In defense against attack, assume the imported public key is not trusted
and check it matches the private key if set.
Added APIs that allow application to explicitly trust public key.
Original APIs default to not trusting public key.
Add a define WOLFSSL_CHECK_MEM_ZERO to turn on code that checks that
memory that must be zeroized before going out of use is zero.
Everytime sensitive data is put into a allocated buffer or stack buffer;
the address, its length and a name is stored to be checked later.
Where the stack buffer is about to go out of use, a call is added to
check that the required parts are zero.
wc_MemZero_Add() adds an address with length and name to a table of
addressed to be checked later.
wc_MemZero_Check() checks that the memory associated with the address is
zeroized where required.
mp_memzero_add() adds mp_int's data pointer with length and name to
table.
mp_memzero_check() checks that the data pointer is zeroized where
required.
Freeing memory will check the address. The length was prepended on
allocation.
Realloction was changed for WOLFSSL_CHECK_MEM_ZERO to perform an
allocate, check, copy, free.
as word64 is not always available, introduce an abstract type and companion
operations. They use a word64 if available and fallback on word32[2] otherwise.
With PR 5170, I added logic that requires a EVP_CTRL_GCM_SET_IV_FIXED command be
issued before a EVP_CTRL_GCM_IV_GEN command. This matches OpenSSL's behavior.
However, OpenSSL also clears the flag enabling EVP_CTRL_GCM_IV_GEN after
EVP_CTRL_GCM_SET_IV_FIXED if EVP_CipherInit is called with a NULL key.
Otherwise, the flag retains its value. We didn't mirror this logic, and that
caused problems in OpenSSH unit testing. This commit aligns our logic with
OpenSSL's and adds a regression test to test_evp_cipher_aes_gcm for this case.
* PKCS7 should use allocated buffer for RSA.
* PKCS7 small stack typo for `keyAlgArray` size in `wc_PKCS7_AddRecipient_KTRI`.
* Fix for use of `free`, which should be `XFREE` in api.c.
* Cleanup old RSA benchmarking MDK5/WINCE code no longer needed with `WC_DECLARE_ARRAY_DYNAMIC_DEC` and `WC_DECLARE_ARRAY_DYNAMIC_EXE`.
- Fix case where message grouping can make CheckAvailableSize return a WANT_WRITE
- CheckAvailableSize in tls13.c will not return a WANT_WRITE since it only does so for DTLS <=1.2
Testing:
./autogen.sh
./configure --enable-psk
make all check
$ ./examples/server/server -j -l ECDHE-PSK-AES128-GCM-SHA256
SSL version is TLSv1.2
SSL cipher suite is TLS_ECDHE_PSK_WITH_AES_128_GCM_SHA256
SSL curve name is SECP256R1
Client message: hello wolfssl!
$ ./examples/client/client -s -l ECDHE-PSK-AES128-GCM-SHA256
SSL version is TLSv1.2
SSL cipher suite is TLS_ECDHE_PSK_WITH_AES_128_GCM_SHA256
SSL curve name is SECP256R1
I hear you fa shizzle!
Discovered the AES-GCM flow using this command didn't work in our OpenSSH port.
This commit makes the behavior match OpenSSL and adds testing using known
OpenSSL-generated test vectors to prevent regressions. This was one of those
problems where two ends of a connection would work fine if they were both using
wolfSSL but not if one was using OpenSSL (i.e. OpenSSH interop with AES-GCM
was broken).
Prior to this commit, if you wanted access to the Finished messages from a
handshake, you needed to turn on the compatibility layer, via one of
OPENSSL_ALL, WOLFSSL_HAPROXY, or WOLFSSL_WPAS. With this commit, defining any
of these causes WOLFSSL_HAVE_TLS_UNIQUE to be defined (a reference to the
tls-unique channel binding which these messages are used for) in settings.h.
This allows a user to define WOLFSSL_HAVE_TLS_UNIQUE to access the Finished
messages without bringing in the whole compat layer.
