The moderator account commands still reached Servatrice_DatabaseInterface through
its concrete class: cmdResetUserPassword ran the privilege lookup inline and
called the concrete-only changeUserPasswordWithLegacyBackup, while the avatar,
banner and blacklisted-banner commands called concrete-only helpers. Promote
those four methods to Server_DatabaseInterface, move RemoveUserBannerResult into
the interface, and add getUserAdminFlags for the reset privilege lookup. The
escalation guard, audit records, had_banner response field and every response
code stay in the handler.
cmdAccountEdit assembled a dynamic UPDATE for the real name, email and country
columns in the socket handler, and cmdAccountImage wrote the avatar blob there
too. Add AccountUpdateRecord, updateUserAccount and setUserAvatar to
Server_DatabaseInterface with fail-closed defaults, implement them with the
previous query text and bind values, and keep the password gate, the email
permission check, the empty-edit early return and the in-memory userInfo updates
in the handler.
cmdBanFromServer, the admin-flag helpers, cmdForceActivateUser and the
admin-notes commands still prepared their statements inline through
Servatrice_DatabaseInterface, and addWarning was only reachable through the
concrete Servatrice class. Add setAdminFlag, getUserToken, getAdminNotes,
setAdminNotes, addBan and getUserNamesByClientId to Server_DatabaseInterface
with fail-closed defaults, promote addWarning, and model the outcomes as
UserFieldResult and AdminFlagUpdateResult. Notifications, the audit record,
the ban disconnect fan-out and every response code stay in the handler.
cmdReportAssign, cmdReportResolve, cmdReportReopen and cmdReportAddComment still
ran their lookups and updates inline through Servatrice_DatabaseInterface. Add
assignReport, resolveReport, reopenReport, getUserNameById,
getReportNotificationTarget, countReportCommentsInLastHour,
getReportCommentTarget and addReportComment to Server_DatabaseInterface with
fail-closed defaults, and model the outcomes as ReportAssignResult and
ReportUpdateResult so "no such report", "wrong state" and "query failed" stay
distinguishable. Authorization checks, audit records, notification recipients
and every response code stay in the handler.
cmdReportStats and cmdReportUserInfo still ran their queries inline through
Servatrice_DatabaseInterface and built the response in the socket handler. Add
ReportStatsRecord, ReportUserInfoRecord and their count records to
Server_DatabaseInterface with fail-closed defaults, implement getReportStats and
getReportUserInfo with the previous statements and query order, and keep
presence flags for the fields the old code set only on a successful query so
the proto2 wire output is unchanged. The stats cache decision and every response
code stay in the handler.
cmdReportList, cmdReportMyList and cmdReportDetails still prepared their SELECTs
inline through Servatrice_DatabaseInterface and mapped the result columns to
protocol records in the socket handler. Add getReportList,
getReportsSubmittedBy, getReportForDetails and getReportComments to
Server_DatabaseInterface with fail-closed defaults and a ReportLookupResult
outcome, implement them with the previous statements, joins and null handling,
and keep the detail authorization check and the tolerated comment-query failure
in the handler in the same order. Response codes are unchanged.
cmdReport and sendPendingReportNotifications still issued their SQL inline
through Servatrice_DatabaseInterface: the daily report cap, the reported-user
lookup, the report insert, both pending-notification selects and the notified
flips. Add the two notification records and seven report methods to
Server_DatabaseInterface with fail-closed defaults, implement them with the
previous statements and binds, and let both functions ask the injected
interface. Check order, response codes and notification payloads are unchanged.
cmdForgotPasswordRequest, continuePasswordRequest, cmdForgotPasswordReset,
cmdForgotPasswordChallenge, cmdRequestPasswordSalt and cmdSubmitPasswordVerifier
still called Servatrice_DatabaseInterface, and two of them validated the account
token and email through the generic validateTableColumnStringData helper. Add
doesForgotPasswordExist, addForgotPassword and getUsableLegacyBackup to
Server_DatabaseInterface with fail-closed defaults, replace the table/column
probes with userTokenMatches and userEmailMatches using the same statement and
case-insensitive comparison, and let the commands ask the injected interface.
Audit records, check order and every response code are unchanged.
cmdRegisterAccount and cmdActivateAccount still called
Servatrice_DatabaseInterface for username validation, user existence, the
per-email account limit, ban checks, registration and activation, and
cmdRegisterAccount issued the activation-email insert inline. Add
addActivationEmail to Server_DatabaseInterface with a false default, implement
it with the same statement and bind value, and let both commands ask the
injected interface. Validation order and every response code are unchanged.
cmdGetServerStats reached through Servatrice_DatabaseInterface for checkSql and
getLatestUptimeSnapshot, a concrete-only method. Move the UptimeSnapshot record
to Server_DatabaseInterface with an invalid default for non-SQL stores, mark the
Servatrice implementation as an override, and let the command ask the injected
interface. The statement, its bind value and the invalid-snapshot response are
unchanged.
createHashForReplay, cmdReplayGetCode, cmdReplaySubmitCode,
cmdReplayDownloadByGameId and cmdGrantReplayAccess still issued their own SQL
through Servatrice_DatabaseInterface. Add checkReplayAccess, getReplayName,
getReplayHashSource, addReplayAccess, getLatestReplayForGame and
grantReplayAccessByName to Server_DatabaseInterface, implement them with the
same statements, bind order and response mapping, and let the commands ask the
injected interface. Hashing and code parsing stay in the handler; the shared
insert is reused by both grant paths.
cmdReplayList, cmdReplayDownload, cmdReplayModifyMatch and cmdReplayDeleteMatch issued their own SQL through Servatrice_DatabaseInterface. Add ReplayMatchRecord/ReplayEntry and the ReplayAccessResult/ReplayWriteResult enums to Server_DatabaseInterface, implement the list, data, do-not-hide and delete operations with the same statements and response mapping, and let the commands ask the injected interface. checkSql stays in the two write commands where it already was.
The five deck-share commands reached through Servatrice_DatabaseInterface for share lookups, and cmdDeckShareCreate still issued the per-user rate-limit count and the folder deck listing inline. Move DeckShareItemRecord and DeckShareSummaryRecord to Server_DatabaseInterface, declare the share operations plus getDeckSharesCreatedToday and getDeckFilesInFolder there with failing defaults, and switch the commands to the injected interface. The MySQL statements and every response code stay as they were, including the -1-on-failure sentinel for the count.
addAuditRecord existed only on Servatrice_DatabaseInterface, so 37 moderator and admin commands reached through the concrete class just to append a log row. Declare it on Server_DatabaseInterface with an empty default, mark the Servatrice implementation as an override, and switch the handler's call sites to the injected interface. No SQL moves; the argument hashes are unchanged.
Deleting a deck, inserting an uploaded deck, and updating one in place were SQL statements in the socket handler, including the dynamic is_public clause and the follow-up visibility read. Add DeckFileData and DeckFileWriteResult to Server_DatabaseInterface with create/update/delete operations, implement them in Servatrice_DatabaseInterface with the same statements and bind order, and let cmdDeckUpload and cmdDeckDel ask the injected interface.
Setting a deck or folder public, creating a folder, and deleting a folder tree were SQL statements in the socket handler. Declare them on Server_DatabaseInterface with a DeckStorageResult so the caller can still tell a missing row from a failed write, implement them in Servatrice_DatabaseInterface with the same statements, and let the three folder commands ask the injected interface. Readiness probing keeps its previous ignore-the-result shape.
Resolving a deck folder path, reading a deck's owner, and walking its folder chain for public visibility were SQL queries living in the socket handler, keyed to the logged-in user through userInfo. Declare the three lookups on Server_DatabaseInterface, implement them in Servatrice_DatabaseInterface with the user id as an explicit parameter, and let the deck commands ask the injected interface.
getMessageLogHistory, getPrivateChatHistory, and getUserIdInDB were reachable only through Servatrice_DatabaseInterface, so the two log-history commands in the socket handler needed the concrete class. Declare the two readers on Server_DatabaseInterface with empty/false defaults, mark the Servatrice implementations as overrides, and resolve the private-chat peer through the already-promoted getUserId.
