1. 1 Practical Techniques for Searches on Encrypted Data Dawn Song, David Wagner, Adrian Perrig

2. 2 Motivation Why searches on encrypted data? –Searching on encrypted e-mails on mail servers –Searching on encrypted files on file servers –Searching on encrypted databases Why is this hard? –Perform computations on encrypted data is often hard –Usual tradeoffs: security and functionality Search query Download emails

3. 3 Sequential Scan and Straw Man Example Search by sequential scan: W i -1 m bits WiWi W i+1 m bits … … W Search for W Naïve approach: E(W i –1 ) m bits E(W i ) m bits E(W i+1 ) m bits … … E(W) Search for W WW E(W)

4. 4 Desired Properties Provable security –Provable secrecy: encryption scheme is provable secure –Controlled search: server cannot search for arbitrary word –Query isolation: search for one word does not leak information about other different words –Hidden queries: does not reveal the search words Efficiency – Low computation overhead – Low space and communication overhead – Low management overhead

5. 5 The Key Idea W i -1 m bits WiWi W i+1 m bits … … S i -1 m bits SiSi S i+1 m bits … … C i -1 CiCi C i+1 … …  W i+1  Search for W i+1

6. 6 Setup and Notations Document: sequence of fixed length words W i -1 m bits WiWi W i+1 m bits … … Pseudorandom Function F and key K : F K maps n bits to m-n bits Pseudorandom Generator G and seed: L  G ( seed ), L i  G i ( seed ) L i-1 n bits … … LiLi L i+1 n bits

7. 7 Basic Scheme (Encryption) WiWi m bits LiLi n bits L i  G i (seed),  CiCi m bits RiRi m-n bits R i  F K ( L i )

8. 8 Basic Scheme (Decryption) m bits n bits  m-n bits m bits WiWi LiLi RiRi CiCi n bits m-n bits C i,L C i,R L i  G i (seed), R i  F K ( L i ) LiLi RiRi  WiWi

9. 9 Basic Scheme (Searches) Search for word W, give server W and K Check: R i ' = F K ( L i ' ) ? Yes  match, ( false positive rate = 1 / 2 m-n ) m bits n bits  m-n bits m bits WiWi LiLi RiRi CiCi W  Li'Li' Ri'Ri' n bitsm-n bits

10. 10 Problems with Basic Scheme Queries are not hidden, server learns word Query isolation is not satisfied, server learns K and can search for arbitrary words

11. 11 Hidden Queries LiLi n bits L i  G i (seed), RiRi m-n bits R i  F K ( L i ) WiWi m bits E(W i ) m bits E(.)  CiCi m bits

12. 12 Controlled Searches and Query Isolation For hidden queries, server can search for word W if it knows E(W) Controlled searches on words Instead of R i  F K ( L i ), R i  F K i ( L i ), where K i = F' K ( W i ) Enhancements –Check for a word in a single chapter/section only –Check only for “word occurs at least once” in document –Check only for “word occurs at least N times” in document

13. 13 Improved Security (Change K) LiLi n bits L i  G i (seed), RiRi m-n bits R i  F K i ( L i ) WiWi m bits E(W i ) m bits E(.) where K i = F' K ( E( W i ))  CiCi m bits

14. 14 Final Scheme (Encryption) LiLi n bits L i  G i (seed),  CiCi m bits RiRi m-n bits R i  F K i ( L i ) WiWi m bits E(W i ) E(.) E 1 (W i ) E 2 (W i ) where K i = F' K ( E 1 ( W i ))

15. 15 Final Scheme (Decryption) C i,L n bits C i,R m-n bits  LiLi E 1 (W i ) RiRi F k i (L i )  E 2 (W i ) LiLi n bits  CiCi m bits RiRi m-n bits WiWi m bits E(W i ) E(.) E 1 (W i ) E 2 (W i )

16. 16 Advanced Search Queries Building blocks for advanced search queries W 1 and W 2, W 1 near W 2, W 1 immediately precedes W 2 Supports variable length words –Same provable security –Similar efficiency

17. 17 Conclusion Provable security –Provable secrecy –Controlled search –Query isolation –Hidden queries Simple and efficient –O(n) stream cipher and block cipher operations per search –Almost no space and communication overhead –Easy to add documents –Convenient key management : user needs only one master key Embedding information in pseudorandom bit streams

18. 18 Discussion Search is one operation on an abstract encrypted data type What other operations on abstract encrypted data types are possible?

19. 19

20. 20 LenE l ( W i )E r (W i ) x bitsn-x bitsm-n bits LiLi RiRi n bitsm–n bits  CiCi m bits L i = G i ( r ),R i = F ( K i, L i ) Variable length words encryption scheme

21. 21 Related Work Secure file servers and memory protection M. Blaze et al, M. Blum et al, P. Devanbu et al. Multiparty computation O. Goldreich et al, R. Canetti et al. Private information retrieval B. Chor et al, C. Cachin et al, Y. Gertner et al.