------------------------------------------------------------------------------- CORE SDI S.A. Buenos Aires, Argentina Security Advisory June 11th, 1998 SSH insertion attack ------------------------------------------------------------------------------- This advisory addresses a vulnerability present in the SSH software package that allows an attacker to execute arbitrary commands on the SSH server or otherwise subvert an encrypted SSH channel with arbitrary data. Problem Description ~~~~~~~~~~~~~~~~~~~ SSH (Secure Shell) is a program that provides strong authentication and secure communications over insecure channels. Its widely used for logging in to remote computers, file transfers and tunneling of other protocols over the encrypted comunications channel. All communications are automatically and transparently encrypted. Encryption is also used for integrity checking purposes although current implementations rely on a 32 bit Cyclic Redundancy Check to perform integrity checks after the decryption of an incoming packet. Encryption is done using one of a list of supported algorithms that is exchanged between client and server. Upon conection establishment client and server perform a protocol negotiation that includes mutual authentication, selection of a cipher supported by both ends for subsequent communications and of a session key to be used with the cipher. Encryption is then turned on using the selected cipher and session key, all further communications are encrypted. Currently supported ciphers are: - Blowfish Bruce Schneier's block cipher using a 128 bit key - IDEA A 128 bit block cipher - DES The Data Encryption Standard 56-bit block cipher - Triple DES (3DES) A three-key triple-DES algorithm with an effective key lenght of 112 bits. - ARCFOUR An RC4 compatible stream cipher using a 128 bit key The use of these algorithms in CBC (Cipher Block Chaining) or CFB (Cipher Feedback 64 bits) modes with the CRC-32 integrity check allows to perform a known plaintext attack (with as few as 16 bytes of known plaintext) that permits the insertion of encrypted packets with any choosen plaintext in the client to server stream that will subvert the integrity checks on the server and decrypt to the given plaintext, thus allowing an attacker to execute arbitrary commands on the server. The attack is equally feasible on the server to client stream , although it just gives the ability to send arbitrary data the user's terminal. The implications of such an attack are probably not as severe as an attack to the server side of the connection but must be taken in consideration in the process of applying fixes. Technical details ~~~~~~~~~~~~~~~~ After the protocol identification phase, where the server sends a plaintext string specifiying its the protocol and software versions, all communication is done encapsulating data in a packet format described as 'The Binary Packet Protocol' [1] The packet layout is as follows: 32 24 16 8 0 +----------+---------+---------+---------+ | data length (bytes) | +----------+---------+---------+---------+ | 1 to 8 bytes of padding | = = +----------+---------+---------+---------+ | type | | +----------+ + | | = data = | | +----------+---------+---------+---------+ | CRC-32 | +----------+---------+---------+---------+ Data length: Length in bytes of the given packet, not including the length field and padding Padding : 8 - (length mod 8) bytes of random data, putting random data at the beginning of the packet is an effort to make known plaintext attacks more difficult. Packet type: An 8-bit unsigned byte. Data : length - 5 data bytes CRC-32 : the four 8-bit check bytes, MSB first. The CRC is computed before any encryption Encryption is done on the padding+type+data+CRC fields, the length field is never encrypted. The encrypted portion of the packet has a length that is always a multiple of 8 bytes. Knowning certain characteristics of the cipher modes being used, i.e. CBC, with a known plaintext an attacker is able to build a custom