README
¶
Security parameter choice
The present Bcrypt security parameter used is 12, which should take about a quarter of a second on midrange consumer hardware (see Benchmarking section below).
For some background into security parameter considerations, see here and here.
Given our security model, where an attacker would need to already have access to a victim's computer and copy the ~/.gaiacli directory (as opposed to e.g. web authentication), this parameter choice seems sufficient for the time being. Bcrypt always generates a 448-bit key, so the security in practice is determined by the length & complexity of a user's password and the time taken to generate a Bcrypt key from their password (which we can choose with the security parameter). Users would be well-advised to use difficult-to-guess passwords.
Benchmarking
To run Bcrypt benchmarks:
go test -v --bench github.com/cosmos/cosmos-sdk/crypto/keys/mintkey
On the test machine (midrange ThinkPad; i7 6600U), this results in:
goos: linux
goarch: amd64
pkg: github.com/cosmos/cosmos-sdk/crypto/keys/mintkey
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-9-4 50 34609268 ns/op
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-10-4 20 67874471 ns/op
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-11-4 10 135515404 ns/op
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-12-4 5 274824600 ns/op
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-13-4 2 547012903 ns/op
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-14-4 1 1083685904 ns/op
BenchmarkBcryptGenerateFromPassword/benchmark-security-param-15-4 1 2183674041 ns/op
PASS
ok github.com/cosmos/cosmos-sdk/crypto/keys/mintkey 12.093s
Benchmark results are in nanoseconds, so security parameter 12 takes about a quarter of a second to generate the Bcrypt key, security param 13 takes half a second, and so on.
Documentation
¶
Index ¶
- Variables
- func ArmorInfoBytes(bz []byte) string
- func ArmorPubKeyBytes(bz []byte, algo string) string
- func EncryptArmorPrivKey(privKey crypto.PrivKey, passphrase string, algo string) string
- func UnarmorDecryptPrivKey(armorStr string, passphrase string) (privKey crypto.PrivKey, algo string, err error)
- func UnarmorInfoBytes(armorStr string) ([]byte, error)
- func UnarmorPubKeyBytes(armorStr string) (bz []byte, algo string, err error)
Constants ¶
This section is empty.
Variables ¶
var BcryptSecurityParameter = 12
Make bcrypt security parameter var, so it can be changed within the lcd test Making the bcrypt security parameter a var shouldn't be a security issue: One can't verify an invalid key by maliciously changing the bcrypt parameter during a runtime vulnerability. The main security threat this then exposes would be something that changes this during runtime before the user creates their key. This vulnerability must succeed to update this to that same value before every subsequent call to the keys command in future startups / or the attacker must get access to the filesystem. However, with a similar threat model (changing variables in runtime), one can cause the user to sign a different tx than what they see, which is a significantly cheaper attack then breaking a bcrypt hash. (Recall that the nonce still exists to break rainbow tables) For further notes on security parameter choice, see README.md
Functions ¶
func EncryptArmorPrivKey ¶
Encrypt and armor the private key.
func UnarmorDecryptPrivKey ¶
func UnarmorDecryptPrivKey(armorStr string, passphrase string) (privKey crypto.PrivKey, algo string, err error)
UnarmorDecryptPrivKey returns the privkey byte slice, a string of the algo type, and an error
func UnarmorInfoBytes ¶
Unarmor the InfoBytes
Types ¶
This section is empty.
Source Files