You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

SHA1.java 21KB

SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
SHA-1: collision detection support Update SHA1 class to include a Java port of sha1dc[1]'s ubc_check, which can detect the attack pattern used by the SHAttered[2] authors. Given the shattered example files that have the same SHA-1, this modified implementation can identify there is risk of collision given only one file in the pair: $ jgit ... [main] WARN org.eclipse.jgit.util.sha1.SHA1 - SHA-1 collision 38762cf7f55934b34d179ae6a4c80cadccbb7f0a When JGit detects probability of a collision the SHA1 class now warns on the logger, reporting the object's SHA-1 hash, and then throws a Sha1CollisionException to the caller. From the paper[3] by Marc Stevens, the probability of a false positive identification of a collision is about 14 * 2^(-160), sufficiently low enough for any detected collision to likely be a real collision. git-core[4] may adopt sha1dc before the system migrates to an entirely new hash function. This commit enables JGit to remain compatible with that move to sha1dc, and help protect users by warning if similar attacks as SHAttered are identified. Performance declined about 8% (detection off), now: MessageDigest 238.41 MiB/s MessageDigest 244.52 MiB/s MessageDigest 244.06 MiB/s MessageDigest 242.58 MiB/s SHA1 216.77 MiB/s (was ~240.83 MiB/s) SHA1 220.98 MiB/s SHA1 221.76 MiB/s SHA1 221.34 MiB/s This decline in throughput is attributed to the step loop unrolling in compress(), which was necessary to easily fit the UbcCheck logic into the hash function. Using helper functions s1-s4 reduces the code explosion, providing acceptable throughput. With detection enabled (default): SHA1 detectCollision 180.12 MiB/s SHA1 detectCollision 181.59 MiB/s SHA1 detectCollision 181.64 MiB/s SHA1 detectCollision 182.24 MiB/s sha1dc (native C) ~206.28 MiB/s sha1dc (native C) ~204.47 MiB/s sha1dc (native C) ~203.74 MiB/s Average time across 100,000 calls to hash 4100 bytes (such as a commit or tree) for the various algorithms available to JGit also shows SHA1 is slower than MessageDigest, but by an acceptable margin: MessageDigest 17 usec SHA1 18 usec SHA1 detectCollision 22 usec Time to index-pack for git.git (217982 objects, 69 MiB) has increased: MessageDigest SHA1 w/ detectCollision ------------- ----------------------- 20.12s 25.25s 19.87s 25.48s 20.04s 25.26s avg 20.01s 25.33s +26% Being implemented in Java with these additional safety checks is clearly a penalty, but throughput is still acceptable given the increased security against object name collisions. [1] https://github.com/cr-marcstevens/sha1collisiondetection [2] https://shattered.it/ [3] https://marc-stevens.nl/research/papers/C13-S.pdf [4] https://public-inbox.org/git/20170223230621.43anex65ndoqbgnf@sigill.intra.peff.net/ Change-Id: I9fe4c6d8fc5e5a661af72cd3246c9e67b1b9fee6
7 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601
  1. /*
  2. * Copyright (C) 2017, Google Inc.
  3. * and other copyright owners as documented in the project's IP log.
  4. *
  5. * This program and the accompanying materials are made available
  6. * under the terms of the Eclipse Distribution License v1.0 which
  7. * accompanies this distribution, is reproduced below, and is
  8. * available at http://www.eclipse.org/org/documents/edl-v10.php
  9. *
  10. * All rights reserved.
  11. *
  12. * Redistribution and use in source and binary forms, with or
  13. * without modification, are permitted provided that the following
  14. * conditions are met:
  15. *
  16. * - Redistributions of source code must retain the above copyright
  17. * notice, this list of conditions and the following disclaimer.
  18. *
  19. * - Redistributions in binary form must reproduce the above
  20. * copyright notice, this list of conditions and the following
  21. * disclaimer in the documentation and/or other materials provided
  22. * with the distribution.
  23. *
  24. * - Neither the name of the Eclipse Foundation, Inc. nor the
  25. * names of its contributors may be used to endorse or promote
  26. * products derived from this software without specific prior
  27. * written permission.