SSH packet (i.e. a type SSH_CMSG_STDIN_DATA packet) with the padding bytes computed in a way such that the next 8-bytes of the encrypted data will decrypt to arbitrary plaintext. In this particular case, the decrypted data will correspond to the type field and 7 data bytes. After the 16 bytes (padding+type+7 data bytes) the attacker would include a variable length of data bytes specifically crafted to produce a valid CRC-32 field for the whole packet once it is decrypted. This attack and several variations using the same technique can be performed due to the usage of weak integrity check schemes, in particular CRC-32 has certain properties that allows the attacker to forge a valid CRC for her corrupted packet. However, for the attack to succeed the attacker must be able to perform an active network attack, by either intercepting the legit SSH connection at any point between the client and server and injecting a forged packet or by performing a TCP session hijack attack. Such an attack is described in [6] and for SSH the two methods of TCP desynchronization can be used. In particular the method described as "Null data desynchronization" can be carried out using packets of type SSH_CMSG_IGNORE. Note that the new revision for the SSH protocol, proposed and published as Internet Drafts [2],[3],[4] [5] makes use of cryptographycally strong message authentication codes for integrity checks that wont fail to these attacks. Its is important to mention that despise the vulnerabilities found in the SSH protocol, it still remains to be a much more secure alternative to telnet, rsh and rlogin applications. [1] "The SSH (Secure Shell) Remote Login Protocol", T. Ylonen Helsinki University of Technology. November 15th 1995 (draft expired on May 15th, 1996) Included as the file ./RFC in the ssh distribution [2] "SSH Protocol Architecture", draft-ietf-secsh-architecture-01.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 [3] "SSH Connection Protocol", draft-ietf-secsh-connect-03.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 [4] "SSH Authentication Protocol", draft-ietf-secsh-userauth-03.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 [5] "SSH Transport Layer Protocol",draft-ietf-secsh-transport-03.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 (drafts expired on May 7th, 1998) All Internet drafts are available at [6] "Simple Active Attack Against TCP", Laurent Joncheray, Merit Networks Inc., 5th USENIX Security Simposium. 1995. Impact: ~~~~~~ An attacker with access to the encrypted SSH stream may insert encrypted blocks in the stream that will decrypt to arbitrary commands to be executed on the SSH server. Fix Information: ~~~~~~~~~~~~~~~ Upgrade to the upcoming SSH protocol version 2. Commercial F-Secure SSH users contact Data Fellows Inc. for information on how to upgrade to F-Secure 2.0 Notice that version 2 of the SSH protocol is not compatible with the previous version, thus you will need to upgrade all the SSH clients as well. In the meantime, upgrade to version 1.2.25 of SSH, which fixes the problem. The SSH 1.2.25 distribution can be obtained from: F-Secure SSH version 1.3.5 fixes this security problem. If you are using the commercial Data Fellows SSH package and you have a support contract, you can obtain the 1.3.5 from your local retailer. Users without a support contract can obtain a patch which fixes this problem from: . A patch for the free SSH 1.2.23 distribution and the complete SSH 1.2.23 package, with the patch applied, can be obtained at: Below are the MD5 hashes for the provided files MD5 (ssh-1.2.23.patch) = 6bdb63d57f893907191986c5ced557ab MD5 (ssh-1.2.23-core.tar.Z) = fffb52122aae26c1f212c051a305a310 MD5 (ssh-1.2.23-core.tar.gz) = f9509ba0f0715637805c6b116adc0869 Vulnerable Systems: ~~~~~~~~~~~~~~~~~ All systems running implementations of SSH using protocol version 1.x are vulnerable. This includes SSH software versions up to 1.2.23 and F-Secure SSH 1.3.4 To obtain the version of the SSH server that is running on a