curl/docs/internals/CREDENTIALS.md
Stefan Eissing 8f71d0fde5
creds: hold credentials
Authorizdation credentials are kept in `struct Curl_creds`. This contains:

* `user`: the username, maybe the empty string
* `passwd`: the password, maybe the empty string
* `sasl_authzid`: the SASL authz value, maybe the empty string
* `oauth_bearer`: the OAUTH bearer token, maybe the empty string
* `source`: where the credentials from from
* `refcount`: a reference counter to link/unkink creds

A `creds` with all values empty is equivalent to NULL, e.g. no `creds`
instance. With reference counting, `creds` can be linked/unlinked
in several places.

See docs/internals/CREDENTIALS.md for use.

Closes #21548
2026-05-12 16:05:15 +02:00

3 KiB

curl creds

Authorization credentials are kept in struct Curl_creds. This contains:

  • user: the username, maybe the empty string
  • passwd: the password, maybe the empty string
  • sasl_authzid: the SASL authz value, maybe the empty string
  • oauth_bearer: the OAUTH bearer token, maybe the empty string
  • source: where the credentials from
  • refcount: a reference counter to link/unlink creds

A creds with all values empty is equivalent to NULL, e.g. no creds instance. With reference counting, creds can be linked in several places.

Two creds are the same if all values are equal apart from source and refcount. The comparison of strings is done via Curl_timestrcmp() to prevent side channel attacks.

creds locations

Credentials are kept in three places:

  • data->state.creds: the credentials to use for the transfer in talking to the origin (see PEERS)
  • conn->creds: the credentials tied to a connection (more below)
  • conn->*_proxy.creds: credentials used to talk to the conn->*_proxy.peer

data->state.creds

This creds instance is created when the transfer starts looking for a suitable connection. For an easy_perform() this may happen several times if, for example, http redirects are followed.

When an easy_perform() starts, the transfer's data->state.initial_origin peer is cleared. When creating the connection, conn->origin is calculated (e.g. who the request talks to). If data->state.initial_origin is not set, the first conn->origin is linked there. Now libcurl knows where the transfer initially talked to on all possible subsequent requests.

Credential information from CURLOPT_* settings is only applicable for the initial origin. Any followup request going to another origin must not use it. Therefore data->state.creds is only created from CURLOPT_* when current origin and initial origin match.

Without credentials from CURLOPT_*, the URL is inspected for user and password and netrc is consulted as well (when built in).

conn->creds

Once data->state.creds is known, the connection credentials are determined. For protocols that tie authorization to everything send on a connection (protocols without flag PROTOPT_CREDSPERREQUEST), conn->creds is linked to data->state.creds. Only connections carrying the same credentials may be reused.

Protocol with flag PROTOPT_CREDSPERREQUEST leave conn->creds empty, as connections for such protocols may be reused with different credentials.

That being said, there are authentication schemes like NTLM and NEGOTIATE that tie credentials to a connection. Those do set conn->creds once they start to operate, preventing connection reuse from then on for transfers with different credentials.

conn->*_proxy.creds

Those are set during connection setup from the CURLOPT_* values. They do not require any "initial origin" handling as the origin of a proxy does not change for a transfer.