  28. *
  29. * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
  30. * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
  31. * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  32. * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  33. * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  34. * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  35. * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  36. * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  37. * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  38. * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  39. * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  40. * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
  41. * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  42. */
  43. package org.eclipse.jgit.util.sha1;
  44. import static java.lang.Integer.lowestOneBit;
  45. import static java.lang.Integer.numberOfTrailingZeros;
  46. import static java.lang.Integer.rotateLeft;
  47. import static java.lang.Integer.rotateRight;
  48. import java.util.Arrays;
  49. import org.eclipse.jgit.lib.MutableObjectId;
  50. import org.eclipse.jgit.lib.ObjectId;
  51. import org.eclipse.jgit.util.NB;
  52. import org.eclipse.jgit.util.SystemReader;
  53. import org.slf4j.Logger;
  54. import org.slf4j.LoggerFactory;
  55. /**
  56. * Pure Java implementation of SHA-1 from FIPS 180-1 / RFC 3174.
  57. *
  58. * <p>
  59. * See <a href="https://tools.ietf.org/html/rfc3174">RFC 3174</a>.
  60. * <p>
  61. * Unlike MessageDigest, this implementation includes the algorithm used by
  62. * {@code sha1dc} to detect cryptanalytic collision attacks against SHA-1, such
  63. * as the one used by <a href="https://shattered.it/">SHAttered</a>. See
  64. * <a href="https://github.com/cr-marcstevens/sha1collisiondetection">
  65. * sha1collisiondetection</a> for more information.
  66. * <p>
  67. * When detectCollision is true (default), this implementation throws
  68. * {@link Sha1CollisionException} from any digest method if a potential
  69. * collision was detected.
  70. *
  71. * @since 4.7
  72. */
  73. public class SHA1 {
  74. private static Logger LOG = LoggerFactory.getLogger(SHA1.class);
  75. private static final boolean DETECT_COLLISIONS;
  76. static {
  77. SystemReader sr = SystemReader.getInstance();
  78. String v = sr.getProperty("org.eclipse.jgit.util.sha1.detectCollision"); //$NON-NLS-1$
  79. DETECT_COLLISIONS = v != null ? Boolean.parseBoolean(v) : true;
  80. }
  81. /** @return a new context to compute a SHA-1 hash of data. */
  82. public static SHA1 newInstance() {
  83. return new SHA1();
  84. }
  85. private final State h = new State();
  86. private final int[] w = new int[80];
  87. /** Buffer to accumulate partial blocks to 64 byte alignment. */
  88. private final byte[] buffer = new byte[64];
  89. /** Total number of bytes in the message. */
  90. private long length;
  91. private boolean detectCollision = DETECT_COLLISIONS;
  92. private boolean foundCollision;
  93. private final int[] w2 = new int[80];
  94. private final State state58 = new State();
  95. private final State state65 = new State();
  96. private final State hIn = new State();
  97. private final State hTmp = new State();
  98. private SHA1() {
  99. h.init();
  100. }
  101. /**
  102. * Enable likely collision detection.
  103. * <p>
  104. * Default is {@code true}.
  105. * <p>
  106. * May also be set by system property:
  107. * {@code -Dorg.eclipse.jgit.util.sha1.detectCollision=true}.
  108. *
  109. * @param detect
  110. * @return {@code this}
  111. */
  112. public SHA1 setDetectCollision(boolean detect) {
  113. detectCollision = detect;
  114. return this;
  115. }
  116. /**
  117. * Update the digest computation by adding a byte.
  118. *
  119. * @param b
  120. */
  121. public void update(byte b) {
  122. int bufferLen = (int) (length & 63);
  123. length++;
  124. buffer[bufferLen] = b;
  125. if (bufferLen == 63) {
  126. compress(buffer, 0);
  127. }
  128. }
  129. /**
  130. * Update the digest computation by adding bytes to the message.
  131. *
  132. * @param in
  133. * input array of bytes.
  134. */
  135. public void update(byte[] in) {
  136. update(in, 0, in.length);
  137. }
  138. /**
  139. * Update the digest computation by adding bytes to the message.
  140. *
  141. * @param in
  142. * input array of bytes.
  143. * @param p
  144. * offset to start at from {@code in}.
  145. * @param len
  146. * number of bytes to hash.
  147. */
  148. public void update(byte[] in, int p, int len) {
  149. // SHA-1 compress can only process whole 64 byte blocks.
