Slashdot Mirror
New ssh Exploit in the Wild
veg writes "In the last few hours there have been several reports of a new ssh bug, with an exploit seemingly in the wild. Oh god not again... The lengths some people will goto to try and damage Theo's pride." Update: 09/17 00:24 GMT by T : friscolr writes "Hot on the heels of rev 1 of the buffer.adv advisory, here is revision 2, which fixes more than revision 1 did. Also see the 3.7.1 release notes."
Best patch and upgr..&*[NO CARRIER]
C:\>
I have to wonder if UsePrivilegeSeparation was enabled. (see the manpage)
One message in the thread indicates it is but this isn't first-hand knowledge. If PrivSep was enabled then is OpenBSD immune to this attack due to other parts of the OS being hardened (much like the zlib hole a few months back)? Also are these default installations or are they "tweaked"? As an aside, PermitRootLogin defaults to enabled, something I always disable as I have no need for it.
Even if this does count as a new remote hole in OpenBSD, it's still a phenomenal track record they can be proud of.
Trolling is a art,
Looks like its time to turn the port forwarding on my router off, and wait for apple to provide a patch.
The advisory itself says The systems in question are FreeBSD, RedHat, Gentoo, and Debian all running the latest versions of OpenSSH. So i'm going to assume that OS X is affected as well.
Look out honey cause I'm usin' technology
Ain't got time to make no apologies
Posting this to slashdot is actually a public service, as the exploit description will be /.'d and unable to effectively be disseminated to the bad actors.
Link to a patch.
[Full-Disclosure] new ssh exploit?
christopher neitzert chris@neitzert.com
Mon, 15 Sep 2003 13:48:34 -0400
More on this;
The systems in question are FreeBSD, RedHat, Gentoo, and Debian all
running the latest versions of OpenSSH.
The attack makes an enormous amount of ssh connections and attempts
various offsets until it finds one that works permitting root login.
I have received numerous messages from folks requesting anonymity or
direct-off-list-reply confirming this exploit;
The suggestions I have heard are:
Turn off SSH and
1. upgrade to lsh
2. add explicit rules to your edge devices allowing ssh from only-known
hosts.
3. put ssh behind a VPN on RFC-1918 space.
On Mon, 2003-09-15 at 12:02, christopher neitzert wrote:
> Does anyone know of or have source related to a new, and unpublished ssh
> exploit? An ISP I work with has filtered all SSH connections due to
> several root level incidents involving ssh. Any information is
> appreciated.
Thank god I'm using something secure like Telnet instead.
Reading the mailing list, it appears that there's nothing confirmed so far. Let's hope its just a false rumour.
There's only one guy that says it its ISP has blocked all incoming SSH connection due to "several root level incidents".
One guy did say that there was a bug somewhere and that a patch existed...No one knows what patch or where it is though.
Let's hope to publish this one quickly before there's any ral damage done.
IP Therefore I am.
Won't having the sshd wrapped (/etc/hosts.allow, /etc/hosts.deny) help offset the damage somewhat?
No sig
Yes, there is a vuln. in 3.6. You need to upgrade to 3.7 which was released today, to be safe (well, 'safer' anyway).
It will be 3.7p1 for us non-OpenBSD people.
It is a patch to one file, buffer.c, which fixes some allocation/offset stuff.
It seems that privilege separation does *not* help here - so get them systems patched (and firewalled)!
Damn trinity and her sshnuke...
How worried should I really be about this? And what steps should I be taking (or ask dad to take)? Since I gather Lindows is similar to Debian, should I just look for a Debian tutorial?
Thanks in advance.
This has already been posted and a fix (upgrade to 3.7) has been posted to www.deadly.org
This guy is way out there
Great, now maybe Redhat will fix their damn openssh RPMs that they fubarred with their last patch!
Anything is possible given time and money.
On the other hand, it's good to have the heads up if something might not be as secure as we think it is. This warning gives those who turn it on occasionally the knowledge they need to turn it off if not needed, and not just leave it on.
It also may give those who need it on something to watch for until a patch does come out.
R: That voice. Where have I heard that voice before? B: In about 365 other episodes. But I don't know who it is either.
A librarian peeked around the corner to see where the noise was coming from, then put her finger to her lips and said, "Ssh!"
The kids ignored her and kept talking, completely and utterly exploiting the hole, and circumventing the 'Ssh'!
Never was I so frightened.
