[go: up one dir, main page]

Skip to content
/ ksum Public

Generate checksums with Keccak-based algorithms, such as SHA-3.

License

Notifications You must be signed in to change notification settings

damaki/ksum

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ksum

ksum is a program to print checksums using algorithms based on Keccak[1] such as SHA-3.

The following algorithms are supported by ksum:

  • SHA-3 and SHAKE as defined in NIST FIPS 202 [2]:
    • SHA3-224, SHA3-256, SHA3-384, and SHA3-512
    • SHAKE128 and SHAKE256
    • RawSHAKE128 and RawSHAKE256
  • Keccak hash functions as defined by the Keccak team:
    • Keccak-224, Keccak-256, Keccak-384, and Keccak-512
  • cSHAKE, KMAC and ParallelHash as defined in NIST SP 800-185 [3]:
    • cSHAKE128 and cSHAKE256
    • KMAC128 and KMAC256
    • ParallelHash128 and ParallelHash256
  • KangarooTwelve and MarsupilamiFourteen as defined by the Keccak authors in [4]

Usage

ksum is invoked by selecting an algorithm (e.g. --sha3-256) and passing in the names of one or more file names. kmac will then output the checksum for each file, in order. For example:

$ ksum --sha3-256 file1 file2 file3
be5215abf72333a73b992dafdf4ab59884b948452e0015cfaddaa0b87a0e4515  file1
006ef4138df934503f34702cfc24b743664b78635dd65844413d464e2867729c  file2
83abc349ca290d8be32afe3d2d1774af58fd799d33afbe8db64afb4572611d39  file3

If - is given on the command line then ksum will read from the standard input. For example:

$ cat file1 | ksum --sha3-256 -
be5215abf72333a73b992dafdf4ab59884b948452e0015cfaddaa0b87a0e4515  -

Checking checksums

The -c or --check switch can be used to read checksums from the FILEs and check them.

For example, to generate a file called checksums.txt that contains the checksums of file1, file2, and file3:

$ ksum --sha3-256 file1 file2 file3 > checksums.txt

Then, to verify the checksums use the --check switch (or its -c short version):

$ ksum --sha3-256 --check checksums.txt

ksum exits with a status of 0 if all checksums are valid. Otherwise, ksum prints diagnostic messages and exits with a non-zero status. Here's an example error output:

$ ksum --sha3-256 --check checksums.txt
file2: FAILED
ksum: WARNING: checksums.txt: 1 computed checksum did NOT match

Note that ksum must be invoked with the same settings that was used to generate the checksums. This is particularly important for algorithms that are customizable, e.g. KMAC, ParallelHash, cSHAKE.

Variable output length

Several of the algorithms supported by ksum can output a variable length checksum. For these algorithms, the -n or --output-size switches control the length of the checksum in bytes. Here's an example using KangarooTwelve (see [4]) to produce different length checksums:

$ ksum --kangarootwelve --output-size=16 file1
7290cf088e4898b86fa4c8cb8f6b5225  file1

$ ksum --kangarootwelve --output-size=32 file1
7290cf088e4898b86fa4c8cb8f6b522599ef37635628165300a52cbb63a0eceb  file1

ParallelHash and KMAC (see [3]) have two different output modes based on the presence of the -x or --xof switch. Without --xof the output checksum depends on the requested output length. I.e. different output lengths will produce different checksums. For example:

$ ksum --parallelhash128 --output-size=16 file1
996a86e8b085365c4b25e4805a58814c  file1

$ ksum --parallelhash128 --output-size=32 file1
867b91a64e1a169adcfee20c3b0a128cc7a45635e9ce8393c4bed74289f3919a  file1

Notice that completely different checksums are produced with different lengths. With the --xof switch the output checksum does not depend on the output length. For example:

$ ksum --parallelhash128 --xof --output-size=16 file1
8f061ea9ea89fd17c258c819a08d58b8  file1

$ ksum --parallelhash128 --xof --output-size=32 file1
8f061ea9ea89fd17c258c819a08d58b8af4bf2f767b15de7c5d39287157f4478  file1

Notice that with --xof the same checksum is produced, just with a different length.

