Two optional sections at the end of the report, each after a marker
byte: every second's packets by peer with its HTTP requests from and to
other servers, and the busiest remote ends with the rest summed.
Reports without them still read (older servers), and older readers stop
before them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
After serviceType the reply now sends major, minor, patch, a flags byte
(build kind in bits 0-1, dirty in bit 2), the first 32 bits of the commit
hash and a u16 length-prefixed build string, instead of the stale fixed
ASCII "0.1.3". unknown stays "DLU3". The 1.10.64 client reads only
netVersion and serviceType and never checks the length, so the extra
bytes are ignored. The build string is optional on read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The guild packets of the 1.10.64 client (docs/Guilds.md) as Serialize/Deserialize structs, laid out the way the client's
packet handlers read them: GUILD_CREATE_RESPONSE, GUILD_INVITE, GUILD_INVITE_INITIAL_RESPONSE, _FINAL_RESPONSE,
_CONFIRM, GUILD_ADD_PLAYER, GUILD_REMOVE_PLAYER, GUILD_LOGIN_LOGOUT and GUILD_DATA (ClientPackets), the world packet
TMP_GUILD_CREATE the create box sends, the chat packets the client sends through its world (GUILD_INVITE,
GUILD_INVITE_RESPONSE, GUILD_LEAVE, GUILD_GET_ALL), DLU's world <-> chat packets (GUILD_CREATE, GUILD_KICK,
GUILD_GET_STATUS as the chat server's guild update to a world, and GUILD_SET_RANK and GUILD_DISBAND appended to
MessageType::Chat after CREATE_TEAM) and the game message DisplayGuildCreateBox (626). Fixed-size names always keep a
NUL, since the client reads them as C strings. Enums eGuildCreateResponse, eGuildInviteResponse,
eGuildInviteFinalResponse, eGuildRank and eGuildLeaveReason. Nothing sends them yet.
Check: GuildPacketsTests compares every packet with the bytes at the client's offsets. Nothing to check in game.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Before private (7), team (8) and local team (10) chat the client asks
the chat server for the minimum chat mode of the channel
(LWOChatComponent::RequestMinimumChatMode; it answers every other
channel itself) and hands the answer to its chat UI. DLU never
answered. Live answered chat mode 0 with the requested channel (74 of
74 team answers in the captures: 53 05 00 39 00 00 00 00 00 08).
The chat server now answers with the lowest chat mode among who reads
the channel (the sender and their online teammates, or the sender and
the recipient), using the GM level as the chat mode as DLU does
everywhere else; the private answer also echoes the recipient's name
and GM level (0 when offline). Inferred: the minimum for GMs (no live
GM sample) and the answer when the recipient is offline.
Check: type in team chat and whisper another player: the messages send
as before and the chat box shows no error; do the same with a GM in
the team.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
World packet 120 (UgcDownloadFailed: resType u32, blueprint ID, status
u32, character ID; lu_packets, 31 bytes after the 0x53 as the client
sends it) was logged as an unknown world packet. The client sends it
from LWOResMgr2Interface::FinishResourceRequest (0x0105e5c0) for every
blueprint file (player model, car or rocket build) whose request did not
end with HTTP 200, including files it did not download at all (status
0). Live: 1525 packets, 1520 with status 0 (all four file types, around
load and FetchModelMetadataRequest), 5 with 404. Live sent nothing back
and the sessions carried on.
The logout the client does on failed UGC downloads
(NET_DISCONNECT_FAILED_DOWNLOAD_UGC, MainThread_LogoutDueToConnectionFailures
0x0102b9c0) is its own decision after repeated connection failures to the
UGC HTTP server; there is no server message that prevents or triggers it.
So the server only records it: a log line for an HTTP failure (for
finding missing files on the UGC server), a debug line for status 0.
MessageType::World gains UGC_DOWNLOAD_FAILED = 120 (appended; the magic
enum range goes to 120).
