KSM and cross-vm attack
This is a multi-part message in MIME format. --------------040006070006020209070007 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Hi, Maybe I should be posting to the kvm mailing list, but I think people here should know a thing or two about it. I just read the following research paper and although the attack was done on VMWare; from what I read about it, it could be possible with KSM on KVM also. If you really need tight security it looks like it would be better to disable KSM. But don't take my word for it as IANAC (I Am Not A Cryptographer). http://soylentnews.org/article.pl?sid=14/06/12/1349234&from=rss Practical Cross-VM AES Full Key Recovery Attack <http://soylentnews.org/article.pl?sid=14/06/12/1349234> posted by janrinok <http://soylentnews.org/%7Ejanrinok/> on Thursday June 12, @02:53PM ** dbot <http://soylentnews.org/%7Edbot/> writes: Researchers from Worcester Polytechnic Institute (Worcester, MA), have published a paper illustrating a practical full Advanced Encryption Standard key recovery from AES operations preformed in one virtual machine, by another VM <http://eprint.iacr.org/2014/435.pdf> [*PDF*] running on the same hardware at the same time. The attack specifically requires memory de-duplication to be enabled, and they target VMWare's VM software. Combining various attacks on memory de-duplication, and existing side channel attacks: In summary, this works: * shows for the first time that de-duplication enables fine grain cross-VM attacks; * introduces a new Flush+Reload based attack that does not require interrupting the victim after each encryption round; * presents the first practical cross-VM attack on AES; the attack is generic and can be adapted to any table-based block ciphers. They target OpenSSL 1.0.1. It will be interesting to see if the suggested countermeasure, flushing the T table cache after each operation (effective against other Flush+Reload attacks), is added to LibreSSL. Will it be left out, in the name of performance - or will they move to a different implementation of AES (not T table-based)? Kind regards, Jorick Astrego Netbulae --------------040006070006020209070007 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit <html> <head> <meta http-equiv="content-type" content="text/html; charset=ISO-8859-1"> </head> <body text="#000000" bgcolor="#FFFFFF"> Hi,<br> <br> Maybe I should be posting to the kvm mailing list, but I think people here should know a thing or two about it.<br> <br> I just read the following research paper and although the attack was done on VMWare; from what I read about it, it could be possible with KSM on KVM also. If you really need tight security it looks like it would be better to disable KSM. <br> <br> But don't take my word for it as IANAC (I Am Not A Cryptographer). <br> <blockquote><a class="moz-txt-link-freetext" href="http://soylentnews.org/article.pl?sid=14/06/12/1349234&from=rss">http://soylentnews.org/article.pl?sid=14/06/12/1349234&from=rss</a><br> <div class="generaltitle"> <div class="title"> <h3> <a href="http://soylentnews.org/article.pl?sid=14/06/12/1349234">Practical Cross-VM AES Full Key Recovery Attack</a> </h3> </div> </div> <div class="body"> posted by <a href="http://soylentnews.org/%7Ejanrinok/"> janrinok</a> on Thursday June 12, @02:53PM <br> <strong></strong> <div class="intro"> <p class="byline"> <a href="http://soylentnews.org/%7Edbot/">dbot</a> writes:</p> <p>Researchers from Worcester Polytechnic Institute (Worcester, MA), have published a paper illustrating a <a href="http://eprint.iacr.org/2014/435.pdf">practical full Advanced Encryption Standard key recovery from AES operations preformed in one virtual machine, by another VM</a> [<b>PDF</b>] running on the same hardware at the same time.</p> <p>The attack specifically requires memory de-duplication to be enabled, and they target VMWare's VM software. Combining various attacks on memory de-duplication, and existing side channel attacks:</p> <blockquote> <div> <p>In summary, this works:</p> <ul> <li>shows for the first time that de-duplication enables fine grain cross-VM attacks;</li> <li>introduces a new Flush+Reload based attack that does not require interrupting the victim after each encryption round;</li> <li>presents the first practical cross-VM attack on AES; the attack is generic and can be adapted to any table-based block ciphers.</li> </ul> </div> </blockquote> <p>They target OpenSSL 1.0.1.</p> <p>It will be interesting to see if the suggested countermeasure, flushing the T table cache after each operation (effective against other Flush+Reload attacks), is added to LibreSSL. Will it be left out, in the name of performance - or will they move to a different implementation of AES (not T table-based)?</p> </div> </div> </blockquote> Kind regards,<br> <br> Jorick Astrego<br> Netbulae<br> </body> </html> --------------040006070006020209070007--
