
      exits cleanly instead of hammering systemd.
    • timeapp language filter. Browsing in a locale no longer auto-locks the order
      language filter (which hid other-language orders and re-appeared on refresh);
      clearing the filter now re-queries immediately.

    Added

    • Fast-sync is the default, and can import a peer snapshot. The install offers
      it [Y/n]; the menu path now offers to stop the indexer (checking the real
      service state, no 90 s wait), offers to discard existing data (no --force
      archaeology), and auto-restarts after. New morphit-ops fast-sync --from-file <snapshot.tar.gz> imports a snapshot handed over from any synced peer — for when
      no fresh on-chain snapshot exists.
    • Harden self-heals a broken run (dead-man's-switch). After a run that changes
      SSH or the firewall, an automatic revert is armed via a transient systemd-run
      timer (independent of your SSH session). Confirm you can still reach the box and
      it cancels; otherwise it restores the exact pre-run SSH + firewall + Docker
      state
      within ~2 minutes. Plus a post-harden self-test that fails loud on a
      lockout.
    • Matrix alerts set themselves up — no hand-editing. morphit-ops matrix setup
      (and the install wizard, and harden) collect the recipient + the bot's
      username/password, mint a fresh token via the Matrix login API, persist it,
      clear the stale crypto store, start the bot, and send a test. Entering a Matrix
      address during setup wires the whole thing.
    • Health report: a "Generating your health report…" spinner; CPU/Memory now
      show free; a "Reachable at" list of every address the node serves (clearnet +
      Tor .onion + I2P .b32.i2p); and the install wizard now has an explicit warrant
      canary step and offers to serve the frontend on a new vanity .loki/.i2p.

    Changed

    • npm's update-notifier is silenced globally (no more "new major version of npm"
      noise on install/upgrade). Italian "commissione" → "commissione (fee)" on the
      order fee-status line. The redundant CryptoCompare price source is dropped
      (it required new auth); a source that's up for one asset but doesn't list another
      now reads "no market" instead of an alarming "down".

    Security

    • Hidden-only instances stay clearnet-free. Audited across all of the above:
      the harden auto-revert restores the exact pre-run firewall rules (never opens
      clearnet ports a hidden-only box didn't have); the post-harden self-test skips
      clearnet checks on a Tor/I2P-only box; the health "Reachable at" list never prints
      a clearnet address the node doesn't serve; and fast-sync's env-sourcing fix
      actually improves hidden-only sync (the bootstrap now gets the Tor/I2P proxy
      config). The Matrix token mint uses the same outbound path as the bot itself; a
      fully air-gapped box uses the paste-a-token path.
    • Matrix token stays 0600. The bot access token is written only to
      matrix-bot.env (0600, owner-only) and preserved from there on re-render; only
      the non-secret recipient MXID lives in the group-readable operator config.

    Notes

    • No protocol/consensus change. All changes are operator-facing safety, recovery,
      and privacy hardening.
    Downloads
  • v1.16.13 74122cb0aa

    agorise released this 2026-09-09 02:02:15 +00:00 | 88 commits to main since this release

    Morphit v1.16.13

    Upgrade robustness on slow/throttled links, an honest update-check, and one more
    self-heal moved onto the release that ships it.

    Fixed

    • Upgrades survive a slow connection. The release download used a fixed 30 s
      total deadline and buffered the whole tarball in memory, so a healthy-but-slow
      link (a throttled or filtered connection) got its download aborted mid-transfer
      ("this operation was aborted"). It now streams to disk with an idle timeout:
      the transfer is only aborted if no bytes arrive for 90 s, so a slow-but-steady
      download completes. Only a genuine stall aborts.
    • A failed update-check no longer looks like "up to date." When the version
      check couldn't reach the release server (e.g. a slow link timed out), the menu
      silently dropped the "update available" marker — indistinguishable from being
      current. It now shows (couldn't check for updates — network) so an outdated
      node is never mistaken for a current one.

    Changed

    • The frontend rebuild now runs in the post-upgrade self-heal phase, from the
      freshly-installed binary (via the re-exec), alongside the WAF and IPFS heals.
      The frontend's nginx.conf is baked into its image, so a config change (e.g. the
      v1.16.12 /v1/broadcast body-size cap) previously only landed on the next
      upgrade; now it applies on the upgrade that ships it. A no-change rebuild is
      cheap (Docker layer cache), and a box without a compose-managed frontend no-ops.

