Re: [RFC/PATCH] receive-pack.c: only accept push-cert if push_cert_nonce was advertised

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 9, 2015, 22:39 UTC
Message-ID: <xmqqsifjbmu2.fsf@gitster.dls.corp.google.com>
In-Reply-To: <1420836437-11068-1-git-send-email-sbeller@google.com>

Stefan Beller <sbeller@google.com> writes:

Show 17 quoted lines

> If the server did not advertise the capability to have signed pushes
> it should not accept signed pushes as stated in
> Documentation/technical/protocol-capabilities.txt:
>
>     Client will then send a space separated list of capabilities it wants
>     to be in effect. The client MUST NOT ask for capabilities the server
>     did not say it supports.
>
>     Server MUST diagnose and abort if capabilities it does not understand
>     was sent.  Server MUST NOT ignore capabilities that client requested
>     and server advertised.  As a consequence of these rules, server MUST
>     NOT advertise capabilities it does not understand.
>
> After rereading the second paragraph I think they should also be reworded to
>
>     Server MUST diagnose and abort if capabilities it did not advertise
>     was sent.

Except for s/was sent/was requested/, I think that rule makes sense very much.

Show 13 quoted lines

> diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
> index 4c069c5..628d13a 100644
> --- a/builtin/receive-pack.c
> +++ b/builtin/receive-pack.c
> @@ -1276,7 +1276,8 @@ static struct command *read_head_info(struct sha1_array *shallow)
>  				use_atomic = 1;
>  		}
>  
> -		if (!strcmp(line, "push-cert")) {
> +		if (push_cert_nonce &&
> +		    !strcmp(line, "push-cert")) {
>  			int true_flush = 0;
>  			char certbuf[1024];

This implementation is somewhat questionable.

The server knows how to parse "push-cert" line, knows that what follows after that line up to "push-cert-end" line are shaped very differently from protocol commands outside the push-cert block. In other words, it knows how to parse the request meant for the capable server; it just wants to refuse to serve that request.

The patched code will make it fail by hoping that queue_command() that only handles "40-hex 40-hex ref" will reject the line that begins with "push-cert". Instead of relying on such a hidden dependency, wouldn't it be cleaner to actually parse the push-cert block and then at the end notice and explictly say "Your requests were syntactically correct, but I am not going to honor your request to use the push-cert extension, because I never told you that I'd offer you that capability", instead of rejecting the request with "I was expecting old/new/ref but you sent a line with 'push-cert' on it; what are you talking about?"