[PATCH] clientloop: die if writing to the sender fails

[Damien Miller]

On Wed, 29 Jan 2020, Mike Frysinger wrote:

> From: Mike Frysinger <vapier at chromium.org>
> 
> The write call here wasn't having its return value checked.  This
> could lead to CPU busy loops when the select() call returns but
> the write attempt fails.  This came up when running under NaCl, so
> I'm not sure how to recreate it in general, but it seems like this
> code should be checking its return value.  There shouldn't be a
> situation where returning an error & ignoring it is wanted.
> 
> I went with fatal() here rather than error()+break becuase the code
> outside the loop will then attempt some writes and then call fatal.

Thanks - I committed a similar patch to use sshpkt_fatal(), which
prints a little more information about the connection endpoint.