RH's non-upstreamed tux patches caused me pain recently(ish). Turns out TUX added a flag to the open syscall named O_ATOMICOPEN. It used the next available O_ flag value after those used in upstream kernels. That flag was not upstreamed. Upstream then added a flag to open called O_CLOEXEC. They (of course) also used hte next available flag value.
So, new binaries which attempted to use O_CLOEXEC (even when they attempted to test for its existence before using it) would fail badly when it returned some completely unexpected random error value...