The ban/warn history, session, alt-account, and moderator-login readers existed only on Servatrice_DatabaseInterface, so the socket handler had to hold the concrete class to call them and to ask checkSql() whether the store was reachable. Declare them on Server_DatabaseInterface with empty-list defaults and a checkSql() that reports ready for stores without a connection, mark the Servatrice implementations as overrides, and let those five commands ask the injected interface.
cmdAddToList/cmdRemoveFromList issued raw SQL against the buddy and ignore tables and looked accounts up through getUserIdInDB, a method that only exists on the MySQL implementation. Add getUserId plus add/remove operations to Server_DatabaseInterface, implement them in Servatrice_DatabaseInterface over the shared single-row helpers, and let the handler ask the injected interface instead of the concrete class.
The socket-handler constructor already takes Servatrice_DatabaseInterface *, so the reinterpret_cast from the base member to the same type adds nothing and would silently hide a wrong implementation if a different database type were ever injected. Initialize sqlInterface from the typed constructor parameter instead.
* [Server] Let users fetch their own private chat history
Players lose their private conversations on reconnect or on another
device. Servatrice can already log private messages, but only
moderators can read the logs.
Add Command_GetPrivateChatHistory / Response_PrivateChatHistory so a
registered user can page through their own logged conversation with one
other user. Operators opt in with logging/allow_user_chat_history
(default off, and log_user_msg_chat must be on). The caller is matched
by account id so reused guest or deleted-account names can't read older
messages; only time, sender name and text are returned. Pages hold
1..200 messages and never split a second, since log_time has 1-second
resolution.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [Server] Harden private chat history identity matching
Match both participants by account id and name so a federated user's remote id cannot collide with a local account, and stop storing foreign ids in the local chat log. Extract the paging assembly into a pure helper and cover the whole-second boundary rules with gtests.
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Lukas Brübach <44814898+BruebachL@users.noreply.github.com>
* [Alts] Server-side search controls, IP counts, adjustable window
Plumbs the match-type chips through to the server: include_email/include_clientid/include_ip skip disabled identifier types in both matching and frontier expansion, so excluded types (e.g. throwaway domains) never enter the graph at any depth. Session lookback becomes adjustable (session_months, clamped 1-24, default 6). Each alt row carries ip_count, the distinct recent session IP count shown in a new IPs column. All proto additions are optional with old defaults, so old clients/servers interoperate.
* Fix remaining review threads on the alts search controls
- Disable keyboard tracking on the depth and window spin boxes so typing a
two-digit value sends one search, not one per digit.
- Reload the user under investigation on a filter change instead of reading
the search box, which could switch to a name never submitted.
- Keep a chip's last match count while it is unchecked; the server omits
unchecked match types, so the label used to drop to "(0)".
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Server] Recursive alt search with match evidence
getUserAlts() now walks a breadth-first graph to the requested depth (clamped 1-3): each level expands frontier accounts over registration email, current client id, recent session IPs (6-month window), and historic session client ids, so identifier changes no longer hide older alts. Every row carries match_reason (email / clientid:<id> / ip:<addr>), depth, and shared_count fan-out so moderators can distinguish real alts from VPN/shared-network overlap. Command_GetUserAlts depth is plumbed through; unset depth defaults to 1, preserving previous single-hop behavior for old clients.
* [Server] Fix alt search review findings
- Seed the BFS with the stored user name so the searched account is not
re-listed as its own alt at depth 1.
- Normalise emails to lower case for the membership and fan-out checks;
SQL compares them case-insensitively.
- Widen the per-level row limit by the visited count so level 2+ cannot
return only already-known accounts.
- Prepare dynamic IN-list queries locally instead of through the shared
prepared-statement cache, whose keys are query texts.
- Count historic client-id fan-out from _sessions, not from the current
_users ids.
- Order capped identifiers by relevance (current email/client id first,
then most recent session IPs) instead of alphabetically.
* Apply suggestion from @ZeldaZach
Co-authored-by: Zach H <zahalpern+github@gmail.com>
* Apply suggestion from @ZeldaZach
Co-authored-by: Zach H <zahalpern+github@gmail.com>
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
Co-authored-by: Zach H <zahalpern+github@gmail.com>
Compiler flags are applied once and for the right compiler:
- Detect the toolchain by CMAKE_CXX_COMPILER_ID (MSVC / GNU / Clang,
warn on an unknown one).
- MSVC: drop /std:c++20 (CMAKE_CXX_STANDARD already adds it) and the
outdated C4251 suppression, add /utf-8, and use /O2 instead of /Ox.
- GCC: drop the no-op -Wno-long-long and the macOS-only -std=gnu++20
override; CMAKE_CXX_EXTENSIONS=OFF now owns the dialect.
- Use add_compile_definitions for SFMT_MEXP instead of add_definitions.
CI build script:
- Only run the schema version check for server builds.
- Accept boolean env flags as 1 and say so in the header.
- Print ccache, ctest and cpack versions next to their log groups.
- Only build the macOS vcpkg triplet when vcpkg is actually used.
- Drop the extra cmake --build verbosity.
- Install NSIS without chocolatey progress noise.
Co-authored-by: toomm <toomm@users.noreply.github.com>
* [Servatrice] Use compile-time checked connections
Replace the remaining SIGNAL/SLOT string connects in servatrice.cpp,
smtpclient.cpp, serversocketinterface.cpp and isl_interface.cpp with
pointer-to-member or lambda connects, and convert the string invokeMethod
calls to the functor overload.
QxtSmtp has overloaded signals, so those need qOverload. Slots inherited as
protected members and the QWebSocket error connection owned by the ISL fix
are left on lambdas/strings as appropriate; the vendored qxtsmtp.cpp is
untouched.
* Fix remaining review threads on the connection syntax
- Connect the IslInterface signals straight to the inherited Server slots
instead of forwarding lambdas.
- Capture servatrice explicitly in the queued admin lambdas so they do not
read this->servatrice after teardown.
- Move the websocket newConnection connect out of the per-pool loop.
- Terminate the logging category and restore pendingKey's indentation.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
checkUserPassword ignored force_password_change on the challenge-response
path, so a reset password stayed valid. Returning PasswordChangeRequired
was not enough on its own: the session was dropped and the client told the
user to ask the administrator for the temporary password they had just used.
The pending session now keeps its identity without registering a user
session or announcing a join, may only ping and change its password, and
the client completes the change in-session before reconnecting.
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
The temporary password was written as a legacy salt+hash row, which makes the
next login look like an authentication downgrade. A client holding a saved
verifier refuses that downgrade, so the reset locked the account out of every
modern client.
Store the verifier and, where security/legacy_backup_key is configured, keep a
sealed legacy hash of the same temporary password as the downgrade backup, so
pre-challenge clients keep the credential path the reset used to provide.
Without the key no row shape serves both: the verifier and the legacy hash
cannot share one column, so those clients have no path at all. Warn at reset
time that the new password has no pre-challenge path, and say so in
servatrice.ini.example instead of promising that a reset gets them back in.
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [CardDB] Drop the CardInfo self pointer
The weak self pointer had to be assigned right after every construction,
and a missed assignment silently stopped change notifications. CardInfo
is always owned by a QSharedPointer, so it now derives from
QEnableSharedFromThis<CardInfo> and asks for sharedFromThis() instead:
nothing to initialize, and the CardSet cycle stays broken.
CardSet::newInstance() also loses the commented-out self-pointer
assignment left over from the same design.
* [CardDB] Pin the cardInfoChanged contract without the self pointer
Dropping CardInfo's self pointer moved change notifications onto
QEnableSharedFromThis, so the signal now rests on the card being owned by
a QSharedPointer. Two tests pin that down:
- A property change on a card owned by a CardDatabase still emits
cardInfoChanged() carrying that card, which is what the card database
views and models listen to.
- clone() starts out unowned, so a change to the clone is reported as the
clone's and must not reach the original's observers.
