Difference between revisions of "Roadmap"

Revision as of 21:27, 18 June 2009

This is the outline of the development roadmap for MIT Kerberos.

Timeline

Target 6 to 12 month cycle. (9 months plus/minus 3)

krb5-1.8

Branch Dec. 2009

• consider shortening testing cycle to two months or less

Release early Mar. 2010

krb5-1.9

Branch Sep. 2010

Release Dec. 2010

Guiding principles

• Code quality
• Modularity
• End-user experience
• Administrator experience
• Performance
• Protocol evolution

Code quality

• Remove krb4 (1.7)
• Move toward test-driven development
• Increase conformance to coding style
  • "The Great Reindent"?
  • Selective refactoring
• Use safer library functions (ongoing)
  • Avoids false positives
  • Avoids need to (probably manually) evaluate "unsafe" calls
  • Stop using strcpy, strcat, sprintf, etc.
    • Mostly done
    • New internal APIs for complex operations
• Reduce commitment to "difficult" platforms
  • See supported platforms
  • Focuses resources more effectively

Modularity

• Crypto (1.8)
  • Native (accelerated) crypto API support
  • Performance optimizations (caching, etc.)
  • New API design 1.7+
• Support readily building subsets (1.8)
  • "Lite" client
  • "Lite" server
  • "GSS-API": separate context establishment from message protection functions, e.g. Solaris user/kernel space split
• GSS-API mechanism glue
  • At least rough form to enable NTLM support (1.7)
  • Possible refinements later (1.8)
• KDC Database (long-term)
  • Does the existing DAL make sense?
  • Make data model less "blobby"
  • Track IETF data model work
  • New API around 1.8
  • New implementation around 1.9
• Secure co-processor ("would be nice")

End-user experience

• Enhanced error messages for GSS-API (done)
• Referrals (1.7)
  • DNS independence via referrals
• Localization of static error strings (1.7+)
• Credential management
  • KIM API (done)
  • Cross-platform CCAPI
    • Done for Mac & Windows
    • UNIX implementation (1.7+)

Administrator Experience

• Incremental propagation (1.7)
  • Integrated; needs cleanup
• Improve key rollover
  • Master key (1.7)
  • Application service keys (1.8)
• Decrease DNS-related fragility
  • Investigate LHA/Apple proposal to store config information in ccache to signal when a realm supports referrals and thus the option to eschew reverse DNS resolution
• Audit support (log all ticket requests) (1.7+)
• Disable DES by default (1.8)
  • Investigate doing this for 1.7 and removing single-DES completely by 1.8
  • Add more versatile facilities for configuring cryptosystems

Performance

• Decrease DNS traffic (1.7)
  • Stop trying to crawl up to the root
• Replay cache ("rcache")
  • Disable on KDC (1.7)
    • Avoids known false-positive issues
  • Collision avoidance (1.7+)
  • Improve implementation (1.7+)
  • Support disabling by service type name (1.7+)
• New crypto API (1.8) facilitates optimizations
• Concurrency

Protocol evolution

• Encryption algorithm negotiation (1.7)
• Microsoft Kerberos extensions (1.7)
• Improved PKINIT support (1.7)
• Anonymous PKINIT (1.8)
• FAST (done in 1.7 for a subset; IETF)
• International strings in protocol (1.8+; IETF)
• Timestamp-independence (1.8, 1.9)
• Replay-proofing protocols (1.8, 1.9)