All checks were successful
PR Checks / plugin-checks (pull_request) Successful in 8s
R2, the game half. The website is the author of record for groups and grants; this plugin makes the framework's own permission store match what it is sent. **One verb, and the plugin does the diffing (D32).** `perm.sync` carries the whole set the site authors for this server. The plugin compares it against the live store and writes only what differs, then answers with a report: what it applied, what was already right, the names it could not resolve, the memberships waiting on a first connection, and every holder the site did not author. `perm.catalogue` is the other half — a live read of what the loaded plugins have registered, which is the option source the authoring form is built from. **The first command whose work is unbounded, so it is drained rather than run.** A community with two thousand linked players sends thousands of operations in one frame, and applying them in the tick the frame arrives is a freeze an operator will blame on the game. A sync compiles into a list of single-store operations and drains two hundred at a time on a timer. That is §5's rule — nothing the far side sends may cost the main thread an unbounded amount of work — applied to the inbound half for the first time. **The pre-check §12.2 rule 1 exists for.** `GrantUserPermission` returns void, throws nothing and logs nothing for a name no plugin registered, so every grant naming an unloaded plugin's permission would disappear without a trace. The plugin checks `PermissionExists` first and reports the name as unresolved; it does NOT register the name itself (D33), because that fabricates a permission the operator never installed. **`UserHasPermission` cannot answer the question the diff asks.** It is true for a member of a group that carries the permission, so a direct grant would look already applied the moment a group covered it — and a revoke already done. `GetPermissionUsers` reads the user's own set, which is what a direct grant is. Six permission hooks join the read path, all void, all in `ExpectedHooks`. They are not the drift record — only the site's desired set can say whether a change is foreign — they are a reason to reconcile sooner, and they are suppressed while this plugin is writing, or every sync would announce itself as drift. `rg.perms` reports the last report from the console, which is the fastest way to tell "no plugin registered that name" from "the store has never seen that player". Protocol bumped to 4 here and in overlay.toml, in the same change as the emitters. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PMH6bw1jXMgbyF3ZWGEzSM
3086 lines
121 KiB
C#
3086 lines
121 KiB
C#
using System;
|
|
using System.Collections.Concurrent;
|
|
using System.Collections.Generic;
|
|
using System.Net.Sockets;
|
|
using System.Text;
|
|
using System.Threading;
|
|
using Newtonsoft.Json;
|
|
using Oxide.Core;
|
|
using Oxide.Core.Plugins;
|
|
|
|
namespace Oxide.Plugins
|
|
{
|
|
/// <summary>
|
|
/// The Runic Gateway bridge. Dials out to a rust-link sidecar on loopback and speaks
|
|
/// newline-delimited JSON over it: events out, commands in.
|
|
///
|
|
/// <para>
|
|
/// Threading contract, which everything later in this plugin depends on:
|
|
/// </para>
|
|
/// <list type="bullet">
|
|
/// <item><description>
|
|
/// <see cref="Emit"/> is called from the server's main thread. It formats nothing,
|
|
/// blocks on nothing, and touches no socket. It enqueues and returns. A slow, wedged,
|
|
/// or absent sidecar cannot stall the game.
|
|
/// </description></item>
|
|
/// <item><description>
|
|
/// One link thread owns the socket. It connects, drains the queue, and reconnects with
|
|
/// backoff. A single writer keeps event ordering intact.
|
|
/// </description></item>
|
|
/// <item><description>
|
|
/// A reader thread parses inbound lines and hands each to the main thread via
|
|
/// <c>Interface.Oxide.NextTick</c>. The reader touches no Unity object, no BasePlayer
|
|
/// and no ConVar — every one of those is main-thread-only, and reading them from here
|
|
/// is the kind of bug that presents as a crash somewhere else entirely.
|
|
/// </description></item>
|
|
/// </list>
|
|
///
|
|
/// <para>
|
|
/// The outbound queue is bounded. On overflow the oldest record is dropped and counted,
|
|
/// because telemetry is worth less than the server's memory.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// <b>Loopback is the trust boundary.</b> There is no token on this link: the plugin and
|
|
/// the sidecar share a host, and the sidecar binds 127.0.0.1. Pointing <c>Host</c> at
|
|
/// anything routable puts an unauthenticated command channel on the network.
|
|
/// </para>
|
|
/// </summary>
|
|
[Info("Runic Gateway", "RunicGateway", "0.1.0")]
|
|
[Description("Bridges this Rust server to a Runic Gateway website through a rust-link sidecar.")]
|
|
internal class RunicGateway : RustPlugin
|
|
{
|
|
/// <summary>
|
|
/// Wire-protocol version. Declared in three places that must agree: here,
|
|
/// <c>PROTOCOL_VERSION</c> in the sidecar, and <c>module.json</c> in Module-Rust. The
|
|
/// overlay's <c>overlay.toml</c> carries this same number so the installer can refuse
|
|
/// to pair a sidecar and an overlay that disagree.
|
|
/// </summary>
|
|
private const int ProtocolVersion = 4;
|
|
|
|
/// <summary>
|
|
/// Loopback reconnects are cheap, so the ceiling is low. A sidecar restart should cost
|
|
/// a few seconds of buffering, not half a minute.
|
|
/// </summary>
|
|
private const int MaxBackoffMs = 5000;
|
|
|
|
/// <summary>The longest line accepted from the sidecar. Over-long lines are discarded.</summary>
|
|
private const int MaxInboundLineBytes = 1024 * 1024;
|
|
|
|
/// <summary>
|
|
/// How long to wait for the TCP connect itself before giving up and backing off.
|
|
///
|
|
/// <para>
|
|
/// It exists because <c>TcpClient.Connect</c> is blocking and has no timeout of its
|
|
/// own, and the OS default is tens of seconds. A host that REFUSES answers instantly —
|
|
/// which is every loopback test, and why this was invisible for two phases — but a host
|
|
/// that <b>drops</b> does not answer at all: a firewall, a wrong address, a machine
|
|
/// that is off. The link thread then sits in the connect, deaf to everything, and
|
|
/// <c>Unload</c> waits out its whole join on the game's MAIN THREAD.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// Generous against any real network and short against a hang. Loopback, which is the
|
|
/// supported deployment, connects in well under a millisecond.
|
|
/// </para>
|
|
/// </summary>
|
|
private const int ConnectTimeoutMs = 3000;
|
|
|
|
/// <summary>
|
|
/// How often the boards are re-sent (PROTOCOL.md §8.3) and the per-player tallies are
|
|
/// flushed (§8.6). One cadence serves both because both answer the same question —
|
|
/// "what has this server been doing lately" — and two timers would only be two things
|
|
/// to get out of step.
|
|
/// </summary>
|
|
private const float BoardCadenceSeconds = 60f;
|
|
|
|
/// <summary>
|
|
/// The side of one map grid cell, in world units. Rust's own grid, as every community
|
|
/// map and every server console uses it.
|
|
/// </summary>
|
|
private const float GridCellSize = 146.3f;
|
|
|
|
private static readonly ConcurrentQueue<string> Outbound = new ConcurrentQueue<string>();
|
|
private static readonly AutoResetEvent Wake = new AutoResetEvent(false);
|
|
|
|
/// <summary>
|
|
/// Set once, by <c>Unload</c>, and never reset. It is what lets a thread blocked on a
|
|
/// CONNECT give up immediately instead of at the end of its timeout.
|
|
///
|
|
/// <para>
|
|
/// It cannot be <see cref="Wake"/>: that handle also means "the queue has something in
|
|
/// it", which <c>Emit</c> signals constantly, so waiting on it during a connect would
|
|
/// abandon the attempt every time a hook fired. Two meanings, two handles.
|
|
/// </para>
|
|
/// </summary>
|
|
private readonly ManualResetEvent _stopping = new ManualResetEvent(false);
|
|
|
|
private PluginConfig _config;
|
|
|
|
private Thread _link;
|
|
private volatile bool _running;
|
|
private volatile bool _connected;
|
|
|
|
/// <summary>Set when the peer goes away, so the writer stops trying.</summary>
|
|
private volatile bool _dead;
|
|
|
|
/// <summary>
|
|
/// Whether the "cannot reach the sidecar" line has been printed for this load. It is
|
|
/// printed once, not per attempt: the retry runs every few seconds for ever.
|
|
/// </summary>
|
|
private volatile bool _loggedConnectFailure;
|
|
|
|
/// <summary>
|
|
/// Incremented per connection attempt. A reader from a previous connection must not be
|
|
/// able to mark a newer one dead — <c>reader.Join</c> can time out, and the stale
|
|
/// thread's finally block would otherwise tear down the connection that replaced it.
|
|
/// </summary>
|
|
private int _epoch;
|
|
|
|
private int _depth;
|
|
private long _sent, _dropped, _received, _connects, _writeErrors;
|
|
|
|
/// <summary>
|
|
/// Identifies this run of the <b>server process</b>. It is how the website tells a game
|
|
/// restart apart from a sidecar reconnect — the same link comes back either way, and
|
|
/// only one of the two means the world it is describing started over.
|
|
///
|
|
/// <para>
|
|
/// It must therefore be stable across a <b>plugin</b> reload, which a fresh
|
|
/// <c>Guid</c> at <c>Init</c> is not: <c>oxide.reload</c> would announce a brand new
|
|
/// boot to a website that is about to reconcile its whole ledger on the strength of it,
|
|
/// while the world it describes never moved. The process's own start time is the only
|
|
/// value here that changes exactly when the thing it names changes.
|
|
/// </para>
|
|
/// </summary>
|
|
private string _bootId;
|
|
|
|
/// <summary>
|
|
/// The current wipe, as <c>w-yyyyMMddTHHmmssZ</c>, derived from the save's creation
|
|
/// time. Null until the server has saved at least once — a server with no save has no
|
|
/// wipe, and PROTOCOL.md §8.2 says that is reported by the field being **absent**
|
|
/// rather than by an empty string that will sort beside every other empty string for
|
|
/// ever.
|
|
///
|
|
/// <para>
|
|
/// Cached rather than read per frame, and re-read on <c>OnNewSave</c>, which is the
|
|
/// only moment it can change. A wipe is a fact about the world; nothing about a
|
|
/// reconnect, a plugin reload or a restart moves it.
|
|
/// </para>
|
|
/// </summary>
|
|
private string _wipeId;
|
|
|
|
/// <summary>
|
|
/// Per-player counters waiting to be flushed as a <c>player.tally</c> frame. Main
|
|
/// thread only — every writer is a game hook and the flush runs on a timer, which is
|
|
/// also the main thread, so this needs no lock and must never acquire one.
|
|
/// </summary>
|
|
private readonly Dictionary<ulong, Tally> _tallies = new Dictionary<ulong, Tally>();
|
|
|
|
/// <summary>
|
|
/// When each connected player connected, in epoch milliseconds, so a disconnect can
|
|
/// report a session length. Main thread only.
|
|
///
|
|
/// <para>
|
|
/// A player already on the server when this plugin loaded has no entry, and
|
|
/// <c>sessionSec</c> is then <b>absent</b> from their disconnect rather than zero: an
|
|
/// unknown session is not a zero-length one, and summing the zeros would under-report
|
|
/// playtime quietly and for ever.
|
|
/// </para>
|
|
/// </summary>
|
|
private readonly Dictionary<ulong, long> _connectedAt = new Dictionary<ulong, long>();
|
|
|
|
/// <summary>
|
|
/// Every hook name this plugin declares, and whether it has fired at least once.
|
|
///
|
|
/// <para>
|
|
/// Both frameworks bind hooks by <b>name and arity, through reflection, with no
|
|
/// compile-time check</b>: a hook Facepunch renames on a Thursday simply stops being
|
|
/// called, and a hook one framework's catalogue does not list may or may not exist.
|
|
/// Neither failure logs anything anywhere. <c>rg.hooks</c> reads this dictionary, which
|
|
/// makes the plugin its own answer to both questions on whichever framework it is
|
|
/// actually running under — and a cheaper answer than either catalogue.
|
|
/// </para>
|
|
/// </summary>
|
|
private readonly Dictionary<string, long> _hookCounts = new Dictionary<string, long>();
|
|
|
|
/// <summary>
|
|
/// The repeating boards/tally timer.
|
|
///
|
|
/// <para>
|
|
/// Typed as the plugin-facing <c>Oxide.Plugins.Timer</c>, which is what
|
|
/// <c>timer.Every</c> hands back — <b>not</b> <c>Oxide.Core.Libraries.Timer.TimerInstance</c>,
|
|
/// which is the library's own type and does not convert. Both frameworks would destroy
|
|
/// this timer for us when the plugin unloads; holding the handle is what lets
|
|
/// <c>Unload</c> stop it before the link thread is joined rather than after.
|
|
/// </para>
|
|
/// </summary>
|
|
private Timer _cadence;
|
|
|
|
// ---- lifecycle ----
|
|
|
|
/// <summary>
|
|
/// Every hook this plugin declares. Seeded into <c>_hookCounts</c> at zero so that
|
|
/// <c>rg.hooks</c> can report a hook that has <b>never</b> fired — which is the whole
|
|
/// question, and which a dictionary that only learns names as they arrive can never
|
|
/// answer.
|
|
/// </summary>
|
|
private static readonly string[] ExpectedHooks =
|
|
{
|
|
"OnServerInitialized", "OnServerShutdown", "OnNewSave",
|
|
"OnPlayerConnected", "OnPlayerDisconnected", "OnPlayerRespawned",
|
|
"OnPlayerDeath", "OnPlayerChat", "OnDispenserGather", "OnEntityDeath",
|
|
"OnPlayerReported", "OnUserBanned", "OnUserUnbanned",
|
|
"CanUserLogin", "OnUserApproved",
|
|
"OnUserPermissionGranted", "OnUserPermissionRevoked",
|
|
"OnUserGroupAdded", "OnUserGroupRemoved",
|
|
"OnGroupPermissionGranted", "OnGroupPermissionRevoked"
|
|
};
|
|
|
|
private void Init()
|
|
{
|
|
foreach (string hook in ExpectedHooks)
|
|
_hookCounts[hook] = 0L;
|
|
|
|
_bootId = ResolveBootId();
|
|
_wipeId = ResolveWipeId();
|
|
_running = true;
|
|
|
|
// Expired link codes are dropped on a timer, not on lookup: an unconfirmed code is
|
|
// never looked up, so nothing else would ever remove one (protocol 3).
|
|
timer.Every(60f, PurgeExpiredCodes);
|
|
|
|
_link = new Thread(LinkLoop)
|
|
{
|
|
Name = "RunicGateway Link",
|
|
IsBackground = true
|
|
};
|
|
|
|
_link.Start();
|
|
}
|
|
|
|
private void Unload()
|
|
{
|
|
// Attempted before `_running` goes false, because `Emit` refuses to enqueue once it
|
|
// does — but be honest about what this achieves: the writer stops on the same flag a
|
|
// few lines below and the queue is cleared after the join, so a frame enqueued here
|
|
// only survives if the writer happens to be mid-drain. It usually is not.
