The disclosed embodiments relate generally to computer communications and, more specifically, to providing secure access to confidential information stored within a server cloud.
Users often communicate with each other over the Internet, and store network-based data, in an unsecure manner. Unsecure communications and data are acceptable for many purposes where the information communicated is of a non-sensitive nature.
However, there are many contexts in which the information communicated is actually or potentially sensitive, such as when communicating and storing confidential business details, conducting e-commerce, and the like. In such contexts, the communicating users should employ systems with cryptographic capabilities that can encrypt and decrypt the communicated information. This prevents intermediate parties such as active eavesdroppers, systems such as routers that make up the Internet architecture, and the systems storing data at remote locations, from being able to obtain the communicated information in its original plaintext form.
The cryptographic infrastructure needed to allow users to easily and transparently secure their communications can be complex. Delegation of cryptographic functionality entails a certain degree of trust of the infrastructure components that provide the functionality. Accordingly, a third-party organization providing the cryptographic functionality may be able to exploit this trust to obtain the original, unencrypted communications. Such a risk may not be acceptable to some users, e.g., when conducting highly sensitive communications.
More specifically, although information is communicated and stored in encrypted form, the mere fact of selecting among several encrypted items in a database may reveal sensitive information. The request which expresses such a selection must therefore be encrypted, as well. While this protects against eavesdroppers on the network, the request itself cannot easily be hidden from the server responding to the query or update request: Human operators of the cloud infrastructure could, for example, gain access to the server's internal memory where the request is decrypted and represented in unencrypted form, and/or to the new content that is to be stored in encrypted form in the cloud.
The figures depict various embodiments of the present invention for purposes of illustration only. One skilled in the art will readily recognize from the following description that other alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the invention described herein.
The client devices 120 and the secure storage servers 101 communicate with each other through a network 140. The network may be any suitable communications network for data transmission. In one embodiment, the network 140 is the Internet and uses standard communications technologies and/or protocols.
It should be understood that although for simplicity
In brief, the secure storage module 124 divides the data across the different, non-colluding secure storage servers 101, combining the different portions of the divided data with random values in such a manner that the divided and obfuscated portions of the divided data can also be read and updated separately on the various servers 101, without the client device 120 needing to obtain the original data in its entirety. This provides greater efficiency relative to operating on the data as a whole, both with respect to computation and with respect to network data transfer, in that only subsets of the data need be transformed and transmitted. The nature of the data distribution and transformation techniques permit the client device to derive the data in its original form, given the various transformed portions of the divided data stored by the various servers 101. Only a single “round-trip” (set of parallel requests from a client device 120 to the set of secure storage servers 101, and reply from the set of secure storage servers 101 to the client device) is required for a read of a particular data portion, and only a single round-trip is needed for a secure update (see
The secure storage servers 101 form a non-colluding server cloud. The servers 101 are designed not to communicate with each other to combine the divided data, which could result in obtaining the original unencrypted data. Additionally, to further lessen the danger of collusion, the various secure storage servers 101 can be operated by different organizations, so that an employee of one organization gaining access to the data in one of the secure storage servers would not allow access to data in other secure storage servers.
The secure storage module 124 comprises a secure distribution module 205 that securely distributes data of the client device 120 to the secure storage servers 101 in an obfuscated manner such that none of the servers 101 has the ability to infer the original value of the data, and such that the client devices can read and update portions of the data in an efficient manner. For purposes of explanation, the data of a client device 120 is hereinafter denoted C (denoting “content”), and the ith secure storage server is denoted Si. Content C may be processed in sub-portions. In the embodiment described hereafter, C is processed in L-bit sub-portions (hereinafter denoted “rows”), and to this purpose is represented as an [N×L] matrix of N rows and L columns (that is, L bits per row), with reads and updates being performed on L-bit rows. Any amount of client data C may be processed; if a row of C is not a multiple of L bits, it may be padded to a multiple of L bits before being processed. In some embodiments, N and L need not be the same for all items of data being securely stored (e.g., files), but rather are adjustable parameters that may vary for each data item.
The secure distribution module 205 distributes the data of C across K secure storage servers 101 (2k=K>=4), with obfuscation, so that no single secure storage server 101 has the entire data of C in original (non-obfuscated) form.