Check in game: nothing changes for the player. With a property that has
models, load it: the world server log has no "Unknown world packet 120"
lines; if a model file is missing on the UGC server there is one "failed
to download blueprint" line naming it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With new binaries in place, master moves everything onto them while the
server keeps running (the dashboard's Live update, /liveupdate or SIGUSR2 to
master):
- database migrations of the new build first; a failure stops there
- UGC finishes the jobs it is running (queued rows stay pending), auth
restarts, chat hands its teams to master for the next chat server; master
starts the new processes and retries ones that don't come back
- once the new chat server is up (CHAT_SERVER_READY) every world connects at
once and sends its players again (LoginSessionNotify resync, no login logged)
- every world instance is replaced with an instance migration: public worlds
and private ones (same password) at once, properties after the old instance
saved and froze the property (MIGRATE_PREPARE: no building, claiming or
saving there any more), activity zones and character select once their
players left or after a wait; empty instances just stop, zones in
prestart_worlds get a new one first
- the dashboard restarts last and picks the status up again
Players land where they stood (position carried in CarriedPlayerState, also on
properties and Moon Base). Draining instances get no new players
(InstanceMigration::AcceptsNewPlayers) and show as "Moving players" in the
world list. The order lives in LiveUpdateMachine.h without master state and is
unit tested; master's glue is LiveUpdateCoordinator. Master itself is not
replaced. Message IDs are appended only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A placed model's client (1.10.64, UGCUSE3DSERVICES=7:0) always loads its
blueprint's NIF and HKX (its BlueprintComponent sets renderUserGen and
physicsUserGen itself) and its LXFML, asks the world for each file's
manifest first and waits for the answer with no timeout.
With ugc_manifest=1 and the new ugc_manifest_models=1 (default 0) the world:
- leaves the models whose mesh the UGC server made out of the LXFML it
sends when a property loads,
- answers their NIF with the UGC server's checksum, their LXFML with the
stored LXFML's (worked out once and kept) and their HKX as not known,
- sends a model that isn't made its LXFML (once for the three requests),
so the client builds it itself and no model is left waiting.
When the UGC server writes a model's mesh with a new checksum it sends
UGC_MODELS_MADE (a new master message, appended) to the master, which
passes it to every world; a world with that model placed sends its
players the new NIF checksum and NotifyClientUGCModelReady (game message
909, the blueprint id), so clients switch to the served mesh.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The dashboard shows worlds master has launched but that aren't connected
yet (Starting) and those shutting down, on the main page's world list
and in /api/worlds (state: starting|stopping), without a shutdown
button. Master's server list now carries each world's state (appended
after the UGC fields, one byte per world) and master pushes the list to
the dashboard whenever a world is launched, becomes ready, is told to
shut down or goes away, instead of the dashboard only seeing it on its
30 second poll. Starting worlds are kept apart from the running ones,
so counts, events and shutdown requests still only see running worlds.
prestart_worlds (masterconfig.ini, zone ids) lists the worlds master
starts when prestart_servers is on, instead of the hardcoded character
select (0) and Venture Explorer (1000), which stay the default when it
is missing or empty. It is on the Settings page as a zone list (restart
only), shown when prestart_servers is on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With UGCUSE3DSERVICES=7:0 (the client's default) the client asks its world for
a blueprint file's MD5 and size (REQUEST_UGC_MANIFEST_INFO, world 27) and then
downloads BrickModels/UserMade/<id % 1000>/<id>.<ext>.sd0. Layouts checked in
the 1.10.64 client: the request is a u64 blueprint id and a u8 resource type;
the answer (UGC_MANIFEST_RESPONSE, client 60) repeats them and adds a u8 valid,
the u32 size and the 16 byte MD5 of the inflated file, 37 bytes after the 0x53
exactly or the client drops it.
- dNet: WorldPackets::RequestUgcManifestInfo, ClientPackets::UgcManifestResponse
(eUgcResourceType), with byte tests against the client's layouts.
- Database: ugc_file_checksums (per model or module combination and file) and
ugc_modular_build.combination_id (migrations 89 / 72), GetUgcFileChecksum
looks a blueprint up as a model, else as a build through its combination.
- UGC server: every download is also written as .sd0 (Sd0::Compress); workers
hand the checksums back and the main thread stores them; old items get their
sd0 icon and checksum, and builds their combination id, once at start-up, a
few per tick; serves <dir>/BrickModels/UserMade/<bucket>/<id>.<ext>.sd0 under
client_path, /<folder>/UserBrickModels and the root (.hkx 404).
- World: UgcManifest answers on the main thread with one indexed query per
request; files not made yet are answered when they are (looked at again
every 5 seconds), and a model in its quiet period is made right away. No
worker threads, HTTP or file reads in the world.
Off by default (ugc_manifest=0): checked in game, the client then downloads
from http://127.0.0.1:80/lwoclient/UserBrickModels/ whatever its boot.cfg says
and logs the player out when it can't connect, so icons need the UGC server on
port 80 of each player's machine. docs/UgcServer.md has the details.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server list now carries whether master starts the UGC server, whether it
is connected and the pid it was started as. Settings reloads reach it like
auth and chat, and shutdown waits for it. The UGC server sends its totals and
storage with its traffic reports.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mail sent to someone who is not in the sender's world (system mail with
no address, and all player-to-player mail) now reaches them: the world
sends a MailNotify (Chat::MAIL, unused until now) to the chat server,
which passes it to the world the receiver is in, and that world sends
the client its unread count with a NewMail NotificationResponse.