|
|
//
|
|
// That is left as it is rather than fixed with a drain-before-exit, because draining
|
|
// means waiting on a socket from `Unload`, which runs on the MAIN THREAD — the exact
|
|
// stall phase 1 removed when a 1.9-second reload showed up as
|
|
// `Calling 'Unload' took 1918ms`. A wedged peer would make it worse than the bug.
|
|
//
|
|
// What is actually lost is under a minute of one player's gathering, and only on a
|
|
// manual plugin reload: a real shutdown flushes at `OnServerShutdown`, and a player
|
|
// leaving flushes at their disconnect.
|
|
try
|
|
{
|
|
FlushTallies();
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
Puts($"could not flush tallies on unload: {ex.Message}");
|
|
}
|
|
|
|
if (_cadence != null)
|
|
{
|
|
_cadence.Destroy();
|
|
_cadence = null;
|
|
}
|
|
|
|
// A half-drained permission sync is abandoned rather than finished, and the site is
|
|
// told nothing — which is right: the report would describe a set that was only partly
|
|
// applied, and the next sync re-derives the whole answer from the store anyway. What
|
|
// must not happen is the timer outliving the plugin and writing during a reload.
|
|
_permJob = null;
|
|
StopPermTimer();
|
|
|
|
_running = false;
|
|
_stopping.Set();
|
|
Wake.Set();
|
|
|
|
Thread t = _link;
|
|
|
|
if (t != null && !t.Join(TimeSpan.FromSeconds(2.0)))
|
|
Puts("link thread did not stop cleanly");
|
|
|
|
_link = null;
|
|
_connected = false;
|
|
_tallies.Clear();
|
|
|
|
// A reload leaves the static queue holding lines addressed to a connection that no
|
|
// longer exists. Clearing it is the difference between a reload and a slow leak.
|
|
string discard;
|
|
while (Outbound.TryDequeue(out discard))
|
|
{
|
|
}
|
|
|
|
Interlocked.Exchange(ref _depth, 0);
|
|
}
|
|
|
|
// ---- configuration ----
|
|
|
|
private class PluginConfig
|
|
{
|
|
/// <summary>Where the sidecar listens. Loopback is the trust boundary; see the class docs.</summary>
|
|
public string Host = "127.0.0.1";
|
|
|
|
public int Port = 7799;
|
|
|
|
/// <summary>
|
|
/// How many outbound lines may wait for a wedged sidecar before the oldest is
|
|
/// dropped. Sized in lines rather than bytes because that is what is cheap to
|
|
/// count; the inbound cap is what bounds a single line's size.
|
|
/// </summary>
|
|
public int QueueCap = 5000;
|
|
|
|
/// <summary>
|
|
/// This server's stable identity across wipes and restarts, as the website knows
|
|
/// it. It is <b>not</b> derived from the hostname: an operator renames a server for
|
|
/// a season and the site must not lose its history for it.
|
|
/// </summary>
|
|
public string ServerId = "main";
|
|
}
|
|
|
|
protected override void LoadDefaultConfig()
|
|
{
|
|
Config.WriteObject(new PluginConfig(), true);
|
|
}
|
|
|
|
private void LoadConfigValues()
|
|
{
|
|
_config = Config.ReadObject<PluginConfig>() ?? new PluginConfig();
|
|
|
|
if (string.IsNullOrEmpty(_config.Host))
|
|
_config.Host = "127.0.0.1";
|
|
|
|
if (_config.Port <= 0 || _config.Port > 65535)
|
|
_config.Port = 7799;
|
|
|
|
// A cap of zero would drop every line on the way in and report a healthy link doing
|
|
// nothing, which is the worst failure available here.
|
|
if (_config.QueueCap < 16)
|
|
_config.QueueCap = 16;
|
|
|
|
if (string.IsNullOrEmpty(_config.ServerId))
|
|
_config.ServerId = "main";
|
|
}
|
|
|
|
private void Loaded()
|
|
{
|
|
LoadConfigValues();
|
|
Puts($"protocol {ProtocolVersion}, serverId '{_config.ServerId}', sidecar {_config.Host}:{_config.Port}");
|
|
}
|
|
|
|
// ---- the outbound queue ----
|
|
|
|
/// <summary>
|
|
/// Main thread. Non-blocking. <paramref name="line"/> must already be a complete JSON
|
|
/// object with no embedded newline; the newline is appended by the writer as the frame
|
|
/// delimiter.
|
|
/// </summary>
|
|
private void Emit(string line)
|
|
{
|
|
if (line == null || !_running)
|
|
return;
|
|
|
|
int cap = _config != null ? _config.QueueCap : 5000;
|
|
|
|
// Drop-oldest. Bound first, then enqueue, so the queue can transiently sit one over the
|
|
// cap but never grows without limit.
|
|
while (Volatile.Read(ref _depth) >= cap)
|
|
{
|
|
string discard;
|
|
|
|
if (!Outbound.TryDequeue(out discard))
|
|
break;
|
|
|
|
Interlocked.Decrement(ref _depth);
|
|
Interlocked.Increment(ref _dropped);
|
|
}
|
|
|
|
Outbound.Enqueue(line);
|
|
Interlocked.Increment(ref _depth);
|
|
|
|
Wake.Set();
|
|
}
|
|
|
|
private void EmitFrame(Dictionary<string, object> frame)
|
|
{
|
|
try
|
|
{
|
|
Emit(JsonConvert.SerializeObject(frame, Formatting.None));
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
// A frame that cannot be serialized must not take the caller down with it — the
|
|
// caller is usually a game hook.
|
|
Puts($"could not serialize frame: {ex.Message}");
|
|
}
|
|
}
|
|
|
|
// ---- the link ----
|
|
|
|
private void LinkLoop()
|
|
{
|
|
int backoffMs = 500;
|
|
|
|
while (_running)
|
|
{
|
|
TcpClient client = null;
|
|
Thread reader = null;
|
|
|
|
int epoch = Interlocked.Increment(ref _epoch);
|
|
|
|
try
|
|
{
|
|
string host = _config != null ? _config.Host : "127.0.0.1";
|
|
int port = _config != null ? _config.Port : 7799;
|
|
|
|
client = new TcpClient();
|
|
client.NoDelay = true;
|
|
Connect(client, host, port);
|
|
|
|
NetworkStream stream = client.GetStream();
|
|
stream.WriteTimeout = 5000; // a wedged peer must surface as an error, not a hang
|
|
|
|
_dead = false;
|
|
_connected = true;
|
|
_loggedConnectFailure = false; // so a LATER outage is reported again
|
|
backoffMs = 500;
|
|
|
|
Interlocked.Increment(ref _connects);
|
|
Puts($"connected to {host}:{port}");
|
|
|
|
// Building the boards reads the world, so it must happen on the main thread.
|
|
// They are sent on every connect rather than once at startup: the sidecar
|
|
// restarts independently of the game, and anything it needs up front has to be
|
|
// re-sent per connection.
|
|
Interface.Oxide.NextTick(SendBoards);
|
|
|
|
NetworkStream localStream = stream;
|
|
reader = new Thread(() => ReadLoop(localStream, epoch))
|
|
{
|
|
Name = "RunicGateway Reader",
|
|
IsBackground = true
|
|
};
|
|
reader.Start();
|
|
|
|
WriteLoop(stream);
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
// Socket exception messages carry EMBEDDED newlines on Windows — the refusal
|
|
// text and the endpoint arrive on separate lines — so trimming the ends is not
|
|
// enough to keep the Oxide log one entry per line.
|
|
string reason = Flatten(ex.Message);
|
|
|
|
if (_connected)
|
|
{
|
|
Puts($"link error: {reason}");
|
|
}
|
|
else if (!_loggedConnectFailure)
|
|
{
|
|
// The FIRST failure to connect, and only the first: a sidecar that is not
|
|
// running yet is normal at boot and would otherwise spam the console every
|
|
// few seconds, but a wrong Host or Port is otherwise silent for ever.
|
|
_loggedConnectFailure = true;
|
|
Puts($"cannot reach the sidecar: {reason} — retrying quietly until it answers");
|
|
}
|
|
}
|
|
finally
|
|
{
|
|
// Captured before it is cleared: a connection that ended because the READER saw
|
|
// EOF leaves the writer to exit cleanly, so nothing throws and the catch above
|
|
// never runs. Without this the link going down is invisible in the game console,
|
|
// which is the one place an operator looks first.
|
|
bool wasConnected = _connected;
|
|
|
|
_connected = false;
|
|
_dead = true;
|
|
|
|
try
|
|
{
|
|
if (client != null)
|
|
client.Close();
|
|
}
|
|
catch
|
|
{
|
|
// Closing a socket that is already gone is not news.
|
|
}
|
|
|
|
if (reader != null)
|
|
reader.Join(TimeSpan.FromSeconds(1.0));
|
|
|
|
if (wasConnected)
|
|
Puts("sidecar link lost; reconnecting");
|
|
}
|
|
|
|
if (!_running)
|
|
break;
|
|
|
|
// Nothing is listening yet, or the sidecar restarted. Both are normal.
|
|
//
|
|
// `Wake`, not `Thread.Sleep`: Unload runs on the MAIN thread and joins this one, so
|
|
// an uninterruptible sleep here is a stall of up to MaxBackoffMs on every reload —
|
|
// Oxide reports it as `Calling 'Unload' took 1918ms`, and the server is frozen for
|
|
// that whole time. Unload sets `_running` false and signals `Wake`, so waiting on
|
|
// the handle turns that stall into an immediate return.
|
|
Wake.WaitOne(backoffMs);
|
|
backoffMs = Math.Min(backoffMs * 2, MaxBackoffMs);
|
|
}
|
|
|
|
_connected = false;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Connects with a bound, and with a way out.
|
|
///
|
|
/// <para>
|
|
/// The wait ends on whichever comes first: the connect completing, the plugin being
|
|
/// unloaded, or <see cref="ConnectTimeoutMs"/>. Without the middle one a reload against
|
|
/// an unreachable sidecar costs the game's main thread the whole join timeout —
|
|
/// observed as <c>hook 'Unload' took longer than 100ms [2002ms]</c> beside
|
|
/// <c>link thread did not stop cleanly</c>, which is phase 1's stall arriving by a
|
|
/// different road.
|
|
/// </para>
|
|
/// </summary>
|
|
private void Connect(TcpClient client, string host, int port)
|
|
{
|
|
IAsyncResult pending = client.BeginConnect(host, port, null, null);
|
|
|
|
int signalled = WaitHandle.WaitAny(
|
|
new WaitHandle[] { pending.AsyncWaitHandle, _stopping }, ConnectTimeoutMs);
|
|
|
|
if (signalled != 0)
|
|
{
|
|
// Closing the client is what abandons the in-flight connect; the socket would
|
|
// otherwise stay open behind an IAsyncResult nobody is going to end.
|
|
try
|
|
{
|
|
client.Close();
|
|
}
|
|
catch
|
|
{
|
|
}
|
|
|
|
throw new System.IO.IOException(
|
|
signalled == 1
|
|
? "unloading"
|
|
: $"connect to {host}:{port} timed out after {ConnectTimeoutMs}ms");
|
|
}
|
|
|
|
// Throws the real socket error — a refusal, an unknown host — which is what the caller
|
|
// logs. Only reached when the connect actually completed or failed on its own.
|
|
client.EndConnect(pending);
|
|
}
|
|
|
|
private void WriteLoop(NetworkStream stream)
|
|
{
|
|
while (_running && !_dead)
|
|
{
|
|
string line;
|
|
|
|
if (!Outbound.TryDequeue(out line))
|
|
{
|
|
Wake.WaitOne(250);
|
|
continue;
|
|
}
|
|
|
|
Interlocked.Decrement(ref _depth);
|
|
|
|
try
|
|
{
|
|
byte[] bytes = Encoding.UTF8.GetBytes(line + "\n");
|
|
stream.Write(bytes, 0, bytes.Length);
|
|
Interlocked.Increment(ref _sent);
|
|
}
|
|
catch (Exception)
|
|
{
|
|
// The record is already off the queue. Count it and let the outer loop
|
|
// reconnect; re-queueing risks an unbounded retry storm against a dead peer.
|
|
Interlocked.Increment(ref _writeErrors);
|
|
_dead = true;
|
|
throw;
|
|
}
|
|
}
|
|
}
|
|
|
|
private void ReadLoop(NetworkStream stream, int epoch)
|
|
{
|
|
byte[] buffer = new byte[8192];
|
|
StringBuilder line = new StringBuilder(512);
|
|
bool discarding = false;
|
|
|
|
try
|
|
{
|
|
while (_running && !_dead)
|
|
{
|
|
int read = stream.Read(buffer, 0, buffer.Length);
|
|
|
|
if (read <= 0)
|
|
break; // clean EOF: the sidecar closed. Normal.
|
|
|
|
for (int i = 0; i < read; i++)
|
|
{
|
|
char c = (char)buffer[i];
|
|
|
|
if (c == '\n')
|
|
{
|
|
if (discarding)
|
|
{
|
|
// The tail of a line already given up on. Swallowing it here is
|
|
// what stops the remainder being parsed as a line of its own.
|
|
discarding = false;
|
|
}
|
|
else
|
|
{
|
|
Dispatch(line.ToString());
|
|
}
|
|
|
|
line.Length = 0;
|
|
}
|
|
else if (c != '\r')
|
|
{
|
|
if (discarding)
|
|
continue;
|
|
|
|
line.Append(c);
|
|
|
|
if (line.Length > MaxInboundLineBytes)
|
|
{
|
|
Puts("inbound line over the cap; discarding to the next newline");
|
|
line.Length = 0;
|
|
discarding = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
catch (System.IO.IOException)
|
|
{
|
|
// Expected when the peer vanishes mid-read.
|
|
}
|
|
catch (ObjectDisposedException)
|
|
{
|
|
// Expected when Unload closes the socket under us.