I appreciate it when Slashdot informs me of a patch I need to apply, but really, I'd rather hear about it once the exploit is actually understood and the patch is available.
Really?
How about hearing about it when you find your machines rooted?
Even though there is no patch available (yet), this heads-up is extremely valuable, as it allows people who cannot afford to be compromised to shut down or appropriately filter SSH on their systems.
Kaa
Kaa's Law: In any sufficiently large group of people most are idiots.
An updated ssh package just hit the Debian security mirrors.
For anyone running debian stable:
apt-get update
apt-get upgrade
A patch for Debian stable is available already. If you're running Debian on a server and have ssh installed, "apt-get update; apt-get upgrade" should pick it up. The new package version is 1:3.4p1-1.1.
-Stephen
... can be found here http://archives.neohapsis.com/archives/fulldisclo
You should note that comment came from Darren Reed (of IPFilter fame), who is not a fan of Theo. After the arguements over the IPFilter licensing, a comment like this coming from him doesn't suprise me at all.
technoid
Two wrongs don't make a right, but 3 lefts do - Lew of GO magazine
- cp openssh-3.6.1_p2.ebuild openssh-3.7_p1.ebuild
- emerge --update openssh
The emerge will fetch the file and complain that there is no digest.- ebuild openssh-3.7_p1.ebuild digest
- emerge --update openssh
Just tested it here, worked fine.Pat
Rather than subject someone's server (like mine!) to a slashdotting, here's the full text of the announcement (slightly mangled to sneak past the lameness filter).
Subject: OpenSSH 3.7 released
Date: Tue, 16 Sep 2003 14:07:00 +0200
From: Markus Friedl
To: openssh-unix-dev _at_ mindrot.org
OpenSSH 3.7 has just been released. It will be available from the mirrors listed at http://www.openssh.com/ shortly.
OpenSSH is a 100% complete SSH protocol version 1.3, 1.5 and 2.0 implementation and includes sftp client and server support.
We would like to thank the OpenSSH community for their continued support to the project, especially those who contributed source and bought T-shirts or posters.
We have a new design of T-shirt available, more info on http://www.openbsd.org/tshirts.html#18
For international orders use http://https.openbsd.org/cgi-bin/order and for European orders, use http://https.openbsd.org/cgi-bin/order.eu
Security Changes:
All versions of OpenSSH's sshd prior to 3.7 contain a buffer management error. It is uncertain whether this error is potentially exploitable, however, we prefer to see bugs fixed proactively.
OpenSSH 3.7 fixes this bug.
Changes since OpenSSH 3.6.1:
* The entire OpenSSH code-base has undergone a license review. As a result, all non-ssh1.x code is under a BSD-style license with no advertising requirement. Please refer to README in the source distribution for the exact license terms.
* Rhosts authentication has been removed in ssh(1) and sshd(8).
* Changes in Kerberos support:
- KerberosV password support now uses a file cache instead of a memory cache.
- KerberosIV and AFS support has been removed.
- KerberosV support has been removed from SSH protocol 1.
- KerberosV password authentication support remains for SSH protocols 1 and 2.
- This release contains some GSSAPI user authentication support to replace legacy KerberosV authentication support. At present this code is still considered experimental and SHOULD NOT BE USED.
* Changed order that keys are tried in public key authentication. The ssh(1) client tries the keys in the following order:
1. ssh-agent(1) keys that are found in the ssh_config(5) file
2. remaining ssh-agent(1) keys
3. keys that are only listed in the ssh_config(5) file
This helps when an ssh-agent(1) has many keys, where the sshd(8) server might close the connection before the correct key is tried.
* SOCKS5 support has been added to the dynamic forwarding mode in ssh(1).
* Removed implementation barriers to operation of SSH over SCTP.
* sftp(1) client can now transfer files with quote characters in their filenames.
* Replaced sshd(8)'s VerifyReverseMapping with UseDNS option. When UseDNS option is on, reverse hostname lookups are always performed.
* Fix a number of memory leaks.
* Support for sending tty BREAK over SSH protocol 2.
* Workaround for other vendor bugs in KEX guess handling.
* Support for generating KEX-GEX groups (/etc/moduli) in ssh-keygen(1).
* Automatic re-keying based on amount of data sent over connection.
* New AddressFamily option on client to select protocol to use (IPv4 or IPv6).
* Experimental support for the "aes128-ctr", "aes192-ctr", and "aes256-ctr" ciphers for SSH protocol 2.
