Date: Thu, 9 Dec 2021 14:10:39 +0100 (CET) From: Frank Lichtenheld Message-ID: <1692147884.739294.1639055439893@...2642...> In-Reply-To: <20211207170211.3275837-8-arne@...1227...> References: <20211207170211.3275837-1-arne@...1227...> <20211207170211.3275837-8-arne@...1227...> MIME-Version: 1.0 Subject: Re: [Openvpn-devel] [PATCH 07/21] Add helper functions to calculate header/payload sizes List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit To: Arne Schwabe , openvpn-devel@lists.sourceforge.net > Arne Schwabe hat am 07.12.2021 18:01 geschrieben: [...] > diff --git a/src/openvpn/mtu.c b/src/openvpn/mtu.c > index 0ab716d7a..25b943722 100644 > --- a/src/openvpn/mtu.c > +++ b/src/openvpn/mtu.c [...] > @@ -51,6 +52,85 @@ alloc_buf_sock_tun(struct buffer *buf, > ASSERT(buf_safe(buf, 0)); > } > > +size_t > +frame_calculate_protocol_header_size(const struct key_type *kt, > + const struct options *options, > + unsigned int payload_size, > + bool occ) > +{ > + /* Sum of all the overhead that reduces the usable packet size */ > + size_t header_size = 0; > + > + /* A socks proxy adds 10 byte of extra header to each packet */ This is only true since we do not support IPv6 UDP over socks. Would it be worth to point that out? > + if (options->ce.socks_proxy_server && proto_is_udp(options->ce.proto)) > + { > + header_size += 10; > + } > + > + /* TCP stream based packets have a 16 bit length field */ > + if (proto_is_tcp(options->ce.proto)) > + { > + header_size += 2; > + } > + > + /* Add the opcode and peerid */ > + header_size += options->use_peer_id ? 4 : 1; > + > + /* Add the crypto overhead */ > + bool packet_id = options->replay; > + bool packet_id_long_form = !tlsmode || cipher_kt_mode_ofb_cfb(kt->cipher); > + > + /* For figuring out the crypto overhead, we need to use the real payload > + * including all extra headers that also get encrypted */ That comment sounds like header size should somehow be a input into calculate_crypto_overhead, which it isn't. Is there an error here or do I just misunderstand the comment? > + header_size += calculate_crypto_overhead(kt, packet_id, > + packet_id_long_form, > + payload_size, occ); > + return header_size; > +} > + > + > +size_t > +frame_calculate_payload_overhead(const struct frame *frame, > + const struct options *options, > + bool extra_tun) > +{ > + size_t overhead = 0; > + > + /* This is the overhead of tap device that is not included in the MTU itself > + * i.e. Ethernet header that we still need to transmit as part of the > + * payload*/ > + if (extra_tun) > + { > + overhead += frame->extra_tun; > + } > + > +#if defined(USE_COMP) > + /* v1 Compression schemes add 1 byte header. V2 only adds a header when it > + * does not increase the packet length. We ignore the unlikely escaping > + * for tap here */ > + if (options->comp.alg == COMP_ALG_LZ4 || options->comp.alg == COMP_ALG_STUB > + || options->comp.alg == COMP_ALG_LZO) > + { > + overhead += 1; > + } > +#endif > +#if defined(ENABLE_FRAGMENT) > + if (options->ce.fragment) > + { > + overhead += 4; Every other statement has an explanation of the number, except this one. Maybe you could add one here as well? > + } > +#endif > + return overhead; > +} > + > +size_t > +frame_calculate_payload_size(const struct frame *frame, const struct options *options) > +{ > + size_t payload_size = options->ce.tun_mtu; > + payload_size += frame_calculate_payload_overhead(frame, options, true); > + return payload_size; > +} > + > void > frame_finalize(struct frame *frame, > bool link_mtu_defined, Regards, Frank -- Frank Lichtenheld