Subject: Re: How do shared libs and kernel modules work?
To: None <>
From: Mike Long <>
List: tech-kern
Date: 06/21/1996 13:01:49
>Date: Thu, 20 Jun 1996 13:20:32 -0500
>From: Travis Hassloch x231 <>

>For example:
>How are shared libs mapped into user processes?  Via mmap?

Yes, shlibs use mmap().  Run a dynamic executable under ktrace(1) to
see for yourself.

>Are there code stubs to implement this, or are they implemented without
>stubs?  Is the memory area used by the stubs reclaimed by the map?

I think it's a jump table.

>Does it involve any self-modifying code?  Can the libraries have
>per-process as well as shared data?  How is that done?

No self-modifying code (.text is read-only).  Data pages are
copy-on-write:  The process uses a shared copy of each page until it
writes to it, at which point that process gets its own copy.

>I'm aware of the memory layout for each process, so I'm interested
>to know if the code segments are mapped above the break, or
>"squeezed" between the code and data, or somehow overlaid on the code.

Wherever mmap() puts 'em.

>Likewise, what is the interface to LKMs?

modload(8) and friends.

>Are we susceptible to the same attacks SunOS is, in that hackers can load
>a stealth-module?

Not if you remove 'options INSECURE' from your kernel config.  Then no
one (not even root) can load LKMs while the machine is multiuser.

>What kind of kernel functionality can be modified by an LKM?

Almost anything.

>What kind of info can survive a warm boot on the different ports?

That depends on the port.  Usually the kernel message buffer survives.

>If this is all in manpages, feel free to just point me there (I can't
>search them from where I am currently).

See, link(7?), ld(1), mmap(2).
Mike Long <>     <URL:>
VLSI Design Engineer         finger for PGP public key
Analog Devices, CPD Division          CCBF225E7D3F7ECB2C8F7ABB15D9BE7B
Norwood, MA 02062 USA       (eq (opinion 'ADI) (opinion 'mike)) -> nil