  150. // Hold partial updates in buffer, whose length is the low bits.
  151. int bufferLen = (int) (length & 63);
  152. length += len;
  153. if (bufferLen > 0) {
  154. int n = Math.min(64 - bufferLen, len);
  155. System.arraycopy(in, p, buffer, bufferLen, n);
  156. p += n;
  157. len -= n;
  158. if (bufferLen + n < 64) {
  159. return;
  160. }
  161. compress(buffer, 0);
  162. }
  163. while (len >= 64) {
  164. compress(in, p);
  165. p += 64;
  166. len -= 64;
  167. }
  168. if (len > 0) {
  169. System.arraycopy(in, p, buffer, 0, len);
  170. }
  171. }
  172. private void compress(byte[] block, int p) {
  173. initBlock(block, p);
  174. int ubcDvMask = detectCollision ? UbcCheck.check(w) : 0;
  175. compress();
  176. while (ubcDvMask != 0) {
  177. int b = numberOfTrailingZeros(lowestOneBit(ubcDvMask));
  178. UbcCheck.DvInfo dv = UbcCheck.DV[b];
  179. for (int i = 0; i < 80; i++) {
  180. w2[i] = w[i] ^ dv.dm[i];
  181. }
  182. recompress(dv.testt);
  183. if (eq(hTmp, h)) {
  184. foundCollision = true;
  185. break;
  186. }
  187. ubcDvMask &= ~(1 << b);
  188. }
  189. }
  190. private void initBlock(byte[] block, int p) {
  191. for (int t = 0; t < 16; t++) {
  192. w[t] = NB.decodeInt32(block, p + (t << 2));
  193. }
  194. // RFC 3174 6.1.b, extend state vector to 80 words.
  195. for (int t = 16; t < 80; t++) {
  196. int x = w[t - 3] ^ w[t - 8] ^ w[t - 14] ^ w[t - 16];
  197. w[t] = rotateLeft(x, 1); // S^1(...)
  198. }
  199. }
  200. private void compress() {
  201. // Method 1 from RFC 3174 section 6.1.
  202. // Method 2 (circular queue of 16 words) is slower.
  203. int a = h.a, b = h.b, c = h.c, d = h.d, e = h.e;
  204. // @formatter:off
  205. e += s1(a, b, c, d,w[ 0]); b = rotateLeft( b, 30);
  206. d += s1(e, a, b, c,w[ 1]); a = rotateLeft( a, 30);
  207. c += s1(d, e, a, b,w[ 2]); e = rotateLeft( e, 30);
  208. b += s1(c, d, e, a,w[ 3]); d = rotateLeft( d, 30);
  209. a += s1(b, c, d, e,w[ 4]); c = rotateLeft( c, 30);
  210. e += s1(a, b, c, d,w[ 5]); b = rotateLeft( b, 30);
  211. d += s1(e, a, b, c,w[ 6]); a = rotateLeft( a, 30);
  212. c += s1(d, e, a, b,w[ 7]); e = rotateLeft( e, 30);
  213. b += s1(c, d, e, a,w[ 8]); d = rotateLeft( d, 30);
  214. a += s1(b, c, d, e,w[ 9]); c = rotateLeft( c, 30);
  215. e += s1(a, b, c, d,w[ 10]); b = rotateLeft( b, 30);
  216. d += s1(e, a, b, c,w[ 11]); a = rotateLeft( a, 30);
  217. c += s1(d, e, a, b,w[ 12]); e = rotateLeft( e, 30);
  218. b += s1(c, d, e, a,w[ 13]); d = rotateLeft( d, 30);
  219. a += s1(b, c, d, e,w[ 14]); c = rotateLeft( c, 30);
  220. e += s1(a, b, c, d,w[ 15]); b = rotateLeft( b, 30);
  221. d += s1(e, a, b, c,w[ 16]); a = rotateLeft( a, 30);
  222. c += s1(d, e, a, b,w[ 17]); e = rotateLeft( e, 30);
