Filled in the stubs for the windows functions:
KeQueryActiveProcessorCount(0) and KeGetCurrentProcessorNumberEx(NULL).
KeQueryActiveProcessorCount(0) maps one to one to num_online_cpus()
KeGetCurrentProcessorNumberEx(NULL) maps to raw_smp_processor_id(),
however this can also be smp_processor_id(), we have to decide based on
preemption, the latter checks for preemption and throws a warning if it
is executed in preemptible code. (I dont know the kernel code well
enough so this may need some discussion)
Code audit of the script engine's scanner/token handling and of the PCI ID
database parser. Each of the issues below was reproduced against the current
code before the fix and re-checked afterwards.
script-engine/scanner.c
* An unterminated string literal ("abc or L"abc) hung the scanner in an
endless loop: sgetc() returns EOF without consuming input, and neither
string loop tested for it, so the token grew until allocation failed.
Both loops now stop at EOF and report the token as UNKNOWN.
script-engine/common.c
* AppendByte()/AppendWchar() doubled Token->MaxLen before checking whether
the larger buffer was actually allocated. After a failed allocation MaxLen
described memory that did not exist and the next append wrote past the end
of the old buffer. MaxLen is now committed only on success.
* CopyToken() allocated strlen(Value) + 1 bytes but carried over the source
token's Len and MaxLen, so the copy's advertised capacity did not match its
allocation, and WSTRING payloads were truncated at their first embedded
null byte. The copy is now sized from Len/MaxLen and copied by length, with
a fallback to the string length for the grammar tokens in parse-table.c,
which only initialize Type and Value.
* NewToken() set MaxLen to the value length, which is zero for an empty
value. The 'Len >= MaxLen - 1' test in the append routines is unsigned, so
a zero MaxLen wrapped and disabled buffer growth entirely.
* IsUnderscore() tested 'c >= '_'', which also accepted the backtick, the
lowercase letters, '{', '|', '}', '~' and DEL. Register scanning uses it,
so '@rax|1' was lexed as one malformed register name instead of a register,
an operator and a number. The pseudo-register path already compared against
'_' directly.
* NewTokenList() did not check the allocation of its Head buffer.
* NewTemp() kept the last handed-out id in a static, so an exhausted temp
list produced a token aliasing a temporary still in use, and it derived
MaxTempNumber from an out-of-range index. It also dereferenced the new
token without a null check.
* FreeTemp() indexed the MAX_TEMP_COUNT-entry map with an unchecked value
parsed out of the token text.
* RotateLeftStringOnce() wrote to str[-1] when handed an empty string.
libhyperdbg/debugger/misc/pci-id.cpp
* The database file was read into a malloc(Length) buffer that was never
null-terminated, while ReadLine() walks it with strchr(). Looking up an
absent vendor scans to the end and reads past the allocation.
* The matched Vendor was allocated with malloc() and its Devices list head
was only assigned once a device line was parsed, so a vendor with no
device entries left it uninitialized and FreeVendor() walked a garbage
pointer.
* FreeVendor() released the device and subdevice lists but never the Vendor
itself, leaking one per lookup for every enumerated PCI device.
* The file handle leaked when the buffer allocation failed, ftell() and
fread() results were unused, and several error paths leaked the Vendor or
the not-yet-linked Device/SubDevice.
* strncmp() compared sizeof(VendorId) bytes, which is the size of the
pointer rather than the length of a vendor id.
* ReadLine() passed an unclamped count to strncpy_s(), which triggers the
invalid parameter handler for a line longer than the destination.
* GetVendorById() ignored the GetModuleFileName() result and overwrote the
tail of the path buffer without checking the room left in it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W3C1DuhHtqK64eEkHjKCHM
The debuggee-side hyperkd fix stops the driver flooding, but the same
delimiter-only framing with a blind retry loop also lives in the
user-mode receivers. Mirror the resync there so an unclean disconnect
does not flood the console either: KdReceivePacketFromDebuggee,
KdReceivePacketFromDebugger, and the ListeningSerialPortInDebuggee loop.
On overflow with no end-of-buffer marker, warn once per episode and
discard to the next frame boundary (bounded by SERIAL_RESYNC_MAX_BYTES)
instead of returning into the still-desynced stream.
Refs #661
When the debugger drops the serial link uncleanly mid-frame (no CLOSE_AND_UNLOAD packet), SerialConnectionRecvBuffer overflowed without an end-of-buffer marker and returned FALSE into the KD dispatch loop retry. The loop re-entered the same desynced stream, so it overflowed again at once and flooded 'buffer exceeds the buffer limitation' (and could wedge the KD driver on the next reconnect). Discard bytes to the next end-of-buffer marker to re-align, start a fresh frame, and log once per desync episode with LogWarning. Framing and protocol are unchanged.
Refs #661
