[Bug 2655] AuthorizedKeysCommand with large output can deadlock

[bugzilla-daemon at bugzilla.mindrot.org]

https://bugzilla.mindrot.org/show_bug.cgi?id=2655

--- Comment #9 from Jim Knoble <jmknoble at pobox.com> ---
(In reply to Damien Miller from comment #7)
> (In reply to Jim Knoble from comment #6)
> 
> > Continuing to consume unneeded output seems like the wrong thing to
> > do here.
> 
> Why?

For large amounts of additional output, it introduces lag in the
processing of authorized keys; which in turn could cause issues in
scalability.

It also seems to "promise" that the output will always be read, whereas
calling out up front that it may not all be read encourages the writers
of AuthorizedKeysCommand's to write code that handles such failures
resiliently. 


(In reply to Damien Miller from comment #8)
> BTW setting a SIGPIPE handler isn't possible as signal handlers are
> reset on exec

You're right, of course. I was thinking of *ignoring* SIGPIPE when I
wrote this, which, although it would survive execve(2), is clearly not
the right thing to do here. Chalk this one up to "it's been too long
since I've done this". Perhaps I should stop armchair
quarterbacking....

-- 
You are receiving this mail because:
You are watching someone on the CC list of the bug.
You are watching the assignee of the bug.