  223. b += s1(c, d, e, a,w[ 18]); d = rotateLeft( d, 30);
  224. a += s1(b, c, d, e,w[ 19]); c = rotateLeft( c, 30);
  225. e += s2(a, b, c, d,w[ 20]); b = rotateLeft( b, 30);
  226. d += s2(e, a, b, c,w[ 21]); a = rotateLeft( a, 30);
  227. c += s2(d, e, a, b,w[ 22]); e = rotateLeft( e, 30);
  228. b += s2(c, d, e, a,w[ 23]); d = rotateLeft( d, 30);
  229. a += s2(b, c, d, e,w[ 24]); c = rotateLeft( c, 30);
  230. e += s2(a, b, c, d,w[ 25]); b = rotateLeft( b, 30);
  231. d += s2(e, a, b, c,w[ 26]); a = rotateLeft( a, 30);
  232. c += s2(d, e, a, b,w[ 27]); e = rotateLeft( e, 30);
  233. b += s2(c, d, e, a,w[ 28]); d = rotateLeft( d, 30);
  234. a += s2(b, c, d, e,w[ 29]); c = rotateLeft( c, 30);
  235. e += s2(a, b, c, d,w[ 30]); b = rotateLeft( b, 30);
  236. d += s2(e, a, b, c,w[ 31]); a = rotateLeft( a, 30);
  237. c += s2(d, e, a, b,w[ 32]); e = rotateLeft( e, 30);
  238. b += s2(c, d, e, a,w[ 33]); d = rotateLeft( d, 30);
  239. a += s2(b, c, d, e,w[ 34]); c = rotateLeft( c, 30);
  240. e += s2(a, b, c, d,w[ 35]); b = rotateLeft( b, 30);
  241. d += s2(e, a, b, c,w[ 36]); a = rotateLeft( a, 30);
  242. c += s2(d, e, a, b,w[ 37]); e = rotateLeft( e, 30);
  243. b += s2(c, d, e, a,w[ 38]); d = rotateLeft( d, 30);
  244. a += s2(b, c, d, e,w[ 39]); c = rotateLeft( c, 30);
  245. e += s3(a, b, c, d,w[ 40]); b = rotateLeft( b, 30);
  246. d += s3(e, a, b, c,w[ 41]); a = rotateLeft( a, 30);
  247. c += s3(d, e, a, b,w[ 42]); e = rotateLeft( e, 30);
  248. b += s3(c, d, e, a,w[ 43]); d = rotateLeft( d, 30);
  249. a += s3(b, c, d, e,w[ 44]); c = rotateLeft( c, 30);
  250. e += s3(a, b, c, d,w[ 45]); b = rotateLeft( b, 30);
  251. d += s3(e, a, b, c,w[ 46]); a = rotateLeft( a, 30);
  252. c += s3(d, e, a, b,w[ 47]); e = rotateLeft( e, 30);
  253. b += s3(c, d, e, a,w[ 48]); d = rotateLeft( d, 30);
  254. a += s3(b, c, d, e,w[ 49]); c = rotateLeft( c, 30);
  255. e += s3(a, b, c, d,w[ 50]); b = rotateLeft( b, 30);
  256. d += s3(e, a, b, c,w[ 51]); a = rotateLeft( a, 30);
  257. c += s3(d, e, a, b,w[ 52]); e = rotateLeft( e, 30);
  258. b += s3(c, d, e, a,w[ 53]); d = rotateLeft( d, 30);
  259. a += s3(b, c, d, e,w[ 54]); c = rotateLeft( c, 30);
  260. e += s3(a, b, c, d,w[ 55]); b = rotateLeft( b, 30);
  261. d += s3(e, a, b, c,w[ 56]); a = rotateLeft( a, 30);
  262. c += s3(d, e, a, b,w[ 57]); e = rotateLeft( e, 30);
  263. state58.save(a, b, c, d, e);
  264. b += s3(c, d, e, a,w[ 58]); d = rotateLeft( d, 30);
  265. a += s3(b, c, d, e,w[ 59]); c = rotateLeft( c, 30);
  266. e += s4(a, b, c, d,w[ 60]); b = rotateLeft( b, 30);
  267. d += s4(e, a, b, c,w[ 61]); a = rotateLeft( a, 30);
  268. c += s4(d, e, a, b,w[ 62]); e = rotateLeft( e, 30);
  269. b += s4(c, d, e, a,w[ 63]); d = rotateLeft( d, 30);
