The AMF3 reader read an array inside an array to any depth and kept
reading values after the data had ended. It now refuses arrays nested
more than 64 deep and throws when the data ends before a value does.
Co-Authored-By: Claude Fable 5.2 <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>
Client-sent AMF (CONTROL_BEHAVIORS) was decoded without limits on the
associative part of arrays, on nesting depth, or on the total number of
values. Associative entries went into an unordered_map with the standard
unseeded string hash, so a client could pick colliding keys and make
insertion quadratic, and deeply nested arrays recursed until the stack
overflowed, crashing the world server.
- The associative map is now an ordered std::map (O(log n) whatever the
keys, deterministic serialization order).
- Each array allows at most 10,000 associative entries, the same as the
existing dense limit, which is now checked before anything is read.
- Arrays may nest at most 32 deep and one deserializer reads at most
100,000 values. Every limit throws, which the only caller already
catches and drops the message.
- Inserting a duplicate key keeps the last value and no longer returns a
reference to a value that was destroyed when the key already held null.
Verified with new unit tests for each limit (including a 100,000 deep
nesting that previously overflowed the stack) and the existing live
packet test, which still decodes.
Fixes#2035
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix: security vulnerabilities
Tested that all functions related to the touched files work
will test sqlite on a CI build
* fix failing test
* ai feedback
* add buffer size checking
* use c_str
* dont log session key
* Try this for a mac definition
* be quiet apple
* Rewrite AMF and behavior logic to use smart pointers, references, and string_views over raw pointers and std::string&
* fix m_BehaviorID initialization
* Fix BlockDefinition member naming
* remove redundant reset()s
* Replace UB forward template declarations with header include
* remove unneeded comment
* remove non-const ref getters
* simplify default behavior id initialization
* Fix invalidated use of Getter to set a value
* Update AddStripMessage.cpp - change push_back to emplace_back
* fix pointer to ref conversion mistake (should not have directly grabbed from the other branch commit)
* deref
* VERY experimental testing of forward declaration of templates - probably will revert
* Revert changes (as expected)
* Update BlockDefinition.h - remove extraneous semicolons
* Update BlockDefinition.h - remove linebreak
* Update Amf3.h member naming scheme
* fix duplicated code
* const iterators
* const pointers
* reviving this branch
* update read switch cases
* Add MSVC optimization flags
* test moving flags to json
* Update CMakePresets.json
* testing
* trying more variations on the flags
* third test
* testing if these even have any effect
* ditto
* final(?) try for now
* ONE MORE TIME
* trying 'init' flags instead
* export the compile commands so I can see if they're having any effect
* move out g++ O2 flag
* add Linux debug preset
* update CMake presets
* edit macos presets
* try adding build types back to mac
* macos refuses to work :(
* try using compiler flags for mac instead
* fix typo in windows preset
* build reorganization and experimental clang support
* temporarily remove macos build for testing purposes
* updated cmake workflows
* unexclude toolchain dir
* update .gitignore
* fix build directory issue
* edit build script
* update cmake configs
* attempted docker fix
* try zero-initializinng this struct to solve docker issue
* try fixing macos build
* one last MacOS try for the night
* try disabling an apple-specific build rule
* more fiddling with mac test builds
* try and narrow down the macos build failure cause
* try stripping out all the custom macos test logic again
* I'm really just throwing everything to the wall and seeing what sticks
* more macos tinkering
* implib
* try manual link directory specification
* save me
* aaaaaaaaa
* paths paths paths
* Revert "paths paths paths"
This reverts commit 9a7d86aa6c.
* Revert "aaaaaaaaa"
This reverts commit 338279c396.
* Revert "save me"
This reverts commit bd73aa21a9.
* Revert "try manual link directory specification"
This reverts commit 0c2d40632e.
* Revert "implib"
This reverts commit d41349d6ed.
* Revert "more macos tinkering"
This reverts commit 829ec35b57.
* Revert "I'm really just throwing everything to the wall and seeing what sticks"
This reverts commit 1a05b027fe.
* Revert "try stripping out all the custom macos test logic again"
This reverts commit cc15a26ce8.
* Revert "try and narrow down the macos build failure cause"
This reverts commit 5fd86833fa.
* Revert "more fiddling with mac test builds"
This reverts commit 0f843c02c9.
* Revert "try disabling an apple-specific build rule"
This reverts commit 45ec66e976.
* back to debug messages
* see if this re-breaks mac
* are these messages actually somehow fixing the issue?
* was not actually fixed
* add debug messages (again)
* debug try 2
* change runtime output dir
* rename gcc to gnu
* expand cmake presets
* fix preset
* change defaults
* altered cmake configuration scripts
* disable /WX on MSVC
* update github actions
* update build presets
* change gnu and clang build directories to enable consistent artifact generation
* add RelWithDebInfo presets and move -Werror flag into presets.json
* use DLU_CONFIG_DIR envvar
* CMakePresets indentation
* temp fix for MSVC debug builds
* Logger: Rename logger to Logger from dLogger
* Logger: Add compile time filename
Fix include issues
Add writers
Add macros
Add macro to force compilation
* Logger: Replace calls with macros
Allows for filename and line number to be logged
* Logger: Add comments
and remove extra define
Logger: Replace with unique_ptr
also flush console at exit. regular file writer should be flushed on file close.
Logger: Remove constexpr on variable
* Logger: Simplify code
* Update Logger.cpp
* Update AMFDeserializeTests.cpp
Redo Amf3 functionality
Overhaul the whole thing due to it being outdated and clunky to use
Sometimes you want to keep the value
Update AMFDeserializeTests.cpp
* Fix enum and constructors
Correct enum to a class and simplify names.
Add a proper default constructor
* Update MasterServer.cpp
* Fix bugs and add more tests
* Refactor: AMF with templates in mind
- Remove hard coded bodge
- Use templates and generics to allow for much looser typing and strengthened implementation
- Move code into header only implementation for portability
Refactor: Convert AMF implementation to templates
- Rip out previous implementation
- Remove all extraneous terminology
- Add proper overloads for all types of inserts
- Fix up tests and codebase
* Fix compiler errors
* Check for null first
* Add specialization for const char*
* Update tests for new template specialization
* Switch BitStream to use references
* Rename files
* Check enum bounds on deserialize
I did this on a phone
Implement GTest as a testing infrastructure.
Make windows output binaries to the build folder instead of the release type folder (potentially issue further down the line)
Add a simple unit test for DestroyableComponent
