Re: possible infinite recursion in notmuch-cli

[David Bremner]

Daniel Kahn Gillmor  writes:


> After some discussion with amdragon on IRC, i believe that this is only
> relevant to notmuch when actively decrypting a message -- OpenPGP's
> ability to embed compression makes it possible to write a PGP/MIME
> message that is a quine: that is, when decompressed, it would expand to
> itself, which would send our parser into an infinite loop.
>
> Since we're not decrypting during indexing, only notmuch-show and
> notmuch-reply are probably affected by this problem. (but if someone
> implements indexing of encrypted messages, then we'd have to worry about
> this in the indexer as well)

This is indeed our current situation.

> The simple and generalized solution would be to limit the recursive
> depth of our walk of the MIME tree; probably a large limit of something
> like 30 or 50 would not trigger any real-world problems, and would halt
> a runaway recursion well before most modern machines ran out of
> resources.

So do I understand correctly that to test this proposed fix, we would
not need to generate a MIME-quine (which sounds challenging) but just a
very deep MIME tree?

___
notmuch mailing list -- notmuch@notmuchmail.org
To unsubscribe send an email to notmuch-le...@notmuchmail.org

possible infinite recursion in notmuch-cli

[Daniel Kahn Gillmor]

Hi notmuch folks--

On oss-security recently, there was a discussion about recursive
compression and the ability to create infinite loops.

 id:20131010013106.GA29693 at openwall.com

After some discussion with amdragon on IRC, i believe that this is only
relevant to notmuch when actively decrypting a message -- OpenPGP's
ability to embed compression makes it possible to write a PGP/MIME
message that is a quine: that is, when decompressed, it would expand to
itself, which would send our parser into an infinite loop.

Since we're not decrypting during indexing, only notmuch-show and
notmuch-reply are probably affected by this problem. (but if someone
implements indexing of encrypted messages, then we'd have to worry about
this in the indexer as well)

The simple and generalized solution would be to limit the recursive
depth of our walk of the MIME tree; probably a large limit of something
like 30 or 50 would not trigger any real-world problems, and would halt
a runaway recursion well before most modern machines ran out of
resources.

A more targeted fix might be to just limit the number of recursive
decryptions that happen, since the MIME spec does not appear to be
capable of permitting other parts to provide both compression and
nesting, which is the root of the problem.  However, it's possible that
our MIME parsing ends up being more permissive than the specs, and is
willing to try to interpret (for example) a gzip-compressed multipart/*
or message/* part.  

So the simple/generalized solution is probably the right way to go.

Sorry i don't have time to implement the fix myself right now, but i
wanted to make sure the active coders in the project are aware of the
issue.

Regards,

  --dkg

references:

 http://mumble.net/~campbell/blag.txt (see 2013-10-08)
 http://www.openwall.com/lists/oss-security/2013/10/10/2
-- next part --
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 965 bytes
Desc: not available
URL: 

possible infinite recursion in notmuch-cli

[Daniel Kahn Gillmor]

Hi notmuch folks--

On oss-security recently, there was a discussion about recursive
compression and the ability to create infinite loops.

 id:20131010013106.ga29...@openwall.com

After some discussion with amdragon on IRC, i believe that this is only
relevant to notmuch when actively decrypting a message -- OpenPGP's
ability to embed compression makes it possible to write a PGP/MIME
message that is a quine: that is, when decompressed, it would expand to
itself, which would send our parser into an infinite loop.

Since we're not decrypting during indexing, only notmuch-show and
notmuch-reply are probably affected by this problem. (but if someone
implements indexing of encrypted messages, then we'd have to worry about
this in the indexer as well)

The simple and generalized solution would be to limit the recursive
depth of our walk of the MIME tree; probably a large limit of something
like 30 or 50 would not trigger any real-world problems, and would halt
a runaway recursion well before most modern machines ran out of
resources.

A more targeted fix might be to just limit the number of recursive
decryptions that happen, since the MIME spec does not appear to be
capable of permitting other parts to provide both compression and
nesting, which is the root of the problem.  However, it's possible that
our MIME parsing ends up being more permissive than the specs, and is
willing to try to interpret (for example) a gzip-compressed multipart/*
or message/* part.  

So the simple/generalized solution is probably the right way to go.

Sorry i don't have time to implement the fix myself right now, but i
wanted to make sure the active coders in the project are aware of the
issue.

Regards,

  --dkg

references:

 http://mumble.net/~campbell/blag.txt (see 2013-10-08)
 http://www.openwall.com/lists/oss-security/2013/10/10/2


pgpeTZ94z3BIt.pgp
Description: PGP signature
___
notmuch mailing list
notmuch@notmuchmail.org
http://notmuchmail.org/mailman/listinfo/notmuch