  270. a += s4(b, c, d, e,w[ 64]); c = rotateLeft( c, 30);
  271. state65.save(a, b, c, d, e);
  272. e += s4(a, b, c, d,w[ 65]); b = rotateLeft( b, 30);
  273. d += s4(e, a, b, c,w[ 66]); a = rotateLeft( a, 30);
  274. c += s4(d, e, a, b,w[ 67]); e = rotateLeft( e, 30);
  275. b += s4(c, d, e, a,w[ 68]); d = rotateLeft( d, 30);
  276. a += s4(b, c, d, e,w[ 69]); c = rotateLeft( c, 30);
  277. e += s4(a, b, c, d,w[ 70]); b = rotateLeft( b, 30);
  278. d += s4(e, a, b, c,w[ 71]); a = rotateLeft( a, 30);
  279. c += s4(d, e, a, b,w[ 72]); e = rotateLeft( e, 30);
  280. b += s4(c, d, e, a,w[ 73]); d = rotateLeft( d, 30);
  281. a += s4(b, c, d, e,w[ 74]); c = rotateLeft( c, 30);
  282. e += s4(a, b, c, d,w[ 75]); b = rotateLeft( b, 30);
  283. d += s4(e, a, b, c,w[ 76]); a = rotateLeft( a, 30);
  284. c += s4(d, e, a, b,w[ 77]); e = rotateLeft( e, 30);
  285. b += s4(c, d, e, a,w[ 78]); d = rotateLeft( d, 30);
  286. a += s4(b, c, d, e,w[ 79]); c = rotateLeft( c, 30);
  287. // @formatter:on
  288. h.save(h.a + a, h.b + b, h.c + c, h.d + d, h.e + e);
  289. }
  290. private void recompress(int t) {
  291. State s;
  292. if (t == 58) {
  293. s = state58;
  294. } else if (t == 65) {
  295. s = state65;
  296. } else {
  297. throw new IllegalStateException();
  298. }
  299. int a = s.a, b = s.b, c = s.c, d = s.d, e = s.e;
  300. // @formatter:off
  301. if (t == 65) {
  302. { c = rotateRight( c, 30); a -= s4(b, c, d, e,w2[ 64]);}
  303. { d = rotateRight( d, 30); b -= s4(c, d, e, a,w2[ 63]);}
  304. { e = rotateRight( e, 30); c -= s4(d, e, a, b,w2[ 62]);}
  305. { a = rotateRight( a, 30); d -= s4(e, a, b, c,w2[ 61]);}
  306. { b = rotateRight( b, 30); e -= s4(a, b, c, d,w2[ 60]);}
  307. { c = rotateRight( c, 30); a -= s3(b, c, d, e,w2[ 59]);}
  308. { d = rotateRight( d, 30); b -= s3(c, d, e, a,w2[ 58]);}
  309. }
  310. { e = rotateRight( e, 30); c -= s3(d, e, a, b,w2[ 57]);}
  311. { a = rotateRight( a, 30); d -= s3(e, a, b, c,w2[ 56]);}
  312. { b = rotateRight( b, 30); e -= s3(a, b, c, d,w2[ 55]);}
  313. { c = rotateRight( c, 30); a -= s3(b, c, d, e,w2[ 54]);}
  314. { d = rotateRight( d, 30); b -= s3(c, d, e, a,w2[ 53]);}
  315. { e = rotateRight( e, 30); c -= s3(d, e, a, b,w2[ 52]);}
  316. { a = rotateRight( a, 30); d -= s3(e, a, b, c,w2[ 51]);}
  317. { b = rotateRight( b, 30); e -= s3(a, b, c, d,w2[ 50]);}
  318. { c = rotateRight( c, 30); a -= s3(b, c, d, e,w2[ 49]);}
  319. { d = rotateRight( d, 30); b -= s3(c, d, e, a,w2[ 48]);}
  320. { e = rotateRight( e, 30); c -= s3(d, e, a, b,w2[ 47]);}
  321. { a = rotateRight( a, 30); d -= s3(e, a, b, c,w2[ 46]);}
  322. { b = rotateRight( b, 30); e -= s3(a, b, c, d,w2[ 45]);}
  323. { c = rotateRight( c, 30); a -= s3(b, c, d, e,w2[ 44]);}