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
Puts($"reader error: {ex.Message}");
|
|
}
|
|
finally
|
|
{
|
|
// Only tear down the connection this reader actually owned.
|
|
if (Volatile.Read(ref _epoch) == epoch)
|
|
{
|
|
_dead = true;
|
|
Wake.Set(); // let the writer notice and fall through to reconnect
|
|
}
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Reader thread. Marshals to the main thread; nothing that touches the game happens
|
|
/// here.
|
|
/// </summary>
|
|
private void Dispatch(string line)
|
|
{
|
|
if (string.IsNullOrEmpty(line))
|
|
return;
|
|
|
|
Interlocked.Increment(ref _received);
|
|
|
|
Interface.Oxide.NextTick(() =>
|
|
{
|
|
try
|
|
{
|
|
HandleCommand(line);
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
// A malformed command must never escape into a game code path.
|
|
Puts($"inbound handler threw: {ex}");
|
|
}
|
|
});
|
|
}
|
|
|
|
// ---- commands ----
|
|
|
|
/// <summary>Main thread. Every world read below is safe only because of that.</summary>
|
|
private void HandleCommand(string line)
|
|
{
|
|
Dictionary<string, object> cmd;
|
|
|
|
try
|
|
{
|
|
cmd = JsonConvert.DeserializeObject<Dictionary<string, object>>(line);
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
Puts($"unparseable command: {ex.Message}");
|
|
return;
|
|
}
|
|
|
|
if (cmd == null)
|
|
return;
|
|
|
|
object verbObj;
|
|
if (!cmd.TryGetValue("cmd", out verbObj) || verbObj == null)
|
|
return;
|
|
|
|
string verb = verbObj.ToString();
|
|
object reqIdObj;
|
|
string reqId = cmd.TryGetValue("reqId", out reqIdObj) && reqIdObj != null
|
|
? reqIdObj.ToString()
|
|
: null;
|
|
|
|
switch (verb)
|
|
{
|
|
case "ping":
|
|
// `control`: neither kept nor forwarded. It exists to move the sidecar's
|
|
// last-seen clock on a server nobody is playing on.
|
|
EmitFrame(Frame("pong", "control"));
|
|
break;
|
|
|
|
case "server.status":
|
|
Dictionary<string, object> status = Frame("server.status", "reply");
|
|
Merge(status, ServerBody());
|
|
|
|
// The correlation id is echoed verbatim, and only when one was supplied. A
|
|
// reply that invents one would be routed to nobody; a reply that omits one the
|
|
// caller sent would leave that caller waiting out its whole timeout.
|
|
if (reqId != null)
|
|
status["reqId"] = reqId;
|
|
|
|
EmitFrame(status);
|
|
break;
|
|
|
|
case "link.confirm":
|
|
// The first command the WEBSITE originates (protocol 3). Everything before it
|
|
// was the game talking, or the sidecar asking the game to say something it
|
|
// already knew.
|
|
HandleLinkConfirm(cmd, reqId);
|
|
break;
|
|
|
|
case "perm.catalogue":
|
|
// Protocol 4. What this server's loaded plugins have registered — the option
|
|
// source the website's authoring form offers, so a grant can only be written
|
|
// against a name that will actually resolve.
|
|
HandlePermCatalogue(reqId);
|
|
break;
|
|
|
|
case "perm.sync":
|
|
// The first command that WRITES to the game (protocol 4, R2). Everything
|
|
// before it read, or spent a code the game itself had minted.
|
|
HandlePermSync(cmd, reqId);
|
|
break;
|
|
|
|
default:
|
|
Puts($"unknown command '{verb}'");
|
|
break;
|
|
}
|
|
}
|
|
|
|
// ---- frames ----
|
|
|
|
/// <summary>
|
|
/// Builds the envelope every outbound frame shares (PROTOCOL.md §8.1). Nothing may
|
|
/// emit a frame it did not get from here.
|
|
///
|
|
/// <para>
|
|
/// <paramref name="type"/> is what the sidecar files on — <c>event</c> is appended to
|
|
/// history, <c>snapshot</c> replaces a board, <c>reply</c> is routed by
|
|
/// <c>reqId</c>, <c>control</c> is neither kept nor forwarded. It is the reason the
|
|
/// sidecar needs no arm per kind and the reason the catalogue can grow for free.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// <c>wipeId</c> is <b>omitted</b> when the server has never saved. A server with no
|
|
/// save has no wipe, and the difference between an absent field and an empty string is
|
|
/// the difference between a fact and a row that sorts beside every other empty string
|
|
/// for ever.
|
|
/// </para>
|
|
/// </summary>
|
|
private Dictionary<string, object> Frame(string kind, string type)
|
|
{
|
|
var frame = new Dictionary<string, object>
|
|
{
|
|
{ "kind", kind },
|
|
{ "type", type },
|
|
{ "t", NowMs() },
|
|
{ "serverId", _config != null ? _config.ServerId : "main" }
|
|
};
|
|
|
|
// Resolved lazily while it is still unknown, which covers the case above on a
|
|
// framework that never fires `OnServerInitialized` — and costs one static property
|
|
// read per frame only until the answer exists.
|
|
if (string.IsNullOrEmpty(_wipeId))
|
|
_wipeId = ResolveWipeId();
|
|
|
|
if (!string.IsNullOrEmpty(_wipeId))
|
|
frame["wipeId"] = _wipeId;
|
|
|
|
return frame;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Main thread. Both boards, in the order a reader wants them: what the server is,
|
|
/// then who is on it.
|
|
///
|
|
/// <para>
|
|
/// Sent on every connect and every cadence tick. A board is current state with exactly
|
|
/// one producer, re-sent rather than reconstructed — which is what makes "a restarted
|
|
/// sidecar is fully populated within one connection" true by construction instead of
|
|
/// by hoping nothing was in flight.
|
|
/// </para>
|
|
/// </summary>
|
|
private void SendBoards()
|
|
{
|
|
SendHello();
|
|
SendPlayersBoard();
|
|
}
|
|
|
|
/// <summary>Main thread.</summary>
|
|
private void SendHello()
|
|
{
|
|
Dictionary<string, object> hello = Frame("server.hello", "snapshot");
|
|
|
|
Merge(hello, ServerBody());
|
|
|
|
hello["protocol"] = ProtocolVersion;
|
|
hello["bootId"] = _bootId;
|
|
hello["plugin"] = Version.ToString();
|
|
|
|
EmitFrame(hello);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Main thread. Who is connected right now, as the board the far side reconciles
|
|
/// against — the connect and disconnect events carry the story in between.
|
|
/// </summary>
|
|
private void SendPlayersBoard()
|
|
{
|
|
var players = new List<object>();
|
|
|
|
if (BasePlayer.activePlayerList != null)
|
|
{
|
|
foreach (BasePlayer player in BasePlayer.activePlayerList)
|
|
{
|
|
if (player == null)
|
|
continue;
|
|
|
|
var row = new Dictionary<string, object>
|
|
{
|
|
{ "steamId", player.UserIDString },
|
|
{ "name", player.displayName },
|
|
{ "sleeping", player.IsSleeping() }
|
|
};
|
|
|
|
long since;
|
|
if (_connectedAt.TryGetValue(player.userID, out since))
|
|
row["connectedAt"] = since;
|
|
|
|
players.Add(row);
|
|
}
|
|
}
|
|
|
|
Dictionary<string, object> frame = Frame("players.online", "snapshot");
|
|
frame["count"] = players.Count;
|
|
frame["players"] = players;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Main thread. The shared body of <c>server.hello</c> and <c>server.status</c> — the
|
|
/// two frames differ in what wraps them, not in what they say about the server, and
|
|
/// letting them drift is how a site ends up showing two different player counts.
|
|
/// </summary>
|
|
private Dictionary<string, object> ServerBody()
|
|
{
|
|
var status = new Dictionary<string, object>
|
|
{
|
|
{ "hostname", ConVar.Server.hostname },
|
|
{ "description", ConVar.Server.description },
|
|
{ "level", ConVar.Server.level },
|
|
{ "seed", ConVar.Server.seed },
|
|
{ "worldSize", ConVar.Server.worldsize },
|
|
{ "maxPlayers", ConVar.Server.maxplayers },
|
|
{ "players", BasePlayer.activePlayerList != null ? BasePlayer.activePlayerList.Count : 0 },
|
|
{ "joining", ServerMgr.Instance != null ? ServerMgr.Instance.connectionQueue.Joining : 0 },
|
|
{ "queued", ServerMgr.Instance != null ? ServerMgr.Instance.connectionQueue.Queued : 0 },
|
|
{ "uptimeSec", (int)UnityEngine.Time.realtimeSinceStartup }
|
|
};
|
|
|
|
// The instant `wipeId` is derived from, kept in its own right: the id is for grouping
|
|
// rows and this is for showing a date, and deriving one back from the other is the
|
|
// kind of parsing nobody should have to do twice.
|
|
try
|
|
{
|
|
status["saveCreatedAt"] = SaveRestore.SaveCreatedTime.ToUniversalTime()
|
|
.ToString("yyyy-MM-ddTHH:mm:ssZ", System.Globalization.CultureInfo.InvariantCulture);
|
|
}
|
|
catch (Exception)
|
|
{
|
|
// A server that has never saved has no created time. Absent, not zero.
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Puts an IP address onto a frame, or leaves it off.
|
|
///
|
|
/// <para>
|
|
/// The game does not always have one. A console <c>banid</c> against an offline id
|
|
/// produces the literal string <b><c>"0"</c></b> rather than an address or a null —
|
|
/// observed on a live server, not guessed — and a database column full of "0" is worse
|
|
/// than one full of nulls, because "0" survives every is-it-missing test a reader
|
|
/// writes and then fails whatever parses it.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// Absent, not zero. The same rule `saveCreatedAt` and `sessionSec` follow.
|
|
/// </para>
|
|
/// </summary>
|
|
private static void PutAddress(Dictionary<string, object> frame, string ip)
|
|
{
|
|
if (string.IsNullOrEmpty(ip) || ip == "0" || ip == "0.0.0.0")
|
|
return;
|
|
|
|
frame["ip"] = ip;
|
|
}
|
|
|
|
private static void Merge(Dictionary<string, object> into, Dictionary<string, object> from)
|
|
{
|
|
foreach (KeyValuePair<string, object> pair in from)
|
|
into[pair.Key] = pair.Value;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Describes who or what killed somebody, onto <paramref name="frame"/>.
|
|
///
|
|
/// <para>
|
|
/// Four outcomes, and collapsing any two of them loses something a killfeed needs:
|
|
/// <c>player</c> (a real kill, and the only one that belongs on a leaderboard),
|
|
/// <c>self</c>, <c>npc</c> and <c>environment</c> — falling, drowning, a helicopter's
|
|
/// fire, or the world simply deciding. `HitInfo` may legitimately be null, which is
|
|
/// the environment case and not an error.
|
|
/// </para>
|
|
/// </summary>
|
|
private static void DescribeAttacker(
|
|
Dictionary<string, object> frame, BasePlayer victim, HitInfo info)
|
|
{
|
|
BaseEntity initiator = info != null ? info.Initiator : null;
|
|
BasePlayer attacker = info != null ? info.InitiatorPlayer : null;
|
|
|
|
string weapon = null;
|
|
|
|
if (info != null)
|
|
{
|
|
if (info.WeaponPrefab != null)
|
|
weapon = info.WeaponPrefab.ShortPrefabName;
|
|
else if (info.Weapon != null)
|
|
weapon = info.Weapon.ShortPrefabName;
|
|
}
|
|
|
|
if (weapon != null)
|
|
frame["weapon"] = weapon;
|
|
|
|
if (initiator == null)
|
|
{
|
|
frame["attackerType"] = "environment";
|
|
return;
|
|
}
|
|
|
|
if (attacker != null && attacker.userID == victim.userID)
|
|
{
|
|
frame["attackerType"] = "self";
|
|
return;
|
|
}
|
|
|
|
if (attacker != null && !attacker.IsNpc)
|
|
{
|
|
frame["attackerType"] = "player";
|
|
frame["attackerId"] = attacker.UserIDString;
|
|
frame["attackerName"] = attacker.displayName;
|
|
}
|
|
else
|
|
{
|
|
frame["attackerType"] = "npc";
|
|
frame["attackerName"] = initiator.ShortPrefabName;
|
|
}
|
|
|
|
float distance = info != null ? info.ProjectileDistance : 0f;
|
|
|
|
if (distance <= 0f && initiator.transform != null && victim.transform != null)
|
|
distance = UnityEngine.Vector3.Distance(
|
|
initiator.transform.position, victim.transform.position);
|
|
|
|
if (distance > 0f)
|
|
frame["distance"] = (float)Math.Round(distance, 1);
|
|
}
|
|
|
|
// ---- the map grid ----
|
|
|
|
/// <summary>
|
|
/// An entity's position as Rust's own map reference (<c>H7</c>), or null if it has no
|
|
/// position to read. A reference, never a coordinate: a grid cell is 146.3 units
|
|
/// across, which is precise enough to say where a fight happened and far too coarse to
|
|
/// walk to somebody's door with.
|
|
/// </summary>
|
|
private static string GridOf(BaseEntity entity)
|
|
{
|
|
if (entity == null || entity.transform == null)
|
|
return null;
|
|
|
|
return Grid(entity.transform.position);
|
|
}
|
|
|
|
private static string Grid(UnityEngine.Vector3 position)
|
|
{
|
|
float size = ConVar.Server.worldsize;
|
|
|
|
if (size <= 0f)
|
|
return null;
|
|
|
|
float half = size / 2f;
|
|
int columns = (int)Math.Ceiling(size / GridCellSize);
|
|
|
|
int x = (int)((position.x + half) / GridCellSize);
|
|
|
|
// Rows are numbered from the NORTH edge downwards, which is why this is a subtraction
|
|
// and not the same expression as the column. Getting it wrong mirrors every grid on
|
|
// the map and is invisible without a second opinion.
|
|
int z = (int)((half - position.z) / GridCellSize);
|
|
|
|
x = Clamp(x, 0, columns - 1);
|
|
z = Clamp(z, 0, columns - 1);
|
|
|
|
return Column(x) + z.ToString(System.Globalization.CultureInfo.InvariantCulture);
|
|
}
|
|
|
|
private static int Clamp(int value, int low, int high)
|
|
{
|
|
if (high < low)
|
|
return low;
|
|
|
|
return value < low ? low : (value > high ? high : value);
|
|
}
|
|
|
|
/// <summary>A column index as Rust spells it: 0 is A, 25 is Z, 26 is AA.</summary>
|
|
private static string Column(int index)
|
|
{
|
|
string name = string.Empty;
|
|
int n = index + 1;
|
|
|
|
while (n > 0)
|
|
{
|
|
int remainder = (n - 1) % 26;
|
|
name = (char)('A' + remainder) + name;
|
|
n = (n - 1) / 26;
|
|
}
|
|
|
|
return name;
|
|
}
|
|
|
|
// ---- identity of the world ----
|
|
|
|
/// <summary>
|
|
/// The current wipe, as <c>w-yyyyMMddTHHmmssZ</c>, or null if this server has never
|
|
/// saved.