KMAC Key

When --kmac128 or --kmac256 is used a variable-length key can be set using the -k or --key switches which take a hexadecimal string specifying the key. Here's an example of using KMAC128 with a 128-bit (16 bytes) key:

$ ksum --kmac128 --key="000102030405060708090a0b0c0d0e0f" file1
e324342e19ef845694570ede5e53a53b  file1

The default key for KMAC is an empty key (length of 0 bytes).

Customization Strings

KangarooTwelve, KMAC, ParallelHash, and cSHAKE all take an optional customization string which may be used to produce domain separation among different uses of the algorithms. I.e. using different customization strings with the same input will produce unrelated outputs. The customization string is set using the -C or --customization switches. Here's an example:

$ ksum --parallelhash128 --customization="hello" file1
52a4c643ad365c44e9552d88c0dd5917  file1

$ ksum --parallelhash128 --customization="world" file1
149122d772260cdb969511ae5aa56a85  file1

The default customization string is the empty string.

ParallelHash Block Size

ParallelHash takes an optional block size parameter which configures the length of each block (in bytes) that are processed in parallel. The block size (in bytes) is configured using the -B or --block-size switches. The output checksum is dependent on the block size. I.e. different block sizes will produce different (and unrelated) output checksums. For example:

$ ksum --parallelhash128 --block-size=8192 file1
996a86e8b085365c4b25e4805a58814c  file1
$ ksum --parallelhash128 --block-size=4096 file1
111b197c98641f63feab7d24390c5b1c  file1

The default block size for ParallelHash is 8192 bytes (8 kiB).

Building ksum

Building ksum requires Alire.

alr build --release

The ksum executable will be placed in the bin directory.

ksum depends on libkeccak. libkeccak can be built with SSE2 and AVX2 instructions for better performance of parallel hashes, if your platform supports them. To build with AVX2 instructions enabled:

alr build --release -- -XLIBKECCAK_ARCH=x86_64 -XLIBKECCAK_SIMD=AVX2

⚠️ AVX2 is not guaranteed to work on Windows since GCC does not ensure 32-byte stack alignment. See GCC Bug #54412

Performance

The performance of ksum depends on the selected algorithm. The following table shows the time taken to process a 1 GiB file filled with data from /dev/urandom on my machine (64-bit Ubuntu 20.04 on an AMD Ryzen 7 5800X) as measured by the time program. For these tests, ksum was built using GNAT 11.2.0, and with AVX2 instructions enabled in libkeccak.

The table also includes the output of other checksum programs from GNU coreutils 8.30, marked in bold.

The fastest out of 3 runs (according to the "user" value from time) is shown.

Program real user sys
ksum --kangarootwelve bigfile 0m0.513s 0m0.444s 0m0.069s
ksum --parallelhash128 bigfile 0m0.736s 0m0.667s 0m0.070s
ksum --parallelhash256 bigfile 0m0.863s 0m0.843s 0m0.020s
sha1sum bigfile 0m1.112s 0m1.032s 0m0.080s
md5sum bigfile 0m1.186s 0m1.115s 0m0.070s
ksum --shake128 bigfile 0m1.623s 0m1.503s 0m0.120s
ksum --cshake128 bigfile 0m1.623s 0m1.543s 0m0.080s
ksum --kmac128 bigfile 0m1.626s 0m1.586s 0m0.040s
sha384sum bigfile 0m1.884s 0m1.775s 0m0.110s
ksum --sha3-224 bigfile 0m1.860s 0m1.780s 0m0.080s
sha512sum bigfile 0m1.882s 0m1.792s 0m0.090s
ksum --kmac256 bigfile 0m1.961s 0m1.871s 0m0.090s
ksum --sha3-256 bigfile 0m1.960s 0m1.880s 0m0.080s
ksum --shake256 bigfile 0m1.960s 0m1.880s 0m0.080s
ksum --cshake256 bigfile 0m1.965s 0m1.885s 0m0.080s
ksum --sha3-384 bigfile 0m2.492s 0m2.422s 0m0.070s
sha256sum bigfile 0m2.764s 0m2.654s 0m0.110s
sha224sum bigfile 0m2.767s 0m2.706s 0m0.060s
ksum --sha3-512 bigfile 0m3.488s 0m3.447s 0m0.040s

Testing

The test suite runs ksum against a set of test vectors in tests/vectors/, and tests of the various CLI options, e.g. quiet mode.

Running the tests requires Python 3. To run the tests:

alr build --validation
cd tests
python run_tests.py

References