* Experimental support for host keys in DNS (draft-ietf-secsh-dns-xx.txt). Please see README.dns in the source distribution for details.
* Portable OpenSSH:
- Replace PAM password authentication kludge with a more correct PAM challenge-response module from FreeBSD.
- PAM support may now be enabled/disabled at runtime using the UsePAM directive.
- Many improvements to the OpenSC smartcard support.
- Regression tests now work with portable OpenSSH. Please refer to regress/README.regress in t
$ find
As opposed to the lengths people will go to to damage Microsoft? But that's ok, right?
An OpenSSH Security Advisory was just posted about this.
So I hear about this ssh exploit the exact same day that my inbox has Markus Friedl's announcement of the release of OpenSSH 3.7.
Either someone on the ssh team is making money from new releases or some black hat, upon downloading 3.7 and seeing the exploit fixed, decided to strike while the iron was still hot (machines weren't yet upgraded).
"Provided by the management for your protection."
Is it? I've successfully exploited my sshd (thank God for easy filtering with PF!)
o mpile/GENERIC
# dmesg | head -n2
OpenBSD 3.4-current (GENERIC) #62: Tue Sep 12 22:49:18 MDT 2003
deraadt@i386.openbsd.org:/usr/src/sys/arch/i386/c
If you don't want to wait for the official ebuild:
/usr/portage/net-misc/openssh/
cd
cp openssh-3.6.1_p2.ebuild openssh-3.7_p1.ebuild
emerge -f openssh-3.7_p1.ebuild
ebuild openssh-3.7_p1.ebuild digest
emerge openssh-3.7_p1.ebuild
Even though there is no patch available (yet)
There is a patch available, as well as it being fixed in 3.7, which was just released this morning. That's the point of all of this. The mention of the bug was in the 3.7 release notes, i believe.
It seems to me that a package that goes through code security audits regularly and is actually finished is infinitely more secure than an incomplete package?
Why are there people suggesting to go from a secure package to an insecure one?
I've released the SOURCE RPM...
you can always grab it and see for yourself..
I only changed buffer.c
Feel free to see for yourself..
I had to make all of these this morning to patch our systems..
ChiefArcher
This is the README from 1998, talking about a beta version of lsh. Don't let age-old doumentation fool you.
lsh has grown mature since then, and has an excellent code quality. I recommend it. Any day over OpenSSH, after having looked at the code of both projects. Up-to-date documentation, as on the web page, or the README inside the tarball, doesn't contain the warning.
you have an email address to...
and a resume www.briangannon.com
and the Source RPMS.
http://stradlin.com/ssh
if you do a diff on the sources, you will see I only edited buffer.c
my intentions are completely noble.
How can you really trust Redhat? One of the disgruntled developers could put a backdoor in a patch?
ChiefArcher
What a troll. Aiming to trick mods into "Informative", I suppose.
Any "linux user" who has openssh open to the world is a huge dumbass. What part of "firewall rules" don't you understand?
How would you suggest it be configured then? Just turn off remote login entirely? Or what other "firewall rule" could help in this situation?
I assume you are suggesting that people only allow ssh access from a specific, previously-known host. That removes much of ssh's utility (no more checking your system from a laptop in the hotel room), and even that sacrifice is not enough to be protective!
An attacker sniffing packets at your ISP can learn exactly which addresses you accept ssh connections over. Then he can spoof from that same address, and go right through the firewall.
The only way to protect yourself from unwanted outside connections is with correct crypto code.
A significant number of changes in 3.7 are removals (Kerberos 4, Kerberos5 in SSH1, AFS, Rhosts auth). Most people agree that simplicity is a wonderful goal... until that means the dropping (or not including) the feature they need or want. Then simplicity versus functionality versus security becomes a balancing act.
/usr/local/sbin/sshd /usr/local/bin/ssh /usr/local/sbin/sshd /usr/local/bin/ssh
To put the size comment in perspective (this is 3.7p1 on Linux/x86):
$ du -ks
272
224
$ find
I just read all these replies (about 15 right now) and all of them are nice and respectfull of the fact that this guy is a newbie!
I must be on the wrong site.
NarratorDan
"If you're not confused by quantum mechanics, you really don't understand it." - Niels Bohr
The bug must center around this line:
It looks like the problem here is that buffer->alloc (which presumably stores the size of the buffer) grows on every try, while the actual size of the buffer grows only on successful tries. So you could have a situation where, after a couple of tries, the buffer is 65536, but buffer->alloc is 98304. This could potentially cause another part of the program to run past the actual end of the buffer.