* [CardDB] Treat an unowned CardInfo as a bug, not a no-op
Every CardInfo is built inside a QSharedPointer (both factory overloads
and clone()), and no constructor notifies anyone, so sharedFromThis() is
always available by the time addToSet() or a setter runs. Assert that and
warn in a released build, instead of silently skipping the notification
or the set append.
The copy constructor keeps its explicit QEnableSharedFromThis
initializer: GCC's -Wextra warns about the uninitialized base in every
translation unit that copies a CardInfo otherwise. A comment there
records that the clone deliberately starts unowned.
* [CardDB] Break the CardInfo <-> CardSet cycle on reload
CardSet is a QList<CardInfoPtr> and CardInfo::addToSet() appends the card
to its sets, while a card keeps a CardSetPtr per printing in
setsToPrintings. The sets therefore hold strong references back to the
cards, and the previous database snapshot stayed alive after a reload, so
RSS grew by roughly a second database on every reload.
Empty the sets before the cards are dropped: CardDatabase::clear() and
swapInDatabaseData() clear the sets before replacing the containers, and
removeCard() detaches the card from its sets, so nothing keeps a card
alive on behalf of a set once the database lets go of it.
Two tests pin the behaviour: a reload and a clear() must drop the last
strong reference to a card from the previous snapshot.
* [CardDB] Keep the set detach in one place
clear() and swapInDatabaseData() both emptied every set before dropping the
cards, so share that as detachSetMembership() instead of two copies that have
to stay in sync.
removeCard() now detaches through the first printing of each set directly, the
same way CardInfo::newInstance() appends, instead of an inner loop that always
breaks after the first item.
Because clear() empties the sets up front, its per-card detach was walking
every card's printings to remove it from sets that were already empty. Split
removeCard() into the per-card detach and unindexCard(), and let clear() call
only the latter.
* [CardDB] Test that removeCard detaches the card from its set
removeCard() is the one teardown path a caller can trigger for a single card
(the token editor uses it), and it now detaches the card from its sets as
well. Without that detach the set's strong CardInfoPtr keeps the removed card
alive forever, which is exactly what the reload and clear tests pin for the
whole-snapshot paths.
The test loads the fixture, removes Cat, and requires the card to be gone from
the CAT set and to be destroyed once the test drops its own reference.
* [Servatrice] Drop the settings cache from the logger
#7602 removed #include "settingscache.h" from server_logger.cpp but left
isLoggingEnabled() reading settingsCache, so master no longer compiles: the
servatrice target fails on the undeclared global and server_logger_test, which
compiles the same file without main.cpp, fails to link it.
The same commit made reloadConfig() the single place that stores writelog and
the log filters, precisely so logging threads stop reading the global QSettings.
isLoggingEnabled() now mirrors logMessage() and reads those members instead: the
console is a sink of its own, and the logfile needs an open file plus writelog.
The mutex is mutable so the const query can take it.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Servatrice] Fix silently broken socket error connections
QAbstractSocket::error() was removed in Qt 6 in favour of errorOccurred(). The ISL interface connected to the old name through SIGNAL(), which compiles but fails at runtime, so catchSocketError never ran. The WebSocket interface used the same name on QWebSocket, where it still exists but is deprecated since 6.5.
Use errorOccurred for both. The WebSocket connect needs a lambda because catchSocketError is a protected base-class slot and a base pointer-to-member is not accessible from the derived class.
* [Servatrice] Guard the WebSocket error connect against Qt 6.4
QWebSocket gained errorOccurred in 6.5 and dropped the error signal with it, so
the typed connect cannot compile against the 6.4 floor that Debian 12 and
Ubuntu 24.04 still use. Keep the typed connect on 6.5+ and fall back to the
old-style connect below it, which is what the code did before.
Verified both branches compile: forcing the guard to the else path and
rebuilding serversocketinterface.cpp succeeds, so the 6.4 spelling is intact.
* [Servatrice] Initialize ISL serverId before authentication
catchSocketError() removes islInterfaces[serverId], but the incoming
connection constructor left serverId uninitialized until initServer()
set it after the SSL handshake and certificate check. A socket error
during waitForEncrypted() therefore read an indeterminate key.
* [Servatrice] Correct the QWebSocket error signal note
error() was deprecated in 6.5, not removed; errorOccurred() was added in
6.5.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Servatrice] Exit non-zero when server initialization fails
initServer() reported every startup failure through a zero exit code: the
database could not be opened, the server printed nothing of its own, and the
process looked like a clean shutdown. Log the two refusal reasons at critical
level and return 1 so supervisors and the Docker exit status can tell the
difference.
* [Servatrice] Keep console logging independent of the logfile
logMessage() returned before the console echo unless a logfile was open and
server/writelog was true, so --log-to-console produced nothing when
server/logfile was empty or could not be opened. That is how a refused database
connection could end in a silent exit.
Print to the console before the logfile gates, drop the duplicate echo from
flushBuffer(), and read writelog/logfilters once in main instead of calling the
non-thread-safe global QSettings from every logging thread.
* [Servatrice] Uninstall the log handler before tearing down globals
The handler reaches into the global logger and settingsCache. Leaving it
installed while main deletes those objects turns any Qt warning emitted during
QCoreApplication/QFactoryLoader teardown into a use-after-free; with
QT_DEBUG_PLUGINS=1 that warning exists, and Servatrice died with SIGSEGV in
QSettings::value() on the way out.
* [Servatrice] Buffer console logs on the logger thread
Address review feedback on the startup failure change:
- Append console lines to a buffer drained by the logger thread, so a
stalled stdout reader cannot block game threads.
- Report islServer->errorString() instead of the literal call text.
- Let the constructor reuse reloadConfig() instead of parsing the filter
list twice.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Build] Use versionless Qt CMake commands
Replace the versioned qt6_add_executable, qt6_add_resources, qt6_add_shaders,\nqt6_add_translations and qt6_wrap_cpp calls with their unversioned forms.\nAll five are documented as available at the project's 6.4 minimum and accept\nthe same arguments.\n\nVerification: full client, Oracle, server and test build passed.
* versionless qt_finalize_target and _project
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
Co-authored-by: tooomm <tooomm@users.noreply.github.com>
* [Network] Do not format discarded debug messages, parse without memmove
Protocol debug logging gets the full protobuf message rendered through
TextFormat and reflection. On the server this happened before every
session/room/game/moderator/admin/developer command, even though the
default Server_ProtocolHandler::logDebugMessage() is a no-op and even
when servatrice has no log file configured. On the clients the same
string was built as an argument to qCDebug(), which evaluates its
arguments regardless of whether the category is enabled.
Server_ProtocolHandler now exposes a virtual logsDebugMessages() hook.
The base implementation returns false, so handlers without a log
destination spend nothing. AbstractServerSocketInterface overrides it
with ServerLogger::isLoggingEnabled(), which checks the log file and the
server/writelog setting that logMessage() itself consults. All debug
message call sites check the hook before formatting. The client
send/receive paths check QLoggingCategory::isDebugEnabled() first, the
same pattern AbstractClient already uses for its slow-command warning.
ISL message decoding does the same.
Byte stream decoding called inputBuffer.remove(0, n) for the length
header and then again for every message. Each call shifts all remaining
bytes, so a read carrying k messages cost O(k * bytes remaining). The
three receive loops (remote client, TCP server interface, ISL interface)
now parse at a read offset and compact the consumed prefix once, when
they stop because a full message is not buffered yet. No buffering or
size-limit behavior changes.
Also in this batch:
- Server_ProtocolHandler::processGameCommandContainer() copied the whole
games map under gameListMutex just to look up one game id. A new
getGameInfo() performs the single lookup and returns the room/player
ids.
- The TCP receive loop's `ok` flag is initialized, so a ParseFromArray
exception can no longer leave it indeterminate before `if (ok)`.
- The exception log prints only the message being parsed instead of
hex-dumping the entire pending buffer.
Full build and ctest (628 tests) pass, including the server tests.
* [Network] Address review comments
- Move getGameInfo() out of the header into server_abstractuserinterface.cpp.
