Re: ldd output for sysinfo DSO

[Jakub Jelinek <jakub at redhat dot com>]

On Sat, Oct 11, 2003 at 11:39:22AM -0400, Daniel Jacobowitz wrote:
> On Tue, Sep 30, 2003 at 01:45:17PM -0700, Roland McGrath wrote:
> > > The leading tab is because of the sysinfo DSO.  Its address is in kernel
> > > space, so writev() stops when it sees the pointer, and nothing gets written.
> > 
> > It still seems to me that the proper solution to this issue is for the
> > kernel to treat user addresses consistently.  That is, what you can access
> > from userland you can access via system calls that use user addresses.
> > 
> > > The other sensible alternative is to skip it entirely.
> > 
> > This might be wise for unrelated reasons.  I don't know what all groks ldd
> > output, but there may well be scripts around that assume that the rhs names
> > a file that can (and should) be found.  linux-gate.so.1 refers to no file.
> 
> I see that Roland's changed it not to set l_libname.  Now we get:
> 
> drow@nevyn:/big/fsf/glibc% ./elf/ld.so --list --library-path .:rt:linuxthreads:/lib /bin/ls
>         linux-gate.so.1 =>  (0xffffe000)
>         librt.so.1 => rt/librt.so.1 (0x40001000)
>         libacl.so.1 => /lib/libacl.so.1 (0x40014000)
>         libc.so.6 => ./libc.so.6 (0x4001b000)
>         libpthread.so.0 => linuxthreads/libpthread.so.0 (0x4014a000)
>         /lib/ld-linux.so.2 => ./elf/ld.so (0x80000000)
>         libattr.so.1 => /lib/libattr.so.1 (0x4019c000)

I think it should print something that scripts can identify as not a
filename. Say /VIRTUAL/ (a directory cannot be a shared library).

	Jakub