summaryrefslogtreecommitdiff
path: root/src/lib/libssl (follow)
Commit message (Collapse)AuthorAgeFilesLines
...
* Expose SSL_R_NO_APPLICATION_PROTOCOL in ssl.htb2021-09-101-3/+1
| | | | ok beck jsing
* Expose SSL_CTRL_GET_TLSEXT_STATUS_REQ_TYPE in ssl.htb2021-09-101-3/+1
| | | | ok beck jsing
* Expose SSL_CTX_get0_privatekey() in ssl.htb2021-09-101-3/+1
| | | | ok beck
* Remove TLS1_get_{,client_}version()tb2021-09-101-9/+1
| | | | ok jsing
* Remove SSL3_RECORD and SSL3_BUFFERtb2021-09-101-25/+1
| | | | with/ok jsing
* Remove TLS1_RT_HEARTBEATtb2021-09-101-2/+1
| | | | ok jsing
* Make SSL opaquetb2021-09-101-2/+4
| | | | with/ok jsing
* Remove struct tls_session_ticket_ext_st and TLS_SESSION_TICKET_EXTtb2021-09-102-2/+6
| | | | | | from public visibility. with/ok jsing
* Do not ignore SSL_TLSEXT_ERR_FATAL from the ALPN callbacktb2021-09-104-7/+31
| | | | | | | | | | | | | | | | | | | As reported by Jeremy Harris, we inherited a strange behavior from OpenSSL, in that we ignore the SSL_TLSEXT_ERR_FATAL return from the ALPN callback. RFC 7301, 3.2 states: 'In the event that the server supports no protocols that the client advertises, then the server SHALL respond with a fatal "no_application_protocol" alert.' Honor this requirement and succeed only on SSL_TLSEXT_ERR_{OK,NOACK} which is the current behavior of OpenSSL. The documentation change is taken from OpenSSL 1.1.1 as well. As pointed out by jsing, there is more to be fixed here: - ensure that the same protocol is selected on session resumption - should the callback be called even if no ALPN extension was sent? - ensure for TLSv1.2 and earlier that the SNI has already been processed ok beck jsing
* Prepare to provide SSL_CTX_get0_privatekey()tb2021-09-102-2/+14
| | | | ok beck
* Provide SSL_SESSION_is_resumable and SSL_set_psk_use_session_callback stubstb2021-09-083-3/+24
| | | | ok jsing
* Prepare to provide API stubs for PHAtb2021-09-082-2/+27
| | | | ok bcook jsing
* Prepare to provide SSL_get_tlsext_status_type()tb2021-09-083-3/+20
| | | | | | Needed for nginx-lua to build with opaque SSL. ok inoguchi jsing
* Prepare to provide SSL_set0_rbio()tb2021-09-082-2/+12
| | | | | | | This is needed for telephony/coturn and telephony/resiprocate to compile without opaque SSL. ok inoguchi jsing
* Factor out the TLSv1.3 code that handles content from TLS records.jsing2021-09-046-80/+238
| | | | | | | | | | | | Currently, the plaintext content from opened TLS records is handled via the rbuf code in the TLSv1.3 record layer. Factor this out and provide a separate struct tls_content, which knows how to track and manipulate the content. This makes the TLSv1.3 code cleaner, however it will also soon also be used to untangle parts of the legacy record layer. ok beck@ tb@
* Refactor ssl_update_cache. This now matches the logic used for TLS 1.3beck2021-09-041-22/+106
| | | | | | | in Openssl 1.1.1 for when to call the session callbacks. I believe it to also generates a lot less eye bleed, confirmed by tb@ ok jsing@ tb@
* Improve DTLS hello request handling code.jsing2021-09-041-2/+8
| | | | | | | Rather than manually checking multiple bytes, actually parse the DTLS handshake message header, then check the values against what we parsed. ok inoguchi@ tb@
* Change dtls1_get_message_header() to take a CBS.jsing2021-09-043-22/+21
| | | | | | The callers know the actual length and can initialise a CBS correctly. ok inoguchi@ tb@
* Improve DTLS record header parsing.jsing2021-09-041-7/+7
| | | | | | | | Rather than pulling out the epoch and then six bytes of sequence number, pull out SSL3_SEQUENCE_SIZE for the sequence number, then pull the epoch off the start of the sequence number. ok inoguchi@ tb@
* Use SSL3_HM_HEADER_LENGTH instead of the magic number 4.jsing2021-09-031-13/+14
| | | | ok beck@
* Ensure that a server hello does not have trailing data.jsing2021-09-031-1/+4
| | | | | | Found by tlsfuzzer. ok beck@
* Ensure that a client hello does not have trailing data.jsing2021-09-031-1/+4
| | | | | | Found by tlsfuzzer. ok beck@
* Set message_size correctly when switching to the legacy stack.jsing2021-09-031-2/+2
| | | | | | | | The message_size variable is not actually the handshake message size, rather the number of bytes contained within the handshake message, hence we have to subtract the length of the handshake message header. ok beck@
* We need to allow for either a CERTIFICATE or CERTIFICATE_STATUS messagebeck2021-09-021-2/+3
| | | | | | here or we break the handshake with BAD_MESSAGE ok tb@
