Busybox make, modify my /dev/null on host

[Denys Vlasenko]

On Sunday 10 February 2008 22:26, Mike Frysinger wrote:
> On Sunday 10 February 2008, Martinb_ARM_NOMMU_KISSDVD wrote:
> > After 2 weeks of the strangest things happening to my /dev/null I did find
> > a relation on this problem and the use of busybox (compiling)

This is a generic UNIX administration problem.
Many programs would inadvertently delete /dev/null
in unusual cases. Heck, even busybox had a bug where
if you would specify that pidfile is "/dev/null",
it was deleting it! Now we protect against that:

void write_pidfile(const char *path)
{
...
        pid_fd = open(path, O_WRONLY|O_CREAT|O_TRUNC, 0666);
        if (pid_fd < 0)
                return;

        /* path can be "/dev/null"! Test for such cases */
        wrote_pidfile = (fstat(pid_fd, &sb) == 0) && S_ISREG(sb.st_mode);
...
}

#define remove_pidfile(path) do { if (wrote_pidfile) unlink(path); } while (0)


From the POV of UNIX admin, though, I find it impractical to ensure
that all installed programs are not buggy. I want to make /dev/null
not deletable instead. ext[23] immutable bit can help, but what to do
on ramfs? :(

> your toolchain is broken.  either dont build as root or upgrade your gcc and 
> binutils.  older versions of both would unlink the output even if it were a 
> character device.

Yes, it probably happens somewhere here:


scripts/trylink:

check_cc() {
    if $CC $1 -shared -o /dev/null -xc /dev/null >/dev/null 2>&1; then
        echo "$1";
    else
        echo "$2";
    fi
}


scripts/Kbuild.include:

as-option = $(shell if $(CC) $(CFLAGS) $(1) -Wa,-Z -c -o /dev/null \
             -xassembler /dev/null > /dev/null 2>&1; then echo "$(1)"; \
             else echo "$(2)"; fi ;)


(these are just examples, there are more such cases).

Add ls -l /dev/null before and after these statements and you
will see which invocation of gcc kills /dev/null.
--
vda