Peter, Vaughn and I were talking the other day about how security configuration for authentication will differ between the current release of Grid Publisher and the future roll out plan. The current release of Grid Publisher uses the existing CDC credential/authentication infrastructure of SDN for issuing user certificates and authenticating users and VeriSign for creating host certificates for nodes. We're doing this because it's simpler initially. But to scale this would be expensive and would also put a large amount of control solely in CDC's hands.
So I created some models on the wiki to try to represent where we want to be with PHGrid in the future. Basically, I see PHGrid supporting multiple CAs that users and nodes can use for certificates. This means that there isn't a single CA for PHGrid but a trust mesh of CAs who can identify participants on the grid.
Of course, nodes still assign rights and privileges to subjects identified by the certificates, but at least we don't get into an expensive component that could limit grid sustainability.
Please let me know your thoughts about how in/valid you think this will be.
Showing posts with label Security. Show all posts
Showing posts with label Security. Show all posts
Monday, April 20, 2009
Tuesday, April 14, 2009
Discussions and progress
Today we had a meeting about some of the ways that the AMDS grid services and AMDS viewers are going to interact in their planned environments. It helped clear up a lot of ambiguities about how they would connect to each other and what sorts of security requirements would be needed and what sorts of security should be expected at different points on the grid.
We also had discussions about ways to handle CAs now and in the future, and how to package things so that we make it easier for other developers to play with stuff should they want to.
I have been making some progress with the NPDSAMDS project, and forgot how handy it is to just write the interface and start pulling in types to wrap your head around all the bits in a service. I really do grok java better than XML Schema still.
Finally, I have been stopping what I was doing to help deal with minor issues in the production deploy of QuickSilver. Said deploy will probably loom for some time, but I can't complain because, hey, code going live is neat!
We also had discussions about ways to handle CAs now and in the future, and how to package things so that we make it easier for other developers to play with stuff should they want to.
I have been making some progress with the NPDSAMDS project, and forgot how handy it is to just write the interface and start pulling in types to wrap your head around all the bits in a service. I really do grok java better than XML Schema still.
Finally, I have been stopping what I was doing to help deal with minor issues in the production deploy of QuickSilver. Said deploy will probably loom for some time, but I can't complain because, hey, code going live is neat!
Monday, March 9, 2009
Dueling for Certificates.
So, It appears that the problem between Rodsadai-web and Amds-web that they are playing musical chairs for the client-certificate.
Felecia (who is developing AMDS-Web) has been helping me debug. We were able to replicate the issues outside of staging on my dev node (handy, since debugging these involves a lot of jboss-restarts). It appears that it is just a mutual exclusion issue: If you started AMDS-web first, it will be able to marshal the needed security credentials, and when Rodsadai-web tries to marshal the credentials, it will not be able-to and the connection to a remote globus server will fail. The behavior is the same in reverse. If the rodsadai-web client is called first, it works fine and then AMDS-web will throw a connection error.
Now, we are not sure if the failure to marshal is because of slightly different needed credentials that are being cached (apparently Rodsadai-web is using tramsaction layer security (TLS) with only privacy and AMDS-web is using TLS with privacy and integrity) or whether there is some sort of explicit locking behavior that prevents different applications from using the same set of proxies.
Either way, it reveals that we need to find a better way for clients to make secure connections to Globus services. Whether it involves some sort of Server-level pooling or having the apps load application-specific certificates, the current method would not allow client applications with slightly different needs to coexist on the same server (or worse, not even allow client applications with the exact same needs to coexist).
Thus, we plan to delve into the innards of globus security to see whether the certs need to change or how they are being marshalled. (Probably both).
Felecia (who is developing AMDS-Web) has been helping me debug. We were able to replicate the issues outside of staging on my dev node (handy, since debugging these involves a lot of jboss-restarts). It appears that it is just a mutual exclusion issue: If you started AMDS-web first, it will be able to marshal the needed security credentials, and when Rodsadai-web tries to marshal the credentials, it will not be able-to and the connection to a remote globus server will fail. The behavior is the same in reverse. If the rodsadai-web client is called first, it works fine and then AMDS-web will throw a connection error.
