Add support for BoringSSL QUIC APIs This is a cherry-pick of 2a4b03a306439307e0b822b17eda3bdabddfbb68 on the master-quic-support2 branch (2019-10-07) Which was a rebase/squash of master-quic-support: * 5aa62ce Add support for more secrets - Todd Short/Todd Short (master-quic-support) * 58e0643 Tweeks to quic_change_cipher_state() - Todd Short/Todd Short * 8169702 Move QUIC code out of tls13_change_cipher_state() - Todd Short/Todd Short * a08cfe6 Correctly disable middlebox compat - Todd Short/Todd Short * 3a9eabf Add OPENSSL_NO_QUIC wrapper - Todd Short/Todd Short * f550eca Add client early traffic secret storage - Todd Short/Todd Short * 1b787ae Quick fix: s2c to c2s for early secret - Todd Short/Todd Short * f97e6a9 Don't process an incomplete message - Todd Short/Todd Short * 81f0ce2 Reset init state in SSL_process_quic_post_handshake() - Todd Short/Todd Short * 5d59cf9 Fix quic_transport constructors/parsers - Todd Short/Todd Short * 5e5f91c Fix INSTALL nit. - Todd Short/Todd Short * bd290ab Fix duplicate word in docs - Todd Short/Todd Short * 699590b fixup! Handle partial handshake messages - Todd Short/Todd Short * a472a8d Handle partial handshake messages - Todd Short/Todd Short * 363cf3d fixup! Use proper secrets for handshake - Todd Short/Todd Short * b03fee6 Use proper secrets for handshake - Todd Short/Todd Short * 2ab1aa0 Move QUIC transport params to encrypted extensions - Todd Short/Todd Short * 0d16af9 Make temp secret names less confusing - Todd Short/Todd Short * abb6f39 New method to get QUIC secret length - Todd Short/Todd Short * 05fdae9 Add support for BoringSSL QUIC APIs - Todd Short/Todd Short This adds a compatible API for BoringSSL's QUIC support, based on the current |draft-ietf-quic-tls|. Based on BoringSSL commit 3c034b2cf386b3131f75520705491871a2e0cafe Based on BoringSSL commit c8e0f90f83b9ec38ea833deb86b5a41360b62b6a Based on BoringSSL commit 3cbb0299a28a8bd0136257251a78b91a96c5eec8 Based on BoringSSL commit cc9d935256539af2d3b7f831abf57c0d685ffd81 Based on BoringSSL commit e6eef1ca16a022e476bbaedffef044597cfc8f4b Based on BoringSSL commit 6f733791148cf8a076bf0e95498235aadbe5926d Based on BoringSSL commit 384d0eaf1930af1ebc47eda751f0c78dfcba1c03 Based on BoringSSL commit a0373182eb5cc7b81d49f434596b473c7801c942 Based on BoringSSL commit b1b76aee3cb43ce11889403c5334283d951ebd37 New method to get QUIC secret length Make temp secret names less confusing Move QUIC transport params to encrypted extensions Use proper secrets for handshake fixup! Use proper secrets for handshake Handle partial handshake messages fixup! Handle partial handshake messages Fix duplicate word in docs Fix INSTALL nit. Fix quic_transport constructors/parsers Reset init state in SSL_process_quic_post_handshake() Don't process an incomplete message Quick fix: s2c to c2s for early secret Add client early traffic secret storage Add OPENSSL_NO_QUIC wrapper Correctly disable middlebox compat Move QUIC code out of tls13_change_cipher_state() Create quic_change_cipher_state() that does the minimal required to generate the QUIC secrets. (e.g. encryption contexts are not initialized). Tweeks to quic_change_cipher_state() Add support for more secrets Fix resumption secret (cherry picked from commit 16fafdf4e0ec6cddd5705f407e5dca26cb30914d) QUIC: Handle EndOfEarlyData and MaxEarlyData QUIC: Increase HKDF_MAXBUF to 2048 Fall-through for 0RTT Some cleanup for the main QUIC changes Try to reduce unneeded whitespace changes and wrap new code to 80 columns. Reword documentation to attempt to improve clarity. Add some more sanity checks and clarifying comments to the code. Update referenced I-D versions. Prevent KeyUpdate for QUIC QUIC does not use the TLS KeyUpdate message/mechanism, and indeed it is an error to generate or receive such a message. Add the necessary checks (noting that the check for receipt should be redundant since SSL_provide_quic_data() is the only way to provide input to the TLS layer for a QUIC connection). Test KeyUpdate rejection For now, just test that we don't generate any, since we don't really expose the mechanics for encrypting one and the QUIC API is not integrated into the TLSProxy setup. Fix out-of-bounds read when TLS msg is split up into multiple chunks Previously, SSL_provide_quic_data tried to handle this kind of situation, but it failed when the length of input data is less than SSL3_HM_HEADER_LENGTH. If that happens, the code might get wrong message length by reading value from out-of-bounds region. Revert "Fix out-of-bounds read when TLS msg is split up into multiple chunks" This reverts commit 18f993cbdae498111c94a075fd9b115bd8367574. Test HKDF with empty IKM Add an extra EVP test that provides empty input key material. It currently fails, since attempting to set a zero-length key on an EVP_PKEY_CTX results in a call to OPENSSL_memdup() with length zero, which returns NULL and is detected as failure. Allow zero-length HKDF keys When making a copy to keep in the EVP_PKEY_CTX, allocate a single byte for the cached key instead of letting memdup return NULL and cause the call to fail. The length still gets set to zero properly, so we don't end up inspecting the allocated byte, but it's important to have a non-NULL pointer set. Buffer all provided quic data Make all data supplied via SSL_provide_quic_data() pass through an internal buffer, so that we can handle data supplied with arbitrary framing and only parse complete TLS records onto the list of QUIC_DATA managed by quic_input_data_head/quic_input_data_tail. This lets us remove the concept of "incomplete" QUIC_DATA structures, and the 'offset' field needed to support them. However, we've already moved the provided data onto the buffer by the time we can check for KeyUpdate messages, so defer that check to quic_get_message() (where it is adjacent to the preexisting ChangeCipherSpec check). To avoid extra memory copies, we also make the QUIC_DATA structures just store offsets into the consolidated buffer instead of having copies of the TLS handshake messages themselves. enforce consistent encryption level for handshake messages The QUIC-TLS spec requires that TLS handshake messages do not cross encryption level boundaries, but we were not previously enforcing this.

Commits

Commits on Dec 15, 2020

deps: update patch and docs for openssl update

deps: upgrade openssl sources to 1.1.1i

deps: various quic patches from akamai/openssl

deps: re-enable OPENSSL_NO_QUIC guards

deps: update archs files for OpenSSL-1.1.1i