Re: The time has come for a kernel interfacing library layer

[Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>]

:Do you mean kernel will directly write errno into userland memory?
:

    No, what I'm working on is a system call 'library' layer which is
    intended to be part of libc without actually being part of libc.
    The way it will work is that libc will declare a special code section
    (whether dynamic or statically linked) which the kernel will recognize
    and remap/override.  In the best case libc's code section will simply
    be a BSS section (not take up any space in the filesystem).

    This section will contain function entry points for system calls at
    fixed offsets.  I'm trying to make this layer completely independant
    of everything else, which means that I would prefer that the kernel
    simply map the appropriate system call library into it without doing
    any further relocations.   So the question becomes... how to deal
    with errno ?  The kernel itself doesn't set it, so the syscall library
    has to.  The easiest way to do that is to either store a pointer to
    errno in the TCB or to store errno itself in the TCB.

    The kernel would select which layer file to map based on the ABI
    version of the program (based on the section name) verses the ABI
    version of the kernel.  Theoretically this means that we can then 
    change system calls in the kernel, like the stat structure, and still
    provide compatibility functions for programs expecting older ABIs.
    And, theoretically, we can also run programs expecting newer ABIs 
    on older systems by providing the appropriate syscall layer to
    translate the ABIs.

					-Matt
					Matthew Dillon 
					<dillon@xxxxxxxxxxxxx>