mirror of
https://github.com/hanzoai/ai.git
synced 2026-11-20 14:13:23 +00:00
hk- is gone. IAM resolves sk-, which authenticates, and pk-, which names an org and authenticates nobody; every other spelling — hk- among them — answers to no live credential and comes back key_unknown, which keyRefusal turns into the one sentence a stranded holder can act on: mint a new one. Nothing here special-cases the retired prefix, because inventing a case for it would put the third family back. That leaves the dispatch unable to tell two families apart by prefix: the key this estate mints and an upstream vendor key wear the same three characters. So it stops trying. The provider table is asked first — an exact lookup, so it can never claim a key it does not hold — and anything it misses is put to IAM. The order matters only for who answers when NEITHER owns the key, and IAM's refusal names the cure where a provider miss can only say "invalid". resolveBillingKey skipped every sk- before it reached the cache, back when sk- meant "vendor key, billed elsewhere". Keeping that would now skip the main credential and cost IAM keys their router-level gate, so only pk- skips. The IAM miss is cached alongside the hit, or a vendor key would buy a fresh round-trip on every request it makes. TestGetUserByAccessKey_RetiredPrefixGetsTheActionableRefusal pins the outcome: an hk- key gets "not recognized … Mint a new one", never the bare "the entity does not exist" relay that sent people looking for a deleted organization, and the refusal carries the prefix only. Co-authored-by: Hanzo Dev <[email protected]>
84 lines
3.5 KiB
Go
84 lines
3.5 KiB
Go
package controllers
|
|
|
|
import (
|
|
"crypto/sha256"
|
|
"encoding/hex"
|
|
"regexp"
|
|
"strings"
|
|
)
|
|
|
|
// Credential redaction for the error log.
|
|
//
|
|
// WHY: the API-error line logged the Authorization header verbatim. Measured on
|
|
// the live cloud pod, 417 full user JWTs were sitting in the log — real,
|
|
// unexpired, replayable bearer tokens for named users. Anyone with log read
|
|
// access (o11y, the datastore, a support export) held those users' sessions.
|
|
//
|
|
// A log line still needs to answer "which caller was this?", so the token is
|
|
// replaced by a FINGERPRINT rather than dropped: scheme plus the first 8 bytes of
|
|
// its SHA-256. That is enough to correlate repeated failures from one credential
|
|
// and to match a report against a session, and it cannot be replayed.
|
|
//
|
|
// WHY NOT routers.maskKey. That helper already exists for the same-shaped job
|
|
// (`apiKey[:6] + "..."`) and reusing it would be the one-way choice, except that
|
|
// prefix truncation is actively WRONG for a bearer JWT: every RS256 token begins
|
|
// with the same encoded header, so "eyJhbG..." is identical for every user in the
|
|
// fleet — it discriminates nothing while still emitting real key material. It
|
|
// remains the right tool where it is used, on short opaque `sk-` API keys. Two
|
|
// credential shapes, two correct treatments; a single helper here would have to
|
|
// branch on shape anyway, and hiding that branch is what makes a leak subtle.
|
|
|
|
// sensitiveJSONKey matches the JSON keys whose values must never reach a log.
|
|
// Passwords are the sharp case: an error on a signup or signin request logs the
|
|
// whole body, so a plaintext password would be written to the warehouse — which
|
|
// the hashing rule exists to prevent, and a log is not an exception to it.
|
|
var sensitiveJSONKey = regexp.MustCompile(
|
|
`(?i)"(password|passwd|pass|secret|token|access_token|refresh_token|id_token|api_key|apikey|client_secret|private_key|otp|code_verifier|card|cvv|cvc|ssn)"\s*:\s*"[^"]*"`)
|
|
|
|
// sensitiveQueryParam matches credential-bearing query parameters. A key in a URL
|
|
// is already the worst place for one, and logging it makes it permanent.
|
|
var sensitiveQueryParam = regexp.MustCompile(
|
|
`(?i)\b(access_token|refresh_token|id_token|api_key|apikey|token|secret|client_secret|password|code|otp)=[^&\s]*`)
|
|
|
|
// redactCredential turns an Authorization header into a correlatable fingerprint.
|
|
// The scheme is kept because it is diagnostic (a "Basic" where a "Bearer" was
|
|
// expected is a real bug) and carries no secret.
|
|
func redactCredential(header string) string {
|
|
h := strings.TrimSpace(header)
|
|
if h == "" {
|
|
return ""
|
|
}
|
|
scheme, rest := "", h
|
|
if i := strings.IndexByte(h, ' '); i > 0 {
|
|
scheme, rest = h[:i], strings.TrimSpace(h[i+1:])
|
|
}
|
|
sum := sha256.Sum256([]byte(rest))
|
|
fp := "sha256:" + hex.EncodeToString(sum[:8])
|
|
if scheme == "" {
|
|
return fp
|
|
}
|
|
return scheme + " " + fp
|
|
}
|
|
|
|
// redactBody removes the values of credential-bearing JSON keys, leaving the rest
|
|
// of the body readable — the point of logging a failing body is the shape of the
|
|
// request, which survives redaction.
|
|
func redactBody(body string) string {
|
|
return sensitiveJSONKey.ReplaceAllStringFunc(body, func(m string) string {
|
|
if i := strings.Index(m, ":"); i > 0 {
|
|
return m[:i+1] + `"[redacted]"`
|
|
}
|
|
return m
|
|
})
|
|
}
|
|
|
|
// redactQuery removes credential values from a raw query string.
|
|
func redactQuery(q string) string {
|
|
return sensitiveQueryParam.ReplaceAllStringFunc(q, func(m string) string {
|
|
if i := strings.IndexByte(m, '='); i > 0 {
|
|
return m[:i+1] + "[redacted]"
|
|
}
|
|
return m
|
|
})
|
|
}
|