  324. { d = rotateRight( d, 30); b -= s3(c, d, e, a,w2[ 43]);}
  325. { e = rotateRight( e, 30); c -= s3(d, e, a, b,w2[ 42]);}
  326. { a = rotateRight( a, 30); d -= s3(e, a, b, c,w2[ 41]);}
  327. { b = rotateRight( b, 30); e -= s3(a, b, c, d,w2[ 40]);}
  328. { c = rotateRight( c, 30); a -= s2(b, c, d, e,w2[ 39]);}
  329. { d = rotateRight( d, 30); b -= s2(c, d, e, a,w2[ 38]);}
  330. { e = rotateRight( e, 30); c -= s2(d, e, a, b,w2[ 37]);}
  331. { a = rotateRight( a, 30); d -= s2(e, a, b, c,w2[ 36]);}
  332. { b = rotateRight( b, 30); e -= s2(a, b, c, d,w2[ 35]);}
  333. { c = rotateRight( c, 30); a -= s2(b, c, d, e,w2[ 34]);}
  334. { d = rotateRight( d, 30); b -= s2(c, d, e, a,w2[ 33]);}
  335. { e = rotateRight( e, 30); c -= s2(d, e, a, b,w2[ 32]);}
  336. { a = rotateRight( a, 30); d -= s2(e, a, b, c,w2[ 31]);}
  337. { b = rotateRight( b, 30); e -= s2(a, b, c, d,w2[ 30]);}
  338. { c = rotateRight( c, 30); a -= s2(b, c, d, e,w2[ 29]);}
  339. { d = rotateRight( d, 30); b -= s2(c, d, e, a,w2[ 28]);}
  340. { e = rotateRight( e, 30); c -= s2(d, e, a, b,w2[ 27]);}
  341. { a = rotateRight( a, 30); d -= s2(e, a, b, c,w2[ 26]);}
  342. { b = rotateRight( b, 30); e -= s2(a, b, c, d,w2[ 25]);}
  343. { c = rotateRight( c, 30); a -= s2(b, c, d, e,w2[ 24]);}
  344. { d = rotateRight( d, 30); b -= s2(c, d, e, a,w2[ 23]);}
  345. { e = rotateRight( e, 30); c -= s2(d, e, a, b,w2[ 22]);}
  346. { a = rotateRight( a, 30); d -= s2(e, a, b, c,w2[ 21]);}
  347. { b = rotateRight( b, 30); e -= s2(a, b, c, d,w2[ 20]);}
  348. { c = rotateRight( c, 30); a -= s1(b, c, d, e,w2[ 19]);}
  349. { d = rotateRight( d, 30); b -= s1(c, d, e, a,w2[ 18]);}
  350. { e = rotateRight( e, 30); c -= s1(d, e, a, b,w2[ 17]);}
  351. { a = rotateRight( a, 30); d -= s1(e, a, b, c,w2[ 16]);}
  352. { b = rotateRight( b, 30); e -= s1(a, b, c, d,w2[ 15]);}
  353. { c = rotateRight( c, 30); a -= s1(b, c, d, e,w2[ 14]);}
  354. { d = rotateRight( d, 30); b -= s1(c, d, e, a,w2[ 13]);}
  355. { e = rotateRight( e, 30); c -= s1(d, e, a, b,w2[ 12]);}
  356. { a = rotateRight( a, 30); d -= s1(e, a, b, c,w2[ 11]);}
  357. { b = rotateRight( b, 30); e -= s1(a, b, c, d,w2[ 10]);}
  358. { c = rotateRight( c, 30); a -= s1(b, c, d, e,w2[ 9]);}
  359. { d = rotateRight( d, 30); b -= s1(c, d, e, a,w2[ 8]);}
  360. { e = rotateRight( e, 30); c -= s1(d, e, a, b,w2[ 7]);}
  361. { a = rotateRight( a, 30); d -= s1(e, a, b, c,w2[ 6]);}
  362. { b = rotateRight( b, 30); e -= s1(a, b, c, d,w2[ 5]);}
  363. { c = rotateRight( c, 30); a -= s1(b, c, d, e,w2[ 4]);}
  364. { d = rotateRight( d, 30); b -= s1(c, d, e, a,w2[ 3]);}
  365. { e = rotateRight( e, 30); c -= s1(d, e, a, b,w2[ 2]);}
  366. { a = rotateRight( a, 30); d -= s1(e, a, b, c,w2[ 1]);}