Specifically, in one embodiment, the set of servers 101 is split in two groups of equal size k, where k>=2. The secure distribution module 205 sends, to each of the k servers (denoted with index j) in each of the two groups (denoted with index i), an obfuscated portion of the bits of C, the obfuscated portion denoted Rij, and having the same size as C itself (i.e., N×L bits). To achieve this, Rij is initialized to random bits (i.e., a randomly-selected 0 or 1 bit, for each of the (N×L) bits), and R2 is initialized to a copy of Rij. (Accordingly, by virtue of the XOR operation, the sum of Rij and R2 is the zero matrix.) For a fixed j, for j in 1 . . . k, each bit of C is added to Rxj (that is, either to R1j or to R2j, randomly selected). Put differently, for each bit of C, the bit is imprinted on R—that is, XOR′ed with the bit at the corresponding position of the selected R, and the result is stored at that position in that R. Imprinting all bits of C to a single R for a given j, or spreading them randomly over both Rs with the same j, are both permissible alternatives employed in different embodiments. The secure distribution module 205 sends Rij to all the servers 101 of the corresponding group i, such that each of the servers receives exactly one of the portions. The servers 101 store the portions Rij in an obfuscated portion repository 255, e.g., implemented as a database. (The term “obfuscated portion” is used for Rij in that each Rij is expected to reflect a portion of the data of C, and the random values to which the portion was initialized provide obfuscation.)
In another embodiment, in order to reduce the risk of collusion attacks, more than two server groups are used. Specifically, the set of servers 101 is split in g groups of size k, where g>=2 and k>=2. The g*k servers receive an obfuscated portion of the bits of C, the obfuscated portion denoted Rij, and having the same size as C itself (i.e., N×L bits). For a fixed j, j in 1 . . . k, and for i in 1 . . . g−1, matrices R1 are independently initialized to random bits, and the remaining one of the Rij is initialized to the sum (bitwise XOR) of the other g−1 matrices Rij. (Accordingly, by virtue of the XOR operation, the sum all Rij for a fixed j is the zero matrix.) For a fixed j, for j in 1 . . . k, each bit of C is added to Rxj (where x is randomly chosen within the range 1 . . . g). Put differently, for each bit of C, the bit is imprinted on R—that is, XOR′ed with the bit at the corresponding position of the selected R, and the result is stored at that position in that R. Imprinting all bits of C to a single R for a given j, or spreading them randomly over all Rs with the same j, are both permissible alternatives employed in different embodiments.
In one embodiment, in order to prevent eavesdroppers on the network 140 from obtaining all the Rij (with the same j) and reassembling C based on them, an encryption module 225 of the client secure storage module 124 encrypts the portions before transmitting them to the secure storage servers 101. In order to achieve the appropriate isolation of the different secure storage servers 101, from each other, the Rij are encrypted differently for the different servers, such as using different cryptographic keys (e.g., different client-server pairs for symmetric key solution such as Kerberos, or a distinct public key per server for asymmetric key systems such as RSA, DSA, or elliptic curve systems, or distinct session keys using a transport layer-based protocol such as SSL). In some embodiments, the client secure storage module 124 sends a randomly-generated key along with each request (e.g., a read request for a data portion) sent to one of the secure storage servers 101, and the servers 101 use the respective randomly-generated key to encrypt the value (e.g., the data portion being read) that it sends back to the client device. The secure storage module 124 additionally comprises a secure read module 210 that obtains the original data of a specified row of C from the set of secure storage servers 101, such that none of the servers can determine which row is being obtained. Thus, not only do the servers 101 not know the actually data bits of the row being obtained (i.e., the data itself), but they additionally don't know which row is being obtained (i.e., the location of the data within C).
Specifically, for each read request, the secure read module 210 initializes k−1 random vectors rj (j=1 . . . k−1), each having N random bits (i.e., each of the N bits is a 0 or 1 selected at random, N being the number of rows of C). Additionally, a vector rk is set to the bitwise sum (XOR) of the k−1 random vectors; accordingly, the bitwise sum of all k vectors rj (j=1 . . . k) is the zero vector.
