Version 5.74.0 du 28/10/2026- Fixed kernel pool out-of-bounds write in Api_CopyStringFromUser, which placed the terminator one character past its allocation and copied two bytes beyond the caller's string (reached through API_UPDATE_CONF)
- Fixed Session_ListLock and the raised IRQL being leaked for the lifetime of the driver when the monitor put handlers faulted on a user mode string
- Fixed API_MONITOR_GET_EX copying eight bytes of the log buffer past each entry, and its bounds check being defeated by a USHORT cast and a signed underflow
- Fixed kernel stack buffer overflow in _wntoi when a box's NetworkAccess rule carried a long port token
- Fixed BlockLocalLoop never blocking IPv4 loopback, because WFP_isLoopback tested the last octet of the address instead of the first
- Fixed kernel pool over-read in the rename/link minifilter path, where FileNameLength out of the caller's information buffer was used as a length without being checked against that buffer
- Fixed the new LPC implementation leaking the per-connection port handle and any pending reassembly buffer on every disconnect
- Fixed concurrent requests on one LPC connection racing the shared reassembly state, which could free it twice
- Fixed stack buffer overflow in MSGID_SBIE_INI_SET_DAT, where the unterminated setting field was appended to a - Fixed size path buffer
- Fixed SbieIniServer abandoning its critical section when a handler faulted, which left every later INI request blocked on it
- Fixed value_len being truncated while scaled to bytes in the INI setting and template handlers, so a large value passed the bounds check and was then read far past the message
- Fixed GUI_MAP_WINDOW_POINTS copying more points out of a request than it actually carried, returning service heap to the caller
- Fixed WndHookRegisterSlave returning while holding m_SlavesLock on two caller reachable rejections, which wedged the GUI proxy for the rest of the session
- Fixed GUI_GET_RAW_INPUT_DEVICE_INFO truncating its size while scaling it to bytes, so an over large cbSize skipped both bounds checks and was handed to the OS as a buffer size
- Fixed the desktop startExplorer and startRun actions launching as LocalSystem rather than with the calling user's token
- Fixed PORT_DATA_INFO being read without requiring the whole structure to lie inside the LPC message
- Fixed the ALPC connect reply returning its uninitialised tail - up to PIPE_MAX_DATA_LEN of service heap - to the caller
- Fixed global ("asterisk") queue names having no session scoping, which let any unsandboxed process squat a well-known name such as the interactive queue
- Fixed the UAC packet being accepted with no minimum length, and its string offsets being used without being checked against it
- Fixed the ICMP echo reply fix-up writing a whole ICMP_ECHO_REPLY past a caller sized reply buffer when the echo itself failed
- Fixed request_size not being checked against the message in the ICMP echo handler, which put up to 64 KB of service heap on the wire
- Fixed m_WorkersList being mutated without taking the lock that exists for it, so concurrent startup requests could close and free the same worker twice
- Fixed UserServer::OpenFile using unvalidated request offsets, and matching the box's file rules against a path that was not canonical
- Fixed a WCHAR/byte length confusion in the SbieDrv API_GET_FILE_NAME handler that over-read the file-object name into the caller's buffer
- Fixed two paged-pool leaks in the SbieDrv API_RENAME_FILE handler (an early return and a faulting user copy) reachable by a sandboxed process
- Fixed the SbieDrv API_OPEN_FILE handler not checking its allocation and leaking it when the user copy faulted
- Fixed the spooler ALPC message filter being dead code on Windows 8.1+ (the driver compared the port id against the wrong literal, so the filter never armed)
- Fixed a kernel NULL-pointer dereference and a NonPagedPool leak in the NetworkAccess firewall-rule parser (netfw.c)
- Fixed a wrong range-overlap test in the firewall-rule parser that could insert overlapping ranges and make a Block rule stop matching
- Fixed two paged-pool leaks on the error paths of the driver configuration variable expander (Conf_Expand_2)
- Fixed the SbieDrv API_IS_BOX_ENABLED handler dereferencing an unprobed user SID pointer in kernel mode
- Fixed proc->file_lock and the raised IRQL being leaked if WFP_UpdateProcess faulted in the SbieDrv API_REFRESH_FILE_PATH_LIST handler
- Fixed a use-after-free of a protected-root list entry in the file minifilter (the node was read after its lock was released)
- Fixed proc->threads_lock being leaked if the output write faulted in the itok/ttok paths of the SbieDrv process-query handler
- Fixed the duplicated-handle access re-validation being skipped for Directory/SymbolicLink/Job objects in Syscall_DuplicateHandle_3 (a loop counter clobbered the type length before the suffix test)
- Fixed the SbieDrv API_OPEN_PROCESS handler dereferencing the terminated-process sentinel with a lock held when asked for PID 0
- Fixed two token references and a handle being leaked if the output write faulted in the SbieDrv API_FILTER_TOKEN handler
- Fixed a 32-bit truncation of the entry size in the driver monitor-log writer that could overflow the log buffer
- Fixed SandMan over-reading an unterminated file path out of a queue reply (MAN_FILE_MIGRATION), which a sandboxed process could use to crash the UI
- Fixed INI section names not being rejected for CR/LF or brackets, which let a caller forge additional sections/settings and bypass a per-section EditAdminOnly lock
- Fixed the SbieSvc USER_SHELL_EXEC handler using an unvalidated, possibly-unterminated wire path offset
- Fixed the SbieSvc MSGID_SBIE_INI_SET_DAT handler using an unvalidated value_len as the write length, reading past the request buffer
- Fixed the INI setting handlers not terminating their - Fixed-size section/setting/password wire fields before use
- Fixed a per-message heap leak in the SbieSvc log drain loop (DriverAssist::LogMessage)
- Fixed GUI_CHANGE_DISPLAY_SETTINGS passing an unterminated device name and an unbounded DEVMODE to the display API
- Fixed GUI_GRANT_HANDLE leaking the two bitmaps GetIconInfo creates on every call
- Fixed GUI_GET_WINDOW_PROP and GUI_FIND_WINDOW treating a caller-chosen value outside the atom range as a raw string pointer
- Fixed GUI_SET_CLIPBOARD_DATA not validating the request length and not bounding the caller-supplied data length
- Fixed the SbieSvc process-info handler disclosing any process's command line/image path to any local user (now restricted to the caller's session or administrators)
- Fixed the SbieSvc WOW64 registry-open handler indexing a - Fixed prefix of KeyPath without checking its length or termination
- Fixed the SbieSvc named-pipe read handler returning the uninitialised tail of its reply buffer (information disclosure)
- Fixed the SbieSvc terminal GetName handler leaking uninitialised stack into its reply and terminating the wrong buffer
- Fixed the four SbieSvc terminal query handlers having no caller gate and taking a session id off the wire, letting any user query other sessions as SYSTEM; QUERY_INFO now also allow-lists the information class
- Fixed the SbieSvc ImBox QueryHandler leaking its critical section (and deadlocking the mount subsystem) when the queried mount raced to NULL
- Fixed the SbieSvc queue server accepting an unbounded number of undrained requests (per-queue cap added)