* Call the ocsp callback if present and we get no response, instead ofbeck2021-09-021-3/+2
| | | | | | succeeding unconditionally. Makes muststaple work with tls1.3 in nc ok tb@
* Correct the is_server flag in the call to the debug callback to be correct.beck2021-09-021-2/+2
| | | | ok tb@
* RFC 6066 section 8 allows the server MAY choose not send the CertificateStatusbeck2021-09-021-3/+37
| | | | | | | | message, even if it has received a "status_request" extension in the client hello message and has sent a "status_request" extention in the server hello message. Genua found a site that is this broken. This makes it work. ok jsing@
* comment out the detailed description of SSL_get_servername(3),schwarze2021-09-011-7/+9
| | | | | | leaving only the basic description in the RETURN VALUES section; tb@ pointed out LibreSSL does not currently provide all those guarantees, and he also OK'ed this diff
* Defragment DTLS.jsing2021-08-312-123/+48
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | In normal TLS, it is possible for record fragments to be sent that contain one byte of alert or handshake message payload. In this case we have to read and collate multiple message fragments before we can decide what to do with the record. However, in the case of DTLS, one record is effectively one packet and while it is possible to send handshake messages across multiple records/packets, the minimum payload is the DTLS handshake message header (plus one byte of data if the handshake message has a payload) - without this, there is insufficient information available to be able to reassemble the handshake message. Likewise, splitting an alert across multiple DTLS records simply does not work, as we have no way of knowing if we're collating the same alert or two different alerts that we lost half of each from (unfortunately, these details are not really specified in the DTLS RFC). This means that for DTLS we can expect to receive a full alert message (a whole two bytes) or a handshake record with at least the handshake message header (12 bytes). If we receive messages with less than these lengths we discard them and carry on (which is what the DTLS code already does). Remove all of the pointless fragment handling code from DTLS, while also fixing an issue where one case used rr->data instead of the handshake fragment. ok inoguchi@ tb@
* Remove a nonsensical s->version == TLS1_VERSION from DTLS code.jsing2021-08-311-6/+1
| | | | ok inoguchi@ tb@ (as part of a larger diff)
* Clean up and simplify info and msg callbacks.jsing2021-08-308-127/+88
| | | | | | | | | The info and msg callbacks result in duplication - both for code that refers to the function pointers and for the call sites. Avoid this by providing typedefs for the function pointers and pulling the calling sequences into their own functions. ok inoguchi@ tb@
* Replace DTLS r_epoch with the read epoch from the TLSv1.2 record layer.jsing2021-08-305-27/+26
| | | | ok inoguchi@ tb@
* Move to an AEAD nonce allocated in the TLSv1.2 record layer.jsing2021-08-301-45/+36
| | | | | | | | There is little to gain by mallocing and freeing the AEAD nonce for each record - move to an AEAD nonce allocated for the record layer, which matches what we do for TLSv1.3. ok inoguchi@ tb@
* sync with OpenSSL 1.1.1, which is still under a free license;schwarze2021-08-301-15/+107
| | | | | | | in particular, this includes new text by Matt Caswell from OpenSSL commit 721eb8f6 Nov 28 12:03:00 2019 +0000 and corrects a wrong argument type that i introduced into the SYNOPSIS; requested by tb@
* Ignore warning alert returns from servername callback in TLSv1.3tb2021-08-301-3/+7
| | | | | | | | | | | | | | | | If a servername callback returns SSL_TLSEXT_ERR_ALERT_WARNING, this results in a fatal error in TLSv1.3 since alert levels are implicit in the alert type and neither close_notify nor user_canceled make sense in this context. OpenSSL chose to ignore this, so we need to follow suit. Found via a broken servername callback in p5-IO-Socket-SSL which returns a Boolean instead of SSL_TLSEXT_ERR_*. This happened to have worked before TLSv1.3 since warning alerts are often ignored. This "fixes" sni.t and sni-verify.t in p5-IO-Socket-SSL. ok beck jsing
* Clean up and simplify ssl3_dispatch_alert() and ssl3_send_alert().jsing2021-08-281-30/+32
| | | | ok inoguchi@ tb@
* SSL_CTX_remove_session() checks for a NULL session, avoid doing it twice.jsing2021-08-041-2/+2
| | | | Noted by tb@ during review of a larger change.