given host you can issue the following commands: $ telnet 22 Trying ... Connected to . Escape character is '^]'. SSH-1.5-1.2.23 \ / \--------- software version |------------ protocol version ^] telnet> close Connection closed. $ exit Additional Information: ~~~~~~~~~~~~~~~~~~~~~~ These vulnerabilities were discovered by Ariel Futoransky and Emiliano Kargieman CORE SDI wishes to thank the SSH maintainers Tatu Ylonen and Tero Kivinen for their quick response to the issues rised by this advisory. Olli Voima of Data Fellows Inc. provided the fix information for the F-Secure products. Comments and questions regarding this advisory should be sent to: Ariel Futoransky Emiliano Kargieman For more information about CORE SDI S.A. contact or visit You can contact CORE SDI S.A. at using the the following PGP key: Type Bits/KeyID Date User ID pub 1024/CF4E0CF5 1998/05/18 CORELABS -----BEGIN PGP PUBLIC KEY BLOCK----- Version: 2.6.3ia Comment: Requires PGP version 2.6 or later. mQCNAzVgfMgAAAEEAJSfJNdvCwIAc4AK0ckeimubLIwzsRVDRhjPQIOYt/7kxxio DZybr53fwMEjyT8cHXRL08i0R9rcuFeCNAez6XcalbhqUKXDcLL/cZK80CCDSCs5 tRCZGGOEBnXQIoyvbvi4gNYhBS5wUvmh3b/mvRFTvhmRrUy9m/nO/LnPTgz1AAUR tCBDT1JFTEFCUyA8Y29yZWxhYnNAY29yZS1zZGkuY29tPokAlQMFEDVgfMn5zvy5 z04M9QEBC6ED/0Szt3f54JTvkZG3ezQ8G60HvAw4/A5Ti6i3oze6jsXxzGp6pA1x i0jaZpKaUSpo0MLc7BcijMKneuUHnN3XtN5YxtFt0aEoot1MIvv4BsdeUb3x257G 3+vr8SxGk44Vm4tfuN8F/2dNo/00yYP9rd3zQ8Tl+gmr5VxnLViZIDuh =ulRg -----END PGP PUBLIC KEY BLOCK----- Copyright Notice: ~~~~~~~~~~~~~~~~ The contents of this advisory are Copyright (C) 1998 CORE SDI S.A., and may be distributed freely provided that no fee is charged for this distribution, and proper credit is given. $Id: ssh-advisory.txt,v 1.8 1998/06/11 22:05:03 iarce Exp $ ------------------------------------------------------------------------------- CORE SDI S.A. Buenos Aires, Argentina General Description July 3rd, 1998 SSH insertion attack -------------------------------------------------------------------------------- Crc32 Compensation attack against ssh-1.5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This file gives a general description of the construction of a crc compensation attack against SSH protocol version 1.5. More information on the scope and characteristics of this attack, and the original advisory is available at The attacker is assumed to have active access to the network where the cipherstream travels. She doesnt know neither the private server keys, nor the simetric session key. The server is trusted, it wont reveal private information and wont give any direct access to the attacker. These attacks works for any authentication method used because they are done after the authentication phase of the protocol negotiation. 1. Feedback modes ~~~~~~~~~~~~~~~~~ SSH is currently using two different feedback modes for its block ciphers: Cipher FeedBack (CFB) and Cipher Block Chaining (CBC). The feedback is applied in the same way in every cipher algorithm SSH uses. An Initialization Vector is not negotiated in the SSH protocol and an IV of all zeros is used. Cipher block chaining is: Ci = Ek ( Pi xor Ci-1 ) Pi = Dk ( Ci ) xor Ci-1 CFB-64 is: Ci = Ek ( Ci-1) xor Pi Pi = Ek ( Ci-1) xor Ci Ci = ith block of the cipher stream. Pi = ith block of the plain stream. Ek(x) = Encryption of block x using k as key. Dk(x) = Decryption of block x using k as key. C0 = IV The use of the feedback strengthens the way the block cipher is used randomizing its input and reducing the information an eavesdropper can get from repeated plaintext blocks. However, using this feedback modes doesnt protect the protocol against replay-attacks or cut-and-paste. If the blocks C1..Cj are inserted in cipher stream, block C1 will be decrypted to garbage, and blocks C2..Cj will be decrypted to P2..Pj. If an active eavesdropper knows some block of the plaintext stream, he may insert any block using the ciphertext stream. The attacker knows the ciphertext stream C1..Cj and she knows also that Ci will decrypt to a particular