|
|
///
|
|
/// <para>
|
|
/// Protocol 1 left this to the website. Protocol 2 derives it here because by now
|
|
/// three components store rows that need it, and the plugin is the only one of the
|
|
/// three that can read the value — the other two would be re-deriving something they
|
|
/// were already told, and two derivations of one fact eventually disagree about a
|
|
/// boundary.
|
|
/// </para>
|
|
/// </summary>
|
|
private static string ResolveWipeId()
|
|
{
|
|
try
|
|
{
|
|
DateTime created = SaveRestore.SaveCreatedTime.ToUniversalTime();
|
|
|
|
// A server that has never saved reports the bottom of the calendar. That is not a
|
|
// wipe in 0001, it is the absence of one.
|
|
if (created.Year < 2000)
|
|
return null;
|
|
|
|
return "w-" + created.ToString(
|
|
"yyyyMMdd'T'HHmmss'Z'", System.Globalization.CultureInfo.InvariantCulture);
|
|
}
|
|
catch (Exception)
|
|
{
|
|
return null;
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Records that a hook fired. See <c>_hookCounts</c> for why this exists at all.
|
|
/// </summary>
|
|
private void MarkHook(string name)
|
|
{
|
|
long count;
|
|
_hookCounts.TryGetValue(name, out count);
|
|
_hookCounts[name] = count + 1;
|
|
}
|
|
|
|
// ---- the read path ----
|
|
//
|
|
// Every hook below obeys three rules, and each of them has a failure behind it.
|
|
//
|
|
// 1. **A read-path hook never vetoes.** Four of these are documented by uMod as
|
|
// "returning a non-null value overrides default behavior" — a bridge that returned
|
|
// something by accident would cancel a death, swallow a player's wood or refuse a
|
|
// login, on somebody's production server at 3am. So they are declared `void`: both
|
|
// frameworks take the method's return value, and a void method has none. The rule is
|
|
// enforced by the signature rather than by remembering to write `return null`, which
|
|
// is the only version of it that survives a year of edits.
|
|
//
|
|
// 2. **A hook that can fire more than once a second per player is a counter, not an
|
|
// event.** `OnDispenserGather` fires on every swing at a tree. It accumulates into
|
|
// `_tallies` and leaves as one `player.tally` frame a minute (PROTOCOL.md §8.6).
|
|
//
|
|
// 3. **Every hook marks itself.** `MarkHook` is what makes `rg.hooks` able to say which
|
|
// of these actually fire on the framework this server is running — the standing
|
|
// answer both to Facepunch renaming a hook and to a framework catalogue that does not
|
|
// list one.
|
|
//
|
|
// All of it is the main thread, which is what makes reading `BasePlayer`, `ConVar` and a
|
|
// transform legal here and illegal in the reader thread.
|
|
|
|
private void OnServerInitialized(bool initial)
|
|
{
|
|
MarkHook("OnServerInitialized");
|
|
|
|
// Re-read the wipe, and this is not belt-and-braces: at a real server
|
|
// start `Init` runs BEFORE the save is loaded, so `SaveCreatedTime` is not yet
|
|
// meaningful and the id resolved there is null for the entire session. This hook is
|
|
// the first moment the world is actually there.
|
|
//
|
|
// It was invisible on a development rig for the reason such things usually are: a
|
|
// hot-reloaded plugin reads an already-loaded world and gets the right answer every
|
|
// time. It took a server that BOOTED with the plugin installed — which is every real
|
|
// one — to show `wipeId=none` on a server with a save sitting on disk.
|
|
_wipeId = ResolveWipeId();
|
|
|
|
// Started here rather than in Init: this hook fires both at the end of server startup
|
|
// and immediately on a hotload, so it is the one place that runs exactly once per
|
|
// load with the world available either way.
|
|
if (_cadence == null)
|
|
_cadence = timer.Every(BoardCadenceSeconds, Cadence);
|
|
|
|
Dictionary<string, object> frame = Frame("server.initialized", "event");
|
|
frame["initial"] = initial;
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnServerShutdown()
|
|
{
|
|
MarkHook("OnServerShutdown");
|
|
FlushTallies();
|
|
EmitFrame(Frame("server.shutdown", "event"));
|
|
}
|
|
|
|
/// <summary>
|
|
/// A new save is the only moment a wipe id can change, so it is the only place that
|
|
/// re-reads it. The frame carries both ids because "which wipe did this replace" is
|
|
/// the question a site asks when it draws the boundary, and it is unanswerable
|
|
/// afterwards.
|
|
/// </summary>
|
|
private void OnNewSave(string filename)
|
|
{
|
|
MarkHook("OnNewSave");
|
|
|
|
string previous = _wipeId;
|
|
_wipeId = ResolveWipeId();
|
|
|
|
// Built after the re-read, so `wipeId` on the envelope is the NEW wipe: this frame
|
|
// belongs to the world it opens, not to the one it closes.
|
|
Dictionary<string, object> frame = Frame("server.wipe", "event");
|
|
|
|
if (!string.IsNullOrEmpty(previous))
|
|
frame["previousWipeId"] = previous;
|
|
|
|
EmitFrame(frame);
|
|
|
|
// Nothing from the old world is still true. The tallies are the only such state this
|
|
// plugin holds; everything else lives on the far side of the link.
|
|
_tallies.Clear();
|
|
_connectedAt.Clear();
|
|
|
|
Puts($"new save; wipe is now {(_wipeId ?? "unknown")}");
|
|
}
|
|
|
|
private void OnPlayerConnected(BasePlayer player)
|
|
{
|
|
MarkHook("OnPlayerConnected");
|
|
|
|
if (player == null)
|
|
return;
|
|
|
|
_connectedAt[player.userID] = NowMs();
|
|
|
|
Dictionary<string, object> frame = Frame("player.connected", "event");
|
|
frame["steamId"] = player.UserIDString;
|
|
frame["name"] = player.displayName;
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnPlayerDisconnected(BasePlayer player, string reason)
|
|
{
|
|
MarkHook("OnPlayerDisconnected");
|
|
|
|
if (player == null)
|
|
return;
|
|
|
|
// Before the frame: a session's last minute of gathering is worth more than the order
|
|
// of two lines, and after this point there is nothing to attribute it to.
|
|
FlushTally(player.userID);
|
|
|
|
Dictionary<string, object> frame = Frame("player.disconnected", "event");
|
|
frame["steamId"] = player.UserIDString;
|
|
frame["name"] = player.displayName;
|
|
|
|
if (!string.IsNullOrEmpty(reason))
|
|
frame["reason"] = Flatten(reason);
|
|
|
|
long since;
|
|
if (_connectedAt.TryGetValue(player.userID, out since))
|
|
{
|
|
frame["sessionSec"] = (int)((NowMs() - since) / 1000L);
|
|
_connectedAt.Remove(player.userID);
|
|
}
|
|
|
|
// A player who was already on when this plugin loaded has no recorded connect, so
|
|
// `sessionSec` is ABSENT rather than zero: a session of unknown length is not a
|
|
// session of no length, and summing zeros silently under-reports playtime for ever.
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnPlayerRespawned(BasePlayer player)
|
|
{
|
|
MarkHook("OnPlayerRespawned");
|
|
|
|
if (player == null)
|
|
return;
|
|
|
|
Dictionary<string, object> frame = Frame("player.respawned", "event");
|
|
frame["steamId"] = player.UserIDString;
|
|
frame["name"] = player.displayName;
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Void, and deliberately: uMod documents a non-null return here as overriding the
|
|
/// death itself. See rule 1 at the top of this region.
|
|
/// </summary>
|
|
private void OnPlayerDeath(BasePlayer player, HitInfo info)
|
|
{
|
|
MarkHook("OnPlayerDeath");
|
|
|
|
if (player == null || player.IsNpc)
|
|
return;
|
|
|
|
Dictionary<string, object> frame = Frame("player.death", "event");
|
|
frame["steamId"] = player.UserIDString;
|
|
frame["name"] = player.displayName;
|
|
frame["sleeping"] = player.IsSleeping();
|
|
|
|
string grid = GridOf(player);
|
|
if (grid != null)
|
|
frame["grid"] = grid;
|
|
|
|
DescribeAttacker(frame, player, info);
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Void, and deliberately: a non-null return here overrides chat delivery — the bridge
|
|
/// would be silently eating messages.
|
|
/// </summary>
|
|
private void OnPlayerChat(BasePlayer player, string message, ConVar.Chat.ChatChannel channel)
|
|
{
|
|
MarkHook("OnPlayerChat");
|
|
|
|
if (player == null || string.IsNullOrEmpty(message))
|
|
return;
|
|
|
|
Dictionary<string, object> frame = Frame("player.chat", "event");
|
|
frame["steamId"] = player.UserIDString;
|
|
frame["name"] = player.displayName;
|
|
frame["channel"] = channel.ToString();
|
|
|
|
// Flattened for the same reason every other string on this wire is: one frame is one
|
|
// line, and a chat message is the one field a player controls the bytes of.
|
|
frame["message"] = Flatten(message);
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Void, and deliberately: a non-null return here replaces what the player receives.
|
|
/// This hook is a counter — see rule 2.
|
|
/// </summary>
|
|
private void OnDispenserGather(ResourceDispenser dispenser, BasePlayer player, Item item)
|
|
{
|
|
MarkHook("OnDispenserGather");
|
|
|
|
if (player == null || player.IsNpc || item == null || item.info == null)
|
|
return;
|
|
|
|
TallyFor(player).Gather(item.info.shortname, item.amount);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Everything that dies which is not a player death: NPC kills, which are a counter,
|
|
/// and raided structures, which are an event with a `staff` class because a
|
|
/// structure's grid is where somebody lives.
|
|
///
|
|
/// <para>
|
|
/// The type checks are ordered cheapest-first and return early, because this hook
|
|
/// fires for every tree, barrel and animal on the map.
|
|
/// </para>
|
|
/// </summary>
|
|
private void OnEntityDeath(BaseCombatEntity entity, HitInfo info)
|
|
{
|
|
MarkHook("OnEntityDeath");
|
|
|
|
if (entity == null)
|
|
return;
|
|
|
|
BasePlayer attacker = info != null ? info.InitiatorPlayer : null;
|
|
bool byPlayer = attacker != null && !attacker.IsNpc;
|
|
|
|
// Nothing below is interesting unless a real player did it. That single test discards
|
|
// decay, despawns and the world killing itself, which is the bulk of this hook.
|
|
if (!byPlayer)
|
|
return;
|
|
|
|
BasePlayer victimPlayer = entity as BasePlayer;
|
|
|
|
if (victimPlayer != null)
|
|
{
|
|
// A real player's death is OnPlayerDeath's, not ours — emitting here too would
|
|
// double every kill in the feed. Only the NPC kind is ours to count.
|
|
if (victimPlayer.IsNpc)
|
|
TallyFor(attacker).NpcKills++;
|
|
|
|
return;
|
|
}
|
|
|
|
if (entity is BaseNpc)
|
|
{
|
|
TallyFor(attacker).NpcKills++;
|
|
return;
|
|
}
|
|
|
|
BuildingBlock block = entity as BuildingBlock;
|
|
|
|
if (block == null || block.OwnerID == 0UL)
|
|
return;
|
|
|
|
TallyFor(attacker).Structures++;
|
|
|
|
Dictionary<string, object> frame = Frame("entity.destroyed", "event");
|
|
frame["ownerId"] = block.OwnerID.ToString();
|
|
frame["prefab"] = block.ShortPrefabName;
|
|
frame["attackerId"] = attacker.UserIDString;
|
|
frame["attackerName"] = attacker.displayName;
|
|
|
|
string grid = GridOf(block);
|
|
if (grid != null)
|
|
frame["grid"] = grid;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnPlayerReported(
|
|
BasePlayer reporter, string targetName, string targetId,
|
|
string subject, string message, string type)
|
|
{
|
|
MarkHook("OnPlayerReported");
|
|
|
|
Dictionary<string, object> frame = Frame("player.reported", "event");
|
|
|
|
if (reporter != null)
|
|
{
|
|
frame["reporterId"] = reporter.UserIDString;
|
|
frame["reporterName"] = reporter.displayName;
|
|
}
|
|
|
|
frame["targetId"] = targetId;
|
|
frame["targetName"] = targetName;
|
|
frame["subject"] = Flatten(subject);
|
|
frame["message"] = Flatten(message);
|
|
frame["reportType"] = type;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnUserBanned(string name, string id, string ipAddress, string reason)
|
|
{
|
|
MarkHook("OnUserBanned");
|
|
|
|
Dictionary<string, object> frame = Frame("player.banned", "event");
|
|
frame["steamId"] = id;
|
|
frame["name"] = name;
|
|
frame["reason"] = Flatten(reason);
|
|
PutAddress(frame, ipAddress);
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnUserUnbanned(string name, string id, string ipAddress)
|
|
{
|
|
MarkHook("OnUserUnbanned");
|
|
|
|
Dictionary<string, object> frame = Frame("player.unbanned", "event");
|
|
frame["steamId"] = id;
|
|
frame["name"] = name;
|
|
PutAddress(frame, ipAddress);
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Void, and this is the most important void in the file: a non-null return here
|
|
/// **kicks the player**, with the returned string as the reason. The bridge is here to
|
|
/// watch logins, never to decide one.
|
|
///
|
|
/// <para>
|
|
/// It fires on every connection attempt, approved or not — so it is an attempt, not a
|
|
/// denial. uMod publishes no <c>OnUserRejected</c>, and the only way to observe a
|
|
/// refusal from this hook is to be the thing refusing. A denial is therefore the
|
|
/// ABSENCE of a matching <c>player.approved</c>, decided by whoever reads the two
|
|
/// frames later; PROTOCOL.md §8.8 records that as a correction to the trigger
|
|
/// catalogue.
|
|
/// </para>
|
|
/// </summary>
|
|
private void CanUserLogin(string name, string id, string ipAddress)
|
|
{
|
|
MarkHook("CanUserLogin");
|
|
|
|
Dictionary<string, object> frame = Frame("player.login.attempt", "event");
|
|
frame["steamId"] = id;
|
|
frame["name"] = name;
|
|
PutAddress(frame, ipAddress);
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
private void OnUserApproved(string name, string id, string ipAddress)
|
|
{
|
|
MarkHook("OnUserApproved");
|
|
|
|
Dictionary<string, object> frame = Frame("player.approved", "event");
|
|
frame["steamId"] = id;
|
|
frame["name"] = name;
|
|
PutAddress(frame, ipAddress);
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
// ---- tallies ----
|
|
|
|
/// <summary>
|
|
/// One player's counters between two flushes. Main thread only; no lock, ever.
