re
/
jwt
forked from mirror/jwt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
342 Commits 17 Branches 24 Tags 968 KiB
Commit Graph

16 Commits

Author SHA1 Message Date
Josh Kline 860640e886
Allocation optimization (#33) 
* Test to ensure ECDSA signature is valid

Add assertions to ensure ECDSA signing methods return valid signatures.

This is probably covered elsewhere as well, but putting it in
ecdsa_test.go makes it more obvious and easier to find.

* Benchmark ECDSA signing methods

Add benchmark coverage of ECDSA signing methods.

Benchmarks are run using the existing helper for comparability with
existing benchmarks.

Sign method is also tested directly, to avoid the overhead of *Token.

Report allocations for all benchmarks.

Allocation count for ES384 and ES512 fluctuate across test runs,
other signing methods consistently report the same number of allocations.

Sample output:
```
$ go test -bench=Bench -run=NONE .
2021/02/26 18:18:30 Listening...
goos: darwin
goarch: amd64
pkg: github.com/dgrijalva/jwt-go
BenchmarkECDSASigning/Basic_ES256-8                     190572      6702 ns/op    4249 B/op      65 allocs/op
BenchmarkECDSASigning/Basic_ES256/sign-only-8            47383     24650 ns/op    3329 B/op      43 allocs/op
BenchmarkECDSASigning/Basic_ES384-8                       1113   1252975 ns/op 1750744 B/op   14474 allocs/op
BenchmarkECDSASigning/Basic_ES384/sign-only-8              286   3937773 ns/op 1746175 B/op   14423 allocs/op
BenchmarkECDSASigning/Basic_ES512-8                        662   1949937 ns/op 3028386 B/op   19608 allocs/op
BenchmarkECDSASigning/Basic_ES512/sign-only-8              170   6856189 ns/op 3025471 B/op   19571 allocs/op
BenchmarkECDSASigning/basic_ES256_invalid:_foo_=>_bar-8 190638      6665 ns/op    4249 B/op      65 allocs/op
BenchmarkHS256Signing-8                                1000000      1024 ns/op    1584 B/op      32 allocs/op
BenchmarkHS384Signing-8                                 917286      1447 ns/op    1969 B/op      32 allocs/op
BenchmarkHS512Signing-8                                 827744      1470 ns/op    2065 B/op      32 allocs/op
BenchmarkRS256Signing-8                                   3037    390077 ns/op   32576 B/op     136 allocs/op
BenchmarkRS384Signing-8                                   2976    379155 ns/op   32684 B/op     136 allocs/op
BenchmarkRS512Signing-8                                   3205    388628 ns/op   32704 B/op     136 allocs/op
```

* Reduce allocations during ECDSA signing

Reduce the number of byte arrays allocated by using big.Int.FillBytes
when calculating ECDSA signature.

After this change, Benchmarks of ES256 signing method consistently
report 4 fewer allocations.

Before:
```
BenchmarkECDSASigning/Basic_ES256-8              190572         6702 ns/op       4249 B/op         65 allocs/op
BenchmarkECDSASigning/Basic_ES256/sign-only-8     47383        24650 ns/op       3329 B/op         43 allocs/op
```

After:
```
BenchmarkECDSASigning/Basic_ES256-8              187682         6725 ns/op       4121 B/op         61 allocs/op
BenchmarkECDSASigning/Basic_ES256/sign-only-8     48656        24446 ns/op       3201 B/op         39 allocs/op
```

* Use base64.RawURLEncoding to avoid padding

JWT uses a non-padded base64 encoding.

Current code uses base64.URLEncoding to generate a padded string and
then removes the padding.
Likewise, current code adds padding before decoding.

Instead, use base64.RawURLEncoding which does not add or require the
padding in the first place.

In addition to making the code cleaner, this reduces memory allocations
as reported by benchmarks.