    Notes

    • No protocol/consensus change. All three are operational robustness only.
    • A node upgrading from a pre-v1.16.11 release (no re-exec phase) picks up the
      self-heal behaviour on its next upgrade, once it is driven by a v1.16.11+
      orchestrator.
    Downloads
  • v1.16.12 b3a8f456c6

    agorise released this 2026-09-08 20:00:53 +00:00 | 90 commits to main since this release

    Morphit v1.16.12

    Two directory-accuracy fixes reported from live instances: a verify.json
    operator_tag that stayed null, and an operator whose contact showed on the
    instances page but not the operators page.

    Fixed

    • verify.json operator_tag no longer stays null. The deploy stamps the
      served verify.json with this instance's tag, but it only read the local config
      — and on an instance whose tag was registered on-chain but never written to the
      local env, that read nothing and left null. It now falls back to the on-chain
      registration the local indexer already serves at /v1/instances (matched to this
      instance's own origin) — the same source that correctly shows the tag in the
      public directory — so a deployed verify.json shows the real tag.
    • Operators page shows a contact even when the on-chain operator record has
      none.
      An operator's contact renders from its on-chain operator_register
      contact, which can be empty even when the operator's instance publishes one
      (e.g. time.relay: a contact on /instances but blank on /operators). The
      operators page now falls back to that instance contact (keyed by operator tag)
      when the on-chain operator contact is empty, using the same scheme-aware policy
      (email / Matrix / XMPP / … render; bare emails repaired).

    Notes

    • No protocol/consensus change; both fixes are display/deploy only. operator_tag
      in verify.json remains informational (fee attribution comes from the runtime
      indexer config, unaffected).
    Downloads
  • v1.16.11 a5758eb73c

    agorise released this 2026-09-08 05:34:07 +00:00 | 91 commits to main since this release

    Morphit v1.16.11

    The hidden-upgrade path is now self-sufficient end to end, and a self-heal
    shipped in a release now takes effect on that upgrade instead of the next one.
    This closes out the zero-clearnet upgrade story: expose → advertise → discover →
    fetch, all automatic, no operator ever hand-signs or hand-configures anything.

    Fixed

    • The /ipns/… release pointer is signed again. The federation's stable IPNS
      name had frozen at an old release because the CI signing step is gated on a
      MORPHIT_IPNS_KEY secret that wasn't set — so every hidden upgrade that resolved
      the release by IPNS got stale bytes and (correctly) rejected them. With the key
      in place, each release re-signs the record and every instance rebroadcasts it, so
      /ipns/… resolves to the current release across the whole federation. The stable
      name is rotated to a fresh identity (k51qzi5uqu5dgkxm…); the site, docs, and the
      on-chain ipns_name all track it (the on-chain value is derived from the key at
      release time, so it can't drift from the site).
    • Hidden peer discovery no longer drops Tor. resolvePeerGateways used an
      else if that only kept a peer's .b32.i2p address whenever it had one, so
      .onion peers were never tried. It now offers both transports per peer.
    • Self-heals apply on the upgrade that ships them. After the upgrade rebuilds
      the new ops-cli bundle it re-execs the new binary for a dedicated self-heal
      phase (__post-upgrade-selfheal), falling back to in-process if that binary is
      too old. So from here on, a fix that self-heals a box lands on the upgrade that
      introduces it — no more "the fix is in, but you have to upgrade twice."

    Added

    • Seeder discoverability nudge. The health report now reminds a node that is
      seeding the release to publish its Tor/I2P address on-chain (via
      morphit-ops register) so hidden-only nodes can actually discover and fetch from
      it — the advertise half of "expose → advertise → discover."

    Notes

    • No protocol/consensus change; clearnet upgrades are unaffected. The IPNS-name
      rotation is transparent to users (the download page and on-chain pointer both
      carry the new name from this release forward). A node still on an older version
      that can only reach stale peers may need one offline hop to reach a version with
      fetch-by-CID (v1.16.10) — after which it upgrades over the hidden network with no
      further manual steps.
    Downloads
  • v1.16.10 4bc2c9c652

    agorise released this 2026-09-08 02:32:28 +00:00 | 92 commits to main since this release

    Morphit v1.16.10

    Every instance is now a Tor/I2P release seeder by default — automatically.
    This makes zero-clearnet upgrades self-sufficient across the federation: a
    hidden-only node fetches the signed release from any peer over .onion/.i2p,
    verified against the on-chain SHA-256, with no hand-signing and no manual gateway
    setup. It's the piece that lets zero-clearnet instances proliferate without an
    operator ever touching IPFS config.