Receivers in the same world are told directly. Player-to-player mail
did not notify the receiver at all before.
Replaces the TODO in Mail::SendMail.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
TrafficStats keeps packets and bytes in and out per second and the busiest
packet and game message types. dServer counts at its send and receive calls
and adds RakNet's connection statistics (datagrams, resends, ping); the report
goes to master as SERVER_TRAFFIC (appended), which passes it to the dashboard.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Nothing in the server writes or reads a packet by hand any more, so the
helpers for doing so go:
- CBITSTREAM, CMSGHEADER, CINSTREAM, CINSTREAM_SKIP_HEADER, SEND_PACKET,
SEND_PACKET_BROADCAST and HEADER_SIZE leave dCommonVars.h;
- the free BitStreamUtils::WriteHeader (LUBitStream::WriteHeader writes the
same bytes) and the unused PacketUtils::SavePacket are deleted.
The last raw reads are replaced: WorldServer builds its input stream
directly, the master packet logs read the header with
LUBitStream::ReadHeader instead of peeking at packet->data[1] and [3], and
MessageInspector reads a sent game message's header with the new
NetGameMsg::ReadPacketHeader (the counterpart of WritePacket) instead of
memcmp/memcpy. packet->data[0] is still compared with RakNet's own
connection IDs.
The frozen oracles keep using the macros verbatim through the test-only
tests/dGameTests/LegacyPacketMacros.h; the HeaderSkip tests, which only
tested CINSTREAM_SKIP_HEADER, are removed.
docs/PacketArchitecture.md: "where we are" now describes the final state
and what still touches raw bytes (RakNet IDs, replica headers, behavior bit
streams), and a new section collects the known wire discrepancies found
during the conversion, with client addresses.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every MASTER service packet is now an LUBitStream struct (docs/PacketArchitecture.md,
PR 14) and the master server's switch is a dispatch map (PacketDispatcher), as are the
master handlers of the world, chat and dashboard servers.
- dNet/MasterPackets.h: RequestZoneTransfer, RequestZoneTransferResponse, ServerInfo,
RequestSessionKey, SetSessionKey, SessionKeyResponse, NewSessionAlert, PlayerAdded /
PlayerRemoved, CreatePrivateZone, RequestPrivateZone (passwords still cut to 50
characters when read), WorldReady, WorldReadyInfo (WORLD_READY to the dashboard),
PrepZone, Shutdown, ShutdownResponse, WorldShutDown (SHUTDOWN_RESPONSE to the
dashboard), ShutdownUniverse, AffirmTransferRequest/Response, RequestServerList,
ServerListResponse, DashboardShutdown, ConfigReload, InstanceShutdown. The Send*
functions are gone; MasterPackets::SendToMaster(msg) and SendTo(sysAddr, msg) send a
struct.
- The dashboard and instance migration structs (PlayerAction, DataChanged, Dashboard
messages, MessageCapture, InstanceMigration) are LUBitStreams of the MASTER service now
and moved to dNet/master/, included by MasterPackets.h. Their payloads are unchanged;
master forwards them by re-serializing the struct instead of copying raw bytes.
- InstanceManager, ZoneInstanceManager, MigrationCoordinator, dServer (server info, zone
transfer response), auth (SET_SESSION_KEY), the world (session keys, player added and
removed, world ready, shutdown response, affirmations, prep zone, shutdown universe) and
the dashboard (server list, instance shutdown, config reload, announcements, player
actions, message capture) send and read structs.
- The login stamps are a `stamps` field of RequestZoneTransfer and
RequestZoneTransferResponse (read leniently as before: a message without them reads as
empty); master adds its stamps in the REQUEST_ZONE_TRANSFER handler and when it answers,
as it did.
- InstanceManager::GetInstanceBySysAddr takes a const address.
Verified: tests/dGameTests/dNetTests/Legacy/MasterPacketsLegacy.h is a verbatim copy of
the old writers and readers; MasterPacketsTests requires identical bytes for a grid of
inputs, checks the old readers read what the structs write, round trips and truncation,
hand written golden packets, that zone transfers without stamps still read, that the dashboard/migration structs write what
"header + Serialize" wrote, and that the dispatcher drops truncated packets. No wire
bytes changed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every packet of the chat service is now an LUBitStream struct in dNet/ChatPackets.h
(docs/PacketArchitecture.md, PR 13), and nothing in the chat server or the chat side of
the world server reads or writes a packet by hand any more.