Pi due to protocol characteristics. Now, since the attacker has the Ci/Pi known-plaintext pair and assuming that X is any arbitrary block the attacker may choose, she may insert the following blocks: CBC: === Cj+1 = X Cj+2 = Ci And will be decrypted to Pj+1 = Cj Xor Dk(X) = ? Pj+2 = X Xor Dk(Ci) = X Xor Ci-1 Xor Pi CFB: === Cj+1 = Ci-1 Cj+2 = X And will be decrypted to Pj+1 = Ci-1 Xor Ek(Cj) = ? Pj+2 = X Xor Ek(C-1) = X Xor Pi Xor Ci The attacker cant control whats in Pj+1, but she can insert any block at Pj+2 because she controls X. It should be considered also that for CBC modes, the IV is plain XORed to the first block, although the IV is always zero and cannot be modified. For CFB-64, the last block is plain XORred to the encription of its previous block and can be modified in an easy manner. 2. Checksum ~~~~~~~~~~~ SSH-1.5 protocol includes a checksum field within each packet that incorporates redundancy that allows the receiver of the packet to verify the integrity and authenticity of the packet with a defined threshold of 2^-32. CRC32, the integrity function, is a linear function in Z2-spaces, the remainder from the division of the input stream (as a polinomy in Z2) with the constant polinomy: (X^32+X^26+X^23+X^22+X^16+X^12+X^11+X^10+X^8+X^7+X^5+X^4+X^2+X^1+X^0) As a result of this definition, the following propierties hold: Crc32(A Xor B) = Crc32(A) Xor Crc32(B) Crc32('0' + A) = Crc32(A) If Crc32(A) = 0 Then Crc32(A + '0') = 0 ( '+' in this context means concatenate. '0' means one 0 Bit) 3. Packet ~~~~~~~~~ SSH-1.5 defines the following packet format: 32 24 16 8 0 +----------+---------+---------+---------+ | data length (bytes) | +----------+---------+---------+---------+ | 1 to 8 bytes of padding | = = +----------+---------+---------+---------+ | type | | +----------+ + | | = data = | | +----------+---------+---------+---------+ | CRC-32 | +----------+---------+---------+---------+ Datalen is unencrypted, the rest is encrypted with the selected blockcipher and feedback mode. The last block of the previous packet is used in the feedback. The padding is added to make the packet size be a multiple of 64-bits, and to include at least one padding byte. The crc32 is applied to padding, type and data. Within a packet, there are two message types that are of particular interest and could be directly used in an attack: SSH_STDIN_DATA and SSH_STDOUT_DATA, as defined in [1]. These type of messages has the following message structure: +-------+------------------------+---------------------------------------------+ | TYPE | 32-Bit Message Length | Message data | +-------+------------------------+---------------------------------------------+ SSH closes the connection if: A) The crc-32 field is not consistent with the packet pad/type/data. B) The 32-bit message len is not equal to the packet len minus fields. C) The message type is invalid Within a single cipher block fits: + 1 Byte type field + 4 Bytes Message length + 3 Bytes data 4. ATTACK ~~~~~~~~~ A) The last block of the packet looks like: +---------+--------+--------+--------+----------+---------+---------+---------+ | LAST 4 BYTES | CRC-32 | +---------+--------+--------+--------+----------+---------+---------+---------+ If CFB-64 is used, the last ciphered block Cj, is not encrypted using the blockcipher but is plain XORed to Ek(Cj-1). Given that Crc(A XOR B) = Crc(A) XOR Crc(B), an attacker may replace the last block of the packet in the cipherstream with the XOR of the existing block and 8-bytes block with a valid crc. Because of the linear characteristics of crc32 the packet will have a valid crc. The last four bytes of the data field in a packet could be modified by an attacker in an easy way. If the attacker knows what is in the last four bytes of the data field, she may replace them with arbitrarily choosen bytes. B) If CBC mode is used, and an attacker have two Pi/Ci known-plaintext/ciphertext blocks she may follow the following steps creating a new valid packet that could be inserted in