- Drop the consumed prefix once after the parse loop instead of on every
early return: the two "not enough data" exits now break out of the loop.
Applied to the remote client, TCP server and ISL receive loops.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
X-Forwarded-For carries a comma-separated list, and constructing a
QHostAddress from the raw header value returns a null address for it, so a
configured X-Forwarded-For or a multi-hop header never resolved and websocket
clients stayed on the proxy's address.
- Add parseForwardedAddress, returning the rightmost usable entry of a
forwarded header list
- Use it for the websocket handshake instead of parsing the raw value
- Exempt loopback from per-address auth rate limits so a same-host proxy with a
customized trusted_sources list cannot lock every user out
- Add forwarded_address_test
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
trusted_sources was checked with QString::contains. A dual-stack listener
reports a local proxy as ::ffff:127.0.0.1, which never matches the default
"127.0.0.1,::1". Behind nginx the forwarded client IP header was then
ignored, every websocket user shared the proxy's address, and logins
failed with "too many concurrent connections" once max_users_per_address
was reached.
Add isTrustedSource(), which parses each entry as a QHostAddress and
compares exact addresses with IPv4-mapped addresses normalized, and use it
at all four call sites. The ban command also read server/trusted_sources
instead of the documented security/trusted_sources; it now uses the same
getter. Entries that are not IP addresses are logged at startup since they
no longer match anything.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [Security] Add challenge-response auth with scrypt verifiers and stop storing plaintext passwords
Challenge-response authentication: the client derives a scrypt verifier
(RFC 7914, EVP_PBE_scrypt, N=32768, r=8, p=1) and authenticates with
HMAC-SHA256(key, nonce), so neither the password nor its hash is
transmitted. The stored format becomes
"$scrypt$<n>$<r>$<p>$<salt>$<verifier>" and Response_PasswordSalt
now carries the cost parameters. Strict servers only accept scrypt
verifiers; legacy accounts are migrated after a successful login.
Fix#344 for challenge-response servers: a saved profile stores the
derived verifier under the password key instead of the plaintext password.
The connect dialog loads it without revealing it, autoconnect passes it
through, the change-password dialog no longer prefills the old password
field with it, and the client only persists the verifier when
"Save password" is checked.
Took 3 minutes
Took 1 minute
Took 10 seconds
Took 7 minutes
* [Security] Harden credential format acceptance and challenge nonce validation
* [Security] Refuse KDF downgrade; trim profile name and warn on silent credential drop
* [Security] Add SupportsAll ServerOptions enumerator to fix enum validation assert
The server advertised both SupportsPasswordHash (1) and
SupportsChallengeResponseAuth (2) via a bitwise-OR that produced the
value 3, which no enumerator declared. Protobuf's proto2 enum
validation rejects the value in _internal_set_server_options, crashing
servatrice on the debug assertion. Add a SupportsAll = 3 enumerator with
an identical wire value so the bitmask reads unchanged while the value
is now a documented, valid member of the enum.
* [Security] Normalize username when issuing challenge nonce
cmdRequestPasswordSalt bound the challenge nonce to the raw, un-simplified
username while cmdLogin verifies it against the simplified name. A user
who typed leading or trailing whitespace in the username box would be
silently rejected during challenge-response login even with the correct
password. Simplify at the nonce-issuing site so both commands agree on
the normalized name.
Also drive-by formatting: collapse the now single-expression
serverOptions initializer onto one line.
* [Security] Keep an encrypted legacy password backup for old-client logins
Migrating an account to scrypt stranded pre-challenge clients, which can
only present the legacy salt+hash: after migration the server served them
a scrypt salt they cannot use, so the only recovery was a password reset.
When security/legacy_backup_key is configured (64 hex chars,
openssl rand -hex 32), submitPasswordVerifier now seals the legacy
credential with AES-256-GCM under that key before overwriting the row,
storing it in password_legacy_backup with the time it was last used. Old
clients keep working: cmdRequestPasswordSalt serves the backup's 16-char
salt in password_salt so they can derive the legacy hash they were
always able to compute, while new clients derive the verifier from the
new scrypt_salt field, which always carries the verifier salt. The
non-challenge login path accepts the plaintext or legacy hash against
the decrypted backup and refreshes its window; a migrated row with no
usable backup fails closed.
Backups age out so a database dump taken after the window closes never
carries a crackable legacy hash: getUsableLegacyBackup purges on first
use past security/legacy_backup_ttl_days (default 90, clamped 1..3650),
and a six-hourly sweep drops backups idle accounts never touch.
changeUserPassword clears the backup since a new credential invalidates
it. Without a configured key nothing is sealed or purged, so a
temporarily missing key cannot destroy existing backups.
The AAD is the simplified user name on both sides, sealing the blob to
its row so it cannot be moved between accounts. Sealing failures
degrade to the previous lockout behaviour rather than blocking
migration. The columns extend the unmerged 0036->0037 migration, so the
schema version stays 37.
* [Security] Address review: proto compat, plaintext verify, AAD case, strict scope
Fix the two blockers and follow-ups from the 2026-10-16 review:
- Advertise challenge-response in a new supports_challenge_response
field instead of the ServerOptions enum and drop SupportsAll:
proto2 enums are closed, so already-released clients parsed
server_options = 3 as NoOptions and fell back to plaintext
passwords. server_options stays exactly SupportsPasswordHash
on the wire; old clients ignore the new field.
- Verify a plaintext password directly against a migrated scrypt
row with verifyPassword() so email changes (cmdAccountEdit)
and old clients sending password keep working without a
legacy backup key. Pre-hashed values are still checked
against the backup, and its TTL window is refreshed only
when the backup is what actually matched.
- Use the stored row name as the AES-GCM AAD on both seal
and open so a login whose case differs from the stored
name (utf8mb4_unicode_ci) still opens its backup.
- Require challenge-response in strict mode only for
registered SQL accounts (userExists gate), leaving
guest/unregistered logins and none/password auth
methods unaffected.
- Document in servatrice.ini.example that a captured
legacy hash can take over a not-yet-migrated account
and that strict only constrains registered SQL
accounts. Schema version is 38.
* [Security] Raise legacy backup TTL default to 180 days
Gives operators and slow-to-upgrade users a longer downgrade
window for pre-challenge clients before unused encrypted
legacy password backups age out.
* [Security] Fix servatrice.cpp formatting per clang-format-18
* [Security] Add missing forward declaration for Command_SubmitPasswordVerifier
The cmdSubmitPasswordVerifier declaration references the proto-generated
type, which otherwise does not name a type in this header and breaks
every C++ build.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
Co-authored-by: BruebachL <bruebachl@cockatrice.example>
* [Servatrice] Relay room channel messages over ISL
* [Servatrice] Skip audit log for unknown ISL channel per review
- externalRoomSay drops messages for channels not in the local
catalog, matching Server_Room::say, instead of logging what was
never delivered
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Servatrice] Log room channel messages and bump schema to 38
* Fixup: show room_channel audit rows and keep migration nullable
The log history query returns room_channel rows, but the log tab
dropped them. Render them in the room table with an extra Channel
column. Also drop the NOT NULL on target_type from the 37->38
migration so migrated schemas match the nullable servatrice.sql
definition (and strict-mode ALTERs cannot fail on legacy NULL rows).
Update the channel tests to the flag-based privilege model: the admin
fixture now carries IsModerator alongside IsAdmin, as the database
layer assigns.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Server] Gate room channels on privilege levels
* [Server] Filter channel catalog per user, split read/write gating, add banners
- Server_Room::getInfo() now takes the requesting user and only lists
channels that user may read. Previously every room member received the
full catalog (e.g. the moderators channel) while message delivery was
already gated, so restricted channels were visible but silent.
- Split channel access into mayReadChannel/mayWriteChannel: history
replay, catalog listing, and event fan-out use the read level
(access_level); Command_RoomChannelSay uses the write level
(write_access_level, falling back to the read level when unset).
- Plumb write_access_level and banner_message through both room sources
(SQL rooms_channels table and config-file channels via writeaccess
and banner keys).