- World <-> chat: LoginSessionNotify, UnexpectedDisconnect, GMLevelUpdate, GMMute,
Announcement (GM announce), CreateTeam, TeamUpdate (TEAM_GET_STATUS to worlds),
AchievementNotify, ShowAllRequest, FindPlayerRequest.
- Client -> world -> chat (friends, ignore list, teams, general and private chat):
GetFriendsList, AddFriendRequest, AddFriendResponse, RemoveFriend, GetIgnoreList,
AddIgnore, RemoveIgnore, GeneralChatMessage, PrivateChatMessage, TeamInvite,
TeamInviteResponse, TeamLeave, TeamKick, TeamSetLeader, TeamSetLoot, TeamGetStatus.
The 77 bytes the handlers skipped are named fields now (the sender block the client
fills in); the 4 unused bytes after the player ID are kept as `unknown`.
- What the client receives. Chat service (ChatPackets::Client): GeneralChatMessage
(replaces SendChatMessage; SendSystemMessage stays as a helper built on it),
PrivateChatMessage. Client service (ClientPackets, as structs go in the file of the
ServiceType in their header): SendCannedText (replaces SendMessageFail), GetFriendsListResponse, AddFriendRequest,
AddFriendResponse, RemoveFriendResponse, UpdateFriendNotify, WhoResponse,
ShowAllResponse, Get/Add/RemoveIgnoreResponse, TeamInvite, TeamInviteInitialResponse,
and the team game messages chat writes (TeamInviteConfirm, TeamGetStatusResponse,
TeamSetLeader, TeamAddPlayer, TeamRemovePlayer, TeamSetOffWorldFlag) as TeamGameMsg
structs, since chat doesn't link dGame's NetGameMsg.
- WORLD_ROUTE_PACKET is ChatPackets::WorldRoutePacket (its own file, dNet/WorldRoutePacket.h,
since it carries packets of other services): the target and the inner packet.
ChatPacketHandler::SendRouted replaces the per-function route headers and
SendRoutedMsg.
Dispatch: dNet/PacketDispatcher.h is a dispatch map from packet ID to (struct, handler
function); packets that fail to Deserialize are logged and dropped. The chat server's
switch and the world's HandlePacketChat switch are now maps. The handlers take the
structs; their logic is unchanged. World code sends to chat with ChatServerLink::Send
(dGame) instead of writing to Game::chatServer by hand. eChatChannel,
eChatMessageResponseCode and eAddIgnoreResponse moved to dCommon/dEnums so the structs
can use them.
Verified: tests/dGameTests/dNetTests/Legacy/ChatPacketsLegacy.h is a verbatim copy of
the old senders and readers; ChatPacketsTests (25 tests) sends a grid of inputs through
both and requires identical bits, bytes and destination, checks the old readers read
what the structs write, round trips every struct, checks truncated packets are refused,
and has hand written golden packets for general chat, canned text, GM mute and a routed
team game message. Breaking one field width (UpdateFriendNotify) and the TeamAddPlayer
zone flag made the tests fail. No wire bytes changed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
WorldPackets now holds what a client sends a world server, one struct per
packet with Serialize/Deserialize: Validation, CharacterListRequest,
CharacterCreateRequest, CharacterLoginRequest, GameMessage,
CharacterDeleteRequest, CharacterRenameRequest, LevelLoadComplete,
PositionUpdate, MailPacket, RoutePacket, StringCheck, GeneralChatMessage,
HandleFunness, UIHelpTop5. WorldServer's switch became a dispatch map of
handlers (each overrides Handle in WorldServer.cpp, logic unchanged:
dashboard hooks, LoadPlayer, chat logging, migration checks, message
inspector capture all still run); a packet that does not deserialize is
logged and dropped. UserManager's create/delete/rename take the structs.
The answers are ClientPackets (the CLIENT service): LoadStaticZone,
CharacterListResponse, CharacterCreateResponse, CharacterRenameResponse,
DeleteCharacterResponse, TransferToWorld, ServerStates, CreateCharacter,
ChatModerationString, MakeGMResponse, HTTPMonitorInfoResponse and
DebugOutput, built at the call sites (world server, UserManager, slash
commands, dashboard actions, components, migration). The world -> chat
forward of a routed packet is ChatPackets::RoutedFromClient. All
WorldPackets::Send* functions, HTTPMonitorInfo and the ClientPackets parse
functions are gone. The architecture doc now states the file rule: a
packet lives in the file of the ServiceType in its header.
