The synchronous client-side SendHolePunchingMessages and the server
worker each implemented the fw_punch config reading, the target
resolution and the message pacing. Move them to a
StunClient::HolePuncher class which sends due messages without
blocking, and express the blocking client-side variant as a loop over
it. The single-message StunClient::SendHolePunchingMessage is no
longer needed.
As a side effect, the client no longer sleeps for fw_punch.delay after
the last message, and skips sending if the server address cannot be
resolved instead of punching an unresolvable address.
Requested by Phosit in the review of #8977.
CNetServer::SendHolePunchingMessage is called on the main thread (from
the lobby's XMPP handler) whenever a lobby client requests to connect.
It called StunClient::SendHolePunchingMessages, which sleeps for
fw_punch.delay (default 200 ms) after each of the fw_punch.num_msg
(default 3) messages. Freezing the main thread for ~600 ms freezes the
hosting player's game, which in turn delays the lockstep turns of every
player in the match.
Instead, queue the request to the network server worker thread (like
lobby auths) and pace the messages from CNetServerWorker::RunStep
without sleeping. As the worker now owns the whole exchange, this also
removes the concurrent use of the server's ENetHost from two threads.
Punching stops as soon as the peer connects, which also gives
num_msg = -1 (send indefinitely) a sane meaning.
Fixes: #7957
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
When using the lobby, after constructing `CNetServer` the public IP has
to be queried. The consructor knows whether the lobby is used, so do it
in the constructor.
The init-attributes can only be changed before the worker runs. It isn't
used before the worker runs. This changes it so that it has to be passed
when the server starts.
With this the `m_InitAttributesQueue` can be removed.
CNetClient and CNetServer are constructed for a single match. The
password of a game can not be changed after creating the match. When the
password is passed to the constructor it's not possible to invoke the
functions in the wrong order and the `ENSURE` can be "checked at
compile-time" so to say.
Make include-what-you-use happy with some files in source and fix what
needs to be fixed.
Ref: #8086
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Make include-what-you-use happy with files in source/network and
fix what needs to be fixed.
Ref: #8086
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Previously the param type was `void*` now the type has to be specified
as template parameter of the `CFsm`.
With this commit some casts can be removed.
Shrinking GCs dump the JITted code, which leads to redundant recompilations, lowers performance, and makes profiling JS more difficult.
They may still happen if the runtime is at risk of OOM.
The UPnP implementation included a combination of two subtle bugs, which
resulted in failure to create port forwardings every time after the
first one.
When using UPnP, the internet gateway to create the port forwardings at
needs to discovered. As that takes a while, the its root descriptor URL
was supposed to be cached after successful discovery in the user config
in "network.upnprootdescurl". However, instead of caching the root
descriptor URL, the control URL got cached. That caused following
requests to the root descriptor URL to fail, as they ended up at the
control URL instead. As such requests might also fail when the network
topology changed, the code was supposed to fall back to discovering the
internet gateway again when the cached one didn't work. However, due to
the inner workings of miniupnpc the request using the cached root
descriptor URL didn't result in an error, so the new discovery was never
triggered. As the wrong value was persisted in the user config there was
also no way to get out of this situation again.
This commit fixes both of these bugs.
As far as I can tell these bugs existed since the introduction of the
caching of the root descriptor URL in 0ba25e9968, which means creating
port forwardings using UPnP has been broken since Alpha 15.
When adding a port forwarding via UPnP, pyrogenesis first needs to
discover the internet gateway to create the forwarding at. To do so, it
utilizes miniupnpc to send an SSDP request and waits 10 seconds until it
processes the responses and continues to add the port forwarding.
As that happens asynchronously it means if players hosting games rely on
UPnP for port forwarding and if their pyrogenesis instance doesn't have
the address of their internet gateway cached, it takes more than 10
seconds after starting to host a game, before players can join. However,
waiting for 10 seconds is completely unnecessary, because the internet
gateway responds immediately to the SSDP request. Therefore, this commit
reduces the time to wait for responses from 10 seconds to 2 seconds,
which is the timeout miniupnpc uses internally as well, when discovering
internet gateways.
Up to know the UPnP logic ignored the port a user was hosting a game on
and always added a port forwarding for the default port UDP 20595. This
commit fixes that, so a port forwarding is added for the actual port a
game is hosted on.
This makes using STUN mandatory for games hosted using the multiplayer
lobby. The motivation for that is a reduction in complexity, because
right now if STUN is disabled we use a home-grown STUN-like logic, which
got implemented before Pyrogenesis got STUN support.
That home-grown logic relies on a custom ejabberd module (mod_ipstamp),
which inserts the external IP-address of a host in the response messages
when a host registers a game. Originally mod_ipstamp was also used to
inform all potential players of a hosts IP-address, however that has
already been removed to let hosts to only share their IP-address with
players actually joining their game.
Removing the home-grown logic and instead always relying on STUN removes
complexity in Pyrogenesis and the lobby server and also eases hosting
games for players, as they don't have to figure out anymore whether they
need to enable STUN or not.
These changes shouldn't negatively impact the ability of Pyrogenesis to
handle different types of NAT or broken networks. There is one
difference though: While the custom logic using mod_ipstamp utilized TCP
as transport protocol, the STUN implementation in Pyrogenesis currently
uses UDP. That doesn't allow hosts with UDP-connectivity issues to
resolve their external IP-address anymore, however without
UDP-connectivity they aren't able to successfully host games anyway, as
the actual game updates are transferred using UDP as well.
