Re: [PATCH 1/3] http: handle proxy authentication failure (error 407)
From: Jeff King <hidden>
Date: 2016-06-15 22:53:49
Subsystem:
the rest · Maintainer:
Linus Torvalds
On Fri, May 11, 2012 at 03:13:40PM +0200, Nelson Benitez Leon wrote:
quoted hunk ↗ jump to hunk
@@ -804,6 +819,15 @@ static int http_request(const char *url, void *result, int target, int options) init_curl_http_auth(slot->curl); ret = HTTP_REAUTH; } + } else if (results.http_code == 407) { /* Proxy authentication failure */ + if (proxy_auth.username && proxy_auth.password) { + credential_reject(&proxy_auth); + ret = HTTP_NOAUTH; + } else { + credential_fill(&proxy_auth); + set_proxy_auth(slot->curl); + ret = HTTP_REAUTH; + }
This part will fill in the username/password based on the proxy URL. But
we never set the proxy URL ahead of time, so there is no chance for
credential helpers to act, and the prompts will be confusing (they will
just say "Password" instead of "Password for ...", which will make it
unclear that we want the proxy password, not the remote server's
password).
So it's OK to drop the environment-parsing bits, but:
1. When we _do_ get the proxy via config, should we still parse it? I
could go either way. It's a nice feature, and I think we don't have
to care about how the environment parsing or NO_PROXY works. On the
other hand, we could just wait for the callback-based
authentication that will come in newer versions of curl, and code
to that, which will be even simpler. In the meantime, people can
just accept it.
2. When we don't know the proxy name beforehand, we should probably
say something to stderr to indicate that it was a proxy
authentication failure.
Also, what about the dumb http-push code-paths? They would need us to
handle http_proactive_auth in the same way. Which obviously won't work
for environment-based proxies, but could work for config-based proxies.
I'm not sure if it's worth caring about. In the long run, the
callback-based authentication is the way forward (though of course it
will take time for that feature to get released in curl, and then for
people to start having a curl that uses it, and so on).
If we just punt on (1) and the proactive auth thing, then I think as a
minimum we can get away with squashing this into your patch:
diff --git a/http.c b/http.c
index 0023119..86e68ee 100644
--- a/http.c
+++ b/http.c@@ -824,6 +824,7 @@ static int http_request(const char *url, void *result, int target, int options) credential_reject(&proxy_auth); ret = HTTP_NOAUTH; } else { + warning(_("http proxy did not accept our credentials, retrying")) credential_fill(&proxy_auth); set_proxy_auth(slot->curl); ret = HTTP_AUTH_RETRY; -Peff