The patch addresses this by only updating buffer->alloc after the new memory has been successfully allocated.
They changed it from 0 to 1 when the last SSH vuln was disclosed. I see no reason that thye wouldn't do it this time. However, it's not afflicting OperBSD anyways...
And as comparison, how many patches do windows users normally need to install over the 'default install' to get it secure and close every hole in the default setup? Methinks slightly more than 1 or 2...
Ssh, don't tell anyone.
Omnis amans amens
None of the rants assumed it couldn't happen.
The nature of _my_ rants at least, include the following:
* UNIX does better at risk minimization (i.e. - chroot jails, more services running as unprivileged users, using processes rather than threads, etc)
* UNIX vulnerabilities are published quickly, and hotfixes are available quickly. In this case, we have a _potential_ vulnerability patched before anyone knows of any way to exploit it. In addition, it made frontpage Slashdot - everyone agrees it's a big deal. This is different from the MS attitude of "sweep it into the next service pack and noone will know".
* I have the source code to the patches, so I can validate whether or not the fix does indeed fix the problem it proposes to, and whether there will be any other impact caused by the patch.
* The patch doesn't require me to reboot anything - I can patch a running system and keep on trucking. Kernel patches should be the only thing that needs a reboot (and, when HURD gets mainstream, we won't even need to then).
* The source code is open to allow more scrutiny. Having the source code available still gives Linux users fewer security-incidents-per-feature than Microsoft while keeping their source closed. Ballmer, I believe, said under oath that giving out the source code to Windows publicly would be a threat to national security.
Nothing about the release of an exploit for Linux changes any of these issues.
Engineering and the Ultimate
Easy, the buffer for a period of time is not as big as the struct indicates. So between the time the buffer size variable is set and the time the realloc is performed, there is a window of opportunity to fill that buffer beyond is alloced boundary if another process/thread goes to put stuff in that buffer at the wrong time.
Also, I'm not sure exactly how 'fatal' the fatal is in there, it could be that it breaks out of the function, but leaves the buffer size parameter at too large a value for the alloc in that buffer.
One or both of those problems is exploitable.
XML is like violence. If it doesn't solve the problem, use more.
It appears that the OpenSSH people found this bug first, and released a fix in version 3.7. People who studied this fix then found the exploit. So it's stupid for this guy to tell people "upgrade to lsh", since the whole reason his buds know about this bug is because 3.7 fixes it.
That's right! It can form remote connections, and generate random keys, and... and... uh, well, that's about it, actually. Form connections, generate session keys.
Public/private key generation? Different program. Managing keys on a local machine? Different program. Transferring files securely? Different (wrapper) program.
Got any concrete suggestions there? Exactly how would you divide the existing tools up? Precisely which tools would you create? In what ways -- details, now -- would they be different from the half-dozen programs that come with ssh now?
You cannot apply a technological solution to a sociological problem. (Edwards' Law)
You are already running windows. You have more serious problems.
...why I always go back and add security holes to all of my programs. If some future (or current) anti-regime hacker needs to be able to break into a local power plant, I want to make sure my code can help!
[I considered signing this post "love, Theo" but then thought better of it.]
You cannot apply a technological solution to a sociological problem. (Edwards' Law)
A demonstration would be nice.
:-)
It'd serve you right if he gave you one.
It appears that *nix systems now have an exploit
Yep, *nix systems have exploits, and an hour or two turnaround between discovery and a fix. I'd like to see Microsoft match that.
"Linux has no exploits that need patching"
People who actually know Linux would never claim that there are no known exploits, just that the time-to-fix is much shorter and that applying these fixes to running systems is usually much easier (in most distributions) than in a Windows system (ie no reboot required, one location for all necessary fixes, better software package management, etc)
I use Linux and BSD at home, but manage Windows machines at work (I have no decision making power, I'm just a grunt) and I must say that Windows management is a royal pain in the ass. We've had no problems with the recent Windows viruses and worms, but I do spend an inordinate amount of time applying patches, rebooting machines, and checking that the new patches did not wipe out the old ones. I don't think that it is unreasonable for Microsoft's customers to demand better patch/upgrade management, a single location for updates to both applications, servers, and the OS, and a better method for confirming that the files included in a patch contain the all of the required fixes (for that file) even if they came from different departments at microsoft.
