Hi Nikolay, Good rule of thumb here - the more design is discussed prior to implementation, the better. There is no specific definition of well- or ill-defined design, this is more about common sense and our general experience. I would say that minimal set of things to be addressed for most major features is (TDE-specific comments are in parantheses): 1) Algorithms (how we encrypt/decrypt, when we encrypt/decrypt, etc) 2) Protocol changes (new messages, changes to old messages) 3) Changes to persistence (where to store encrypted keys, changes to WAL) 4) Public API and usability (key management interfaces, how to enable/disable, activation flow, etc)
Without it we are at risk of wasting time in future. On Mon, Apr 16, 2018 at 3:24 PM, Nikolay Izhikov <nizhi...@apache.org> wrote: > Hello, Vladimir. > > > community is not aware of concrete architecture and proposed public API > > Concrete architecture is described in IEP-18 [1]. > Please, tell me, what else you want to be written. > > I think answers to all questions have to be addressed(and discussed with > community!) when we crack into implementation. > Personally, I don't believe in any kind of more detailed design from eagle > height. > So I want to move from one part of this feature to another and make > concrete choices just before implementation. > > Thoughts? > > Public API proposition will be part of an implementation and, of course, > will be discussed. > I don't think we have to do specific choices at the moment. > > > We need to continue discussion and come up with detailed design. > > Can you give me the check-list of detailed design? > As far as I can see many of other IEP(already implemented) doesn't include > any details you mentioned here [2], [3]. > > Anyway, let's make this IEP some kind of "reference implementation" for > other IEP's :). > > > 1) How encryption is configured - public API is needed > > Similarly to other SPI. > > > 2) How encrypted cluster is started > > I study links you provide. > I asked DB administrators I know. > Looks like we can implement automatic TDE activation. > > > 3) How and where encrypted cache keys are stored > > I want to follow your advice and store CEK's near cache. > > > 4) Protocol of key's exchange between nodes > > 5) Detailed description of encryption flow - what algorithm is used? > will> we use CBC? > > 6) What changes would be required to WAL logic (at the very least -> > enumerate affected types of records)? > > I don't have specific answers to this questions. > Do you think we must answer this questions before starting any > implementation? > > > 7) Would it be possible to enable/disable encryption on specific> > cache/table in runtime? > > No, for the first implementation. > Cache has to be created as encrypted. > If cache created as decrypted it can't become encrypted. > > In the future we can extend encryption mechanism and add this feature to > the product. > > > 8) Would these changes affect compression design anyhow? > > No. > > [1] https://cwiki.apache.org/confluence/display/IGNITE/IEP- > 18%3A+Transparent+Data+Encryption > [2] https://cwiki.apache.org/confluence/display/IGNITE/IEP- > 4+Baseline+topology+for+caches > [3] https://cwiki.apache.org/confluence/display/IGNITE/IEP- > 7%3A+Ignite+internal+problems+detection > > В Пн, 16/04/2018 в 13:28 +0300, Vladimir Ozerov пишет: > > Hi Nikolay, > > > > I noticed that some tickets have been created for this feature. > Hopefully, > > you haven't started implementation yet, because at this point community > is > > not aware of concrete architecture and proposed public API. We need to > > continue discussion and come up with detailed design. Of most importance > we > > need to understand the following: > > 1) How encryption is configured - public API is needed (how to setup key > > providers, how to enable encryption for specific caches, etc) > > 2) How encrypted cluster is started - normally, I do not expect any > > additional manual actions from user once everything is configured. > > Especially, in presence of auto-activation capability. > > 3) How and where encrypted cache keys are stored > > 4) Protocol of key's exchange between nodes > > 5) Detailed description of encryption flow - what algorithm is used? will > > we use CBC? > > 6) What changes would be required to WAL logic (at the very least - > > enumerate affected types of records)? > > 7) Would it be possible to enable/disable encryption on specific > > cache/table in runtime? > > 8) Would these changes affect compression design anyhow? > > > > Vladimir. > > > > On Tue, Apr 10, 2018 at 5:30 PM, Vladimir Ozerov <voze...@gridgain.com> > > wrote: > > > > > Hi NIkolay, > > > > > > Regarding system caches, rule of thumb here - do not use them. Keys > should > > > be stored near cache. > > > > > > As far as password: > > > 1) Oracle auto-login wallet [1] > > > 2) MySQL- password may be set inside configuration [2] > > > > > > I do not think that any kind of prompts are needed here out of the box. > > > May be we could consider them in future, but at the moment it looks > > > redundant to me. > > > > > > [1] https://docs.oracle.com/cd/E11882_01/network.112/ > > > e40393/asotrans.htm#CHDCCHBH > > > [2] https://dev.mysql.com/doc/refman/5.7/en/keyring- > > > encrypted-file-plugin.html > > > > > > On Mon, Apr 9, 2018 at 5:42 PM, Nikolay Izhikov <nizhi...