> > This means the call is rather useless... > > Sorry? A syscall whose only effect is to return a value, and which does not return the value it should, is not too useful. (Just think what would happen if someone accidentally swapped getpid() and getppid() in some mainstream versions of the kernel or libc.) Sure it would be possible to add autoconf tricks to select the version, but then the resulting binary would break on the next system upgrade. Or a runtime version test? Ugly. > SO_SNDBUF has nothing common with "application" send buffer. > It is internal kernel buffer. What I mean is: I want to allocate an application buffer which keeps data until I can be sure, by way of a checkpointing mechanism implemented by the application, that it has been delivered to the other end. What I want to know is how big this buffer should be. But the whole design is in such an early stage, perhaps I'll go for a buffer list or something. Olaf - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