|
|
/// </summary>
|
|
private class Tally
|
|
{
|
|
public readonly Dictionary<string, int> Gathered = new Dictionary<string, int>();
|
|
public int NpcKills;
|
|
public int Structures;
|
|
public string Name;
|
|
|
|
public void Gather(string shortname, int amount)
|
|
{
|
|
if (string.IsNullOrEmpty(shortname) || amount <= 0)
|
|
return;
|
|
|
|
int have;
|
|
Gathered.TryGetValue(shortname, out have);
|
|
Gathered[shortname] = have + amount;
|
|
}
|
|
|
|
public bool IsEmpty()
|
|
{
|
|
return Gathered.Count == 0 && NpcKills == 0 && Structures == 0;
|
|
}
|
|
}
|
|
|
|
private Tally TallyFor(BasePlayer player)
|
|
{
|
|
Tally tally;
|
|
|
|
if (!_tallies.TryGetValue(player.userID, out tally))
|
|
{
|
|
tally = new Tally();
|
|
_tallies[player.userID] = tally;
|
|
}
|
|
|
|
tally.Name = player.displayName;
|
|
return tally;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Emits one <c>player.tally</c> per player with anything to report, and clears what it
|
|
/// emitted.
|
|
///
|
|
/// <para>
|
|
/// A tally frame is a <b>delta, not a running total</b>: it says what happened since
|
|
/// the last flush, so the far side sums rather than diffs. A missed frame then costs
|
|
/// one interval instead of corrupting the whole series, which is the failure mode a
|
|
/// running total has over a link that is allowed to drop frames (and this one is —
|
|
/// the outbound queue is drop-oldest by design).
|
|
/// </para>
|
|
/// </summary>
|
|
private void FlushTallies()
|
|
{
|
|
if (_tallies.Count == 0)
|
|
return;
|
|
|
|
// Materialised, because emitting mutates nothing here but clearing does, and iterating
|
|
// a dictionary while clearing it is a different bug on every runtime.
|
|
var ids = new List<ulong>(_tallies.Keys);
|
|
|
|
foreach (ulong id in ids)
|
|
FlushTally(id);
|
|
}
|
|
|
|
private void FlushTally(ulong userId)
|
|
{
|
|
Tally tally;
|
|
|
|
if (!_tallies.TryGetValue(userId, out tally))
|
|
return;
|
|
|
|
_tallies.Remove(userId);
|
|
|
|
if (tally.IsEmpty())
|
|
return;
|
|
|
|
Dictionary<string, object> frame = Frame("player.tally", "event");
|
|
frame["steamId"] = userId.ToString();
|
|
|
|
if (!string.IsNullOrEmpty(tally.Name))
|
|
frame["name"] = tally.Name;
|
|
|
|
if (tally.Gathered.Count > 0)
|
|
frame["gathered"] = tally.Gathered;
|
|
|
|
if (tally.NpcKills > 0)
|
|
frame["npcKills"] = tally.NpcKills;
|
|
|
|
if (tally.Structures > 0)
|
|
frame["structures"] = tally.Structures;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>
|
|
/// The one repeating job: re-send the boards and flush the tallies.
|
|
///
|
|
/// <para>
|
|
/// It does nothing at all while the link is down. That is not an optimisation — the
|
|
/// outbound queue is drop-oldest, so a disconnected server would spend its whole
|
|
/// outage pushing minute-old boards over the top of the events that actually matter.
|
|
/// A reconnect re-sends the boards anyway, which is the whole point of a board.
|
|
/// </para>
|
|
/// </summary>
|
|
private void Cadence()
|
|
{
|
|
if (!_connected)
|
|
return;
|
|
|
|
FlushTallies();
|
|
SendBoards();
|
|
}
|
|
|
|
/// <summary>
|
|
/// Collapses every run of whitespace <b>and control characters</b> to one space, so an
|
|
/// exception message stays one log entry.
|
|
///
|
|
/// <para>
|
|
/// The control half is not defensive padding. Mono's <c>SocketException.Message</c> on
|
|
/// Windows comes back from a fixed-size OS buffer and is <b>NUL-padded</b> — a couple
|
|
/// of hundred <c>\0</c> bytes sitting in the middle of the sentence. <c>\0</c> is not
|
|
/// whitespace, so <c>Trim</c> does not touch it and neither does a whitespace-only
|
|
/// collapse; the line looks like it contains a huge run of spaces and no amount of
|
|
/// trimming removes it.
|
|
/// </para>
|
|
/// </summary>
|
|
private static string Flatten(string text)
|
|
{
|
|
if (string.IsNullOrEmpty(text))
|
|
return string.Empty;
|
|
|
|
var sb = new StringBuilder(text.Length);
|
|
bool lastWasSpace = false;
|
|
|
|
foreach (char c in text)
|
|
{
|
|
if (char.IsWhiteSpace(c) || char.IsControl(c))
|
|
{
|
|
if (!lastWasSpace && sb.Length > 0)
|
|
sb.Append(' ');
|
|
|
|
lastWasSpace = true;
|
|
}
|
|
else
|
|
{
|
|
sb.Append(c);
|
|
lastWasSpace = false;
|
|
}
|
|
}
|
|
|
|
return sb.ToString().TrimEnd();
|
|
}
|
|
|
|
/// <summary>
|
|
/// The server process's start instant, as <c>boot-yyyyMMddTHHmmssZ</c>.
|
|
///
|
|
/// <para>
|
|
/// <c>Process.StartTime</c> is exact and identical on every read, which is what makes
|
|
/// the id survive a plugin reload. The fallback derives the same instant from Unity's
|
|
/// monotonic clock, and is only reached if the platform refuses the process query; it
|
|
/// is rounded to the second deliberately, because the subtraction jitters in the last
|
|
/// decimal and an id that changes on rounding is worse than one a second off.
|
|
/// </para>
|
|
/// </summary>
|
|
private static string ResolveBootId()
|
|
{
|
|
DateTime startUtc;
|
|
|
|
try
|
|
{
|
|
startUtc = System.Diagnostics.Process.GetCurrentProcess().StartTime.ToUniversalTime();
|
|
}
|
|
catch (Exception)
|
|
{
|
|
double up = UnityEngine.Time.realtimeSinceStartup;
|
|
startUtc = DateTime.UtcNow.AddSeconds(-Math.Round(up));
|
|
}
|
|
|
|
startUtc = new DateTime(
|
|
startUtc.Year, startUtc.Month, startUtc.Day,
|
|
startUtc.Hour, startUtc.Minute, startUtc.Second,
|
|
DateTimeKind.Utc);
|
|
|
|
return "boot-" + startUtc.ToString(
|
|
"yyyyMMdd'T'HHmmss'Z'", System.Globalization.CultureInfo.InvariantCulture);
|
|
}
|
|
|
|
private static long NowMs()
|
|
{
|
|
return (long)(DateTime.UtcNow - new DateTime(1970, 1, 1, 0, 0, 0, DateTimeKind.Utc))
|
|
.TotalMilliseconds;
|
|
}
|
|
|
|
// ---- account linking ----
|
|
|
|
/// <summary>
|
|
/// The alphabet a link code is drawn from.
|
|
///
|
|
/// <para>
|
|
/// <b>No O, 0, I or 1.</b> A player reads this code off their screen and types it into
|
|
/// a browser, often on a phone, sometimes reading it aloud to somebody else. Those four
|
|
/// glyphs are the ones that get transcribed wrong, and a code that fails for a reason
|
|
/// the player cannot see is worse than a longer one.
|
|
/// </para>
|
|
/// </summary>
|
|
private const string CodeAlphabet = "ABCDEFGHJKLMNPQRSTUVWXYZ23456789";
|
|
|
|
private const int CodeLength = 6;
|
|
|
|
/// <summary>
|
|
/// How long a code is good for. Short on purpose: R1 made this an authorization path,
|
|
/// not a convenience — the site becomes the author of who may do what in game, so a
|
|
/// code that lingers is a privilege waiting to be picked up.
|
|
/// </summary>
|
|
private static readonly TimeSpan CodeTtl = TimeSpan.FromMinutes(5);
|
|
|
|
/// <summary>How long a player must wait between <c>/link</c> requests.</summary>
|
|
private static readonly TimeSpan RequestCooldown = TimeSpan.FromSeconds(30);
|
|
|
|
/// <summary>
|
|
/// An unconfirmed code, and the Steam identity it will bind.
|
|
///
|
|
/// <para>
|
|
/// <b>These live in memory and nowhere else</b>, which is the same call the UO bridge
|
|
/// made and it is the right one for the same reason: the secret exists only on the
|
|
/// server that minted it, for five minutes, and nothing writes it to disk. A plugin
|
|
/// reload drops every pending code — and phase 7b's config editor reloads plugins
|
|
/// routinely — but the cost of that is a player typing <c>/link</c> again, which is
|
|
/// cheaper than a credential in a second process.
|
|
/// </para>
|
|
/// </summary>
|
|
private sealed class PendingLink
|
|
{
|
|
public string SteamId;
|
|
public string Name;
|
|
public DateTime Expires;
|
|
}
|
|
|
|
/// <summary>code -> pending link. Main thread only, like every world read here.</summary>
|
|
private readonly Dictionary<string, PendingLink> _codes =
|
|
new Dictionary<string, PendingLink>(StringComparer.OrdinalIgnoreCase);
|
|
|
|
/// <summary>steam id -> when they last asked, so <c>/link</c> cannot be spammed.</summary>
|
|
private readonly Dictionary<string, DateTime> _lastLinkRequest =
|
|
new Dictionary<string, DateTime>(StringComparer.Ordinal);
|
|
|
|
private readonly System.Random _codeRandom = new System.Random();
|
|
|
|
/// <summary>
|
|
/// <c>/link</c> — mint a one-time code for the caller and tell the sidecar about it.
|
|
///
|
|
/// <para>
|
|
/// <b>Chat, not console, and the reply is private.</b> Both frameworks consume a
|
|
/// <c>/</c> command rather than broadcasting it, and <c>SendReply</c> addresses one
|
|
/// player — so neither the request nor the code reaches anybody else's chat. That is
|
|
/// load-bearing rather than polish: a code read off a stream is a code somebody else
|
|
/// can spend.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// The plugin does <b>not</b> record who is linked. The website is the author of
|
|
/// record for that (R2), and a second copy here would be a second thing to keep in
|
|
/// step across a wipe. So there is no "you are already linked" check on this path —
|
|
/// the site refuses a Steam id it already holds, where it can say which account holds
|
|
/// it.
|
|
/// </para>
|
|
/// </summary>
|
|
[ChatCommand("link")]
|
|
private void CmdLinkChat(BasePlayer player, string command, string[] args)
|
|
{
|
|
if (player == null)
|
|
return;
|
|
|
|
string steamId = player.UserIDString;
|
|
if (string.IsNullOrEmpty(steamId))
|
|
return;
|
|
|
|
DateTime last;
|
|
if (_lastLinkRequest.TryGetValue(steamId, out last) && DateTime.UtcNow - last < RequestCooldown)
|
|
{
|
|
SendReply(player, "Please wait a moment before asking for another link code.");
|
|
return;
|
|
}
|
|
|
|
// One outstanding code per player: drop any earlier one so only the newest works.
|
|
// Without this a player who ran /link three times leaves three live codes behind.
|
|
DropCodesFor(steamId);
|
|
|
|
string code = MintCode();
|
|
_codes[code] = new PendingLink
|
|
{
|
|
SteamId = steamId,
|
|
Name = player.displayName,
|
|
Expires = DateTime.UtcNow + CodeTtl
|
|
};
|
|
_lastLinkRequest[steamId] = DateTime.UtcNow;
|
|
|
|
Dictionary<string, object> frame = Frame("account.link.requested", "event");
|
|
frame["steamId"] = steamId;
|
|
frame["name"] = player.displayName;
|
|
frame["ttlSec"] = (long)CodeTtl.TotalSeconds;
|
|
|
|
// **The code is NOT in this frame, and that is the point.** The event exists so an
|
|
// operator can see that linking is being used and so the site can rate-limit a player
|
|
// who is fishing; it is not how the code travels. The code travels through the player,
|
|
// which is what makes typing it proof that they are the one who asked.
|
|
EmitFrame(frame);
|
|
|
|
SendReply(player, "Your link code is <color=#e0b34d>" + code + "</color>.");
|
|
SendReply(
|
|
player,
|
|
"Enter it on the website within " + (int)CodeTtl.TotalMinutes +
|
|
" minutes to link this Steam account. It works once.");
|
|
}
|
|
|
|
/// <summary>
|
|
/// <c>/unlink</c> — ask the website to sever this Steam id's link.
|
|
///
|
|
/// <para>
|
|
/// <b>It emits rather than deletes</b>, because the plugin holds no link to delete. The
|
|
/// site owns the record and applies this like any other event off the feed.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// It exists because the site <b>refuses</b> to move a Steam id that another account
|
|
/// already holds. Without a way out, a player who linked the wrong account while logged
|
|
/// in as it would need staff; with this they walk to a computer they are already sitting
|
|
/// at. The authority is the Steam id itself — whoever is connected as it is who it is.
|
|
/// </para>
|
|
/// </summary>
|
|
[ChatCommand("unlink")]
|
|
private void CmdUnlinkChat(BasePlayer player, string command, string[] args)
|
|
{
|
|
if (player == null)
|
|
return;
|
|
|
|
string steamId = player.UserIDString;
|
|
if (string.IsNullOrEmpty(steamId))
|
|
return;
|
|
|
|
DropCodesFor(steamId);
|
|
|
|
Dictionary<string, object> frame = Frame("account.unlinked", "event");
|
|
frame["steamId"] = steamId;
|
|
frame["name"] = player.displayName;
|
|
frame["origin"] = "in-game";
|
|
EmitFrame(frame);
|
|
|
|
SendReply(
|
|
player,
|
|
"Asked the website to unlink this Steam account. If it was linked, it is not now.");
|
|
}
|
|
|
|
/// <summary>
|
|
/// Inbound <c>link.confirm</c> — the website has been given a code and wants to know
|
|
/// whose it is.
|
|
///
|
|
/// <para>
|
|
/// <b>This is the first command the website originates</b> in this bridge; everything
|
|
/// before protocol 3 was the game talking. The reply carries the Steam id and nothing
|
|
/// the caller did not already prove — there is no session here, no account creation,
|
|
/// and no write to the game.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// A code is consumed on the <b>first</b> lookup whether or not it turns out to be
|
|
/// expired, so a code cannot be probed twice. That is why the removal happens before
|
|
/// the expiry check rather than after it.