@apache.org> > > > wrote: > > > > > > > Hello, Vladimir. > > > > > > > > > 1) Why do you propose to store CEK in separate cache? > > > > > > > > All CEKs data should be available on all cluster nodes. > > > > We want to use system cache to get data synchronization feature "for > > > > free". > > > > > > > > > We consider storing any metadata in system caches as antipattern > from > > > > > > > > our previous experience. > > > > > > > > Very interesting! > > > > Can you share your experience? > > > > I didn't know that. > > > > I think we can use any convinient storage for CEK. > > > > Current design doesn't couple to system caches, so we can change this > > > > part at any moment. > > > > > > > > > 2) I do not think that decryption process should require any > > > > > > > > "administrator" role and passwords. > > > > > Instead, we can have a kind of pluggable interface which will > provide > > > > > > > > decrypted MEK on demand when it is needed. > > > > > This should be pre-configured in advance on server node(s). > > > > > AFAIK this is how a number of other vendors work. > > > > > > > > Can't agree with you, for now. > > > > Can you please send me some info about implementation you refer to? > > > > We study following papers to make current design: > > > > > > > > 1. ORACLE [1] > > > > > > > > To enable encryption one has to execute following statement: > > > > > > > > `ALTER SYSTEM SET ENCRYPTION KEY IDENTIFIED BY *clear_text_password*` > > > > Or after database server restart - `ALTER SYSTEM SET WALLET OPEN > > > > IDENTIFIED BY *clear_text_password*` > > > > So yes, administrator is involved into server restart. > > > > And no, it's not all automatic. > > > > > > > > 2. MSSQL [4] - AFAIK uses some Windows OS internal key storage to > write > > > > and read master key. > > > > DB server process should have administrator(system) priveleges to > access > > > > that storage. > > > > > > > > > Instead, we can have a kind of pluggable interface which will > provide > > > > > > > > decrypted MEK on demand when it is needed. > > > > > > > > We, for sure, will have pluggable interface for providing MEK. > > > > But, from my point of view, default implementation should use some > > > > default java features. > > > > I think we should use java key store. > > > > It requires to have a clear text password to be loaded [3] > > > > > > > > [1] https://docs.oracle.com/cd/B19306_01/network.102/b14268/asot > > > > rans.htm#BABJJAIG > > > > [2] https://technet.microsoft.com/en-us/library/bb934049(v=sql. > 110).aspx > > > > [3] https://docs.oracle.com/javase/8/docs/api/java/security/ > > > > KeyStore.html#load-java.io.InputStream-char:A- > > > > [4] https://technet.microsoft.com/en-us/library/bb934049(v=sql. > 110).aspx > > > > > > > > > 3) CEK decryption should not be tied to MEK decryption. > > > > > > > > MEK will be available from the moment it obtained to the node stop. > > > > So, we can decrypt existing or encrypt newly created CEK whenever it > > > > necessary. > > > > > > > > > 4) I do not think that SSL should be a strict requirement. It is > up to > > > > > > > > the> user to asses the risks. > > > > > > > > I'm fully OK with it. > > > > Let's remove that restriction. > > > > > > > > В Пн, 09/04/2018 в 11:35 +0300, Vladimir Ozerov пишет: > > > > > Hi Nikolay, > > > > > > > > > > First of all thank you for excellent summary. Two-tiered key > management > > > > > > > > is > > > > > well respected technique and makes perfect sense to me. However, > several > > > > > questions regarding architecture arises: > > > > > 3) CEK decryption should not be tied to MEK decryption. Main > reason - > > > > > > > > CEK > > > > > could be required during dynamic cache/table creation. So there > should > > > > > > > > be > > > > > no coupling between activation and CEK processing. > > > > > 4) I do not think that SSL should be a strict requirement. It is > up to > > > > > > > > the > > > > > user to asses the risks. > > > > > > > > > > On Fri, Apr 6, 2018 at 9:59 PM, Denis Magda <dma...@apache.org> > wrote: > > > > > > > > > > > Nikolay, Dmitriy R., > > > > > > > > > > > > Thanks for the research and for writing down a summary in the IEP > > > > > > > > form. > > > > > > > > > > > > Please answer several high-level questions: > > > > > > > > > > > > - Is it necessary to have CEP keys for every cache? Not sure > how > > > > > > > > all the > > > > > > keys will be managed if the user wants to encrypt 10-100 > caches. > > > > > > > > Is it > > > > > > possible to use a single CEP by default with an option of > having a > > > > > > unique > > > > > > one for a cache with more sensitive information? > > > > > > - It's not written, but I guess it would be up to me which > caches > > > > > > > > to > > > > > > encrypt, right? In practice, you don't need to have all the > data > > > > > > encrypted. > > > > > > Usually, companies look to hide personal, payments history, > etc. > > > > > > - Should we think of procedures of CEP keys regeneration? A > key > > > > > > > > can be > > > > > > lost or stolen. > > > > > > - Similar question goes for MEP key. > > > > > > > > > > > > -- > > > > > > Denis > > > > > > > > > > > > On Thu, Apr 5, 2018 at 2:15 PM, Dmitriy Setrakyan < > > > > > > > > dsetrak...@apache.org> > > > > > > wrote: > > > > > > > > > > > > > Here is a correct link to IEP: > > > > > > > https://cwiki.apache.org/confluence/display/IGNITE/IEP- > > > > > > > 18%3A+Transparent+Data+Encryption > > > > > > > > > > > > > > On Thu, Apr 5, 2018 at 12:01 PM, Nikolay Izhikov < > > > > > > > > nizhi...@apache.org> > > > > > > > wrote: > > > > > > > > > > > > > > > Hello, Igniters. > > > > > > > > > > > > > > > > Based on previous discussion [1] we've created "IEP-18: > > > > > > > > Transparent > > > > > > > > > > > > Data > > > > > > > > Encryption" [2] > > > > > > > > I've planned to start implementation of TDE in few weeks. > > > > > > > > I will create JIRA ticket for each piece of implementation. > > > > > > > > > > > > > > > > So, please, see IEP-18 and give us feedback. > > > > > > > > > > > > > > > > Dima Ryabov, huge thanks for pushing TDE IEP forward. > > > > > > > > > > > > > > > > [1] http://apache-ignite-developers.2346864.n4.nabble. > > > > > > > > com/Transparent-Data-Encryption-TDE-in-Apache- > Ignite-td18957.html > > > > > > > > [2] https://cwiki.apache.org/confluence/pages/viewpage. > > > > > > > > action?pageId=75979078 > > > > > > > > > > > > > > >