The secure read module 210 encodes in k query vectors qj the index (denoted b) of the desired row. In one embodiment, in order to perform the encoding, the secure read module 210 randomly selects one of the k vectors rj and encodes b within that selected rj. To do so, the vector eb is formed, eb being an all-zero vector of N bits, except for a bit set in the position of the vector corresponding to b (e.g., if b is 1, corresponding to the first row, eb has a 1 in the first element, and a 0 in the remaining N−1 elements). (It is appreciated that, in general, the choice of 0 or 1 as representing set/unset is not significant; equally, 1 could represent unset, and 0 could represent set.) To encode the desired row in the selected qj, eb is bitwise added (XOR′ed) to the corresponding rj. All other query vectors qj are simply a copy of the respective rj. In other embodiments, eb may be added to any odd number of rjs, rather than to only a single rj; due to the nature of bitwise XOR, the result will be the same as if eb had been added to a single rj, in that every pair of eb additions will cancel each other out.
The secure read module 210 sends each qj to the jth of the k servers 101 of group i, such that each server of this group receives exactly one of the vectors qj. In one embodiment, the secure read module 210 performs encryption on the vectors qj before sending them to the servers 101, so as to prevent eavesdroppers from obtaining all the qj and bitwise adding them to obtain b, as represented by eb.
Upon receiving the vector qj as part of the request for the row data, a query vector application module 260 of the corresponding secure storage server Sij 101ij of server group i computes a query result component aij, the product of the obfuscated portion Rij stored by Sij and the query vector qj: When computing the matrix-vector product Rij*qij, if the mth element of qij is set, row m from Rij is selected, and all selected rows are summed up using XOR, resulting in the partial query result vector aij with L bits:
a
ij=Σm(qijm*rowm(Ri))=qijT*Rij (matrix-vector product)
The query vector application module 260 sends the computed partial query result vectors aij back to the client secure read module 210. In one embodiment, in order to prevent eavesdroppers on the network 140 from obtaining all the aij (of a group i) and reassembling the retrieved row based on them, the secure storage server Sij 101ij encrypts the result component aij before transmitting it to the client secure storage 124. If SSL transport was used for the query, the encryption of SSL itself may be used for this purpose; alternatively, if the client device 120 provided the secure storage servers 101 with different randomly-generated keys, the servers 101 may use those keys when providing aij back to the client device.
The client secure read module 210 accordingly obtains the qij for each server Sij. For a fixed j in the range 1 . . . k, the secure read module 210 computes aj=Σi aij, the bitwise sum (XOR) of all the received aij for this j. Due to way in which the various Rij were computed, transformed, and distributed so as to securely represent C, the result of a=Σ aj, which is computed by the secure read module 210, is the value of row b. Note that:
a
j=Σiaij=ΣiRij*qj=qjT*Σj=qjT*C and therefore
a=Σ
j
a
j=ΣjqjT*C=ebT*C=rowb(C)
That is, the original value of C is obtained from the combination of the obfuscated R1, and because the query vectors qi were selected to sum to eb, eb can be used to select the desired row from the original data, C.
The client secure storage module 124 further comprises a secure update module 215 that allows a client device 120 to update a portion of the data C, as stored by the secure storage servers 101, to a new value, while preventing any of the servers 101 from determining which portion is being updated. Specifically, the secure update module 215 sets the value of a particular row of C with index b to a given value (i.e., set of L bits) while concealing the value of b from the servers 101.
In order to request the servers 101 on which C is securely stored to update the row b to a new value (set of bits) w, the secure update module 215 uses the secure read module 210 to obtain the current value (denoted v) of the row b as collectively securely stored by the servers 101.
The secure update module 215 computes an update delta value u as the bitwise sum (XOR) of the current value v and the desired value w (i.e., u=v+w).
The secure update module 215 then sends different update messages <t,u> to the two (or more, if thus configured) groups of servers. That is, the secure update module 215 initializes a vector r of N random bits (N being the number of rows in C). For each group i of k servers 101, the secure update module 215 generates a “randomized component” ti, which is set either to the random vector r, or to the combination of r with the row ID vector eb (e.g., the bitwise sum of r and eb). Specifically, ti=r+eb for one of the groups, and ti=r for the other groups; the group for which ti=r+eb is selected at random. For each of the groups, the secure update module 215 sends an update message mi=<ti, u> to each of that group's k servers. The value ti (either r, or r+eb) is referred to as a “randomized component”. Thus, eb is not provided in original form, but rather as transformed by a random vector t—so the individual servers receiving the update messages cannot determine eb from ti.