|
|
/// </para>
|
|
/// </summary>
|
|
private void HandleLinkConfirm(Dictionary<string, object> cmd, string reqId)
|
|
{
|
|
object codeObj;
|
|
string code = cmd.TryGetValue("code", out codeObj) && codeObj != null
|
|
? codeObj.ToString().Trim()
|
|
: null;
|
|
|
|
if (string.IsNullOrEmpty(code))
|
|
{
|
|
EmitLinkError(reqId, "malformed");
|
|
return;
|
|
}
|
|
|
|
PendingLink pending;
|
|
if (!_codes.TryGetValue(code, out pending))
|
|
{
|
|
EmitLinkError(reqId, "unknown");
|
|
return;
|
|
}
|
|
|
|
_codes.Remove(code);
|
|
|
|
if (DateTime.UtcNow > pending.Expires)
|
|
{
|
|
EmitLinkError(reqId, "expired");
|
|
return;
|
|
}
|
|
|
|
Dictionary<string, object> ok = Frame("link.ok", "reply");
|
|
ok["steamId"] = pending.SteamId;
|
|
ok["name"] = pending.Name;
|
|
if (reqId != null)
|
|
ok["reqId"] = reqId;
|
|
|
|
EmitFrame(ok);
|
|
}
|
|
|
|
/// <summary>
|
|
/// A refusal, with a reason the website can turn into a sentence.
|
|
///
|
|
/// <para>
|
|
/// <c>unknown</c> and <c>expired</c> are deliberately <b>separate</b> here and
|
|
/// deliberately <b>the same</b> to the player: an operator reading a log wants to know
|
|
/// whether codes are being guessed or merely going stale, and a stranger typing codes
|
|
/// must not learn which of the two they hit.
|
|
/// </para>
|
|
/// </summary>
|
|
private void EmitLinkError(string reqId, string reason)
|
|
{
|
|
Dictionary<string, object> frame = Frame("link.error", "reply");
|
|
frame["reason"] = reason;
|
|
if (reqId != null)
|
|
frame["reqId"] = reqId;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
/// <summary>Drops every outstanding code belonging to one Steam id.</summary>
|
|
private void DropCodesFor(string steamId)
|
|
{
|
|
List<string> stale = null;
|
|
|
|
foreach (KeyValuePair<string, PendingLink> entry in _codes)
|
|
{
|
|
if (entry.Value != null && entry.Value.SteamId == steamId)
|
|
{
|
|
if (stale == null)
|
|
stale = new List<string>();
|
|
|
|
stale.Add(entry.Key);
|
|
}
|
|
}
|
|
|
|
if (stale == null)
|
|
return;
|
|
|
|
foreach (string key in stale)
|
|
_codes.Remove(key);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Drops expired codes and stale cooldowns.
|
|
///
|
|
/// <para>
|
|
/// Run on a timer rather than only on lookup, because an unconfirmed code is never
|
|
/// looked up — a player who runs <c>/link</c> and wanders off leaves an entry that
|
|
/// nothing else would ever remove.
|
|
/// </para>
|
|
/// </summary>
|
|
private void PurgeExpiredCodes()
|
|
{
|
|
DateTime now = DateTime.UtcNow;
|
|
List<string> dead = null;
|
|
|
|
foreach (KeyValuePair<string, PendingLink> entry in _codes)
|
|
{
|
|
if (entry.Value == null || now > entry.Value.Expires)
|
|
{
|
|
if (dead == null)
|
|
dead = new List<string>();
|
|
|
|
dead.Add(entry.Key);
|
|
}
|
|
}
|
|
|
|
if (dead != null)
|
|
{
|
|
foreach (string key in dead)
|
|
_codes.Remove(key);
|
|
}
|
|
|
|
List<string> cold = null;
|
|
|
|
foreach (KeyValuePair<string, DateTime> entry in _lastLinkRequest)
|
|
{
|
|
if (now - entry.Value > RequestCooldown)
|
|
{
|
|
if (cold == null)
|
|
cold = new List<string>();
|
|
|
|
cold.Add(entry.Key);
|
|
}
|
|
}
|
|
|
|
if (cold == null)
|
|
return;
|
|
|
|
foreach (string key in cold)
|
|
_lastLinkRequest.Remove(key);
|
|
}
|
|
|
|
/// <summary>
|
|
/// A code from <see cref="CodeAlphabet"/>, retried until it does not collide.
|
|
///
|
|
/// <para>
|
|
/// <c>System.Random</c> rather than a cryptographic source, and the reason it is
|
|
/// defensible is the shape of the guess rather than the shape of the generator: a code
|
|
/// lives five minutes, is consumed on first lookup, and the website rate-limits the
|
|
/// confirm endpoint. An attacker who could predict the sequence would still have to
|
|
/// reach a code before the player does.
|
|
/// </para>
|
|
/// </summary>
|
|
private string MintCode()
|
|
{
|
|
for (int attempt = 0; attempt < 16; attempt++)
|
|
{
|
|
var chars = new char[CodeLength];
|
|
|
|
for (int i = 0; i < CodeLength; i++)
|
|
chars[i] = CodeAlphabet[_codeRandom.Next(CodeAlphabet.Length)];
|
|
|
|
string candidate = new string(chars);
|
|
|
|
if (!_codes.ContainsKey(candidate))
|
|
return candidate;
|
|
}
|
|
|
|
// Sixteen collisions in a row against a table this small is not bit-space exhaustion,
|
|
// it is something being wrong. A code that cannot collide is better than a loop that
|
|
// might not end.
|
|
return Guid.NewGuid().ToString("N").Substring(0, CodeLength).ToUpperInvariant();
|
|
}
|
|
|
|
// ---- site-owned permissions (protocol 4) ----
|
|
//
|
|
// R2: the website is the author of record for groups and grants, and this framework's own
|
|
// permission store is an ENFORCEMENT CACHE. Every third-party plugin honours a site grant
|
|
// with no adapter, because they all already call `UserHasPermission`.
|
|
//
|
|
// ── One verb, and the plugin does the diffing (D32) ───────────────────────
|
|
//
|
|
// The site sends the whole set it authors for this server; this plugin compares it against
|
|
// the live store and writes only what differs. It is the same shape as a board, pointed the
|
|
// other way: one producer, re-sent on connect, reconciled by the receiver.
|
|
//
|
|
// The alternative — shipping this server's permission store to the website and diffing
|
|
// there — was rejected for two reasons. The store is the bigger of the two sets and would
|
|
// cross the wire constantly, and a website holding a copy of it has a second source of
|
|
// truth that is stale the moment it lands.
|
|
//
|
|
// ── The site must not be able to stall the game, either ───────────────────
|
|
//
|
|
// This is the first command whose work is UNBOUNDED: a community with two thousand linked
|
|
// players has thousands of grants in one frame, and applying them in the tick the frame
|
|
// arrives is a freeze the operator will blame on the game. So a sync is compiled into a
|
|
// list of single-store operations and drained a few hundred at a time on a timer, and the
|
|
// report goes back when the last one lands.
|
|
//
|
|
// That is the same rule the outbound queue lives under (§5), applied to the inbound half:
|
|
// **nothing the far side sends may cost the main thread an unbounded amount of work.**
|
|
//
|
|
// ── What it reports, and what it will not do ──────────────────────────────
|
|
//
|
|
// • A permission this server has not REGISTERED is reported `unresolved` and skipped.
|
|
// `GrantUserPermission` silently no-ops for an unregistered name — void, no throw, no
|
|
// log (PLAN.md §12.2 rule 1) — so without the pre-check a re-push into a server missing
|
|
// one plugin is a silent partial. Registering the name here instead would fabricate a
|
|
// permission the operator never installed (D33).
|
|
// • A holder the site does not author is reported `foreign` and never touched (D31).
|
|
// Somebody typing `oxide.grant` during an incident is drift, not an error, and the
|
|
// website offers an operator two answers to it.
|
|
// • A group member who has never connected is reported `pending`. The store has no user
|
|
// record to put in a group yet (§12.2 rule 4), and the membership lands on their first
|
|
// connection — a direct grant would have reached them immediately, and pretending the
|
|
// two are the same thing is how a site promises access it did not give.
|
|
|
|
/// <summary>
|
|
/// How many store operations one timer step performs.
|
|
///
|
|
/// <para>
|
|
/// Each is a dictionary lookup and, at most, one write to an in-memory structure the
|
|
/// framework flushes on its own schedule — microseconds. Two hundred of them is well
|
|
/// under a frame at 60fps, and it drains ten thousand operations in about two and a
|
|
/// half seconds, which is inside the sidecar's reply timeout.
|
|
/// </para>
|
|
/// </summary>
|
|
private const int PermOpsPerStep = 200;
|
|
|
|
/// <summary>How often the sync drain runs while a job is in flight.</summary>
|
|
private const float PermStepSeconds = 0.05f;
|
|
|
|
/// <summary>
|
|
/// The ceiling on one sync, in operations.
|
|
///
|
|
/// <para>
|
|
/// A frame larger than this is refused whole rather than applied for a while and then
|
|
/// abandoned. The website has its own limit below this one, so reaching it means
|
|
/// something is wrong rather than something is big — and a half-applied permission set
|
|
/// is the state nobody can reason about.
|
|
/// </para>
|
|
/// </summary>
|
|
private const int PermMaxOps = 20000;
|
|
|
|
/// <summary>The operations a compiled sync is made of. One store question each.</summary>
|
|
private enum PermOpKind
|
|
{
|
|
EnsureGroup,
|
|
EnsureGroupPermission,
|
|
EnsureMember,
|
|
EnsureGrant,
|
|
RetireGroupPermission,
|
|
RetireMember,
|
|
RetireGrant,
|
|
RetireGroup,
|
|
ScanPermission,
|
|
ScanGroup
|
|
}
|
|
|
|
private struct PermOp
|
|
{
|
|
public PermOpKind Kind;
|
|
|
|
/// <summary>A Steam id, or a group name — whichever the op is about.</summary>
|
|
public string Subject;
|
|
|
|
/// <summary>A permission name, a group name, or null.</summary>
|
|
public string Object;
|
|
|
|
/// <summary>Only <see cref="PermOpKind.EnsureGroup"/> uses these.</summary>
|
|
public string Title;
|
|
|
|
public int Rank;
|
|
}
|
|
|
|
/// <summary>
|
|
/// One sync in flight. There is at most one: a second arriving while this one drains is
|
|
/// answered `busy` rather than interleaved, because two writers reconciling the same
|
|
/// store against two different desired sets is a race with no correct outcome.
|
|
/// </summary>
|
|
private sealed class PermSyncJob
|
|
{
|
|
public string ReqId;
|
|
public string SetId;
|
|
public List<PermOp> Ops = new List<PermOp>();
|
|
public int Next;
|
|
|
|
public int Grants;
|
|
public int Revokes;
|
|
public int GroupsCreated;
|
|
public int GroupsRemoved;
|
|
public int GroupPermissions;
|
|
public int GroupPermissionsRemoved;
|
|
public int Members;
|
|
public int MembersRemoved;
|
|
public int AlreadyCorrect;
|
|
public int Absent;
|
|
|
|
public readonly List<string> Unresolved = new List<string>();
|
|
public readonly List<object> Foreign = new List<object>();
|
|
public readonly List<string> Pending = new List<string>();
|
|
|
|
/// <summary>Permission names already reported unresolved, so each is reported once.</summary>
|
|
public readonly HashSet<string> UnresolvedSeen = new HashSet<string>();
|
|
|
|
/// <summary>Who the site says should hold each managed permission, for the foreign scan.</summary>
|
|
public readonly Dictionary<string, HashSet<string>> DesiredHolders =
|
|
new Dictionary<string, HashSet<string>>();
|
|
|
|
/// <summary>Who the site says is in each managed group, and what it should carry.</summary>
|
|
public readonly Dictionary<string, HashSet<string>> DesiredMembers =
|
|
new Dictionary<string, HashSet<string>>();
|
|
|
|
public readonly Dictionary<string, HashSet<string>> DesiredGroupPermissions =
|
|
new Dictionary<string, HashSet<string>>();
|
|
}
|
|
|
|
private PermSyncJob _permJob;
|
|
private Timer _permTimer;
|
|
|
|
/// <summary>
|
|
/// True while this plugin is writing to the permission store.
|
|
///
|
|
/// <para>
|
|
/// The framework raises <c>OnUserPermissionGranted</c> and friends for every write,
|
|
/// including ours — so without this flag every sync would announce itself as drift and
|
|
/// the website would chase its own tail. It is set around each drain step rather than
|
|
/// for the whole job, so a write by somebody else BETWEEN steps is still seen.
|
|
/// </para>
|
|
/// </summary>
|
|
private bool _applyingPermissions;
|
|
|
|
/// <summary>The last report, for <c>rg.perms</c>. Diagnostics only.</summary>
|
|
private string _lastPermSummary = "none yet";
|
|
|
|
/// <summary>
|
|
/// Main thread. Everything this server knows about permissions and groups — the option
|
|
/// source the website's authoring form is built from (D33).
|
|
///
|
|
/// <para>
|
|
/// It is a live read rather than anything cached, because the answer changes when an
|
|
/// operator loads a plugin and the whole point of asking is to offer names that will
|
|
/// actually resolve.
|
|
/// </para>
|
|
/// </summary>
|
|
private void HandlePermCatalogue(string reqId)
|
|
{
|
|
Dictionary<string, object> reply = Frame("perm.catalogue", "reply");
|
|
|
|
var permissions = new List<object>();
|
|
string[] registered = permission.GetPermissions();
|
|
|
|
if (registered != null)
|
|
{
|
|
foreach (string name in registered)
|
|
{
|
|
if (!string.IsNullOrEmpty(name))
|
|
permissions.Add(name);
|
|
}
|
|
}
|
|
|
|
var groups = new List<object>();
|
|
string[] groupNames = permission.GetGroups();
|
|
|
|
if (groupNames != null)
|
|
{
|
|
foreach (string name in groupNames)
|
|
{
|
|
if (string.IsNullOrEmpty(name))
|
|
continue;
|
|
|
|
var row = new Dictionary<string, object>
|
|
{
|
|
{ "name", name },
|
|
{ "title", permission.GetGroupTitle(name) },
|
|
{ "rank", permission.GetGroupRank(name) }
|
|
};
|
|
|
|
var granted = new List<object>();
|
|
string[] groupPerms = permission.GetGroupPermissions(name, false);
|
|
|
|
if (groupPerms != null)
|
|
{
|
|
foreach (string perm in groupPerms)
|
|
{
|
|
if (!string.IsNullOrEmpty(perm))
|
|
granted.Add(perm);
|
|
}
|
|
}
|
|
|
|
row["permissions"] = granted;
|
|
groups.Add(row);
|
|
}
|
|
}
|
|
|
|
reply["permissions"] = permissions;
|
|
reply["groups"] = groups;
|
|
|
|
if (reqId != null)
|
|
reply["reqId"] = reqId;
|
|
|
|
EmitFrame(reply);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Main thread. Compiles a desired set into operations and starts draining it.