Only the sender and the recipients receive the chat messages.
This commit only has an affecto on messages where the addressee(s) are
selected through the dropdown. Addressee(s) selected with a "/" command
are still sent to evevyone and filteret by the receiver.
This patch addresses issues concerning a1796ed71f:
Allow for a more elegant implementation of observer flares.
And still display flares even if the sender is lagging behind:
Split off flares from simulation commands.
Remove the new, problematic 'observer commands' entirely.
Provide an engine function 'SendNetworkFlare' to the JS interface.
-> which sets off the (pretty ordinary) transmission process.
Add a new type of net messages exclusively for flares
-> contains the flare's position and its sender's GUID.
This replaces all links pointing to trac with their corresponding links
to gitea. Also replace http with https while at it.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
The user doesn't have to fiddle with `std::shared_ptr`.
And two (more unrelated) things: use `std::unordered_map`, use a
`std::find_if` in the callback.
Comments By: @vladislavbelov, @Stan
Differential Revision: https://code.wildfiregames.com/D5239
This was SVN commit r28048.
The `StateSet` is only used to check if a state is valid. That's useless
since it wouldn't be found in the `TransitionMap` - With the same
outcome.
Accepted By: @vladislavbelov
Differential Revision: https://code.wildfiregames.com/D5236
This was SVN commit r28039.
Avoid cases of filenames
Update years in terms and other legal(ish) documents
Don't update years in license headers, since change is not meaningful
Will add linter rule in seperate commit
Happy recompiling everyone!
Original Patch By: Nescio
Comment By: Gallaecio
Differential Revision: D2620
This was SVN commit r27786.
This fixes packet loss issues on some VPN solutions.
Patch By: sera
Differential Revision: https://code.wildfiregames.com/D4967
This was SVN commit r27599.
Switch some logic from C++ to JS in PREGAME for player assignments. Refs
#3049Fixes#6204
Reported by: Imarok
Tested By: Imarok
Differential Revision: https://code.wildfiregames.com/D4092
This was SVN commit r25699.
The purpose of our client-side hashing for lobby game passwords is to
prevent malicious hosts from getting valuable passwords from clients
(e.g. accidentally typing their lobby password instead of the game, or
even their email password, etc).
However, the hashing was deterministic (and rather simple), making it
possible to compute rainbow tables and recover user passwords anyways.
By adding more variation, including some that cannot so easily be
controlled by the host (the client name), this becomes impractical. The
password hashing function used is rather fast, but given the base low
probability of mistypes, this seems fine.
Differential Revision: https://code.wildfiregames.com/D3459
This was SVN commit r25459.
Instead of using platform-specific sockets, use enet_socket* functions
(which ends up doing the same).
Clean up some confusing APIs, removing the distinction between finding
the public IP for the host/join.
Fix endianness support & use simpler code.
Refs D364 / 61261d14fc (and some subsequent fixing diffs).
Differential Revision: https://code.wildfiregames.com/D3970
This was SVN commit r25453.
This allows joining a lobby game hosted on the same network (behind the
same NAT gateway).
This is relatively primitive to keep things simple: if the server and
the client have the same public IP, it is assumed that they are on the
same network and the client instead requests the local IP.
Differential Revision: https://code.wildfiregames.com/D3944
This was SVN commit r25448.
Follows 34b1920e7b.
JSON functions and ToString are movec to their own headers.
Also clean out a few PersistentRooted usage to use the 2-phase init to
clean up scriptInterface usage.
With these functions split off, we can finally clean out headers and
remove ScriptInterface.h from most of them, in favour of smaller and
more precise headers.
Take the opportunity to clarify some comments regarding Mutability.
Differential Revision: https://code.wildfiregames.com/D3961
This was SVN commit r25434.
Follows 34b1920e7b.
This splits off the object-related functions, such as
[Set/Get/Has]Property, CreateObject, CreateArray, FreezeObject.
It also puts the definitions in the header itself, which might end up
with faster code here & there, though perhaps slower compilation time
(somewhat doubtful since we already included most things anyways).
Differential Revision: https://code.wildfiregames.com/D3956
This was SVN commit r25430.
Observers no longer lag the game for players. There is still some time
to serialise the game when sending it to a joining observer, and
depending on the chosen 'max lag' the game may stop while observers
sufficiently catch up, but this impact too is reduced.
- Make the NetServerTurnManager ignore players marked as 'observers' for
the purpose of ending a turn, effectively making it possible for
observers to lag without it affecting the players in any way.
- Add a config option (network.observermaxlag) that specifies how many
turns behind the live game observers are allowed to be. Default to 10
turns, or 2 seconds, to keep them 'largely live'.
- The controller is not treated as an observer.
- Implement a simple UI to show this delay & allow the game to speed up
automatically to try and catch up. This can be deactivated via
network.autocatchup.
- Move network options to the renamed 'Network / Lobby' options page.
- Do not debug_warn/crash when receiving commands from the past -
instead warn and carry on, to avoid DOS and "coop play" issues.
Refs #5903, Refs #4210
Differential Revision: https://code.wildfiregames.com/D3737
This was SVN commit r25156.