    Changed

    • IPFS gateway exposed over Tor/I2P by default. morphit_ipfs_gateway_expose
      now defaults on, and the upgrade self-heals it on every existing instance —
      no admin step. It sets Gateway.NoFetch=true first (so the gateway serves
      ONLY the release CIDs this node has pinned — never an arbitrary CID, never an
      open proxy), then binds the gateway bridge-reachable, restarts Kubo, ensures the
      IPNS rebroadcaster is running, and verifies the gateway is live. Best-effort and
      idempotent; a node without IPFS hosting simply no-ops.
    • Hidden-only nodes seed the canonical release too. The offline bundle now
      ships the canonical morphit-<ver>.tar.gz under .canonical-release/, and an
      offline upgrade seeds that — so a zero-clearnet node produces the CID that
      matches the on-chain anchor and becomes a real Tor/I2P origin host, not just a
      consumer.

    Fixed

    • Hidden upgrade fetches by the on-chain CID, not a peer's IPNS. It now tries
      /ipfs/<ipfs_cid>/… first — content-addressed, so it pulls the EXACT canonical
      bytes from any peer that has the CID pinned, regardless of that peer's IPNS
      freshness — with IPNS as a fallback. This is what let a stale/divergent peer
      break the fetch before; now a peer serving old bytes just loses the race to a
      canonical one. Every seeder also runs the IPNS rebroadcaster (anti-stale on the
      publisher side).
    • Deterministic seed across Kubo versions. The ipfs add DAG params
      (--chunker=size-262144 --raw-leaves) are now pinned in both the CI anchor
      computation and the node seed, so every node — on any Kubo version — reproduces
      the same CID as the on-chain anchor. (No-op for modern Kubo, which already
      matched; fixes older ones.)

    Notes

    • No protocol/consensus change. Clearnet upgrades are unchanged (git mirrors +
      on-chain SHA + HTTPS primary). The gateway exposure only affects a box that runs
      IPFS hosting and is safe by construction (NoFetch); the anchor CID for a given
      release only changes for releases cut after this one (v1.16.9's anchor is
      already published), so the deterministic-seed pinning takes effect going forward.
    Downloads
  • v1.16.9 e2b22a29ea

    agorise released this 2026-09-07 22:42:22 +00:00 | 93 commits to main since this release

    Morphit v1.16.9

    Installs and upgrades self-heal the reverse-proxy WAF, and hidden/offline
    upgrades no longer need a hand-signed tarball.
    This release closes a cluster of
    real operator-reported issues — a JSON-API firewall that was 4xx-ing legitimate
    avatar/order broadcasts, a warrant canary that wouldn't publish on a same-box
    setup, and several UI papercuts — and hardens the BunkerWeb self-heal so an admin
    never has to run a diagnostic for something the upgrade can detect and fix itself.