Hi, it's kind of you to let those know about these attacks who do not already know them, but this should be well understood by every professional by know. Shared resources are never secure, if you can not control the access from third parties to shared memory. this does not just affect KSM (or similar techniques from vmware, xen and microsoft) but also L3-Caches of modern CPUs. If you are interested in these topics, here are some papers: L3-Side-Channel attack to recover private GPG-Keys from another VM: http://eprint.iacr.org/2013/448.pdf Correlation attack against openssl, polarssl and libgcrypt on xen and vmware: https://eprint.iacr.org/2014/248.pdf I don't know if IBMs PowerVM is vulnerable to such attacks, as it's LPAR architecture is certified EAL 4+ (which might not tell anything about this attack vector). But you always need to have in mind, what attack scenario you talk about: These attacks are about a malicious vm (this could be a hacked/hijacked vm) which recovers parts of the shared memory from a known other instance to attack. if you have high security concerns you might want _not_ to share your physical server with third party controlled vms, or with vms which might be the target of getting hacked (or which runs software, which is known to be vulnerable). I still consider this scenario not as that relevant today, as there are many more low hanging fruits (sadly). This means in short: For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm. There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them). As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys. HTH -- Mit freundlichen Grüßen / Regards Sven Kieske Systemadministrator Mittwald CM Service GmbH & Co. KG Königsberger Straße 6 32339 Espelkamp T: +49-5772-293-100 F: +49-5772-293-333 https://www.mittwald.de Geschäftsführer: Robert Meyer St.Nr.: 331/5721/1033, USt-IdNr.: DE814773217, HRA 6640, AG Bad Oeynhausen Komplementärin: Robert Meyer Verwaltungs GmbH, HRB 13260, AG Bad Oeynhausen
This is a multi-part message in MIME format. --------------030204050405060907020301 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Hi Sven, Thanks for you response, I will read some more. But as you say it has been known for a while and I was aware of it for many years although never diving into the specifics. I always thought it was not a practical attack vector What caught my attention was that it was so fast it can be done in less then a minute: Lightning-Fast Attack: *Even in the worst case scenario (cross-VM) the attack** **succeeds in less than a minute. To the best of our knowledge, no faster attack** **has been implemented against AES in a realistic cloud-like setting. This also** **means that just one minute of co-location with the encryption server suffices to** **recover the key.*
For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm.
So that was the part that worries me, if I have a public cloud offering and someone doesn't hack a vm but simply rents one. He can then spawn a new VM every couple of minutes and it will probably be on a different host each time..... with different neighbours. You could hack every vulnerable customer VM in a couple of hours this way and it would all be undetected.
There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them).
As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys. I'm sure there are automatic tools being built as we speak but they will not be generally available.
Kind regards, Jorick Astrego Netbulae B.V. On 06/13/2014 09:38 AM, Sven Kieske wrote:
Hi,
it's kind of you to let those know about these attacks who do not already know them, but this should be well understood by every professional by know.
Shared resources are never secure, if you can not control the access from third parties to shared memory.
this does not just affect KSM (or similar techniques from vmware, xen and microsoft) but also L3-Caches of modern CPUs.