  367. { b = rotateRight( b, 30); e -= s1(a, b, c, d,w2[ 0]);}
  368. hIn.save(a, b, c, d, e);
  369. a = s.a; b = s.b; c = s.c; d = s.d; e = s.e;
  370. if (t == 58) {
  371. { b += s3(c, d, e, a,w2[ 58]); d = rotateLeft( d, 30);}
  372. { a += s3(b, c, d, e,w2[ 59]); c = rotateLeft( c, 30);}
  373. { e += s4(a, b, c, d,w2[ 60]); b = rotateLeft( b, 30);}
  374. { d += s4(e, a, b, c,w2[ 61]); a = rotateLeft( a, 30);}
  375. { c += s4(d, e, a, b,w2[ 62]); e = rotateLeft( e, 30);}
  376. { b += s4(c, d, e, a,w2[ 63]); d = rotateLeft( d, 30);}
  377. { a += s4(b, c, d, e,w2[ 64]); c = rotateLeft( c, 30);}
  378. }
  379. { e += s4(a, b, c, d,w2[ 65]); b = rotateLeft( b, 30);}
  380. { d += s4(e, a, b, c,w2[ 66]); a = rotateLeft( a, 30);}
  381. { c += s4(d, e, a, b,w2[ 67]); e = rotateLeft( e, 30);}
  382. { b += s4(c, d, e, a,w2[ 68]); d = rotateLeft( d, 30);}
  383. { a += s4(b, c, d, e,w2[ 69]); c = rotateLeft( c, 30);}
  384. { e += s4(a, b, c, d,w2[ 70]); b = rotateLeft( b, 30);}
  385. { d += s4(e, a, b, c,w2[ 71]); a = rotateLeft( a, 30);}
  386. { c += s4(d, e, a, b,w2[ 72]); e = rotateLeft( e, 30);}
  387. { b += s4(c, d, e, a,w2[ 73]); d = rotateLeft( d, 30);}
  388. { a += s4(b, c, d, e,w2[ 74]); c = rotateLeft( c, 30);}
  389. { e += s4(a, b, c, d,w2[ 75]); b = rotateLeft( b, 30);}
  390. { d += s4(e, a, b, c,w2[ 76]); a = rotateLeft( a, 30);}
  391. { c += s4(d, e, a, b,w2[ 77]); e = rotateLeft( e, 30);}
  392. { b += s4(c, d, e, a,w2[ 78]); d = rotateLeft( d, 30);}
  393. { a += s4(b, c, d, e,w2[ 79]); c = rotateLeft( c, 30);}
  394. // @formatter:on
  395. hTmp.save(hIn.a + a, hIn.b + b, hIn.c + c, hIn.d + d, hIn.e + e);
  396. }
  397. private static int s1(int a, int b, int c, int d, int w_t) {
  398. return rotateLeft(a, 5)
  399. // f: 0 <= t <= 19
  400. + ((b & c) | ((~b) & d))
  401. + 0x5A827999 + w_t;
  402. }
  403. private static int s2(int a, int b, int c, int d, int w_t) {
  404. return rotateLeft(a, 5)
  405. // f: 20 <= t <= 39
  406. + (b ^ c ^ d)
  407. + 0x6ED9EBA1 + w_t;
  408. }
  409. private static int s3(int a, int b, int c, int d, int w_t) {
  410. return rotateLeft(a, 5)
  411. // f: 40 <= t <= 59
  412. + ((b & c) | (b & d) | (c & d))
  413. + 0x8F1BBCDC + w_t;
  414. }
  415. private static int s4(int a, int b, int c, int d, int w_t) {
  416. return rotateLeft(a, 5)
  417. // f: 60 <= t <= 79
  418. + (b ^ c ^ d)
  419. + 0xCA62C1D6 + w_t;
  420. }
  421. private static boolean eq(State q, State r) {
  422. return q.a == r.a
  423. && q.b == r.b
  424. && q.c == r.c
  425. && q.d == r.d
  426. && q.e == r.e;
  427. }
  428. private void finish() {
  429. int bufferLen = (int) (length & 63);
  430. if (bufferLen > 55) {
  431. // Last block is too small; pad, compress, pad another block.