Read, L
you can restrict access in your /etc/sshd_config (wherever you have it) like so until you can get the patched version, if you allow access from anywhere:
DenyUsers *
AllowUsers you@your_ip_address
(and restart sshd)
You can also firewall the port off. I've done a hodge-podge of these solutions on different systems I admin until I can actually get the 3.7p1 source from the mirrors (they dont' seem to have it yet).
-- http://frobnosticate.com
The only really secure server is buried in concrete, unlugged and at the bottom of the deepest trench in the ocean. It's *probably* secure there, but I wouldn't bet my life on it.
That's okay, I will.
I bet this guy's life that a server on the bottom of the ocean is secure.
Obviously the *NIX side of the world isn't bulletproof either. Now perhaps we might be spared (at least for a day or two) about the anti-M$ rants about insecure M$ code. It can happen, and it can happen regardless of OS platform.
The MS rants are well deserved.
While your statement about security bugs can happen on any platform is technically correct, unintended bugs are not the only thing that causes security problems. Both MS and *NIX can have unintentional bugs, which lead to security problems. In this case, MS should not be blamed for "insecure" code.
Where the MS rants are well deserved is when a system is insecure by design. It may not have been a design goal, but the design can still be insecure. Just one past example: IIS runs under the SYSTEM account. It is installed by default and turned on by default. These kinds of problems deserve to be ranted about, and MS deserves the resulting reputation. Apache may or may not be installed and/or turned on by default, depending on distribution, but even if it could be compromised, it runs as "nobody" or "wwwrun" or some other unprivileged account.
Those who would give up liberty in exchange for security and DRM should switch to Microsoft Palladium!
I'm on a couple of the lists that should have been informed. As one example, NetBSD's security officer has received no information from the openssh team at all. I'm unaware of other groups having received official word.
If you are aware of a security team that was informed officially, I'd be interested to hear about it.
The systems in question are FreeBSD, RedHat, Gentoo, and Debian all
running the latest versions of OpenSSH.
The attack makes an enormous amount of ssh connections and attempts
various offsets until it finds one that works permitting root login.
Odd. I run Debian on all of my systems and PermitRootLogin is set to no on all of them. Sarge and Sid also have UsePrivilegeSeparation set to yes by default.
http://www.securityfocus.com/archive/1/337662/2003 -09-13/2003-09-19/0
No.
Besides, what have they "swept under the carpet"? What do you mean "you have probably"?Just because you seem to have something personal with Theo going on, we're supposed to take your word for this "deceit"?
Unlimited growth == Cancer.
Amazing how a newsworthy point about a ssh bug becomes an attack on an entire operating system and/or person.
"Given that the default install has ssh turned on, will they change it to "two remote holes" ?"
Yes, if they confirm the exploit. They've changed this notice in the past. It went from 0 to 1.
"Lets make some noise and force Theo to finally update that!"
Why? Just to piss off the developers? The openssh code is open and subject to review by anyone.
I think since you didn't catch this bug, we should all be asses and target you for harrassment.
"If you follow misc@ carefully you have probably seen it done before."
Bullshit. If you follow misc@, most of the exploits discussed hit previous unpatched versions of OpenBSD. The point of OBSD is to catch bugs and bad code ahead of time; it undergoes near constant review.
A lot of folks want OBSD to add to this count stuff OBSD noticed may have been exploitable, then patched it anyways, frequently weeks or months or years ahead of a known exploit. When the known exploit comes out, they point to the OBSD version 6 months ago.
Exploits are counted that can violate current, stable systems, not OBSD 2.8.
This is like blaming MS for the exploit that allowed slammer to spread; if people patched their systems when they were supposed to, they wouldn't have been inconvenienced. OTOH, MS should have caught the bug ahead of time.
I feel OBSD falls into the latter category, not the former. They are more than likely ahead of the game. Given what I've seen of security reports on Linux and FreeBSD over the past 2 years, OBSD tends to play catchup in coming up with fixes. Rather, they tend to fight the tide that their "policy" in reporting exploits is wrong.
Oddly, I think that is more a testament to them doing things right as opposed to your attitude that they are being purposefully deceitful.
Yeah those "NO CARRIER" jokes just aren't fun@~%4!.z^%r#$% NO CARRIER
Life is too short to proofread.
No, it would still be an ssh vulnerability.
