Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 19 Jun 2017 22:39:33 +0200
From: Solar Designer <>
Subject: Re: Qualys Security Advisory - The Stack Clash

On Mon, Jun 19, 2017 at 09:40:28AM -0600, wrote:
> I just want to publicly thank Qualys for working with the Open Source
> community so we (Linux and *BSD) could all get this fixed properly.
> There was a lot of work from everyone involved and it all went pretty
> smoothly.


It's excellent research by Qualys, building upon yet exceeding what was
previously known about this vulnerability class.  The quality of Qualys'
writing is also rare these days.

That said, we owe apologies to the community for violating the published
distros list policy regarding the maximum embargo duration.  Personally
and as distros list admin, I do apologize for letting this happen.

I think we shouldn't have let it happen.

The stated argument for extending the embargo duration beyond list
policy's maximum was that fixes presumably wouldn't be ready.

However, there were also arguments in favor of disclosing on the
originally planned date, which was within list policy: interim
workarounds and mitigations could be ready in time for that, the fixes
that would be ready by the later date (thus, presumably are ready now)
are not the end of the story anyway (specifically, gcc -fstack-check
wasn't expected to be ready, and isn't - let alone having distros'
userlands rebuilt with that option), the general issue wasn't new, and
all the usual arguments against long embargoes (some of the affected
parties being left out of the discussion, inconvenient and maybe less
productive discussion in the smaller private community, risk of leaks,
risk of rediscovery).

Apparently, the workarounds and mitigations just were not enterprisey
enough for the bigger players.  Releasing interim updates first would
mean the vendors would have to acknowledge that some (I think obscure)
functionality might be temporarily(?) gone, until a later final update
that would include the more invasive fixes for the underlying issues
while optionally removing the previously introduced workarounds and
mitigations (although personally I would prefer to make them standard,
as security hardening).  The vendors could also provide a knob (e.g., a
sysctl) to restore the questionable functionality right away (in this
case, it's things like limited debugging of dynamic linking on SUID/SGID
exec, and multi-megabyte command-line argument lists to SUID/SGID
programs) for those customers who prefer no risk of breakage over
security.  I understand it's a tough trade-off to expose to a paying
customer.  Yet I think it would have been most reasonable from a purely
technical and community friendliness points of view.

As Qualys advisory says, relevant security hardening for Linux kernel is
already found in grsecurity (per my quick check, some of it since 2012,
some earlier).  Relevant security hardening for glibc has been in Owl's
and ALT Linux's packages (always enabled - no knob) since at least 2005
(some or all of it probably also since 2000-2001, but I didn't re-review
how complete older revisions of that patch were).  Merging some of those
changes (at least the least invasive ones, sufficient to deal with the
most promising and most immediate attacks - short of the currently more
hypothetical remote attacks on system services) into distros' kernels
and glibc could surely be done in 14 days, including QA.

Apparently, companies disclosing security issues are concerned of being
held to a higher standard (than individuals) with regard to being (or
appearing) responsible.  If a company's disclosure is deemed
irresponsible by some (like we've seen happen before), that company (and
others) might become more cautious next time (which might be what
happened here), or worse - they might no longer fund security research
(luckily not the case here).

When I say these things, I don't mean to blame anyone, nor any company.
Everyone was probably doing their best under their circumstances.  I am
merely sharing my thoughts with the community.

Qualys first informed the distros list about this upcoming set of issues
on May 3.  This initial notification didn't say Stack Clash nor anything
like that, but merely expressed intent to disclose the issues and
concern that the list's maximum embargo duration of 14 to 19 days might
not be sufficient in this case.  In the resulting discussion, I agreed
to consider extending the embargo beyond list policy should there be
convincing reasons for that.  In retrospect, I think I shouldn't have
agreed to that.

Qualys posted the detail to distros on May 17 with public disclosure
planned for May 30, which was within list policy.  Due in part to
requests by Red Hat (who were going to do much of the work needed by
other distros, in particular on glibc) and presumably Qualys' internal
reasoning, on May 23 Qualys unilaterally said they were extending the
embargo until June 19 (the date previously requested by Red Hat, and
IIRC one that Qualys had said was also known good for Oracle Solaris,
although Oracle was OK with others publishing on May 30), and they said
they were already in the process of informing others (those they had
notified beyond the distros list membership) of this extension.