If you are interested in these topics, here are some papers:
L3-Side-Channel attack to recover private GPG-Keys from another VM:
http://eprint.iacr.org/2013/448.pdf
Correlation attack against openssl, polarssl and libgcrypt on xen and vmware:
https://eprint.iacr.org/2014/248.pdf
I don't know if IBMs PowerVM is vulnerable to such attacks, as it's LPAR architecture is certified EAL 4+ (which might not tell anything about this attack vector).
But you always need to have in mind, what attack scenario you talk about:
These attacks are about a malicious vm (this could be a hacked/hijacked vm) which recovers parts of the shared memory from a known other instance to attack.
if you have high security concerns you might want _not_ to share your physical server with third party controlled vms, or with vms which might be the target of getting hacked (or which runs software, which is known to be vulnerable).
I still consider this scenario not as that relevant today, as there are many more low hanging fruits (sadly).
This means in short:
For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm.
There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them).
As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys.
HTH
--------------030204050405060907020301 Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: 7bit <html> <head> <meta content="text/html; charset=UTF-8" http-equiv="Content-Type"> </head> <body text="#000000" bgcolor="#FFFFFF"> Hi Sven,<br> <br> Thanks for you response, I will read some more.<br> <br> But as you say it has been known for a while and I was aware of it for many years although never diving into the specifics. I always thought it was not a practical attack vector<br> <br> What caught my attention was that it was so fast it can be done in less then a minute:<br> <br> <blockquote>Lightning-Fast Attack: <b>Even in the worst case scenario (cross-VM) the attack</b><b><br> </b><b>succeeds in less than a minute. To the best of our knowledge, no faster attack</b><b><br> </b><b>has been implemented against AES in a realistic cloud-like setting. This also</b><b><br> </b><b>means that just one minute of co-location with the encryption server suffices to</b><b><br> </b><b>recover the key.</b><br> </blockquote> <br> <pre wrap=""><blockquote type="cite"><pre wrap="">For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm.</pre> </blockquote> </pre> So that was the part that worries me, if I have a public cloud offering and someone doesn't hack a vm but simply rents one. He can then spawn a new VM every couple of minutes and it will probably be on a different host each time..... with different neighbours.<br> <br> You could hack every vulnerable customer VM in a couple of hours this way and it would all be undetected. <br> <br> <blockquote type="cite"> <pre wrap="">There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them). As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys.</pre> </blockquote> I'm sure there are automatic tools being built as we speak but they will not be generally available.<br> <br> <br> Kind regards,<br> <br> Jorick Astrego<br> Netbulae B.V.<br> <br> <br> <br> <div class="moz-cite-prefix">On 06/13/2014 09:38 AM, Sven Kieske wrote:<br> </div> <blockquote cite="mid:539AAAC8.3080601@mittwald.de" type="cite"> <pre wrap="">Hi, it's kind of you to let those know about these attacks who do not already know them, but this should be well understood by every professional by know. Shared resources are never secure, if you can not control the access from third parties to shared memory. this does not just affect KSM (or similar techniques from vmware, xen and microsoft) but also L3-Caches of modern CPUs. If you are interested in these topics, here are some papers: L3-Side-Channel attack to recover private GPG-Keys from another VM: <a class="moz-txt-link-freetext" href="http://eprint.iacr.org/2013/448.pdf">http://eprint.iacr.org/2013/448.pdf</a> Correlation attack against openssl, polarssl and libgcrypt on xen and vmware: <a class="moz-txt-link-freetext" href="https://eprint.iacr.org/2014/248.pdf">https://eprint.iacr.org/2014/248.pdf</a> I don't know if IBMs PowerVM is vulnerable to such attacks, as it's LPAR architecture is certified EAL 4+ (which might not tell anything about this attack vector). But you always need to have in mind, what attack scenario you talk about: These attacks are about a malicious vm (this could be a hacked/hijacked vm) which recovers parts of the shared memory from a known other instance to attack. if you have high security concerns you might want _not_ to share your physical server with third party controlled vms, or with vms which might be the target of getting hacked (or which runs software, which is known to be vulnerable). I still consider this scenario not as that relevant today, as there are many more low hanging fruits (sadly). This means in short: For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm. There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them). As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys. HTH </pre> </blockquote> <br> </body> </html> --------------030204050405060907020301--
On Fri, Jun 13, 2014 at 11:05:37AM +0200, Jorick Astrego wrote:
Hi Sven,
Thanks for you response, I will read some more.