In one embodiment, in order to reduce the risk of collusion attacks, more than one group is sent the randomized component t=r+eb, provided that the number of these groups is odd and the total number of groups is even. In case of the system having only two groups, exactly one group must receive a message containing t=r+eb, as was explained above.
Upon receiving the messages m=<ti,u> as part of the request to update the row b to value w, the secure storage servers 101 collectively replace the current value v in row b with the value w. To effect this replacement, a transformation module 565 of each of the secure storage servers 101 uses an outer product of the received update message t and update value u to imprint a server-specific mask on the obfuscated portion Rij already stored by the server, such that operation of the secure read module 220 will return the correct update value, w.
That is, the transformation module 565 computes and stores, instead of its current obfuscated portion Rij, a transformation (denoted Rij′) of Rij that factors in the values ti and u (the bitwise sum of v and w) received as part of the update message for group i. Specifically, the transformation module 265 computes the outer product of ti and u, which serves as a mask used to transform the stored obfuscated portion Rij. The transformation module bitwise adds (XOR) the mask to Rij. That is, Rij′=Rij+(ti⊗u). The outer product is a [N×L] matrix U that is constructed by inserting a row at position m with value u each time the mth element of ti is set, or inserting a row of zeros otherwise. The transformation module 565 then saves Rij′ in place of Rij. (Note that the use of the various vectors ti prevents the identity of the row being transformed from being known to any of the individual servers 101, yet the identity of the row contributes to the transformation effected by at least one of the tis which then goes into the outer products (ti⊗u)).
Due to the design of the update transformations, when the client secure storage module requests the value of row b from the servers 101 via the secure read module 210, after the securely stored row b is updated by the secure updated module 215 to the value w, it will correctly receive w. Recall that retrieving a row from the system involves the construction of query vectors qj which apply to all servers at position j, in each of the i groups. Moreover, aj=qjT*C. First consider retrieving one of the rows m that were not updated (but the Rij nevertheless changed). Because the number of groups is even, and the rows m of the Rij were either not modified at all (tm was not set) or tm was set and the value u had to be added to all rows, which cancel out each other because of the even group count, the value of am remains the same:
a
m
′=a
m when m≠b
Second, in the case where m is equal to b (index of the row that changed), the number of Rijs to which u was added is odd: If the random vector r underlying t was not set at position b, an odd number of “special” Rijs receive update u, by construction. If r was set at position b, because eb reverses this, tb is not set and an odd number of special Rijs do not receive update u, which means that the number of Rijs which receive update u is also odd as the number g of groups is even. Overall, when m equals b:
a
b
′=a
b
+n*u (where count n is odd)
which means aj′=aj+u because of the XOR addition. Because u is the bitwise sum of v and w (i.e., u=v+w), and aj=v (old value), this results in aj′=v+u=v+(v+w)=w. The final result a=Σj aj will also have value w because of the construction of the query vectors qj which select exactly the value at position b.
The client secure storage module 124 further comprises a fully secure update module 220 that allows a client device 120 to update a portion of the data C, as stored collectively by the secure storage servers 101, to a new value, while preventing any of the servers 101 from determining which portion is being updated, and while additionally preventing any of the servers 101 from even determining an update delta value that is added to effect the transform of the row data to the desired new value. Specifically, the fully secure update module 220 sets the value of a particular row with index b to a given value (i.e., set of L bits) while concealing the value of b from the servers 101, and while also concealing the value of an update delta u from the servers, thus providing confidentiality guarantees for all stored items and update values at all times.
In one embodiment, the update delta u can be made confidential by doubling the amount of servers 101 in the system. To this end, the fully secure update module 220 stores C distributed across two sets of g*k servers, and sends different update messages to the two sets of g*k servers, where g is the number of groups and k the size of the groups.