|
|
///
|
|
/// <para>
|
|
/// Nothing is written here. Compiling is pure bookkeeping over the frame, so a
|
|
/// malformed or enormous sync is refused before it has touched the store — there is no
|
|
/// state to unwind because none was made.
|
|
/// </para>
|
|
/// </summary>
|
|
private void HandlePermSync(Dictionary<string, object> cmd, string reqId)
|
|
{
|
|
if (_permJob != null)
|
|
{
|
|
EmitPermError(reqId, "busy");
|
|
return;
|
|
}
|
|
|
|
var job = new PermSyncJob
|
|
{
|
|
ReqId = reqId,
|
|
SetId = StringField(cmd, "setId")
|
|
};
|
|
|
|
// ── The desired set ────────────────────────────────────────────────
|
|
//
|
|
// Order matters and it is not cosmetic: a group has to exist before anything can be
|
|
// put in it, and a permission has to be on the group before a member inherits it.
|
|
|
|
foreach (Dictionary<string, object> group in Rows(cmd, "groups"))
|
|
{
|
|
string name = Lower(StringField(group, "name"));
|
|
if (string.IsNullOrEmpty(name))
|
|
continue;
|
|
|
|
job.Ops.Add(new PermOp
|
|
{
|
|
Kind = PermOpKind.EnsureGroup,
|
|
Subject = name,
|
|
Title = StringField(group, "title") ?? name,
|
|
Rank = IntField(group, "rank")
|
|
});
|
|
|
|
HashSet<string> wanted = Bucket(job.DesiredGroupPermissions, name);
|
|
|
|
foreach (string perm in Strings(group, "permissions"))
|
|
{
|
|
string lowered = Lower(perm);
|
|
if (string.IsNullOrEmpty(lowered))
|
|
continue;
|
|
|
|
wanted.Add(lowered);
|
|
job.Ops.Add(new PermOp
|
|
{
|
|
Kind = PermOpKind.EnsureGroupPermission,
|
|
Subject = name,
|
|
Object = lowered
|
|
});
|
|
}
|
|
|
|
foreach (string member in Strings(group, "members"))
|
|
{
|
|
if (string.IsNullOrEmpty(member))
|
|
continue;
|
|
|
|
Bucket(job.DesiredMembers, name).Add(member);
|
|
job.Ops.Add(new PermOp
|
|
{
|
|
Kind = PermOpKind.EnsureMember,
|
|
Subject = member,
|
|
Object = name
|
|
});
|
|
}
|
|
|
|
// A group with no members still has to be scanned: the interesting case is a
|
|
// member the site did not put there.
|
|
Bucket(job.DesiredMembers, name);
|
|
}
|
|
|
|
foreach (Dictionary<string, object> grant in Rows(cmd, "grants"))
|
|
{
|
|
string steamId = StringField(grant, "steamId");
|
|
if (string.IsNullOrEmpty(steamId))
|
|
continue;
|
|
|
|
foreach (string perm in Strings(grant, "permissions"))
|
|
{
|
|
string lowered = Lower(perm);
|
|
if (string.IsNullOrEmpty(lowered))
|
|
continue;
|
|
|
|
Bucket(job.DesiredHolders, lowered).Add(steamId);
|
|
job.Ops.Add(new PermOp
|
|
{
|
|
Kind = PermOpKind.EnsureGrant,
|
|
Subject = steamId,
|
|
Object = lowered
|
|
});
|
|
}
|
|
}
|
|
|
|
// ── What the site has withdrawn ────────────────────────────────────
|
|
//
|
|
// The site remembers what it put here, so a grant it has since removed arrives as an
|
|
// explicit retirement. It cannot be inferred from the desired set: a name that is not
|
|
// in it is either something the site withdrew or something a human granted, and those
|
|
// two have opposite correct answers (D31).
|
|
|
|
foreach (Dictionary<string, object> row in Rows(cmd, "retire"))
|
|
{
|
|
string kind = StringField(row, "kind");
|
|
string subject = StringField(row, "subject");
|
|
string target = Lower(StringField(row, "object"));
|
|
|
|
if (string.IsNullOrEmpty(kind) || string.IsNullOrEmpty(subject))
|
|
continue;
|
|
|
|
switch (kind)
|
|
{
|
|
case "grant":
|
|
job.Ops.Add(new PermOp { Kind = PermOpKind.RetireGrant, Subject = subject, Object = target });
|
|
break;
|
|
case "member":
|
|
job.Ops.Add(new PermOp { Kind = PermOpKind.RetireMember, Subject = subject, Object = target });
|
|
break;
|
|
case "group-permission":
|
|
job.Ops.Add(new PermOp { Kind = PermOpKind.RetireGroupPermission, Subject = Lower(subject), Object = target });
|
|
break;
|
|
case "group":
|
|
job.Ops.Add(new PermOp { Kind = PermOpKind.RetireGroup, Subject = Lower(subject) });
|
|
break;
|
|
}
|
|
}
|
|
|
|
// ── The namespace the site claims ──────────────────────────────────
|
|
//
|
|
// Foreign holders can only be found for names the site says it manages. A permission it
|
|
// does not claim is somebody else's business entirely and is never enumerated — which
|
|
// also bounds this scan by the site's own set rather than by the size of the store.
|
|
|
|
foreach (string perm in Strings(cmd, "managed"))
|
|
{
|
|
string lowered = Lower(perm);
|
|
if (string.IsNullOrEmpty(lowered))
|
|
continue;
|
|
|
|
Bucket(job.DesiredHolders, lowered);
|
|
job.Ops.Add(new PermOp { Kind = PermOpKind.ScanPermission, Subject = lowered });
|
|
}
|
|
|
|
foreach (string group in job.DesiredMembers.Keys)
|
|
job.Ops.Add(new PermOp { Kind = PermOpKind.ScanGroup, Subject = group });
|
|
|
|
if (job.Ops.Count > PermMaxOps)
|
|
{
|
|
EmitPermError(reqId, "too-large");
|
|
Puts($"permission sync refused: {job.Ops.Count} operations exceeds {PermMaxOps}");
|
|
return;
|
|
}
|
|
|
|
_permJob = job;
|
|
_permTimer = timer.Every(PermStepSeconds, StepPermSync);
|
|
|
|
// Run the first slice now rather than waiting out a tick: a small sync — which is every
|
|
// ordinary one — then finishes inside the call the website is waiting on.
|
|
StepPermSync();
|
|
}
|
|
|
|
/// <summary>
|
|
/// Main thread. Drains up to <see cref="PermOpsPerStep"/> operations and, on the last
|
|
/// one, sends the report.
|
|
/// </summary>
|
|
private void StepPermSync()
|
|
{
|
|
PermSyncJob job = _permJob;
|
|
|
|
if (job == null)
|
|
{
|
|
StopPermTimer();
|
|
return;
|
|
}
|
|
|
|
_applyingPermissions = true;
|
|
|
|
try
|
|
{
|
|
int budget = PermOpsPerStep;
|
|
|
|
while (budget-- > 0 && job.Next < job.Ops.Count)
|
|
{
|
|
PermOp op = job.Ops[job.Next++];
|
|
|
|
try
|
|
{
|
|
Apply(job, op);
|
|
}
|
|
catch (Exception ex)
|
|
{
|
|
// One bad operation must not abandon the set. The report still goes back,
|
|
// and the next sync retries what this one could not do.
|
|
Puts($"permission op {op.Kind} on '{op.Subject}' threw: {ex.Message}");
|
|
}
|
|
}
|
|
}
|
|
finally
|
|
{
|
|
_applyingPermissions = false;
|
|
}
|
|
|
|
if (job.Next < job.Ops.Count)
|
|
return;
|
|
|
|
_permJob = null;
|
|
StopPermTimer();
|
|
SendPermReport(job);
|
|
}
|
|
|
|
private void StopPermTimer()
|
|
{
|
|
if (_permTimer == null)
|
|
return;
|
|
|
|
_permTimer.Destroy();
|
|
_permTimer = null;
|
|
}
|
|
|
|
/// <summary>Main thread. One store question, and at most one store write.</summary>
|
|
private void Apply(PermSyncJob job, PermOp op)
|
|
{
|
|
switch (op.Kind)
|
|
{
|
|
case PermOpKind.EnsureGroup:
|
|
if (permission.GroupExists(op.Subject))
|
|
{
|
|
job.AlreadyCorrect++;
|
|
}
|
|
else
|
|
{
|
|
permission.CreateGroup(op.Subject, op.Title, op.Rank);
|
|
job.GroupsCreated++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.EnsureGroupPermission:
|
|
if (!Registered(job, op.Object))
|
|
break;
|
|
|
|
if (permission.GroupHasPermission(op.Subject, op.Object))
|
|
{
|
|
job.AlreadyCorrect++;
|
|
}
|
|
else
|
|
{
|
|
permission.GrantGroupPermission(op.Subject, op.Object, this);
|
|
job.GroupPermissions++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.EnsureMember:
|
|
if (permission.UserHasGroup(op.Subject, op.Object))
|
|
{
|
|
job.AlreadyCorrect++;
|
|
break;
|
|
}
|
|
|
|
permission.AddUserGroup(op.Subject, op.Object);
|
|
|
|
// Read back, because this is the call that can do nothing at all. A player the
|
|
// store has never seen has no user record to put in a group (§12.2 rule 4), and
|
|
// the API says so by returning void and changing nothing.
|
|
if (permission.UserHasGroup(op.Subject, op.Object))
|
|
job.Members++;
|
|
else
|
|
job.Pending.Add(op.Subject + ":" + op.Object);
|
|
|
|
break;
|
|
|
|
case PermOpKind.EnsureGrant:
|
|
if (!Registered(job, op.Object))
|
|
break;
|
|
|
|
if (HoldsDirectly(op.Subject, op.Object))
|
|
{
|
|
job.AlreadyCorrect++;
|
|
}
|
|
else
|
|
{
|
|
permission.GrantUserPermission(op.Subject, op.Object, this);
|
|
job.Grants++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.RetireGrant:
|
|
if (HoldsDirectly(op.Subject, op.Object))
|
|
{
|
|
permission.RevokeUserPermission(op.Subject, op.Object);
|
|
job.Revokes++;
|
|
}
|
|
else
|
|
{
|
|
job.Absent++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.RetireMember:
|
|
if (permission.UserHasGroup(op.Subject, op.Object))
|
|
{
|
|
permission.RemoveUserGroup(op.Subject, op.Object);
|
|
job.MembersRemoved++;
|
|
}
|
|
else
|
|
{
|
|
job.Absent++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.RetireGroupPermission:
|
|
if (permission.GroupHasPermission(op.Subject, op.Object))
|
|
{
|
|
permission.RevokeGroupPermission(op.Subject, op.Object);
|
|
job.GroupPermissionsRemoved++;
|
|
}
|
|
else
|
|
{
|
|
job.Absent++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.RetireGroup:
|
|
if (permission.GroupExists(op.Subject))
|
|
{
|
|
permission.RemoveGroup(op.Subject);
|
|
job.GroupsRemoved++;
|
|
}
|
|
else
|
|
{
|
|
job.Absent++;
|
|
}
|
|
|
|
break;
|
|
|
|
case PermOpKind.ScanPermission:
|
|
ScanPermission(job, op.Subject);
|
|
break;
|
|
|
|
case PermOpKind.ScanGroup:
|
|
ScanGroup(job, op.Subject);
|
|
break;
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Everybody holding a managed permission DIRECTLY, minus everybody the site says
|
|
/// should — the rest is a hand edit, and it is reported rather than undone.
|
|
/// </summary>
|
|
private void ScanPermission(PermSyncJob job, string perm)
|
|
{
|
|
if (!permission.PermissionExists(perm))
|
|
return;
|
|
|
|
HashSet<string> desired;
|
|
if (!job.DesiredHolders.TryGetValue(perm, out desired))
|
|
desired = new HashSet<string>();
|
|
|
|
string[] holders = permission.GetPermissionUsers(perm);
|
|
|
|
if (holders == null)
|
|
return;
|
|
|
|
foreach (string entry in holders)
|
|
{
|
|
string id = UserId(entry);
|
|
|
|
if (string.IsNullOrEmpty(id) || desired.Contains(id))
|
|
continue;
|
|
|
|
job.Foreign.Add(new Dictionary<string, object>
|
|
{
|
|
{ "kind", "grant" },
|
|
{ "subject", id },
|
|
{ "object", perm }
|
|
});
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// The same question about a managed group, asked twice: who is in it that the site did
|
|
/// not put there, and what does it carry that the site did not put on it.
|
|
/// </summary>
|
|
private void ScanGroup(PermSyncJob job, string group)
|
|
{
|
|
if (!permission.GroupExists(group))
|
|
return;
|
|
|
|
HashSet<string> members;
|
|
if (!job.DesiredMembers.TryGetValue(group, out members))
|
|
members = new HashSet<string>();
|
|
|
|
string[] present = permission.GetUsersInGroup(group);
|
|
|
|
if (present != null)
|
|
{
|
|
foreach (string entry in present)
|
|
{
|
|
string id = UserId(entry);
|
|
|
|
if (string.IsNullOrEmpty(id) || members.Contains(id))
|
|
continue;
|
|
|
|
job.Foreign.Add(new Dictionary<string, object>
|
|
{
|
|
{ "kind", "member" },
|
|
{ "subject", id },
|
|
{ "object", group }
|
|
});
|
|
}
|
|
}
|
|
|
|
HashSet<string> wanted;
|
|
if (!job.DesiredGroupPermissions.TryGetValue(group, out wanted))
|
|
wanted = new HashSet<string>();
|
|
|
|
string[] carried = permission.GetGroupPermissions(group, false);
|
|
|
|
if (carried == null)
|
|
return;
|
|
|
|
foreach (string perm in carried)
|
|
{
|
|
string lowered = Lower(perm);
|
|
|
|
if (string.IsNullOrEmpty(lowered) || wanted.Contains(lowered))
|
|
continue;
|
|
|
|
job.Foreign.Add(new Dictionary<string, object>
|
|
{
|
|
{ "kind", "group-permission" },
|
|
{ "subject", group },
|
|
{ "object", lowered }
|
|
});
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Whether this server has a plugin that registered the name, recording the answer once
|
|
/// per sync when it has not.
|
|
///
|
|
/// <para>
|
|
/// This is the pre-check PLAN.md §12.2 rule 1 exists for. Without it the grant below
|
|
/// would return void, throw nothing, log nothing and do nothing, and the website would
|
|
/// record a privilege it never gave.