Now, we are not sure if the failure to marshal is because of slightly different needed credentials that are being cached (apparently Rodsadai-web is using tramsaction layer security (TLS) with only privacy and AMDS-web is using TLS with privacy and integrity) or whether there is some sort of explicit locking behavior that prevents different applications from using the same set of proxies.
Either way, it reveals that we need to find a better way for clients to make secure connections to Globus services. Whether it involves some sort of Server-level pooling or having the apps load application-specific certificates, the current method would not allow client applications with slightly different needs to coexist on the same server (or worse, not even allow client applications with the exact same needs to coexist).
Thus, we plan to delve into the innards of globus security to see whether the certs need to change or how they are being marshalled. (Probably both).
Tuesday, March 3, 2009
Security thoughts continued.
Had a lot of people in and out of the lab today.
But I still managed to make some headway on the security stuff. Brian and I switched out from a servlet listener to a servlet filter, moved things we wanted authenticated into the "auth" directory, and then we got the SHA1 algorithm parsing the entered passwords.
After that, I got the parsing to put out readable text using the Apache Commons Codec library to base64 encode the output parse. Then I set up some methods for the login.jsp page to allow for checking and setting the proper variable for actually marking authentication in the session.
Tomorrow, I will set up the database connectivity to a user/passwordHash table. I will probably set up the appropriate table in PostGres and then start wiring it to the SQL Server. I will also be doing some updating of connection code inside Gmap-polygon and moving some code out of the static load for a servlet inside gmap-poly-web and npdsgmaps-web.
All these epiphanies and suggestions from sitting with another coder and having them look at what I've written for about an hour. We need to do this more often.
Otherwise, I met with Will Duck again today, and I think we are settling on Flot as the next thing to draw the NPDS charts. As you can see from the examples there are a lot of really neat things that Flot can do, it's all java script, and it allows for things like zooming and hover-over tips.
It's pretty too.
But I still managed to make some headway on the security stuff. Brian and I switched out from a servlet listener to a servlet filter, moved things we wanted authenticated into the "auth" directory, and then we got the SHA1 algorithm parsing the entered passwords.
After that, I got the parsing to put out readable text using the Apache Commons Codec library to base64 encode the output parse. Then I set up some methods for the login.jsp page to allow for checking and setting the proper variable for actually marking authentication in the session.
Tomorrow, I will set up the database connectivity to a user/passwordHash table. I will probably set up the appropriate table in PostGres and then start wiring it to the SQL Server. I will also be doing some updating of connection code inside Gmap-polygon and moving some code out of the static load for a servlet inside gmap-poly-web and npdsgmaps-web.
All these epiphanies and suggestions from sitting with another coder and having them look at what I've written for about an hour. We need to do this more often.
Otherwise, I met with Will Duck again today, and I think we are settling on Flot as the next thing to draw the NPDS charts. As you can see from the examples there are a lot of really neat things that Flot can do, it's all java script, and it allows for things like zooming and hover-over tips.
It's pretty too.
Thursday, January 15, 2009
SST Security Settings
The SST has been configured to use the grid-map file for user credential verification. Here is an excerpt from the log that demonstrates an unauthorized user attempting to access the SST.
Log File:
Jan 13, 2009 3:20:18 PM org.globus.wsrf.impl.security.authorization.GridMapAuthorization isPermitted
WARNING: Gridmap authorization failed: peer "/O=Messed/OU=updomain.net/OU=Some Organization/OU=updomain.net/CN=Bubba Gump" not in gridmap file.
Jan 13, 2009 3:20:18 PM org.globus.wsrf.impl.security.authorization.ServiceAuthorizationChain authorize
WARNING: "/O=Messed/OU=updomain.net/OU=Some Organization/OU=updomain.net/CN=Bubba Gump" is not authorized to use operation: {http://transfer.blahdomain.com/TransferService}storeFile on this service
We are currently researching a method for doing this via the Globus / Tomcat configuration files. This configuration is currently built into the SST gar file.