  432. buffer[bufferLen++] = (byte) 0x80;
  433. Arrays.fill(buffer, bufferLen, 64, (byte) 0);
  434. compress(buffer, 0);
  435. Arrays.fill(buffer, 0, 56, (byte) 0);
  436. } else {
  437. // Last block can hold padding and length.
  438. buffer[bufferLen++] = (byte) 0x80;
  439. Arrays.fill(buffer, bufferLen, 56, (byte) 0);
  440. }
  441. // SHA-1 appends the length of the message in bits after the
  442. // padding block (above). Here length is in bytes. Multiply by
  443. // 8 by shifting by 3 as part of storing the 64 bit byte length
  444. // into the two words expected in the trailer.
  445. NB.encodeInt32(buffer, 56, (int) (length >>> (32 - 3)));
  446. NB.encodeInt32(buffer, 60, (int) (length << 3));
  447. compress(buffer, 0);
  448. if (foundCollision) {
  449. ObjectId id = h.toObjectId();
  450. LOG.warn("possible SHA-1 collision " + id.name()); //$NON-NLS-1$
  451. throw new Sha1CollisionException(id);
  452. }
  453. }
  454. /**
  455. * Finish the digest and return the resulting hash.
  456. * <p>
  457. * Once {@code digest()} is called, this instance should be discarded.
  458. *
  459. * @return the bytes for the resulting hash.
  460. * @throws Sha1CollisionException
  461. * if a collision was detected and safeHash is false.
  462. */
  463. public byte[] digest() throws Sha1CollisionException {
  464. finish();
  465. byte[] b = new byte[20];
  466. NB.encodeInt32(b, 0, h.a);
  467. NB.encodeInt32(b, 4, h.b);
  468. NB.encodeInt32(b, 8, h.c);
  469. NB.encodeInt32(b, 12, h.d);
  470. NB.encodeInt32(b, 16, h.e);
  471. return b;
  472. }
  473. /**
  474. * Finish the digest and return the resulting hash.
  475. * <p>
  476. * Once {@code digest()} is called, this instance should be discarded.
  477. *
  478. * @return the ObjectId for the resulting hash.
  479. * @throws Sha1CollisionException
  480. * if a collision was detected and safeHash is false.
  481. */
  482. public ObjectId toObjectId() throws Sha1CollisionException {
  483. finish();
  484. return h.toObjectId();
  485. }
  486. /**
  487. * Finish the digest and return the resulting hash.
  488. * <p>
  489. * Once {@code digest()} is called, this instance should be discarded.
  490. *
  491. * @param id
  492. * destination to copy the digest to.
  493. * @throws Sha1CollisionException
  494. * if a collision was detected and safeHash is false.
  495. */
  496. public void digest(MutableObjectId id) throws Sha1CollisionException {
  497. finish();
  498. id.set(h.a, h.b, h.c, h.d, h.e);
  499. }
  500. /**
  501. * Check if a collision was detected.
  502. *
  503. * <p>
  504. * This method only returns an accurate result after the digest was obtained
  505. * through {@link #digest()}, {@link #digest(MutableObjectId)} or
  506. * {@link #toObjectId()}, as the hashing function must finish processing to
  507. * know the final state.
  508. *
  509. * @return {@code true} if a likely collision was detected.
  510. */
  511. public boolean hasCollision() {
  512. return foundCollision;
  513. }
  514. /**
  515. * Reset this instance to compute another hash.
  516. *
  517. * @return {@code this}.
  518. */
  519. public SHA1 reset() {
  520. h.init();
  521. length = 0;
  522. foundCollision = false;
  523. return this;
  524. }
  525. private static final class State {
  526. int a;
  527. int b;
  528. int c;
  529. int d;
  530. int e;
  531. final void init() {
  532. // Magic initialization constants defined by FIPS180.
  533. save(0x67452301, 0xEFCDAB89, 0x98BADCFE, 0x10325476, 0xC3D2E1F0);
  534. }
  535. final void save(int a1, int b1, int c1, int d1, int e1) {
  536. a = a1;
  537. b = b1;
  538. c = c1;
  539. d = d1;
  540. e = e1;
  541. }
  542. ObjectId toObjectId() {
  543. return new ObjectId(a, b, c, d, e);
  544. }
  545. }
  546. }