- Schema 38 -> 39: add write_access_level (NULL means inherit the read
level) and banner_message columns; also widen access_level to the
full six-value privilege enum to match the 37 -> 38 migration.
* Fixup: evaluate channel access per user flag instead of ladder
User flags are independent bits: IsDeveloper does not imply
IsModerator/IsAdmin, so developer-only users must not reach
moderator/admin channels. Replace the userPrivilegeLevel ladder with
hasChannelPrivilege: moderator channels need IsModerator (admins hold
it alongside IsAdmin), admin channels need IsAdmin, developer channels
need IsDeveloper or IsAdmin, judge channels admit IsJudge/IsModerator.
Covers catalog filtering, history replay, write gating and fan-out,
and corrects the ini example and proto docs.
* [Server] Make channel privilege helpers static per review
* [Server] Fail closed on unknown channel access levels per review
- roomChannelAccessFromKey returns nullopt for unknown keywords
instead of silently falling back to Public
- skip channels with unknown read/write levels with qWarning, empty
read defaults to Public, empty write means no override
- mayWriteChannel now also requires read access so write-below-read
configs cannot post into invisible channels
* [Server] Remove stray conflict marker in channel replay comment
* [Server] Fix servatrice.cpp formatting per clang-format-18
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
Co-authored-by: BruebachL <bruebachl@cockatrice.example>
* [Server] Add room chat channels with per-channel fan-out and history
* Fixup: redact all channel histories on message removal
removeSaidMessages only redacted the main history, so banned/warned
content stayed visible to late joiners on every other channel.
Drop the unused channelId parameter and redact up to amount messages
per channel across the whole channelChatHistory map.
* [Server] Validate room channel ids and fix catalog schema per review
- servatrice.sql: expand rooms_channels access_level enum to full set
to match 0037->0038 migration
- servatrice.cpp: skip empty, reserved (__main__) and duplicate channel
ids when loading from ini and DB with qWarning
- server_protocolhandler.cpp: clarify catalog replay comment (filtering
happens in moderator stack)
- server_room: drop duplicate channelIndex, linear search preserves
catalog order
* [Server] Fix server_room.cpp formatting per clang-format-18
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
Co-authored-by: BruebachL <bruebachl@cockatrice.example>
* [Server] Apply card art rules to playmats
Card art DENY rules were only enforced for user card art
(SetCardArtParams) while playmats using the same art bypassed them.
Move the rule lookup to Servatrice_DatabaseInterface::isCardArtAllowed
with a default-allow base on Server_DatabaseInterface, delegate the
session check to it, reject DENY-listed Command_SetPlaymat with
RespFunctionNotAllowed, and strip DENY-listed deck playmats on
DeckSelect so banned art is never broadcast.
* [Server] Address review nits on playmat card art rules
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Security] Add per-address rate limiting for auth endpoints
Introduce a thread-safe RateLimiter that tracks attempts per key
(IP address) within a sliding time window, and wire it into the
authentication endpoints:
- Login: failed login attempts from an address are counted; once the
configured maximum is exceeded within the window, further logins
from that address are rejected with RespTooManyRequests. A
successful login clears the failed attempts for that address.
- Registration: implement the previously stubbed
tooManyRegistrationAttempts, limiting how many accounts can be
created per address per window.
- Forgot-password: throttle both the email-request and the
email-challenge paths per address.
New [security] settings with defaults:
max_login_attempts_per_ip=5 / login_attempt_window_seconds=900
max_registrations_per_ip=2 / registration_window_seconds=3600
max_forgot_password_requests_per_ip=3 / forgot_password_window_seconds=3600
Adds unit tests for the RateLimiter (window limit, over-limit
blocking, clearing, per-key independence).
Took 3 minutes
* [Server] Fix rate limiter: disable semantics, draining window, bounded keys, pre-work login gate
- recordAttempt()/isBlocked() return false when maxAttempts <= 0 or windowSeconds <= 0, so the documented 'set to 0 to disable' actually disables instead of blocking every address permanently
- Keys at their limit are rejected without recording, so repeated attempts at the attacker's pace cannot keep sliding the window and holding the lockout open; a block now clears once the recorded attempts age out of windowSeconds
- Attempts map is bounded: an opportunistic prune (every 60s) erases keys whose attempts have fully aged out
- Login is throttled before the work: isLoginRateLimited() (a new virtual on Server, backed by isBlocked()) gates loginUser(), so a locked-out address no longer burns a database round trip and password verification per attempt
- recordAttemptAt()/isBlockedAt() time seams make the sliding-window behaviour deterministically testable; tests rewritten for the real semantics (no reliance on the maxAttempts=0 bug, no sleeps)
* [Server] Inform client of login/registration rate limit timeout
- Add retryAfterSeconds/retryAfterSecondsAt to RateLimiter to compute
seconds until oldest attempt exits the sliding window
- Expose loginRateLimitRemainingSeconds via Server/Servatrice
- Include denied_end_time (absolute epoch) in RespTooManyRequests for
login pre-auth and when transitioning to lockout after failed attempt
- Include denied_end_time in registration rate limit response
- Client: show retry time for login RespTooManyRequests
- Client: simplify registration RespTooManyRequests message to show
retry time when provided
- Add test for retryAfterSecondsAt
* [Client] Restore registration guidance and drop duplicated rate limit reason
The registration RespTooManyRequests case now only shows the retry-after
copy when the server actually sent a denied_end_time; the per-email account
cap path (no end time) falls back to the previous guidance about already
having an account and contacting the server operator.
Drop the canned denied_reason_str from the new login rate limit branches so
the client no longer prints the same sentence twice.
* [Server] Exempt trusted_sources from auth rate limits
Addresses in security/trusted_sources bypass per-address limits, matching
the existing max_users_per_address bypass in initSession. Without this,
Webatrice behind local nginx (all clients as 127.0.0.1) would lock out
every web user after 5 mistyped passwords, and cap the whole server at
2 registrations / 3 forgot-password requests per window.
Adds Servatrice::isRateLimitExempt helper; recordFailedLogin,
isLoginRateLimited, recordRegistrationAttempt and
recordForgotPasswordAttempt all check it.
# Conflicts:
# servatrice/src/servatrice.cpp
* [Server] Key login rate limit by (address, username)
A successful login cleared the whole per-IP failure history, so 4 bad
guesses plus 1 self-login (or guest UnknownUser login) from the same
address bypassed the limiter. Failures are now bucketed per
(address, lowercase username) and a success clears only its own pair.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [SMTP] Add reliable email delivery with retry and database-backed queue
Rework the SMTP client from fire-and-forget sending into a persistent
queue with exponential backoff and per-message retry budget, backed by
the existing database rows as the durable source of truth.
- Keep pending emails in an in-memory queue with attempts/backoff
tracking; enqueue without deleting database rows, and only remove the
row once delivery is confirmed via the mailDelivered signal.
- Add exponential backoff (doubling, capped at 30 min) between delivery
attempts, bounded by new smtp/maxretries and smtp/retrydelay settings.
- Track connection-layer failures (connect/auth/encrypt) with their own
backoff; route failures to recordFailed/recordConnectionFailure so
undeliverable mail eventually drops from the queue instead of
retrying forever.
- Treat rejected recipients/senders as permanent failures.
- Rebuild the SMTP buffer from the authoritative queue before each send
via a new QxtSmtp::reset() so reconnects never duplicate deliveries.
- Start the idle timeout after authentication, not on TCP connect, so
slow SSL/STARTTLS handshakes are not cut short.
