[Fwd: Re: [Tinycc-devel] re: Selinux patch fixup]

[grischka]

---------- Forwarded to the list ----------

From: Henry Kroll III <address@hidden>

On Mon, 2010-04-26 at 01:08 +0200, grischka wrote:
> Anyway, do what you need to feel comfortable, and don't forget to tell
> us what new features we now have and how people can try them out, at
> some point.

The new options (*nix only)
* (Option) --with-selinux
* (Option) --disable-static

Todo: make sure these "options" don't interfere with what's already
there...

--disable-static

        I should really add a line that says these options won't work
        with win32 builds. It could be built as a .dll however, that is
        over my head at the moment.

        --disable-static is not a very good descriptive name. I think it
        should be --enable-shared, but people will expect
        --disable-static because so many other packages use it. :|

        On *nix, these patches make static library, libtcc.a become a
        shared library, libtcc.so.1.0 with links to libtcc.so and
        libtcc.so.1. These extra links are apparently necessary for
        linking against the library and loading it dynamically. The
        file, libtcc_test is built dynamically against this library and
        apparently works, loading the library, which now optionally
        works under Selinux with --with-selinux, also known by me as the
        mmap patch...

--with-selinux
        Also incompatible with win32, this patch works by (optionally)
        using mmap instead of malloc...return (*prog_main)(argc, argv).
        Essentially, a /tmp file is created +wx and then immediately
        unlinked, so no files remain if the process decides to take a
        dump. Some memory is allocated for reading and writing through
        this (non-existent) file pointer. After that is done, memory can
        be accessed through 2 randomized pointers, one of them can be
        written. The other can be executed. The docs on this say an
        attacker will have a tough time writing and executing the same
        block of memory, thus making programs written for tcc -run more
        secure...I guess they are right. I certainly feel more secure
        after wading through all these docs.

        At least it gets around errors that Selinux pops up when tcc
        -run is attempted on Fedora. The errors basically say that I am
        trying to write to and execute the same block of memory and that
        it has been disallowed. These changes work around these errors
        in the recommended way.