But as you say it has been known for a while and I was aware of it for many years although never diving into the specifics. I always thought it was not a practical attack vector
What caught my attention was that it was so fast it can be done in less then a minute:
Lightning-Fast Attack: *Even in the worst case scenario (cross-VM) the attack** **succeeds in less than a minute. To the best of our knowledge, no faster attack** **has been implemented against AES in a realistic cloud-like setting. This also** **means that just one minute of co-location with the encryption server suffices to** **recover the key.*
For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm.
So that was the part that worries me, if I have a public cloud offering and someone doesn't hack a vm but simply rents one. He can then spawn a new VM every couple of minutes and it will probably be on a different host each time..... with different neighbours.
You could hack every vulnerable customer VM in a couple of hours this way and it would all be undetected.
There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them).
As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys. I'm sure there are automatic tools being built as we speak but they will not be generally available.
It would be relatively simple to disable KSM per VM. This way, a customer that values security more than density, could pay more to keep his memory pages unscanned by KSM. Anyone cares to open an RFE for that? I remember that the idea was discussed, but I do not find a formal bug that tracks this.
----- Original Message -----
From: "Dan Kenigsberg" <danken@redhat.com> To: "Jorick Astrego" <j.astrego@netbulae.eu>, alitke@redhat.com, doron@redhat.com Cc: users@ovirt.org Sent: Friday, June 13, 2014 1:36:17 PM Subject: Re: [ovirt-users] KSM and cross-vm attack
On Fri, Jun 13, 2014 at 11:05:37AM +0200, Jorick Astrego wrote:
Hi Sven,
Thanks for you response, I will read some more.
But as you say it has been known for a while and I was aware of it for many years although never diving into the specifics. I always thought it was not a practical attack vector
What caught my attention was that it was so fast it can be done in less then a minute:
Lightning-Fast Attack: *Even in the worst case scenario (cross-VM) the attack** **succeeds in less than a minute. To the best of our knowledge, no faster attack** **has been implemented against AES in a realistic cloud-like setting. This also** **means that just one minute of co-location with the encryption server suffices to** **recover the key.*
For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm.
So that was the part that worries me, if I have a public cloud offering and someone doesn't hack a vm but simply rents one. He can then spawn a new VM every couple of minutes and it will probably be on a different host each time..... with different neighbours.
You could hack every vulnerable customer VM in a couple of hours this way and it would all be undetected.
There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them).
As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys. I'm sure there are automatic tools being built as we speak but they will not be generally available.
It would be relatively simple to disable KSM per VM. This way, a customer that values security more than density, could pay more to keep his memory pages unscanned by KSM.
Anyone cares to open an RFE for that? I remember that the idea was discussed, but I do not find a formal bug that tracks this.
Indeed we had an old bz for it. A brand new ovirt RFE will help. Also, it is possible to disable ksm for a cluster today. So it's already possible to have an optimized cluster and a more strict one in the same DC.
----- Original Message -----
From: "Doron Fediuck" <dfediuck@redhat.com> To: "Dan Kenigsberg" <danken@redhat.com> Cc: users@ovirt.org Sent: Friday, June 13, 2014 2:14:30 PM Subject: Re: [ovirt-users] KSM and cross-vm attack
----- Original Message -----
From: "Dan Kenigsberg" <danken@redhat.com> To: "Jorick Astrego" <j.astrego@netbulae.eu>, alitke@redhat.com, doron@redhat.com Cc: users@ovirt.org Sent: Friday, June 13, 2014 1:36:17 PM Subject: Re: [ovirt-users] KSM and cross-vm attack
On Fri, Jun 13, 2014 at 11:05:37AM +0200, Jorick Astrego wrote:
Hi Sven,
Thanks for you response, I will read some more.