the stream regardless of the blockcipher and key used. The known plaintext/ciphertext pair is Pi/Ci, Pj, Cj The packet is at Cl..Cm and Pl..Pm. I. The attacker choose the packet length to be a multiple of 64-bits, so the padding will be 64-bits. II. She starts the packet with two cipher blocks: Cl = X Cl+1 = Ci X is choosen to make Pl+1 a block with a valid type and 7-bytes of selected data. Pl is unknown to the attacker, but its the padding field of the packet, and is not used in the protocol (except for the crc32 computation). III. The attacker appends 32 64-bits blocks to the cipher stream Cl+3..Cl+34. Each block is either X (The same choosen in step II) or Ci in a particular combination that makes the final crc on Pl+3..Pl+34 independent on the value of Dk(X). That is, for each possible value of Dk(X) the crc of those blocks is the same. This combination exists, and can be found, because of the characteristics of the crc function mentioned on point 2. IV. The attacker appends 32 64-bits blocks to the cipher stream Cl+35..Cl+67. Each block is either Ci (Known Ciphertext A) or Cj (known ciphertext B) making a particular pattern that fix the crc result to the value Pi XOR Ci (paying attention to the last two blocks appended in step V below). V. The attacker adds two Ci blocks to the cipherstream appending the crc32 field to the packet. Cm-1 = Ci Cm = Ci The final encrypted packet looks like: X [block] X/Ci X/Ci X/Ci .... Ci/Cj Ci/Cj Ci/Cj Ci/Cj ... Ci Ci Where 'block' is the constructed block with type/messagelen/data. C) If CFB-64 mode is used and an attacker has one Pi/Ci Known plaintext/ciphertext pair she may follow the following steps to construct a valid packet that will subvert the integrity mechanism. The known plaintext/ciphertext pair is Pi/Ci, 'I' is the last block of the previous packet. The new packet is to be inserted at Cl..Cm and Pl..Pm. I. The attacker chooses the packet length to be a multiple of 64-bits, so the padding will be 64-bits. II. Two blocks are appended to the cipher stream: Cl = Ci Cl+1 = X The attacker selects X to make Pl+1 a valid type and 7 data bytes block. Pl is unknown but is entirely contained in the padding field. III. The attacker inserts 32 64-bit blocks to the cipher stream. Each block is either 'I' or Ci defining a particular pattern that will make the final crc for the packet independent of Ek(I). This step is needed because in cfb-64 what is feeded into to the new packet is Ek(I), as oppossed to CBC mode where 'I' unencripted is feeded. IV. The attacker appends 32 64-bit blocks to the cipher stream. Each block is either Ci or X in a particular combination that defines a pattern for which the CRC on the decrypted stream does not depend on the value of Dk(X). This is the same process used in Step III of attack B. V. The attacker adds two blocks to end the cipher stream: Ci and Y. Ci is the ciphertext of the known plain/ciphertext. Y is the last block of the cipherstream for that packet, it's directly XORed to Ek(Ci). She must select Y to make the crc32 field consistent with the constructed packet. The crc of the packet could be calculated because it doesnt depend on the unknown Ek(X) value nor on Ek(I). The final packet (encrypted) looks like: Ci X Ci/I Ci/I Ci/I .... Ci/X Ci/X Ci/X .......... Ci Y 5. FINAL CONSIDERATIONS ~~~~~~~~~~~~~~~~~~~~~~~ The last two described attacks B, C, defined attacks that inserts 1 selected block of 8 bytes with arbitrary choosen data thru the cipher stream that decrypts to a valid type/messagelen/data with 3 bytes of data that will be inserted into the input stream of the running program or into the client terminal. More that 3 bytes could be inserted by appending more than one Ci, X (X, Ci for CBC) pair at the beginning of the packet, and compensating it's crc later. Most shells include special meta-characters that could help the attacker deal with 8 unknown characters between each 8-bytes choosen blocks (i.e. ` or '). The calculation of the X/Ci pattern that doesnt depend