Re: [bug-inetutils] memory exhaustion in ftpd/ftpd.c

[Dan Kegel]

Eric Schmeddes wrote:
> I would like report a memory exhaustion problem occurring when
> getting a binary file larger than the available memory from a
> machine running ftpd using mmap().
>
> The problem occurs in the function send_data() in ftpd/ftpd.c,
> when using mmap() to access the file. It seems that when the
> current get position in the mmapped buffer exceeds the currently
> available machine memory, the running ftpd process gets killed
> by the kernel.
>
> I have only be able to reproduce this error when running without
> a swap partition.

This is normal and expected for mmap-based servers.
(And it's not very smart, I agree.  mmap is probably not
the right way to serve files if VM is at all tight!)

Fortunately, you can probably work around this by doing

export ac_cv_func_mmap_fixed_mapped=no

before configuring ftpd.  This will cause it to use its
fallback, getc-based server code path.

You might consider using a different server... betaftpd
(when properly patched) behaves pretty well in embedded systems,
at least for what I've done.
http://www.pengutronix.de/software/ptxdist/patches-cvs/betaftpd-0.0.8pre17/generic/betaftpd-errhand.patch
is the patch we use.  (And if you're interested in performance,
there are a couple of patches there to convert betaftpd to
use sigio or epoll, vroom vroom...)
- Dan