No wire change. Verified against frozen verbatim copies of the old code
(tests/dGameTests/dNetTests/Legacy/WorldPacketsLegacy.h): every response
is sent through the old function and the struct over grids of inputs
(all enum values, strings of every length class, IDs, >64 moderation
segments, big XML) and must match byte for byte; every request is read
by the old code and the struct and must give the same values; plus
golden bytes, round trips, truncation checks and a field width mutation
that made the tests fail. Only differences: malformed requests are
dropped instead of handled with partial values, and LevelLoadComplete now
reads the zone ID the client sends after it (lu_packets and captures show
the 1.10.64 client always sends it; DLU ignored it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The login's Stamps now go with auth's REQUEST_ZONE_TRANSFER to master and
come back in REQUEST_ZONE_TRANSFER_RESPONSE, so master stamps the steps it
performs itself, with its own time: got the request
(WORLD_PACKET_RECEIVED, zone), the world is still starting and the request
waits (IM_LOGIN_QUEUED, instance), answered with a world
(WORLD_SESSION_CONFIRM_TO_AUTH, instance). Auth sends what comes back in
the login response.
Server to server wire change (message IDs unchanged): both messages end
with a Stamps list (u32 16 * count + 4, then the stamps); it is empty
(4 bytes) when a world server asks for a transfer. Touched:
- MasterPackets::SendZoneTransferRequest / SendZoneTransferResponse: take
and write the stamps (default empty)
- MasterServer.cpp REQUEST_ZONE_TRANSFER: reads them, stamps, keeps them
in the PendingInstanceRequest (new stamps member, InstanceManager.h)
- InstanceManager::ReadyInstance / AffirmTransfer: stamp and send them back
- ZoneInstanceManager: HandleRequestZoneTransferResponse reads them; a
RequestZoneTransfer overload takes stamps and a callback that gets them
- AuthPackets LoginRequest::Handle uses that overload
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The login response's stamps were a fixed list: START and CLIENT_OS at the
start, an ERROR or two, and IM_LOGIN_START / WORLD_COMMUNICATION_FINISH
appended to every response whatever happened. Now a Stamps list (new
dNet/Stamps.{h,cpp}: eStamps, Stamp and Stamps with Serialize /
Deserialize in the login response's layout, so server messages can carry
it too) is created when the login request arrives, and each step adds its
own stamp, with the time, where it happens: the request read (value: the
client's OS), the account lookup (found or not), the password check,
failed checks, the closed server and play key checks, database writes,
asking master for a world, master's answer, handing the session key to
master, and sending the player on. The response serializes whatever was
stamped; the auth server logs the list like the client does.
What the client does with stamps (1.10.64): PacketHandler_MSG_CLIENT_
LOGIN_RESPONSE @ 00b32f90 reads them (LoginResponse::ReadStamps @ 005ed3c0,
count = (size - 4) / 16) and only logs each with its name from
StampLookup @ 017e6e88 (the 39 eStamps names) and its time relative to the
first and previous stamp. Documented in Stamps.h.
Behaviour change: on success, master gets the session key just before the
client gets the response instead of just after, so the handoff can be
stamped (and master knows the key before the client can reach a world).
The client-facing layout is unchanged; tests pin the stamp bytes, check
the steps of a failed login in order, and round trip the list.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds CommonPackets (ServiceType::COMMON): ClientVersionConfirm (the
client's handshake, with the handler that logs it and answers),
ServerVersionConfirm, DisconnectNotify and GeneralNotify (layout from the
1.10.64 client's PacketHandler_MSG_SERVER_GENERAL_NOTIFY; DLU does not send
it yet). AuthPackets::LoginRequest reads the login and keeps the old login
logic in its Handle; ClientPackets::LoginResponse (with the Stamp list)
replaces the hand written response, and AuthPackets::SendLoginResponse
fills it from the settings as before. dServer::Disconnect writes a
DisconnectNotify. AuthServer's if-chain and WorldServer's COMMON case
become CommonPackets::Handle / AuthPackets::Handle, dispatch maps that
read the struct, drop and log it if it does not deserialize, then Handle.
HandleHandshake and SendHandshake are gone.
No wire change. Verified against a frozen verbatim copy of the old
functions (tests/dGameTests/dNetTests/Legacy): the old handshake and login
handlers and the new dispatchers are fed the same packets and must send
identical bytes to the same address (same RNG seed for the session key),
over grids of versions, service types, ports, response codes, error
messages, IPs and stamp lists; plus hand computed golden bytes, round
trips, truncation checks, and a deliberate field width mutation that made
the tests fail. The only behaviour difference: truncated handshake and
login packets are now dropped instead of handled with whatever was read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