Log File:
Jan 13, 2009 3:20:18 PM org.globus.wsrf.impl.security.authorization.GridMapAuthorization isPermitted
WARNING: Gridmap authorization failed: peer "/O=Messed/OU=updomain.net/OU=Some Organization/OU=updomain.net/CN=Bubba Gump" not in gridmap file.
Jan 13, 2009 3:20:18 PM org.globus.wsrf.impl.security.authorization.ServiceAuthorizationChain authorize
WARNING: "/O=Messed/OU=updomain.net/OU=Some Organization/OU=updomain.net/CN=Bubba Gump" is not authorized to use operation: {http://transfer.blahdomain.com/TransferService}storeFile on this service
We are currently researching a method for doing this via the Globus / Tomcat configuration files. This configuration is currently built into the SST gar file.
Friday, November 7, 2008
Security Policy Document - New Draft
Raja and Joseph put out a new draft of the PHGrid Security Policy Document.
It's available on the wiki for your review.
It's available on the wiki for your review.
Tuesday, October 21, 2008
new techs on old demo
Today I managed to finish sprucing up the poicondai demo with better logging, and better handling of the new zipcode lists. I also started my research of the JQuery show/hide toggle so I could show/hide the raw zip-code list unless data was to be specified.
Tomorrow I hope to complete the show/hide, and get a prelim set of geographical outlays in poicondai with count bubbles. It will look a lot like rodsadai.
Otherwise, today we had a cool meeting on security, and we are going to doing a lot of data and application classifications and start laying out all of the different standards and policies... and just looking upon our data with more of a "security" eye. Thus, I am meeting a lot of new security focused folks who ask the important, retrospectively obvious questions about what services are present and what sorts of fields are used and whether they might be dangerous when boxed up and shipped somewhere to someone who hasn't seen them before.
Tomorrow I hope to complete the show/hide, and get a prelim set of geographical outlays in poicondai with count bubbles. It will look a lot like rodsadai.
Otherwise, today we had a cool meeting on security, and we are going to doing a lot of data and application classifications and start laying out all of the different standards and policies... and just looking upon our data with more of a "security" eye. Thus, I am meeting a lot of new security focused folks who ask the important, retrospectively obvious questions about what services are present and what sorts of fields are used and whether they might be dangerous when boxed up and shipped somewhere to someone who hasn't seen them before.
Tuesday, August 12, 2008
GridMedlee Progress
Making progress on GridMedlee testing. I have altered the initial client code to add security options. I have gotten it to the point to where I am receiving a error of Unknown CA. In speaking with my team members this issue should be resolvable by running the code from a controlled node rather than my local laptop. I am in the process of moving the code over to the controlled node and will (hopefully) get it running from there.
Along the way I picked up some knowledge with regard to the GlobusCredential class. When the Introduce framework created the test client for GridMedlee there were several versions of the GridMedleeClient method that were created, some included a parameter for a GlobusCredential object. Upon further investigation I found that the GlobusCredential class can be used to create secure credentials.
I will have another update later today after I get the GridMedlee client built on the controlled node.
Along the way I picked up some knowledge with regard to the GlobusCredential class. When the Introduce framework created the test client for GridMedlee there were several versions of the GridMedleeClient method that were created, some included a parameter for a GlobusCredential object. Upon further investigation I found that the GlobusCredential class can be used to create secure credentials.
I will have another update later today after I get the GridMedlee client built on the controlled node.
Friday, August 8, 2008
Testing, Testing and more testing
- I spent the day working with the GridMedLee services and I feel I have made a lot of progress. I modified the client code sent to me to invoke the GridMedLee service. Currently working through what appears to be a security issue with getting a status back from the GridMedLeeContext service. I have some ideas of a few things to do on my end to try to make sure my set up is functioning properly.
- I also went back and re-ran the client for the ZipCode service for Utah and I am getting the same message from their service that I am receiving with GridMedlee.
- I also installed WireShark locally to be able to actually view the contents of data being sent between the client and the server.
Stay tuned friends more to come as I delve deeper into Globus security.