* [SMTP] Fix retry accounting, backoff reset, and token dedupe
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Report] Add report reopen action for closed reports
- New Command_ReportReopen moderator command (extension 1206) with server
handler that reopens resolved/dismissed reports, resets resolution state
and notified, and records the reopening note in the audit log
- Reopen... button in the report queue action bar, gated to resolved and
dismissed reports, with optional note prompt and tooltip
- Success confirmations for Assign, Resolve/Dismiss and Reopen now survive
the list refresh via a pending status message instead of being clobbered
- Dispatch/auth coverage in server_moderator_role_test
* [Report] Surface resolution audit info in report details
- Add resolved_by_name to ServerInfo_Report and join resolved_by users in
the report list, my-list and details queries
- Render resolution note and resolved-at/by in the report detail pane
instead of duplicating the description into the chat log context
* [Report] Add context menu to report queue rows
- Open the standard UserContextMenu on right-click/Menu key for the
reported user, selecting the row first so the details panel matches
the menu target
- Carry the reported user's level in ServerInfo_Report so the menu
shows the same actions as the user list, with an IsUser fallback for
deleted or unregistered users
- Wire the Private chat action through TabReport to TabSupervisor and
retranslate the menu on live language switches
* [Server] Fix reported_user_level column indices in report queries
The SELECT lists resolved_by_name at index 10, shifting room_id to 11,
replay_id to 12, reported_user_id to 13 and reported_user_admin to 14.
The code read 12/13, so 'user exists' meant 'a replay exists' and the
admin flags came from the user id (e.g. id 5 -> admin+judge; users
without a replay reported as unregistered, hiding registered-only
actions). Fixed in cmdReportList and cmdReportMyList, plus the column
comment.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
- Add resolved_by_name to ServerInfo_Report and join resolved_by users in
the report list, my-list and details queries
- Render resolution note and resolved-at/by in the report detail pane
instead of duplicating the description into the chat log context
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
- New Command_ReportReopen moderator command (extension 1206) with server
handler that reopens resolved/dismissed reports, resets resolution state
and notified, and records the reopening note in the audit log
- Reopen... button in the report queue action bar, gated to resolved and
dismissed reports, with optional note prompt and tooltip
- Success confirmations for Assign, Resolve/Dismiss and Reopen now survive
the list refresh via a pending status message instead of being clobbered
- Dispatch/auth coverage in server_moderator_role_test
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Tests] Register each test case with CTest instead of each test binary
Registering a test cost four edits across a dozen CMakeLists files: add_test,
add_executable, a conditional add_dependencies on gtest, and target_link_libraries.
The conditional branch was the trap -- it is only needed when GTest was fetched and
add_subdirectory'd, and omitting it yields undefined gtest symbols at link time --
so it was easy to leave out and impossible to notice until the link failed.
Fold all four steps into cockatrice_add_test() and migrate every test directory
onto it, with a cockatrice_add_manual_test() wrapper for the three benchmarks that
are built but deliberately kept out of ctest. The calls now read as a declaration
of what a test is rather than a transcript of how to register one.
While rewriting the registrations, replace add_test(NAME x COMMAND x) with
gtest_discover_tests(). Naming the binary as a single CTest entry meant a CI failure
reported only which binary failed, and ctest -R Suite.Case could not select an
individual case. Discovery raises the suite from 35 opaque entries to 469 real ones:
ctest -R ThemeMigrationTest.IsIdempotent now runs exactly that case, and
ctest -L server / -LE bench work off labels applied at registration.
Give every test a 60s TIMEOUT. CTest's implicit default is 1500s, so one hung
socket wait would stall CI for 25 minutes, and single_instance_manager_test already
does real QLocalSocket waits.
Dropping per-case granularity costs wall-clock time, since the suite becomes
process-spawn bound rather than CPU bound: serial ctest went from 3.8s to 30.8s. Run
it with -j in CI to recover it -- 7.6s at -j4, and 3.5s at -j8, both at or below
the old serial time. The binaries are parallel-safe: the only test claiming a fixed
resource scopes its local socket name to its own pid, so concurrent copies cannot
collide.
Also delete dummy_test, which asserted 1 == 1 and only proved the harness was
wired up -- something the other 468 cases already establish.
No test is added, removed, or changed by this commit; the 469 cases are the same
ones the 35 binaries contained.
* [Tests] Make QtTest available to test targets
QtTest was not among the modules test targets could link, so the Qt model and
signal contracts had no way to be verified from a test: QAbstractItemModelTester
and QSignalSpy both live there.
Add Test to _TEST_NEEDED in cmake/FindQtRuntime.cmake. That is the single place the
union of test-only Qt modules is declared, so this is the only file that needs to
change to make the module reachable from every test target.
Nothing uses it yet. This lands the capability ahead of the model-layer tests that
need it, so those can be added as tests alone rather than as test plus build-system
plumbing.
* [Tests] Correct the note on the oracle test compression defines
add_definitions() populates the directory's COMPILE_DEFINITIONS property, which is
consumed at generate time, so it reaches every target in the directory no matter
where it is called relative to add_executable(). The previous note claimed the
flags stayed confined to the benchmark because the find_package block came after
it; oracle_importer_test, created above that block, carries both defines too.
Verified harmless rather than a live bug: HAS_ZLIB and HAS_LZMA gate code only in
oracle/src/pages.cpp and oracle/src/pagetemplates.cpp, and no test target compiles
either. The ordering and the leaked defines are both pre-existing, so this only
corrects the explanation.
* [Tests] Carry one CTest label per test instead of several
gtest_discover_tests() can hand CTest exactly one label. GoogleTestAddTests.cmake
accumulates its generated script with `foreach(arg ${ARGN})`, so a label list reaches
the CTestTestfile as bare `LABELS ci server`, and CTest reads only the first value of a
property and drops the rest without a word.
The `ci` label that is appended to every registered test was therefore absent from every
test that also asked for `server` or `bench`: `ctest -L ci` reported 400 of 469 cases,
and `ctest --print-labels` never listed a second label anywhere in the build.
Reduce each test to exactly one of `ci`, `server`, or `bench`, which is all the
selections need. The three now partition the suite exactly, 400 + 64 + 5 = 469, and
`ctest -LE bench` still means everything but benchmarks. A second label is now a
configure-time FATAL_ERROR rather than a line that quietly means something else.
Anything orthogonal to a test's kind belongs in a helper flag beside the label, so it
composes with the group instead of competing for the single slot.
Also raise DISCOVERY_TIMEOUT off GoogleTestAddTests' 5s default. A plain debug binary
answers --gtest_list_tests in 0.04s, but that budget is thin once several binaries
discover concurrently behind a -j build, and being over it fails at the *link* step with
a bare "Process terminated due to timeout" that points nowhere near the cause.
* [Tests] Link gtest_main on the fetched-gtest path
The downloaded-gtest branch set GTEST_BOTH_LIBRARIES to `gtest` alone, while
FindGTest sets it to both halves. Every test target links that list rather than
picking an entry point, so on this path nothing provides main() and any test
source that does not define its own fails at link with `undefined reference to
main` pointing at crt1.o, naming neither gtest nor the line responsible.
That stayed latent while every test carried its own four-line main(). A test that
omits it now fails, and the error points at the C runtime rather than at gtest,
which is a long way from the cause. Restoring both halves makes the two paths
agree and drops the boilerplate requirement on new test sources.
The 18 sources that still define main() keep linking: gtest_main.cc defines
nothing but main, so the linker leaves its object in the archive unused rather
than reporting a duplicate.
* [Build] Add an ENABLE_SANITIZERS option for ASan and UBSan
Adds -DENABLE_SANITIZERS, off by default, building with AddressSanitizer and
UndefinedBehaviorSanitizer. The bulk of the work is in the three ways the obvious
implementation silently does the wrong thing.
Probe the flags on the link step, not just the compile step. check_cxx_compiler_flag()
only compiles, so every sanitizer flag comes back unsupported, because linking an ASan
object without -fsanitize=address leaves __asan_init undefined. The result is not a
build failure but a build that appears instrumented and is not, which is worse. Use
check_cxx_source_compiles() with CMAKE_REQUIRED_LINK_OPTIONS instead, and treat an
unsupported sanitizer pair as fatal since the cosmetic flags are allowed to degrade.
Opt vendored sources out per source, through a cockatrice_disable_sanitizers_for()
helper rather than a list of set_source_files_properties() calls at the root. The
helper has to exist because CMake resolves a source-file property against the directory
scope its target was declared in: setting the property from the root with absolute paths
is a no-op, silently, which is why the call sites sit next to the add_library and
add_executable that pull the vendored code in. SFMT, minizip, the LZMA SDK and Qxt SMTP
are excluded. oracle/src/qt-json is in the history but not in the tree, removed in
#7214.