Remember, we're supposed to seperate the OS and the apps that have the holes...remember?
Or are we still using the term "Windows hole" when referring to Outlook?
"Sufferin' succotash."
It kind of does. Seems that fatal() calls fatal_cleanup() [cf fatal.c], which runs thru a list of callback procedures before exiting. Somewhere in those callbacks the function buffer_free() [cf buffer.c] might be called on the offending buffer. That function does a memset(buffer->buf, 0, buffer->alloc), and there's your overrun.
QED.
The dog ate my
Redhat has finally posted patched RPMS on their errata pages. Scroll down and select your release.
Redhat just released an advisory with links to updated RPMS: RHSA-2003-279
The FreeBSD team has released a related Security Advisory and issued patches for affected FreeBSD versions as well as OpenSSH in the ports tree.
Corrected:
2003-09-16 16:24:02 UTC (RELENG_4)
2003-09-16 16:27:57 UTC (RELENG_5_1)
2003-09-16 17:34:32 UTC (RELENG_5_0)
2003-09-16 16:24:02 UTC (RELENG_4_8)
2003-09-16 16:45:16 UTC (RELENG_4_7)
2003-09-16 17:44:15 UTC (RELENG_4_6)
2003-09-16 17:45:23 UTC (RELENG_4_5)
2003-09-16 17:46:02 UTC (RELENG_4_4)
2003-09-16 17:46:37 UTC (RELENG_4_3)
2003-09-16 12:43:09 UTC (ports/security/openssh)
2003-09-16 12:43:10 UTC (ports/security/openssh-portable)
Ooops, I had totally forgotten about that old copy of the README file in the ftp archive. After it was pointed out to me in private mail, I've replaced it with the current README. /Niels (LSH author)
I've read the patch on SecurityFocus. It's very simple. It looks like there is a buffer that at some point needs to be reallocated to fit more data. However, if that reallocation would put the buffer over 32KB, then the reallocation is not performed. The bug was that the recorded size of the buffer would be increased, even if the allocation was not performed, but not reset back to the original size. The patch only increases the recorded buffer size if the buffer really is grown.
An exploit of this hole would have to find a way to trigger this buffer allocation, and get it to overflow 32KB. Since the allocation would not take place, but the recorded buffer size is still large, there would be an unallocated memory area that could be referenced by code that uses that buffer. One would need to use that to overwrite program code with exploit code.
A solution to the problem with music today
missing000's comment is quite correct, there's a mistake in my original post. Omit the DenyUsers line, it will override the AllowUsers line. Just use the AllowUsers line by itself.
Sorry.
AllowUsers you@your_ip_address
Remember, always test making a new ssh connection before logging out of your existing one, after restarting sshd.
-- http://frobnosticate.com
Why are they bothering with proper cleanup? This is FATAL CONDITION! ABANDON SHIP!
Only guessing, but how about to ensure that the freed memory isn't handed over to a subsequently-run app, still stuffed full of cryptographically-sensitive information?
Tubal-Cain smokes the white owl.
iptables:
#!/bin/sh
insmod iptables
iptables -F INPUT
iptables -P INPUT ACCEPT
iptables -A INPUT -j ACCEPT -p tcp --destination-port 22 -s 1.2.3.4
iptables -A INPUT -j DROP -p tcp --destination-port 22
iptables -A INPUT -j DROP -p udp --destination-port 22
ipchains:
#!/bin/sh
insmod ipchains
ipchains -F input
ipchains -P input ACCEPT
ipchains -A input -j ACCEPT -p tcp --destination-port 22 -s 1.2.3.4
ipchains -A input -j DENY -p tcp --destination-port 22
ipchains -A input -j DENY -p udp --destination-port 22
OpenSSH refuses to allocate a buffer over a certain size, but doesn't set the buffer size value back to its previous value. When this occurs, OpenSSH immediately calls fatal() to clean up and exit. The connection is closed, and the buffer is not reused.
The problem is that the cleanup code will, in some cases, attempt to zero the block at the larger size, resulting in OpenSSH crashing.
Because no data other than zeroes is every written to this buffer after the failed allocation (unless the FreeBSD folks missed something), this bug cannot be exploited except as a denial-of-service attack unless combined with some other exploit that allowed you to overwrite the exception handling vectors and add arbitrary code (in which case, there are probably much easier ways to perform the exploit).
120 character sigs suck. Make it 250.