But as you say it has been known for a while and I was aware of it for many years although never diving into the specifics. I always thought it was not a practical attack vector
What caught my attention was that it was so fast it can be done in less then a minute:
Lightning-Fast Attack: *Even in the worst case scenario (cross-VM) the attack** **succeeds in less than a minute. To the best of our knowledge, no faster attack** **has been implemented against AES in a realistic cloud-like setting. This also** **means that just one minute of co-location with the encryption server suffices to** **recover the key.*
For the most parts, it's easier to hack you machine directly or social-engineer your way into it, than it is to hack/get access to a different vm on the same system and than hack another vm.
So that was the part that worries me, if I have a public cloud offering and someone doesn't hack a vm but simply rents one. He can then spawn a new VM every couple of minutes and it will probably be on a different host each time..... with different neighbours.
You could hack every vulnerable customer VM in a couple of hours this way and it would all be undetected.
There are also still no automatic tools for this, which I'm aware of (if they are, I'd like to be pointed to them).
As soon as automatic attack tools will cover this scenario I'm pretty sure we'll see an increase in hacked vms and sniffed private keys. I'm sure there are automatic tools being built as we speak but they will not be generally available.
It would be relatively simple to disable KSM per VM. This way, a customer that values security more than density, could pay more to keep his memory pages unscanned by KSM.
Anyone cares to open an RFE for that? I remember that the idea was discussed, but I do not find a formal bug that tracks this.
Indeed we had an old bz for it. A brand new ovirt RFE will help. Also, it is possible to disable ksm for a cluster today. So it's already possible to have an optimized cluster and a more strict one in the same DC.
One more thing; For future reference, please us the procedures details here: http://www.ovirt.org/Security For anything which may impact other users as well. Thanks, Doron
This site misses public keys for sending encrypted mails. That's not that good for a security related mail. I'm sure it just isn't mentioned in the wiki, could one use the same keys as for redhat security mailings? Am 13.06.2014 13:16, schrieb Doron Fediuck:
One more thing;
For future reference, please us the procedures details here: http://www.ovirt.org/Security For anything which may impact other users as well.
-- Mit freundlichen Grüßen / Regards Sven Kieske Systemadministrator Mittwald CM Service GmbH & Co. KG Königsberger Straße 6 32339 Espelkamp T: +49-5772-293-100 F: +49-5772-293-333 https://www.mittwald.de Geschäftsführer: Robert Meyer St.Nr.: 331/5721/1033, USt-IdNr.: DE814773217, HRA 6640, AG Bad Oeynhausen Komplementärin: Robert Meyer Verwaltungs GmbH, HRB 13260, AG Bad Oeynhausen
Done that: https://bugzilla.redhat.com/show_bug.cgi?id=1109157 Am 13.06.2014 12:36, schrieb Dan Kenigsberg:
It would be relatively simple to disable KSM per VM. This way, a customer that values security more than density, could pay more to keep his memory pages unscanned by KSM.
Anyone cares to open an RFE for that? I remember that the idea was discussed, but I do not find a formal bug that tracks this.
-- Mit freundlichen Grüßen / Regards Sven Kieske Systemadministrator Mittwald CM Service GmbH & Co. KG Königsberger Straße 6 32339 Espelkamp T: +49-5772-293-100 F: +49-5772-293-333 https://www.mittwald.de Geschäftsführer: Robert Meyer St.Nr.: 331/5721/1033, USt-IdNr.: DE814773217, HRA 6640, AG Bad Oeynhausen Komplementärin: Robert Meyer Verwaltungs GmbH, HRB 13260, AG Bad Oeynhausen
participants (4)
-
Dan Kenigsberg -
Doron Fediuck -
Jorick Astrego -
Sven Kieske