Specifically, to initially store C across the two sets of g*k servers, two matrices C1 and C2 are created, such that the bitwise sum (C1+C2)=C; for example, C1 could be C and C2 could be the 0-matrix, or any other distribution of the bits of C across C1 and C2 such that C1 and C2 sum to C. C1 is stored by the secure distribution module 205 across one of the two sets of g*k servers, and C2 is stored across the other set of g*k servers. In order to update the row of C as thus stored, the fully secure update module 220 initializes a vector p of L random bits (L being the number of bits in a row of C). The fully secure update module 220 uses the secure update module 215 to update row b in one set of the g*k servers 101 with the value (p+v), and the other set of the g*k servers with the value (p+w) for row b. Thus, the row update values sent as part of the update are always combined with a random p, so that the true row update values are not determinable by the processing servers. The added degree of security afforded by also concealing the update delta u may in some contexts be considered worth the doubling of servers that are used by the secure update module 215, in that this also prevents possible reverse-engineering of the original value v or updated value w based on value u.
When the client secure storage module requests the value of row b from the two sets of servers 101 via the secure read module 210 and then adding them up (with bitwise XOR), after the securely stored row b is updated by the fully secure updated module 220 to the value w, it will correctly receive w. Assuming that before the update the first set of the servers stored v1 and the second set of the servers stored v2 at the position of interest such that v1+v2=v, then updating will lead to the first set of the servers storing v1+p+v and the second set of the servers storing v2+p+w. Retrieving these two values and adding them up (with bitwise XOR) yields a=(v1+p+v)+(v2+p+w)=v+v+p+p+w=w.
In one embodiment, the set of servers used to conceal the update value is at least doubled (in order to increase to effort necessary for servers 101 to collude). That is, if the secure update module 215 uses g*k servers 101 (where g is the number of server groups and k the size of each group), the fully secure update module 220 uses f*g*k servers, f>=2. The minimum configuration is 8 servers, since the secure update module 215 uses at least 4 servers (g=k=2).
The fully secure update module 220 performs two operations: The first operation is to initialize f−1 rows with random bits and one row with their sum such that when adding them all up results in a row of zeros; the second operation is to spread the bits of w and v over the f rows and adding them with XOR at the corresponding position. These f rows will be used to update each of the f sets independently. Because all these f rows will be summed up in the retrieval phase, the overall sum will be the old content of the f rows plus the value v+w, which is the update value u. In other words, all distribution of the update value to f rows is undone at the client device 120A. It is appreciated that there are many ways of distributing the bits of w and v: fully randomly, regularly (e.g. the first bit of w to row 1, second bit of w to row 2 etc. and restarting at row 1 after reaching the f+1 bit of w), or w fully in one row and v spread over all other rows, and the like.
The use of the fully secure update module 220, relative to the secure update module 215, additionally addresses the problem of preventing the value of the update delta from being determined, but uses double the number of servers 101. Alternatively, in some embodiments in which all the client devices 120 that access given data stored on the secure servers can obtain a shared key, the shared key is used to encrypt the update delta value so that the servers 101 cannot determine it. This avoids the need for doubling the number of servers 101 to provide the obfuscation of the update delta value via the common random vector p; for example, in one embodiment only the same number of servers 101 used by the secure update module 315 is required, plus a (possibly distinct) server to distribute a shared key to all the client devices 120 that are authorized to view the data in question. In these embodiments, the operation of the secure update module 215 is modified to encrypt the update value, w, using the key, resulting in the encrypted value we. Since the row values are stored in encrypted form in this embodiment, the read of the current row value as performed by the secure read module 210 is ve, the encrypted form of the original value. The update messages sent to the servers then include, as the update delta value, the bitwise sum (XOR) of the encrypted versions of the old and new values—that is, u′ is the value included in the update messages sent to the servers, where u′=ve+we.
