[parthod0x]

receivePacketWithHeaderAs computes
remainingSize = header.size() - sizeof(RequestResponseHeader) from a
node-controlled value, then runs
std::fill_n(mBuffer.begin(), remainingSize, 0) before any bound check.

RequestResponseHeader::size() returns INT32_MAX when the 24-bit size field
is zero (the code comments "size is never zero, zero means broken packets"). So
a node that sends a response header with a zero size field produces
remainingSize == 2147483639, and fill_n writes ~2.1 GB into the 16 MB
mBuffer — well past its end.

Confirmed with AddressSanitizer: global-buffer-overflow, WRITE of size 2147483639. A malicious or malfunctioning node can corrupt the memory of any
client that queries it (the client trusts the node only for its response, but
this happens before the size is validated).

Fix

Require remainingSize > 0 and reject a remainingSize larger than the receive
buffer before fill_n. receiveData already throws on an oversized read,
but fill_n ran first; this closes that window.

Verified: the oversize case overflows under ASAN on current main; with the
guard it is rejected before fill_n, and g++ -std=c++20 -fsyntax-only is clean.

This is the response-parsing path, distinct from the user-input buffer fixes in
#133 and #134, and is the more security-relevant of the set because it is
triggerable by the remote node rather than by local input.

[fnordspace]

Good catch, need slight adjustments. Also please keep comments short and concise and only where the code is not self-explanatory.

Thanks for the submission!

[parthod0x]

Adopted your approach: fill only sizeof(T) (the bytes read back into result) and let receiveData()'s existing sz > buffer.size() guard reject an oversized — or underflowed — size. Dropped my explicit check and the verbose comment. Rebased onto current main (the argparser.h hunks are already there via #133). Compiles clean under the CMake /WX build. Thanks!