Before:
```
BenchmarkECDSASigning/Basic_ES256-8                     191396         6917 ns/op       4121 B/op         61 allocs/op
BenchmarkECDSASigning/Basic_ES256/sign-only-8            49347        25039 ns/op       3201 B/op         39 allocs/op
BenchmarkECDSASigning/basic_ES256_invalid:_foo_=>_bar-8 190668         6586 ns/op       4121 B/op         61 allocs/op
BenchmarkHS256Signing-8                                1260060         1131 ns/op       1585 B/op         32 allocs/op
BenchmarkHS384Signing-8                                 861378         1387 ns/op       1969 B/op         32 allocs/op
BenchmarkHS512Signing-8                                 896745         1463 ns/op       2065 B/op         32 allocs/op
BenchmarkRS256Signing-8                                   3086       355769 ns/op      32576 B/op        136 allocs/op
BenchmarkRS384Signing-8                                   3414       353570 ns/op      32694 B/op        136 allocs/op
BenchmarkRS512Signing-8                                   3235       349394 ns/op      32706 B/op        136 allocs/op
```

After:
```
BenchmarkECDSASigning/Basic_ES256-8                     176617         6827 ns/op       4021 B/op         58 allocs/op
BenchmarkECDSASigning/Basic_ES256/sign-only-8            48038        24213 ns/op       3169 B/op         38 allocs/op
BenchmarkECDSASigning/basic_ES256_invalid:_foo_=>_bar-8 194352         6928 ns/op       4021 B/op         58 allocs/op
BenchmarkHS256Signing-8                                1000000         1127 ns/op       1488 B/op         29 allocs/op
BenchmarkHS384Signing-8                                 972552         1369 ns/op       1873 B/op         29 allocs/op
BenchmarkHS512Signing-8                                 780751         1368 ns/op       1969 B/op         29 allocs/op
BenchmarkRS256Signing-8                                   3014       387326 ns/op      32475 B/op        133 allocs/op
BenchmarkRS384Signing-8                                   3044       361411 ns/op      32591 B/op        133 allocs/op
BenchmarkRS512Signing-8                                   3273       355504 ns/op      32607 B/op        133 allocs/op
```

Benchmarks of signing methods ES384 and ES512 are omitted because their
allocations are not consistent.
 2021-07-13 08:31:42 +02:00
Kévin Dunglas 42625203e8
chore: code cleanup (#10) 2021-05-27 19:26:21 -04:00
Dave Grijalva 317b82a681 Merge remote-tracking branch 'origin/master' into release_3_0_0 2016-06-06 18:20:35 -07:00
Benjamin Ruston e1403b0ab2 Fix typo in KeyFunc documentation 2016-05-27 14:22:35 +01:00
Dave Grijalva 5e270fa6cd changed argument order to put claims type before keyfunc. this is easier to read when keyfunc is an inline closure 2016-04-12 16:25:25 -07:00
Dave Grijalva bc13ee82c3 Merge branch 'release_3_0_0' into dg/request 2016-04-08 11:57:11 -07:00
Dave Grijalva 4ec621a2d1 Merge branch 'master' into release_3_0_0 2016-03-31 11:19:33 -07:00
Dave Grijalva e0e3b433f5 WIP on migrating request parsing stuff 2016-01-14 14:09:27 -08:00
Snorre lothar von Gohren Edwin fea509ebfe pushed a test change 2015-12-23 09:45:17 +01:00
Snorre lothar von Gohren Edwin ca46641b15 PR updated, faster string method and more reasonable message feedback 2015-12-23 09:43:00 +01:00
Snorre lothar von Gohren Edwin 57b1269c41 modifications on PR. Added a space in the bearer string check so that we unexpectly dont experience an base64url encoding because bearer is technically part of a valid endcoding, we think. Also moved it into a failed decoding to get a better feedback for the developer, but not do unessecary amount of string checks 2015-12-22 15:30:57 +01:00
Snorre lothar von Gohren Edwin 1f970af1f8 added right amount of return 2015-12-20 09:25:50 +01:00
Snorre lothar von Gohren Edwin 5d11392aac no breakpoints 2015-12-19 23:58:27 +01:00
Snorre lothar von Gohren Edwin b863883b96 token.go: did some changes to the checks so that it will give better error feedback for noobs who write the authorization bearer value wrong 2015-12-19 23:49:37 +01:00
Dave Grijalva 56c7810ac4 Merge branch 'master' into release_3_0_0 2015-11-16 11:59:51 -08:00
Dave Grijalva 7ac27fb6ac renamed files to match their purpose 2015-11-02 11:26:07 -08:00