In an alternative embodiment of the fully secure update module 220, only the g*k servers of the secure update module 315 are used (instead of the f*g*k servers of the above-described embodiment), with per-row shared keys being stored in the g*k servers themselves without the need for additional key servers, and without the need for key distribution or storage of keys in the client devices 120. In this embodiment, the security depends on the quality of the encryption algorithm that uses the keys. More specifically, the content of each row is extended by two fields, s and r. In case of a symmetric encryption algorithm, the first field (s) serves as the seed for an encryption key, and the second field (r) is a pool of random bits. Initialization differs from the initialization of the g*k servers discussed above in that it prepares a wider content matrix C′ from of C as follows: For each row c of C, two fields of fixed width (i.e., s and r) are filled with random bits to which the encrypted form of the content row c is appended, i.e. s|r|enc(c), where “|” denotes appending, and enc(c) denotes the encrypted value of c. (It is appreciated that any other order of combining s, r, and enc(c) would also work, provided that a mechanism is available to separate the appended fields.) An encryption key is derived from the first field s, optionally also using (depending on the properties of the chosen encryption algorithm) a portion of the second field r. This key, together with the other portion of the second field (that can serve as initialization vector, for example), is used to encrypt the original data row c and this ciphertext enc(c) used in C′. The matrix C′ is then distributed to the g*k servers as before. Information retrieval is accomplished by securely retrieving an expanded row, extracting from it the key seed and the second field r, deriving from them the decryption key, and decrypting the bits that follow the two fields. Update is implemented by first securely retrieving the old content of the expanded row, assigning new random bits to the second field r′, encrypting the new data cnew and forming a new expanded row s|r′|enc′(cnew) as described for the initialization, and then updating the g*k servers with the XOR between the retrieved old expanded row and the new expanded row. Note that none of the g*k servers ever obtains the key used to encrypt the row's content, the key seed, the old data value, or the new data value. Rather, the servers only obtain the XOR differences of the ciphertexts and the (random) change of the second field r. The random bits r ensure that each update results in a different bit pattern (via new derived key and fresh initialization vector, for example). In one embodiment using AES and SHA256, for example, the first field (key seed) is 256 bits long, to which 128 bits from the second field are appended before hashing them, and this hash is used as key together with another 128 bits, also from the second field, as the initialization vector for this row.
The secure distribution module 205 of the client device 120A produces 302 a set of K obfuscated portions Ri (e.g., N rows of L bits each) and sends 304 the obfuscated portions to the corresponding K secure storage servers 101. Each secure storage server Si 101 stores its received obfuscated portion Ri in its obfuscated portion storage 255.
The secure read module 210 of the client device 120A produces 312 a set of K query vectors qi that collectively represent the row b whose value is desired, but with randomization such that no individual query vector reveals the identity of the row b. The secure read module 210 distributes 314 the query vectors to the corresponding secure storage servers Si 101, as described above with respect to the secure read module 210.
Each server 101 computes a query result component ai, as described above with respect to the query vector application module 260, and provides 316 the ai to the secure storage module 124 of the client device 120A.
The secure storage module 124 combines 318 the query result components ai using a bitwise sum to derive the desired portion: v, the present value of the row b.
The secure storage module 124 of the client device 120A obtains 330 the current value, v, of the given row to be updated (b) using the secure read module 210 (which involves obtaining 329 the partial query result vectors ai), as illustrated by
The secure storage module 124 computes 332 the update delta value, u, and a set of randomized components t, and generates 334 and sends different update messages containing <t, u> to the two halves of the 2k servers, as described above with respect to the secure update module 215.
Each server Si 101 that receives an update message uses its transformation module 565 to transform its current stored obfuscated portion of C, Rj, using a mask based on the t and u values from the provided update message data. When taken across the set of servers 101 as a whole, this represents the updating of the desired row b to value w, without any given server 101 being able to determine which row b was updated due to the use of the randomized components t to obfuscate the row (though each server does at least know the update delta value u).
At step 338, in response to secure reading of the value of the row b by the client device 120A (as illustrated by
The secure storage module 124 of the client device 120A obtains 350 the current value, v, of the given row to be updated (b) using the secure read module 210, as illustrated by
The secure storage module 124 computes 352 a first update value, to =(p+v), and computes 353 a second update value, u2=(p+w), and then uses the secure update module 215 to update 356 the row b of half of the servers 101 with the first update value and to update 357 the row b of the other half of the servers with the second update value, as described above with respect to the secure update module 215. (That is, in terms of
As a consequence of the way in which the updating of steps 352-357 is performed, when the secure storage module 124 uses the secure read module 210 to obtain 360 the current value by querying the 4k servers, it obtains the value w to which row b was securely updated.