* We have defines for alert levels - use them instead of magic numbers.jsing2021-07-312-7/+5
|
* Dedup dtls1_dispatch_alert()/ssl3_dispatch_alert().jsing2021-07-266-65/+26
| | | | | | | | The code for dtls1_dispatch_alert() and ssl3_dispatch_alert() is largely identical - with a bit of reshuffling we can use ssl3_dispatch_alert() for both protocols and remove the ssl_dispatch_alert function pointer. ok inoguchi@ tb@
* Remove DTLS processed_rcds queue.jsing2021-07-213-50/+22
| | | | | | | | | | | | | | | | | | | When DTLS handshake records are received from the next epoch, we will potentially queue them on the unprocessed_rcds queue - this is usually a Finished message that has been received without the ChangeCipherSuite (CCS) message (which may have been dropped or reordered). After the epoch increments (due to the CCS being received), the current code processes all records on the unprocessed queue and immediate queues them on the processed queue, which dtls1_get_record() then pulls from. This form of processing only adds more complexity and another queue. Instead, once the epoch increments, pull a single record from the unprocessed queue and process it, allowing the contents to be consumed by the caller. We repeat this process until the unprocessed queue is empty, at which point we go back to consuming messages from the wire. ok inoguchi@ tb@
* Silently discard invalid DTLS records.jsing2021-07-211-4/+11
| | | | | | | | | | | Per RFC 6347 section 4.1.2.1, DTLS should silently discard invalid records, including those that have a bad MAC. When converting to the new record layer, we inadvertantly switched to standard TLS behaviour, where an invalid record is fatal. This restores the previous behaviour. Issue noted by inoguchi@ ok inoguchi@
* Mop up dtls1_get_ccs_header() and struct ccs_header_st.jsing2021-07-193-21/+3
| | | | | | | | All this code does is read one byte from memory with an unknown length, potentially being a one byte overread... and then nothing is actually done with the value. ok tb@
* Inline DTLS1_CCS_HEADER_LENGTH rather than having a single use variable.jsing2021-07-191-3/+2
| | | | ok tb@
* Do a first pass clean up of SSL_METHOD.jsing2021-07-034-103/+14
| | | | | | | | | The num_ciphers, get_cipher_by_char and put_cipher_by_char function pointers use the same function for all methods - call ssl3_num_ciphers() directly, absorb ssl3_get_cipher_by_char() into SSL_CIPHER_find() and remove the unused ssl3_put_cipher_by_char() code. ok inoguchi@ tb@
* Merge SSL_METHOD_INTERNAL into SSL_METHOD.jsing2021-07-0115-195/+126
| | | | | | | Now that SSL_METHOD is opaque and in internal headers, we can remove SSL_METHOD_INTERNAL by merging it back into SSL_METHOD. ok tb@
* Prepare to provide SSL_get_signature_nid() and friends.jsing2021-06-302-2/+81
| | | | | | | | | | | This adds functionality for SSL_get_signature_nid(), SSL_get_peer_signature_nid(), SSL_get_signature_type_nid() and SSL_get_peer_signature_type_nid(). This is not currently publicly visible and will be exposed at a later date. ok inoguchi@ tb@
* Move some structs from public to private headers.jsing2021-06-303-125/+108
| | | | | | | | Move struct ssl_cipher_st, struct ssl_method_st, struct ssl_session_st and struct ssl3_state_st from public to private headers. These are already under #ifdef LIBRESSL_INTERNAL and are no longer publicly visible. ok inoguchi@ tb@
* Correct sigalg hash usage when signing content for client verify.jsing2021-06-301-3/+2
| | | | This was inadvertently broken during sigalgs refactoring.
* Pull sigalg selection up into ssl3_send_client_verify().jsing2021-06-291-14/+11
| | | | | | | | This means that we do sigalg selection for all cases, including those where are are not sending sigalgs. This is needed in order to track our signature type in legacy cases. ok tb@
* Track sigalg used by ourselves and our peer in the legacy stack.jsing2021-06-292-2/+6
| | | | This is needed for upcoming API additions.
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-03-31 14:45:39 +0000