Also mark those sources SKIP_PRECOMPILE_HEADERS. The PCH is built with the sanitizer
flags, so it defines __SANITIZE_ADDRESS__; a translation unit compiled with
-fno-sanitize=all does not and GCC then refuses the PCH, which is a fatal
-Werror=invalid-pch in the servatrice and oracle builds.
-fno-sanitize=vptr is not optional. Qt constructs QObject subclasses into raw storage
and writes the vptr afterwards, so the vptr check reports on essentially every QObject
instantiation, and it has no value here because nothing outside moc-generated code
inherits QObject.
Verified on the resulting build: our own translation units carry ASan and UBSan calls,
SFMT.o carries neither, and the full suite of 464 cases passes clean under
ASAN_OPTIONS=detect_leaks=0:abort_on_error=1 with UBSAN_OPTIONS=halt_on_error=0.
The test suite needs two adjustments to make this option usable at all, both of which
land here because they are consequences of COCKATRICE_SANITIZERS existing rather than
independent changes.
DISCOVERY_TIMEOUT rises from 60s to 300s under sanitizers. An instrumented binary takes
2.8s to answer --gtest_list_tests against 0.04s uninstrumented, and discovery runs
several binaries at once behind a -j build.
One target is excluded via a NO_SANITIZE flag, server_developer_role_test. Reaching
google::protobuf::TextFormat::Printer's destructor aborts on an absl assertion inside
system protobuf 33.4 whenever the consumer translation unit is built without NDEBUG,
which a Debug sanitizer build is. This reproduces with no Cockatrice code at all, just
"TextFormat::Printer p; p.PrintToString(msg, &out);" built with -fsanitize=address, so it
is upstream rather than ours. It is a flag and not a fourth label because "skip this in
the sanitizer leg" has to compose with a test's group rather than compete for the single
label slot, and because a flag excludes by not registering the cases at all, which
cannot rot the way a label silently can. The target still builds and still runs on the
legs without sanitizers.
* [CI] Run the suite once under ASan and UBSan
Adds --sanitize to .ci/compile.sh, mirroring --test, and turns it on for one entry of
the existing Linux Docker matrix: Ubuntu 26.04, which already had a test-enabled leg.
That reuses the image, the ccache layer and the existing step shape, so the CI diff is a
matrix key plus the argument plumbing. Debian 13 and Fedora 44 keep covering the
uninstrumented suite, so nothing is lost by instrumenting one leg.
Sanitizers stay off the release and package legs, which build with --release for
packaging: instrumenting those would change the shipped binaries' optimization profile
and build time for no benefit. macOS and Windows stay uninstrumented too, since ASan is
unreliable against the Qt build there and MSVC has no equivalent without /fsanitize=address.
--sanitize does not imply --test, so an instrumented build can be produced locally
without paying for ctest. When both are given, ctest runs with detect_leaks=0 because Qt
and the sanitizer runtimes both leak at exit by design and LeakSanitizer's default buries
real findings under hundreds of process-teardown reports, and with UBSan halt_on_error=0 so
one run collects many findings instead of stopping at the first. ASan keeps
abort_on_error=1, since a bad access leaves the process unusable and continuing only
produces a second, more confusing report from the wreckage.
* [CI] Publish a gcovr line-coverage report from the test run
Sanitizers tell us what a run did wrong; they say nothing about what it never
reached. This adds the other half: an instrumented build whose report says which
of our own lines no test executes.
Add an ENABLE_COVERAGE option that compiles with --coverage -O0 -g, mirroring the
ENABLE_SANITIZERS opt-in including the per-source carve-out that keeps vendored
trees out of the denominator. -O0 is deliberate: at higher optimisation levels GCC
merges and relocates lines, so the report would describe code that no longer exists
in that form. Only GCC is supported, since Clang's intermediate format needs
llvm-cov; asking for coverage under Clang is a hard error rather than a build that
silently produces unreadable .gcno files.
Keep the filters in a committed gcovr.cfg rather than in the invocation. The tests
themselves are excluded because they are compiled with the same instrumentation
and are only ever executed by the run doing the measuring, so leaving them in would
report a healthier project than we have. The process entry points and the client's
window chrome go too: each needs a display, an interactive terminal or a live
database before it does anything a headless test could observe, so they would
otherwise sit at 0% forever and bury real movement.
Give coverage its own matrix leg rather than a second flag on the sanitizer leg.
The two answer different questions, and combining them slows both down without
telling us more. The new variant key gives the leg its own Docker image and ccache
entry so the two legs cannot race on the same tag.
The leg publishes an HTML report with per-file pages, LCOV, Cobertura XML and a
JSON summary. gcovr is a report-time dependency rather than a build dependency, so
a missing one warns and skips instead of failing an otherwise green build.
* [CI] Resolve the coverage source root before the build dir cd
The previous commit computed SRC_ROOT after `cd "$BUILD_DIR"`, under a comment
claiming it ran before. `$0` is relative to the directory the script was invoked
from, so from inside the build dir `dirname "$0"` no longer resolves and the
subshell's `cd` fails. Under `set -e` that is fatal rather than an empty variable,
which took down every leg in the matrix, coverage or not, including the ones this
change does not otherwise touch.
Move it above the cd and guard it on MAKE_COVERAGE, since the coverage report is
the only consumer. Resolving it unconditionally meant a coverage-only detail was
paid for by every build in the fleet, and gave every one of them a way to fail.
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
* [Tests] Register each test case with CTest instead of each test binary
Registering a test cost four edits across a dozen CMakeLists files: add_test,
add_executable, a conditional add_dependencies on gtest, and target_link_libraries.
The conditional branch was the trap -- it is only needed when GTest was fetched and
add_subdirectory'd, and omitting it yields undefined gtest symbols at link time --
so it was easy to leave out and impossible to notice until the link failed.
Fold all four steps into cockatrice_add_test() and migrate every test directory
onto it, with a cockatrice_add_manual_test() wrapper for the three benchmarks that
are built but deliberately kept out of ctest. The calls now read as a declaration
of what a test is rather than a transcript of how to register one.
While rewriting the registrations, replace add_test(NAME x COMMAND x) with
gtest_discover_tests(). Naming the binary as a single CTest entry meant a CI failure
reported only which binary failed, and ctest -R Suite.Case could not select an
individual case. Discovery raises the suite from 35 opaque entries to 469 real ones:
ctest -R ThemeMigrationTest.IsIdempotent now runs exactly that case, and
ctest -L server / -LE bench work off labels applied at registration.
Give every test a 60s TIMEOUT. CTest's implicit default is 1500s, so one hung
socket wait would stall CI for 25 minutes, and single_instance_manager_test already
does real QLocalSocket waits.
Dropping per-case granularity costs wall-clock time, since the suite becomes
process-spawn bound rather than CPU bound: serial ctest went from 3.8s to 30.8s. Run
it with -j in CI to recover it -- 7.6s at -j4, and 3.5s at -j8, both at or below
the old serial time. The binaries are parallel-safe: the only test claiming a fixed
resource scopes its local socket name to its own pid, so concurrent copies cannot
collide.
Also delete dummy_test, which asserted 1 == 1 and only proved the harness was
wired up -- something the other 468 cases already establish.
No test is added, removed, or changed by this commit; the 469 cases are the same
ones the 35 binaries contained.
* [Tests] Make QtTest available to test targets
QtTest was not among the modules test targets could link, so the Qt model and
signal contracts had no way to be verified from a test: QAbstractItemModelTester
and QSignalSpy both live there.
Add Test to _TEST_NEEDED in cmake/FindQtRuntime.cmake. That is the single place the
union of test-only Qt modules is declared, so this is the only file that needs to
change to make the module reachable from every test target.
Nothing uses it yet. This lands the capability ahead of the model-layer tests that
need it, so those can be added as tests alone rather than as test plus build-system
plumbing.