It is appreciated that the particular order of operations described in
The use of the client secure storage module 124 and the secure storage servers 101 permit storage of client data on the servers (e.g., as cloud storage) in a manner that maintains privacy of the data, both with respect to the servers and to other parties, such as network eavesdroppers. The data stored may be of any type, such as messages, trade secrets (e.g., source code or business plans), images, and/or documents.
For example, a user wishing to share the source code for his computer program with other members of his team can securely store the code on the servers and make the credentials required to access it available to his team, such that his team can read and update the code, but none of the servers can determine the contents of the code, nor can a third party snooping on the network.
For another example, a family sharing its contacts in the cloud can use the service to retrieve and update the database without the servers being able to see which contact is looked up most frequently, which contacts recently changed their addresses and whom a contact represents.
For another example, social networking communication can be offered where the servers cannot deduce who has read the posts from whom and who added a comment to what message.
In step 410, the source computer distributes a data item C to a plurality of secure storage servers 101, as discussed above with respect to the secure distribution module 205 of
In step 420, the source computer identifies a row of the data item—that is, a sub-portion of L bits of the data item—to be updated from a current value to a new value. For example, if the data item is a bitmap image securely stored on the servers 101 but read and edited locally on the source computer, and a user locally edits the bitmap image to change the colors of one of the pixels, and then indicates that the revised bitmap image should be securely updated, image editing software executing on the source computer could identify the changed pixel of the bit image, and correspondingly identify which row of the bitmap image data that pixel corresponds to.
In step 430, the source computer computes an update delta value for the row based on the current value of the row (which includes the bits encoding the color value of the pixel stored before the pixel color was changed) and the new value of the row (which includes the bits encoding the color value of the pixel after the pixel color was changed). For example, the update delta value can be the bitwise sum of the current and new values of the row.
In step 440, the source computer distributes update messages to the plurality of servers that cause the servers collectively to update the value of the row from the current value to the new value by updating their corresponding stored obfuscated portions Rij, as described above with respect to the secure update module 215 and the operations of
In step 450, after the update of step 440, the source computer (or another source computer that is authorized to read the data item) requests the updated value of the row. For example, the source computer may re-load the file for the data item at some later time and request the updated value of the row (and of the other rows of the data item) from the servers 101. To obtain the updated value of the row, the source computer generates and distributes a plurality of query vectors to the servers 101, as described above with respect to the secure read module 210 and the operations of
In step 460, the source computer determines the new (updated) value of the row in question using the partial query result vectors aij that the servers 101 send back in response to receiving the query vectors, as described above with respect to the secure read module 210 and the operations of
The storage device 508 is any non-transitory computer-readable storage medium, such as a hard drive, compact disk read-only memory (CD-ROM), DVD (Digital Versatile Disc), or a solid-state memory device. The memory 506 holds instructions and data used by the processor 502. The pointing device 514 may be a mouse, track ball, or other type of pointing device, and is used in combination with the keyboard 510 to input data into the computer 500. The graphics adapter 512 displays images and other information on the display 518. The network adapter 516 couples the computer 500 to a local or wide area network.
As is known in the art, a computer 500 can have different and/or other components than those shown in
As is known in the art, the computer 500 is adapted to execute computer program modules for providing functionality described herein. As used herein, the term “module” refers to computer program logic utilized to provide the specified functionality. Thus, a module can be implemented in hardware, firmware, and/or software. In one embodiment, program modules are stored on the storage device 508, loaded into the memory 506, and executed by the processor 502.
Reference in the specification to “one embodiment” or to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiments is included in at least one embodiment. The appearances of the phrase “in one embodiment” in various places in the specification are not necessarily all referring to the same embodiment.
It should be noted that the process steps and instructions are embodied in software, firmware or hardware, and when embodied in software, can be downloaded to reside on and be operated from different platforms used by a variety of operating systems.
The operations herein may also be performed by an apparatus. Furthermore, the computers referred to in the specification may include a single processor or may be architectures employing multiple processor designs for increased computing capability. It will be appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any references below to specific languages are provided for disclosure of enablement and best mode of the present invention.
While the invention has been particularly shown and described with reference to a preferred embodiment and several alternate embodiments, it will be understood by persons skilled in the relevant art that various changes in form and details can be made therein without departing from the spirit and scope of the invention.
Finally, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, the disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.