Monday, July 14, 2008
Well, I got some documentation updated and MySQL on a server
Today I did a lot more research, tried getting tomcat secure access working (I did) to see if it would make secure client OGSA-DAI access from tomcat work (it didn't). I am out of ideas, but I sent out more feelers to see if other people could help me or provide working examples.
Otherwise, I documented and cleaned up a bunch of the RODSAdai code, and I am setting up a dedicated MySQL server just so we have a MySQL database in the lab that is not eaten by globus/vdt.
Otherwise, I documented and cleaned up a bunch of the RODSAdai code, and I am setting up a dedicated MySQL server just so we have a MySQL database in the lab that is not eaten by globus/vdt.
Tuesday, July 8, 2008
RODSAdai secure access from in the test context
Today I ran into the familiar syndrome... lovingly known as "But it works in the ogsadai directory".
So, secure access of the RODS database does work from the ogsadai directory... which has a setenv.sh file that sets the java classpath up with all sorts of jars and property files from globus and ogsadai.
However, when I try to make a secure call from a test context... I get the dreaded error of "unable to find valid certification path to requested target".. I am pretty sure this means that I need to include some more jars from the globus context, hopefully those jars will have something that imports the globus security infrastructure. If it does, then it might very well be included in the war file that eventually gets moved to tomcat and it will solve the "secure access doesn't work from tomcat" problem.
Many thanks to Dr. Jeremy for helping me get over my vacillations on what to try next. My next step will be to get the test for secure access running. Then I will start working on getting these clients called securely from within tomcat. Dr. Espino pointed out that everything should test and pass before moving on to other projects/discoveries... and if it's going to take some painful jar hunting to get it working, then so be it.
So, secure access of the RODS database does work from the ogsadai directory... which has a setenv.sh file that sets the java classpath up with all sorts of jars and property files from globus and ogsadai.
However, when I try to make a secure call from a test context... I get the dreaded error of "unable to find valid certification path to requested target".. I am pretty sure this means that I need to include some more jars from the globus context, hopefully those jars will have something that imports the globus security infrastructure. If it does, then it might very well be included in the war file that eventually gets moved to tomcat and it will solve the "secure access doesn't work from tomcat" problem.
Many thanks to Dr. Jeremy for helping me get over my vacillations on what to try next. My next step will be to get the test for secure access running. Then I will start working on getting these clients called securely from within tomcat. Dr. Espino pointed out that everything should test and pass before moving on to other projects/discoveries... and if it's going to take some painful jar hunting to get it working, then so be it.
Monday, July 7, 2008
Secure traffic for rodsadai
Today we got Secure traffic for rodsadai working. Dan helped with several of the hostname issues (many thanks to him for that) and I got the client factory to pass the secure client instead of the generic sql client.
Tomorrow, I will be adjusting all of the tests to take into account proper spatial and time series formatting, and probably do a bit of re-architecting to turn secure access into a separate test (this will most likely involve passing the query client into the processor as opposed to having the processor call its own).
That and documenting, putting in descriptive errors, that sort of thing.
Tomorrow, I will be adjusting all of the tests to take into account proper spatial and time series formatting, and probably do a bit of re-architecting to turn secure access into a separate test (this will most likely involve passing the query client into the processor as opposed to having the processor call its own).
That and documenting, putting in descriptive errors, that sort of thing.
Monday, June 16, 2008
Security Evaluations
I had a call with a NCPHI security steward to ask about some of the security evaluations that we'll be performing with the Secure Reliable Grid Messaging PoC project.
She pointed us specifically to the FIPS 140-2, NIST SP 800-53 and eAuthentication guidelines.
It's a lot of reading, but we'll need to evaluate the algorithm type and strength used by Reliable GridFTP, integrity checking of the message payload, and certificate usage.
She pointed us specifically to the FIPS 140-2, NIST SP 800-53 and eAuthentication guidelines.
It's a lot of reading, but we'll need to evaluate the algorithm type and strength used by Reliable GridFTP, integrity checking of the message payload, and certificate usage.
Subscribe to:
Posts (Atom)