-
Notifications
You must be signed in to change notification settings - Fork 4.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add cached OCSP client support to Cert Auth #17093
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
A bunch of nits but overall 👍
Yeah, I expected a lot of nits given the sprint of coding. |
1 similar comment
Yeah, I expected a lot of nits given the sprint of coding. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've gotta take another look at the trusted cert stuff and make sure I understand that, but this is a first pass review on stuff.
…ed values to stay under the storage entry limit
…l testing of multiple servers
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
A few nits on the non-OCSP responder stuff. I think the last one might be a bug.
Co-authored-by: Alexander Scheel <[email protected]>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Some initial thoughts.
I think two things are worth addressing:
- The behavior within
doRequest
in the OCSP client, - that
nextUpdate
is an optional field.
the rest I don't think is critical AFAICT.
} | ||
if len(rest) > 0 { | ||
// extra bytes to the end | ||
return nil, err |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This returns nil, nil
, wouldn't we want to return a non-nil error here?
|
||
// isInValidityRange checks the validity | ||
func isInValidityRange(currTime, nextUpdate time.Time) bool { | ||
return !nextUpdate.IsZero() && !currTime.After(nextUpdate) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
My reading of https://datatracker.ietf.org/doc/html/rfc6960#section-3.2 seems to state that nextUpdate
is an optional field, i.e., if the OCSP responder doesn't know, they can leave this field blank.
This matches the ASN.1:
nextUpdate [0] EXPLICIT GeneralizedTime OPTIONAL,
If that's the case, I'd probably just sub in the default cache expiry period to make things simple, tbh.
case ocsp.Good: | ||
return &ocspStatus{ | ||
code: ocspStatusGood, | ||
err: nil, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can probably elide err: nil
for consistency.
return returnOCSPStatus(ocspRes), nil | ||
} | ||
|
||
func returnOCSPStatus(ocspRes *ocsp.Response) *ocspStatus { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Seems weird that we're translating from one internal form to another internal form?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is this possibly for their file caching?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, wondering if perhaps it is that. Then they're stable internal identifiers rather than external identifiers that might potentially change and break their cached entries.
But still, I don't think Go upstream would make that type of constants change, except maybe if they import x/ocsp into the main series.
if cacheValue == nil { | ||
return &ocspStatus{ | ||
code: ocspMissedCache, | ||
err: fmt.Errorf("miss cache data. subject: %v", subjectName), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Subject... is just used to have something to log?
producedAt: float64(ocspRes.ProducedAt.UTC().Unix()), | ||
thisUpdate: float64(ocspRes.ThisUpdate.UTC().Unix()), | ||
nextUpdate: float64(ocspRes.NextUpdate.UTC().Unix()), | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Don't we need to set status here, out of curiosity?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would think so
if !verifiedChains[i][numberOfNoneRootCerts].IsCA || string(verifiedChains[i][numberOfNoneRootCerts].RawIssuer) != string(verifiedChains[i][numberOfNoneRootCerts].RawSubject) { | ||
// Check if the last Non Root Cert is also a CA or is self signed. | ||
// if the last certificate is not, add it to the list | ||
rca := c.caRoot[string(verifiedChains[i][numberOfNoneRootCerts].RawIssuer)] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Root subjects are not necessarily unique, but we can fix this later I suppose :-)
sdk/helper/ocsp/client.go
Outdated
return nil | ||
} | ||
|
||
func (c *Client) ValidateWithCacheForAllCertificates(verifiedChains []*x509.Certificate) (bool, error) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Just curious why this is exported?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Earlier version of the code called this. Fixed.
sdk/helper/ocsp/client.go
Outdated
return true, nil | ||
} | ||
|
||
func (c *Client) ValidateWithCache(subject, issuer *x509.Certificate) (*ocspStatus, []byte, *certIDKey, error) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
As above. :-)
} | ||
|
||
// insecureOcspTransport is the transport object that doesn't do certificate revocation check. | ||
func newInsecureOcspTransport(extraCas []*x509.Certificate) *http.Transport { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should this just go in the test helpers?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Now that we have them, yes.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Actually, no, it is used by the OCSP request logic itself.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ah, d'oh, sorry.
I don't like the name then, its not any more insecure than the default Go helper, perhaps just newTransportWithoutOcsp
?
Co-authored-by: Alexander Scheel <[email protected]>
sdk/helper/ocsp/client.go
Outdated
|
||
// cache key | ||
type certIDKey struct { | ||
HashAlgorithm crypto.Hash |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Per discussion on Slack, we can probably remove this field and clarify that NameHash
is the Issuer's Name (for uniqueness) and the SerialNumber
is the leaf's serial number; combination of (IssuerName, IssuerKey)
should uniquely identify the issuer, and LeafSerialNumber
should uniquely identify the leaf within that parent issuer.
Consensus seems to be that the hash algorithm doesn't matter (as it is whatever the OCSP responder uses and doesn't really change our behavior if we want to prefer other hashes) even if it has changed, doesn't really impact our treating previously cached entries as valid or not, given the other three fields are still unique.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think anything left blocks merging!
return returnOCSPStatus(ocspRes), nil | ||
} | ||
|
||
func returnOCSPStatus(ocspRes *ocsp.Response) *ocspStatus { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, wondering if perhaps it is that. Then they're stable internal identifiers rather than external identifiers that might potentially change and break their cached entries.
But still, I don't think Go upstream would make that type of constants change, except maybe if they import x/ocsp into the main series.
Can we please get configurable ocsp request timeouts? Imagine ocsp responder being down (firewalled, incorrectly configured, etc), your ocsp client would be banging it 5 times before giving up. And the cert auth would hang for ~1 min. This isn't the way to do things. Thanks. |
Using a modified version of an Apache licensed OCSP client for Go, modified
to use Vault's storage and have better more consistent error handling,
support for custom configured CAs and overridden OCSP servers.
Mostly for early consideration.