This was against what I had said previously, where I offered merely to
consider extending the embargo, not to unconditionally accept a decision
like this from Qualys nor anyone else.  At this point, I was still in my
right to insist on the originally planned date, and to make it happen
technically - just post the detail to oss-security myself on May 30, as
that wouldn't have violated any agreement I had.  (Of course, people
would be mad at me.)  Instead:

On Tue, May 23, 2017 at 11:21:01AM +0200, Solar Designer wrote:
> That's really bad.  I don't support you in your decision as from my
> perspective it's wrong, but I also don't want to go against it.
> So I reluctantly accept it

I didn't say it at the time, but frankly part of my reasoning for the
reluctant acceptance was that I had no idea what might have been going
on inside Qualys, nor wanted to have that inside info.  I understand
it's rare for companies to do quality security research, and I didn't
want my action to have hampered the stream of quality security research
we're seeing from Qualys lately.

Distro vendors saying they wouldn't be ready wasn't as important to me.
In fact, most distros hadn't yet expressed a clear opinion by then and I
was still NAK'ing Red Hat's requests at the time of Qualys' decision.

I am really not blaming Red Hat here.  They were the ones making the
request largely because they were also one of two distros, the other
being SUSE, doing much of the work on shared upstream code, for benefit
of most other Linux distros.

Anyhow, with the unfortunate embargo extension in place, my remaining
goal was to minimize the damage.  I helped disentangle and push out two
of the sub-issues - with Sudo and ISC/Vixie Cron - on dates that are
within list policy (no more than 14 days since initial disclosure of
vulnerability detail on the distros list).  I also used the opportunity
to test which of the distros are actually reading the lengthy thread
(resulting in one distro, which was represented by just one person,
getting removed from the distros list).  Finally, it's also in this
context that I decided to accept some non-distro volunteers to help with
patch review (unfortunately, this hasn't worked well enough yet -
although some patch review did occur and changes were made, an important
issue with the first Sudo patch was missed).


May 3 - preliminary notification by Qualys, no detail
May 17 - detail posted by Qualys on Stack Clash and Sudo
May 23 - embargo extension
May 26 - detail posted by Qualys on Cron
May 30 - initially planned public disclosure date for all Qualys issues
May 30 - Sudo issue public:
May 31 - Sudo incomplete fix fixed in 1.8.20p2
June 2 - Sudo incomplete fix and its fix announced:
June 8 - Cron minor issue public (along with some fixes):
June 19 - Stack Clash public

Since we were making this public in pieces like that, I have to say: no,
there's nothing else left to publish as part of this series of Qualys'
findings.  Everything Qualys brought to distros so far is now public.

Given this experience, I am not going to give any impression I could
agree to an embargo extension beyond list policy again.  I am going to
enforce the stated list policy.  Unfortunately, this means that next
time someone or some company who's extra careful about being/appearing
responsible wants to inform distros of an issue, they might opt to work
with individual distros off-list or through some other channel.  That's
life.  At least the distros list would not be contributing to that.
Having this discussion explode on the distros list was not great in
other aspects anyway - it's too many too frequent messages for too long
to conveniently handle on an encrypted list.  This wasn't worth it.

I am also going to remove the "up to 19 days" option, where an embargo
longer than 14 days (up to 19) could be applied for issues posted to
distros too close to or on a weekend, so that the embargo time could be
"rounded up" to expire next Tuesday after the normal maximum of 2 weeks.
Part of the reason for this removal, besides desire to shorten embargoes
in general and on average, is to remove the wrong incentive to report
issues to distros close to a weekend just to have (and give distros)
more time (much of that being an extra weekend).  The original issue
this option tried to address can also be addressed by "rounding down"
(e.g., to last Wednesday or Thursday before 14 days would run out on a
weekend), so let's be doing that.

This wrong incentive didn't play a role this time, but it was mentioned.


P.S. While I am at it, for those reading this in web archives of the
list here's a link to a portion of this thread on gcc issues, which I
think inadvertently broke off:

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.