* [Tests] Correct the note on the oracle test compression defines
add_definitions() populates the directory's COMPILE_DEFINITIONS property, which is
consumed at generate time, so it reaches every target in the directory no matter
where it is called relative to add_executable(). The previous note claimed the
flags stayed confined to the benchmark because the find_package block came after
it; oracle_importer_test, created above that block, carries both defines too.
Verified harmless rather than a live bug: HAS_ZLIB and HAS_LZMA gate code only in
oracle/src/pages.cpp and oracle/src/pagetemplates.cpp, and no test target compiles
either. The ordering and the leaked defines are both pre-existing, so this only
corrects the explanation.
* [Tests] Carry one CTest label per test instead of several
gtest_discover_tests() can hand CTest exactly one label. GoogleTestAddTests.cmake
accumulates its generated script with `foreach(arg ${ARGN})`, so a label list reaches
the CTestTestfile as bare `LABELS ci server`, and CTest reads only the first value of a
property and drops the rest without a word.
The `ci` label that is appended to every registered test was therefore absent from every
test that also asked for `server` or `bench`: `ctest -L ci` reported 400 of 469 cases,
and `ctest --print-labels` never listed a second label anywhere in the build.
Reduce each test to exactly one of `ci`, `server`, or `bench`, which is all the
selections need. The three now partition the suite exactly, 400 + 64 + 5 = 469, and
`ctest -LE bench` still means everything but benchmarks. A second label is now a
configure-time FATAL_ERROR rather than a line that quietly means something else.
Anything orthogonal to a test's kind belongs in a helper flag beside the label, so it
composes with the group instead of competing for the single slot.
Also raise DISCOVERY_TIMEOUT off GoogleTestAddTests' 5s default. A plain debug binary
answers --gtest_list_tests in 0.04s, but that budget is thin once several binaries
discover concurrently behind a -j build, and being over it fails at the *link* step with
a bare "Process terminated due to timeout" that points nowhere near the cause.
* [Tests] Link gtest_main on the fetched-gtest path
The downloaded-gtest branch set GTEST_BOTH_LIBRARIES to `gtest` alone, while
FindGTest sets it to both halves. Every test target links that list rather than
picking an entry point, so on this path nothing provides main() and any test
source that does not define its own fails at link with `undefined reference to
main` pointing at crt1.o, naming neither gtest nor the line responsible.
That stayed latent while every test carried its own four-line main(). A test that
omits it now fails, and the error points at the C runtime rather than at gtest,
which is a long way from the cause. Restoring both halves makes the two paths
agree and drops the boilerplate requirement on new test sources.
The 18 sources that still define main() keep linking: gtest_main.cc defines
nothing but main, so the linker leaves its object in the archive unused rather
than reporting a duplicate.
* [Build] Run the test suite under ASan and UBSan (#7400)
* [Build] Add an ENABLE_SANITIZERS option for ASan and UBSan
Adds -DENABLE_SANITIZERS, off by default, building with AddressSanitizer and
UndefinedBehaviorSanitizer. The bulk of the work is in the three ways the obvious
implementation silently does the wrong thing.
Probe the flags on the link step, not just the compile step. check_cxx_compiler_flag()
only compiles, so every sanitizer flag comes back unsupported, because linking an ASan
object without -fsanitize=address leaves __asan_init undefined. The result is not a
build failure but a build that appears instrumented and is not, which is worse. Use
check_cxx_source_compiles() with CMAKE_REQUIRED_LINK_OPTIONS instead, and treat an
unsupported sanitizer pair as fatal since the cosmetic flags are allowed to degrade.
Opt vendored sources out per source, through a cockatrice_disable_sanitizers_for()
helper rather than a list of set_source_files_properties() calls at the root. The
helper has to exist because CMake resolves a source-file property against the directory
scope its target was declared in: setting the property from the root with absolute paths
is a no-op, silently, which is why the call sites sit next to the add_library and
add_executable that pull the vendored code in. SFMT, minizip, the LZMA SDK and Qxt SMTP
are excluded. oracle/src/qt-json is in the history but not in the tree, removed in
#7214.
Also mark those sources SKIP_PRECOMPILE_HEADERS. The PCH is built with the sanitizer
flags, so it defines __SANITIZE_ADDRESS__; a translation unit compiled with
-fno-sanitize=all does not and GCC then refuses the PCH, which is a fatal
-Werror=invalid-pch in the servatrice and oracle builds.
-fno-sanitize=vptr is not optional. Qt constructs QObject subclasses into raw storage
and writes the vptr afterwards, so the vptr check reports on essentially every QObject
instantiation, and it has no value here because nothing outside moc-generated code
inherits QObject.
Verified on the resulting build: our own translation units carry ASan and UBSan calls,
SFMT.o carries neither, and the full suite of 464 cases passes clean under
ASAN_OPTIONS=detect_leaks=0:abort_on_error=1 with UBSAN_OPTIONS=halt_on_error=0.
The test suite needs two adjustments to make this option usable at all, both of which
land here because they are consequences of COCKATRICE_SANITIZERS existing rather than
independent changes.
DISCOVERY_TIMEOUT rises from 60s to 300s under sanitizers. An instrumented binary takes
2.8s to answer --gtest_list_tests against 0.04s uninstrumented, and discovery runs
several binaries at once behind a -j build.
One target is excluded via a NO_SANITIZE flag, server_developer_role_test. Reaching
google::protobuf::TextFormat::Printer's destructor aborts on an absl assertion inside
system protobuf 33.4 whenever the consumer translation unit is built without NDEBUG,
which a Debug sanitizer build is. This reproduces with no Cockatrice code at all, just
"TextFormat::Printer p; p.PrintToString(msg, &out);" built with -fsanitize=address, so it
is upstream rather than ours. It is a flag and not a fourth label because "skip this in
the sanitizer leg" has to compose with a test's group rather than compete for the single
label slot, and because a flag excludes by not registering the cases at all, which
cannot rot the way a label silently can. The target still builds and still runs on the
legs without sanitizers.
* [CI] Run the suite once under ASan and UBSan
Adds --sanitize to .ci/compile.sh, mirroring --test, and turns it on for one entry of
the existing Linux Docker matrix: Ubuntu 26.04, which already had a test-enabled leg.
That reuses the image, the ccache layer and the existing step shape, so the CI diff is a
matrix key plus the argument plumbing. Debian 13 and Fedora 44 keep covering the
uninstrumented suite, so nothing is lost by instrumenting one leg.
Sanitizers stay off the release and package legs, which build with --release for
packaging: instrumenting those would change the shipped binaries' optimization profile
and build time for no benefit. macOS and Windows stay uninstrumented too, since ASan is
unreliable against the Qt build there and MSVC has no equivalent without /fsanitize=address.
--sanitize does not imply --test, so an instrumented build can be produced locally
without paying for ctest. When both are given, ctest runs with detect_leaks=0 because Qt
and the sanitizer runtimes both leak at exit by design and LeakSanitizer's default buries
real findings under hundreds of process-teardown reports, and with UBSan halt_on_error=0 so
one run collects many findings instead of stopping at the first. ASan keeps
abort_on_error=1, since a bad access leaves the process unusable and continuing only
produces a second, more confusing report from the wreckage.
* [Build] Fix sanitizer flag leak into release package and make UBSan fail CI
- compile.sh always passes -DENABLE_SANITIZERS explicitly so the
release-package reconfigure in the same build dir cannot inherit ON
from the sanitized debug step (which shipped libasan/libubsan DEB deps).
- Sanitizers.cmake resets COCKATRICE_SANITIZERS to OFF when the option is
off (CACHE INTERNAL otherwise sticks forever, keeping NO_SANITIZE skips
and the 300s discovery timeout).
- UBSan now fails the job: -fno-sanitize-recover=undefined at compile time
plus UBSAN_OPTIONS halt_on_error=1 (recover mode exits 0, so the old
halt_on_error=0 could never fail).
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
---------
Co-authored-by: Lukas Brübach <Bruebach.Lukas@bdosecurity.de>