|
|
/// </para>
|
|
/// </summary>
|
|
private bool Registered(PermSyncJob job, string perm)
|
|
{
|
|
if (string.IsNullOrEmpty(perm))
|
|
return false;
|
|
|
|
if (permission.PermissionExists(perm))
|
|
return true;
|
|
|
|
if (job.UnresolvedSeen.Add(perm))
|
|
job.Unresolved.Add(perm);
|
|
|
|
return false;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Whether a player holds a permission through a grant of their own, as opposed to
|
|
/// through a group.
|
|
///
|
|
/// <para>
|
|
/// <c>UserHasPermission</c> cannot answer this — it is true for a member of a group
|
|
/// that carries the permission, so using it would make every direct grant look already
|
|
/// applied the moment a group covered it, and a revoke look already done.
|
|
/// <c>GetPermissionUsers</c> reads the user's own set, which is the question.
|
|
/// </para>
|
|
/// </summary>
|
|
private bool HoldsDirectly(string steamId, string perm)
|
|
{
|
|
string[] holders = permission.GetPermissionUsers(perm);
|
|
|
|
if (holders == null)
|
|
return false;
|
|
|
|
foreach (string entry in holders)
|
|
{
|
|
if (UserId(entry) == steamId)
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
/// <summary>
|
|
/// The id out of a store listing entry.
|
|
///
|
|
/// <para>
|
|
/// Both <c>GetPermissionUsers</c> and <c>GetUsersInGroup</c> answer with the id and the
|
|
/// player's last known name stuck together — <c>76561198…(Wanderer)</c> — and the
|
|
/// spacing differs between the two calls and between the frameworks. So the id is
|
|
/// everything before the first bracket, and a name containing one cannot confuse it
|
|
/// because only the first is looked for.
|
|
/// </para>
|
|
/// </summary>
|
|
private static string UserId(string entry)
|
|
{
|
|
if (string.IsNullOrEmpty(entry))
|
|
return null;
|
|
|
|
int bracket = entry.IndexOf('(');
|
|
string id = bracket >= 0 ? entry.Substring(0, bracket) : entry;
|
|
|
|
return id.Trim();
|
|
}
|
|
|
|
private void SendPermReport(PermSyncJob job)
|
|
{
|
|
var applied = new Dictionary<string, object>
|
|
{
|
|
{ "grants", job.Grants },
|
|
{ "revokes", job.Revokes },
|
|
{ "groupsCreated", job.GroupsCreated },
|
|
{ "groupsRemoved", job.GroupsRemoved },
|
|
{ "groupPermissions", job.GroupPermissions },
|
|
{ "groupPermissionsRemoved", job.GroupPermissionsRemoved },
|
|
{ "members", job.Members },
|
|
{ "membersRemoved", job.MembersRemoved }
|
|
};
|
|
|
|
Dictionary<string, object> report = Frame("perm.report", "reply");
|
|
report["applied"] = applied;
|
|
report["alreadyCorrect"] = job.AlreadyCorrect;
|
|
report["absent"] = job.Absent;
|
|
report["unresolved"] = job.Unresolved;
|
|
report["foreign"] = job.Foreign;
|
|
report["pending"] = job.Pending;
|
|
report["operations"] = job.Ops.Count;
|
|
|
|
if (!string.IsNullOrEmpty(job.SetId))
|
|
report["setId"] = job.SetId;
|
|
|
|
if (job.ReqId != null)
|
|
report["reqId"] = job.ReqId;
|
|
|
|
_lastPermSummary =
|
|
$"ops={job.Ops.Count} grants={job.Grants} revokes={job.Revokes} " +
|
|
$"groups={job.GroupsCreated} members={job.Members} same={job.AlreadyCorrect} " +
|
|
$"unresolved={job.Unresolved.Count} foreign={job.Foreign.Count} " +
|
|
$"pending={job.Pending.Count} at={DateTime.UtcNow:HH:mm:ss}Z";
|
|
|
|
EmitFrame(report);
|
|
}
|
|
|
|
/// <summary>A refusal of the whole sync, with a reason the website turns into a sentence.</summary>
|
|
private void EmitPermError(string reqId, string reason)
|
|
{
|
|
Dictionary<string, object> frame = Frame("perm.error", "reply");
|
|
frame["reason"] = reason;
|
|
|
|
if (reqId != null)
|
|
frame["reqId"] = reqId;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
// ---- drift, as it happens ----
|
|
//
|
|
// Both frameworks raise a hook for every permission write, ours included. They are NOT the
|
|
// drift record — the sync report is, because only the site's desired set can say whether a
|
|
// grant is foreign. What they are is a reason to reconcile SOON: an operator typing
|
|
// `oxide.grant` at the console shows up on the website in seconds instead of at the next
|
|
// scheduled audit, and the authoritative answer still comes from a sync.
|
|
//
|
|
// That division is what makes them safe to trust at exactly the weight they deserve. A hook
|
|
// that stops firing on a framework upgrade costs latency, not correctness.
|
|
|
|
private void OnUserPermissionGranted(string id, string permName)
|
|
{
|
|
MarkHook("OnUserPermissionGranted");
|
|
EmitPermDrift("granted", id, permName, null);
|
|
}
|
|
|
|
private void OnUserPermissionRevoked(string id, string permName)
|
|
{
|
|
MarkHook("OnUserPermissionRevoked");
|
|
EmitPermDrift("revoked", id, permName, null);
|
|
}
|
|
|
|
private void OnUserGroupAdded(string id, string groupName)
|
|
{
|
|
MarkHook("OnUserGroupAdded");
|
|
EmitPermDrift("group-added", id, null, groupName);
|
|
}
|
|
|
|
private void OnUserGroupRemoved(string id, string groupName)
|
|
{
|
|
MarkHook("OnUserGroupRemoved");
|
|
EmitPermDrift("group-removed", id, null, groupName);
|
|
}
|
|
|
|
private void OnGroupPermissionGranted(string name, string permName)
|
|
{
|
|
MarkHook("OnGroupPermissionGranted");
|
|
EmitPermDrift("group-permission-granted", null, permName, name);
|
|
}
|
|
|
|
private void OnGroupPermissionRevoked(string name, string permName)
|
|
{
|
|
MarkHook("OnGroupPermissionRevoked");
|
|
EmitPermDrift("group-permission-revoked", null, permName, name);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Main thread. Says that the store changed under us, and by how much it is worth
|
|
/// caring.
|
|
///
|
|
/// <para>
|
|
/// Staff class (§8.5): it names a Steam id beside a privilege, which is a fact about a
|
|
/// person's standing rather than about what happened on the server.
|
|
/// </para>
|
|
/// </summary>
|
|
private void EmitPermDrift(string action, string steamId, string perm, string group)
|
|
{
|
|
// Our own writes raise these hooks too. Suppressing them here rather than filtering on
|
|
// the website is the only version that works: the site cannot tell its own grant from
|
|
// a human's by looking at one, which is the entire reason drift needs a desired set.
|
|
if (_applyingPermissions)
|
|
return;
|
|
|
|
Dictionary<string, object> frame = Frame("perm.drift", "event");
|
|
frame["action"] = action;
|
|
|
|
if (!string.IsNullOrEmpty(steamId))
|
|
frame["steamId"] = steamId;
|
|
|
|
if (!string.IsNullOrEmpty(perm))
|
|
frame["permission"] = perm;
|
|
|
|
if (!string.IsNullOrEmpty(group))
|
|
frame["group"] = group;
|
|
|
|
EmitFrame(frame);
|
|
}
|
|
|
|
// ---- reading a command's fields ----
|
|
//
|
|
// `JsonConvert` hands back `object`, and every field below arrives as one. These are the
|
|
// three shapes this protocol uses, read defensively: a malformed field is a missing field,
|
|
// never an exception on the main thread.
|
|
|
|
private static string StringField(Dictionary<string, object> row, string key)
|
|
{
|
|
object value;
|
|
return row != null && row.TryGetValue(key, out value) && value != null
|
|
? value.ToString()
|
|
: null;
|
|
}
|
|
|
|
private static int IntField(Dictionary<string, object> row, string key)
|
|
{
|
|
object value;
|
|
|
|
if (row == null || !row.TryGetValue(key, out value) || value == null)
|
|
return 0;
|
|
|
|
int parsed;
|
|
return int.TryParse(value.ToString(), out parsed) ? parsed : 0;
|
|
}
|
|
|
|
/// <summary>The objects in an array field, skipping anything that is not one.</summary>
|
|
private static IEnumerable<Dictionary<string, object>> Rows(Dictionary<string, object> row, string key)
|
|
{
|
|
object value;
|
|
|
|
if (row == null || !row.TryGetValue(key, out value))
|
|
yield break;
|
|
|
|
var list = value as Newtonsoft.Json.Linq.JArray;
|
|
|
|
if (list == null)
|
|
yield break;
|
|
|
|
foreach (Newtonsoft.Json.Linq.JToken item in list)
|
|
{
|
|
var obj = item as Newtonsoft.Json.Linq.JObject;
|
|
|
|
if (obj == null)
|
|
continue;
|
|
|
|
yield return obj.ToObject<Dictionary<string, object>>();
|
|
}
|
|
}
|
|
|
|
/// <summary>The strings in an array field.</summary>
|
|
private static IEnumerable<string> Strings(Dictionary<string, object> row, string key)
|
|
{
|
|
object value;
|
|
|
|
if (row == null || !row.TryGetValue(key, out value))
|
|
yield break;
|
|
|
|
var list = value as Newtonsoft.Json.Linq.JArray;
|
|
|
|
if (list == null)
|
|
yield break;
|
|
|
|
foreach (Newtonsoft.Json.Linq.JToken item in list)
|
|
{
|
|
if (item == null || item.Type == Newtonsoft.Json.Linq.JTokenType.Null)
|
|
continue;
|
|
|
|
yield return item.ToString();
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Permission and group names are compared case-insensitively because both frameworks
|
|
/// store them lowered. Comparing what the site sent against what the store holds
|
|
/// without this makes `Kits.VIP` and `kits.vip` two different privileges, one of which
|
|
/// is granted for ever and never found again.
|
|
/// </summary>
|
|
private static string Lower(string value)
|
|
{
|
|
return value == null ? null : value.Trim().ToLowerInvariant();
|
|
}
|
|
|
|
private static HashSet<string> Bucket(Dictionary<string, HashSet<string>> map, string key)
|
|
{
|
|
HashSet<string> bucket;
|
|
|
|
if (!map.TryGetValue(key, out bucket))
|
|
{
|
|
bucket = new HashSet<string>();
|
|
map[key] = bucket;
|
|
}
|
|
|
|
return bucket;
|
|
}
|
|
|
|
// ---- diagnostics ----
|
|
|
|
/// <summary>
|
|
/// The link's own counters, from the server console. This is how an operator answers
|
|
/// "is the bridge working" without reading a log, and it is the first thing to ask for
|
|
/// when the website says a server is offline.
|
|
/// </summary>
|
|
[ConsoleCommand("rg.link")]
|
|
private void CmdLink(ConsoleSystem.Arg arg)
|
|
{
|
|
arg.ReplyWith(
|
|
$"protocol={ProtocolVersion} serverId={(_config != null ? _config.ServerId : "?")} " +
|
|
$"connected={_connected} depth={Volatile.Read(ref _depth)} " +
|
|
$"sent={Interlocked.Read(ref _sent)} dropped={Interlocked.Read(ref _dropped)} " +
|
|
$"received={Interlocked.Read(ref _received)} connects={Interlocked.Read(ref _connects)} " +
|
|
$"writeErrors={Interlocked.Read(ref _writeErrors)} bootId={_bootId} " +
|
|
$"wipeId={(_wipeId ?? "none")}");
|
|
}
|
|
|
|
/// <summary>
|
|
/// What the last permission sync did, from the server console.
|
|
///
|
|
/// <para>
|
|
/// It is the fastest way to tell the two silent failures apart on a rig: a sync that
|
|
/// reported `unresolved` names granted nothing because no loaded plugin registered
|
|
/// them, and a sync that reported `pending` members could not put somebody in a group
|
|
/// because the store has never seen them. Both look like "the website says they have
|
|
/// it and the game disagrees" from the other end.
|
|
/// </para>
|
|
/// </summary>
|
|
[ConsoleCommand("rg.perms")]
|
|
private void CmdPerms(ConsoleSystem.Arg arg)
|
|
{
|
|
string[] groups = permission.GetGroups();
|
|
string[] registered = permission.GetPermissions();
|
|
|
|
arg.ReplyWith(
|
|
$"protocol={ProtocolVersion} registered={(registered != null ? registered.Length : 0)} " +
|
|
$"groups={(groups != null ? groups.Length : 0)} " +
|
|
$"running={(_permJob != null)}" + Environment.NewLine +
|
|
$"last sync: {_lastPermSummary}");
|
|
}
|
|
|
|
/// <summary>
|
|
/// Which of this plugin's hooks have actually fired, and how often.
|
|
///
|
|
/// <para>
|
|
/// Both frameworks bind hooks by name and arity through reflection, with no
|
|
/// compile-time check and no warning when a name matches nothing. So a hook Facepunch
|
|
/// renames on a Thursday, and a hook one framework's catalogue does not list, both
|
|
/// present identically: silence. This command is the standing answer to both, on
|
|
/// whichever framework the server is actually running — and it outranks either
|
|
/// catalogue, because it is a measurement rather than a document.
|
|
/// </para>
|
|
///
|
|
/// <para>
|
|
/// A zero is not automatically a fault: <c>OnNewSave</c> fires on wipe day and
|
|
/// <c>OnUserBanned</c> only when somebody is banned. It is a fault when the thing it
|
|
/// names has demonstrably happened.
|
|
/// </para>
|
|
/// </summary>
|
|
[ConsoleCommand("rg.hooks")]
|
|
private void CmdHooks(ConsoleSystem.Arg arg)
|
|
{
|
|
var fired = new List<string>();
|
|
var silent = new List<string>();
|
|
|
|
foreach (string hook in ExpectedHooks)
|
|
{
|
|
long count;
|
|
_hookCounts.TryGetValue(hook, out count);
|
|
|
|
if (count > 0L)
|
|
fired.Add($"{hook}={count}");
|
|
else
|
|
silent.Add(hook);
|
|
}
|
|
|
|
string firedText = fired.Count > 0 ? string.Join(" ", fired.ToArray()) : "(none)";
|
|
string silentText = silent.Count > 0 ? string.Join(" ", silent.ToArray()) : "(none)";
|
|
|
|
arg.ReplyWith(
|
|
$"protocol={ProtocolVersion} hooks={ExpectedHooks.Length} " +
|
|
$"fired={fired.Count} silent={silent.Count}" + Environment.NewLine +
|
|
$"fired: {firedText}" + Environment.NewLine +
|
|
$"silent: {silentText}");
|
|
}
|
|
}
|
|
}
|