on Dk(X) and the calculation of Ci/Cj for a particular crc can be done easily and efficiently with basic linear algebra. This doesnt imply 2^32 calculations, although 2^32 is within the power reach of most microcomputers. The known plaintext/ciphertext pair could easily be found on ssh-protocol stream given that the attacker knows something about the terminal type or the destination system that the client is trying to connect to. SSH includes this kind of information at fixed positions on particular messages exchanged during the initial negotiation that could be identified by its size (the packet length is sent unencrypted on the network stream). Other sources of known-plaintext are for example the /etc/motd that is displayed at logon on most unixes. 16 and 8 bytes of known plaintext is enough for attacks B and C respectively. Attack A is less practical than B and C because the known plaintext block must be fixed at a particular place within the packet, and it cannot be extended for more than 4 bytes. However attack A needs no known plaintext at all. The attacker needs active access to the network stream of the ssh protocol. That means that she must have access to one machine that's connected to any segment on the connection's path and has sniffing/spoofing capabilities on the network. 6. REFERENCES ~~~~~~~~~~~~~ [1] "The SSH (Secure Shell) Remote Login Protocol", T. Ylonen Helsinki University of Technology. November 15th 1995 (draft expired on May 15th, 1996) Included as the file ./RFC in the ssh distribution [2] "SSH Protocol Architecture", draft-ietf-secsh-architecture-01.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 [3] "SSH Connection Protocol", draft-ietf-secsh-connect-03.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 [4] "SSH Authentication Protocol", draft-ietf-secsh-userauth-03.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 [5] "SSH Transport Layer Protocol",draft-ietf-secsh-transport-03.txt.gz T. Ylonen, T. Kivinen, M. Saarinen. SSH. November 7th, 1997 (drafts expired on May 7th, 1998) All Internet drafts are available at [6] "Simple Active Attack Against TCP", Laurent Joncheray, Merit Networks Inc., 5th USENIX Security Simposium. 1995. [7] "Problem areas for the IP Security Protocols", Steve Bellovin, USENIX 1996. [8] "On Message Integrity in Cryptographic Protocols", S. Stubblebine and V. Gligor, IEEE Computer Society Symposium on Research in Security and Privacy, Oakland, CA, May, 1992, pp. 85-104. [9] "Protocol Design for Integrity Protection", S. Stubblebine and V. Gligor, IEEE Computer Society Symposium on Research in Security and Privacy, Oakland, CA, May, 1993, pp. 41-53. Additional Information: ~~~~~~~~~~~~~~~~~~~~~~~ These vulnerabilities were discovered by Ariel Futoransky and Emiliano Kargieman Comments and questions regarding this attack should be sent to: Ariel Futoransky Emiliano Kargieman For more information about CORE SDI S.A. contact or visit You can contact CORE SDI S.A. at using the the following PGP key: Type Bits/KeyID Date User ID pub 1024/CF4E0CF5 1998/05/18 CORELABS -----BEGIN PGP PUBLIC KEY BLOCK----- Version: 2.6.3ia Comment: Requires PGP version 2.6 or later. mQCNAzVgfMgAAAEEAJSfJNdvCwIAc4AK0ckeimubLIwzsRVDRhjPQIOYt/7kxxio DZybr53fwMEjyT8cHXRL08i0R9rcuFeCNAez6XcalbhqUKXDcLL/cZK80CCDSCs5 tRCZGGOEBnXQIoyvbvi4gNYhBS5wUvmh3b/mvRFTvhmRrUy9m/nO/LnPTgz1AAUR tCBDT1JFTEFCUyA8Y29yZWxhYnNAY29yZS1zZGkuY29tPokAlQMFEDVgfMn5zvy5 z04M9QEBC6ED/0Szt3f54JTvkZG3ezQ8G60HvAw4/A5Ti6i3oze6jsXxzGp6pA1x i0jaZpKaUSpo0MLc7BcijMKneuUHnN3XtN5YxtFt0aEoot1MIvv4BsdeUb3x257G 3+vr8SxGk44Vm4tfuN8F/2dNo/00yYP9rd3zQ8Tl+gmr5VxnLViZIDuh =ulRg -----END PGP PUBLIC KEY BLOCK----- Copyright Notice: ~~~~~~~~~~~~~~~~~ The contents of this document are Copyright (C) 1998 CORE SDI S.A., and may be distributed freely provided that no fee is charged for this distribution, and proper credit is given. $Id: attack.txt,v 1.2 1998/07/03 21:22:27 iarce Exp $ ============================================================================ CORE