    Fixed

    • BunkerWeb WAF no longer blocks the JSON APIs — self-healed on upgrade, three
      ways.
      Three independent BunkerWeb layers were rejecting a legitimate
      POST /v1/broadcast (e.g. an avatar upload): (1) MAX_CLIENT_SIZE defaulted
      below the ~8 KB broadcast body → 413; (2) bad-behavior counted the routine
      400s a JSON API returns when it rejects an invalid/expired broadcast, banning
      the client IP → 403 on everything; (3) OWASP CRS false-positived on the
      base64 payload → 403. The upgrade now self-heals all three: sets
      MAX_CLIENT_SIZE=1m (the relay's 64 KB stays the real cap), removes 400 from
      the bad-behavior triggers, and exempts /v1/+/relay/ from ModSecurity. Each
      fix is applied more than one way (the ModSec exemption as both an env var and
      a mounted config file), the reload uses a fallback chain, and the result is
      verified against the running container — no admin diagnostics, no
      back-and-forth. Fresh installs get the same settings from the Ansible template.
    • Offline / hidden-only upgrades verify against the on-chain SHA-256 — no more
      hand-signing.
      A zero-clearnet or air-gapped node can now apply an offline
      tarball by checking it against the release hash @morphit published on-chain
      (read from the node's own indexer over its hidden RPC), instead of requiring a
      hand-made GPG .asc. The release also anchors the -offline bundle's own
      offline_sha256, and decideTrust gained an onchain-anchored-sha256 proof.
    • Offline-bundle upgrades no longer print a scary "CID MISMATCH." A
      -offline bundle is a different artifact than the standard tarball the on-chain
      IPFS CID anchors, so the self-seed can never match it — the upgrade now skips
      the seed cleanly with a one-line note instead.
    • operator_tag is stamped into the served verify.json on every deploy, so
      a prebuilt-tarball upgrade no longer serves operator_tag: null.
    • Operators-page contact links render every scheme. The operators page used
      an https-only validator that silently dropped email / Matrix / XMPP / Discord-
      invite / nostr / session / cwtch contacts; it now uses the shared, scheme-aware
      contact policy (and repairs a bare email to mailto:), matching the instances
      page.
    • Warrant canary: same-box setup publishes immediately. Local (same-box)
      canary setup now hands the served build/ dir over before the first publish
      (so it isn't a root-owned "permission denied"), and verifies the file actually
      landed. In the health report, a not-yet-published canary is now a yellow
      "pending" ⚠ with the exact publish command, not an alarming red ✗ — only an
      expired canary is red.
    • Compare-orderbooks page handles zero-clearnet + http/https, with accurate
      errors.
      The instance-URL validator now accepts http:// and https:// (and
      defaults bare .onion/.i2p/.loki hosts to http://), so zero-clearnet
      instances can be compared. The stale "could not fetch" error now clears the
      moment you edit the field, so it never describes a previous input.
    • Sign-in key field resists Firefox autofill. The import page's posting-key
      field starts read-only (unlocked on focus) with a unique name, so Firefox no
      longer dumps a saved site password into it on load.
    • The sliding "⇨" link arrow has breathing room. A small text-facing gap
      (logical, so it's correct in RTL too) keeps the glyph from jamming against the
      label.

    Added

    • Two FAQ articles, in all 10 locales, for the two things operators kept
      asking about: why login asks for your posting key instead of the password you
      set, and why a brand-new chat/order can take a moment to appear (indexer catch-
      up), with search-friendly wording.

    Notes

    • No protocol/consensus change. Clearnet instances are unaffected by the
      offline/hidden verification change. The WAF self-heal only touches a BunkerWeb
      deployment and no-ops elsewhere; it is idempotent and skips its reload when the
      settings are already in place.
    Downloads
  • v1.16.8 97b0ef6768

    agorise released this 2026-09-07 04:45:29 +00:00 | 94 commits to main since this release

    Morphit v1.16.8

    Hidden-only upgrades can now actually find their peers. v1.16.6 made a
    hidden-only node correctly resolve its upgrade over Tor/I2P and refuse clearnet —
    but a field-name bug meant it never actually saw any federation peer's hidden
    address, so every hidden-only upgrade dead-ended at "no federation peer exposes a
    hidden IPFS gateway yet." This fixes the discovery.

    Fixed

    • Hidden-only upgrade now discovers peer gateways. The upgrade reads the
      federation directory (/v1/instances) to find peers' Tor/I2P addresses, but the
      resolver read i.tor / i.reg_alt_networks, while the response actually nests
      them under i.alt_networks (alt_networks.tor, alt_networks.i2p_b32). So it
      found zero peers and always reported "no hidden gateway." It now reads
      alt_networks (with the others as fallbacks), so a hidden-only node sees every
      peer that advertises a hidden address and fetches the release over Tor/I2P.
    • Clearer fail-closed messaging. When no peer serves the release (as opposed
      to none advertising an address), the error now names the fix: a federation peer
      must expose its IPFS gateway over Tor/I2P (morphit-ops → Web firewall / IPFS gateway), instead of a generic "no gateway."

    Notes

    • No protocol/consensus change; clearnet instances are unaffected (they never use
      the hidden path). hidden-upgrade-detection-smoke now pins that the resolver
      reads alt_networks so this field-name mismatch can't recur.
    • Operator note for hidden-only nodes: for a hidden-only node (e.g. one that
      eliminates clearnet entirely) to upgrade over Tor/I2P, at least one federation
      peer with a Tor/I2P address must expose its IPFS gateway over that hidden
      service. Enable it on a well-connected clearnet peer (one that already has a
      .onion/.i2p) so hidden-only nodes have a private source for releases.
    • A node currently on the buggy v1.16.6/v1.16.7 hidden resolver cannot pull this
      fix over Tor/I2P (its own resolver can't find peers) — deliver v1.16.8 to it
      once via the offline / local-tarball upgrade path over a channel you control;
      every hidden upgrade after that works automatically.
    Downloads