SDI S.A. Buenos Aires, Argentina SSH insertion attack detection ============================================================================ Crc32 Compensation attack detector ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This document describes the scope and characteristics of SSH crc32 compensation attack detector engine. It will inspect an encrypted SSH protocol 1.5 stream for suspicious patterns. If either the server or client is under attack, the detector closes the conection and logs the attack. Attack characteristics ~~~~~~~~~~~~~~~~~~~~~ An attacker, with sniffing and spoofing capabilities on the SSH network stream, may perform an active network attack by constructing a packet using a known cipher/plaintext pair and computing addditional data to fill the packet in a way that will produce a valid CRC-32 field and pass as a valid packet when decryped and integrity checked on the server side. If the attack succeeds, arbitrary commands will get executed on the server. This attack doesnt affect the confidential security characteristics of the products using this protocols. However, integrity and authentication of packets can no longer be trusted. Solution ~~~~~~~ This is not an implementation bug, but a protocol design vulnerability. The protocol must be updated to use cryptographic strong message authentication codes. SSH Protocol version 2 as published in [1] includes the use of a real strong MAC. However, this will require to update all clients and servers simultaneusly while disabling compatibility with protocol version 1.5. On many installations this is not practical. A mid-term solution exists, that will help protecting the server from the auth/integrity vulnerabilities. On most scenarios it will be enough to upgrade only the servers. This approach is based on the fact that an attacker will need to exploit some of the linear characteristics of the integrity function in order to implement the attack. In that context, the constructed packets will follow certain patterns that could be detected on the encrypted stream. This is not a general solution, but a countermeasure against a model of all of the possible attacks that exploits this vulnerabilities. The provided patch looks for patterns of repeated ciphered blocks in each SSH packet received, the performance loss in speed is less 2% in an uncompressed stream compared to an unpatched server. Memory usage increases in about 8k per conection. The chances of reporting a false attack in a 32GB file transfer is around 1 in 2**51. Contact information ~~~~~~~~~~~~~~~~~~ These vulnerabilities were discovered by Ariel Futoransky and Emiliano Kargieman. Comments and questions regarding this vulnerability and the fix should be sent to: Ariel Futoransky Emiliano Kargieman For more information about CORE SDI S.A. contact or visit For more information about this problem and related ones visit: You can contact CORE SDI S.A. at using the the following PGP key: -----BEGIN PGP PUBLIC KEY BLOCK----- Version: 2.6.3ia Comment: Requires PGP version 2.6 or later. mQCNAzVgfMgAAAEEAJSfJNdvCwIAc4AK0ckeimubLIwzsRVDRhjPQIOYt/7kxxio DZybr53fwMEjyT8cHXRL08i0R9rcuFeCNAez6XcalbhqUKXDcLL/cZK80CCDSCs5 tRCZGGOEBnXQIoyvbvi4gNYhBS5wUvmh3b/mvRFTvhmRrUy9m/nO/LnPTgz1AAUR tCBDT1JFTEFCUyA8Y29yZWxhYnNAY29yZS1zZGkuY29tPokAlQMFEDVgfMn5zvy5 z04M9QEBC6ED/0Szt3f54JTvkZG3ezQ8G60HvAw4/A5Ti6i3oze6jsXxzGp6pA1x i0jaZpKaUSpo0MLc7BcijMKneuUHnN3XtN5YxtFt0aEoot1MIvv4BsdeUb3x257G 3+vr8SxGk44Vm4tfuN8F/2dNo/00yYP9rd3zQ8Tl+gmr5VxnLViZIDuh =ulRg -----END PGP PUBLIC KEY BLOCK----- Copyright ~~~~~~~~ This file, deattack.c and deattack.h are copyright (c) 1998 CORE SDI S.A., Buenos Aires, Argentina. All rights reserved. and should be redistributed or modified for non-commercial purposes only provided that this copyright notice is retained. THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES ARE DISCLAIMED. IN NO EVENT SHALL CORE SDI S.A. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES RESULTING FROM THE USE OR MISUSE OF THIS SOFTWARE.