GnuTLS before 2017-02-20 has an out-of-bounds write caused by an integer overflow and heap-based buffer overflow related to the cdk_pkt_read function in opencdk/read-packet.c. This issue (which is a subset of the vendor's GNUTLS-SA-2017-3 report) is fixed in 3.5.10.
Double free vulnerability in the gnutls_x509_ext_import_proxy function in GnuTLS before 3.3.26 and 3.5.x before 3.5.8 allows remote attackers to have unspecified impact via crafted policy language information in an X.509 certificate with a Proxy Certificate Information extension.
The stream reading functions in lib/opencdk/read-packet.c in GnuTLS before 3.3.26 and 3.5.x before 3.5.8 allow remote attackers to cause a denial of service (out-of-memory error and crash) via a crafted OpenPGP certificate.
Stack-based buffer overflow in the cdk_pk_get_keyid function in lib/opencdk/pubkey.c in GnuTLS before 3.3.26 and 3.5.x before 3.5.8 allows remote attackers to have unspecified impact via a crafted OpenPGP certificate.
Multiple heap-based buffer overflows in the read_attribute function in GnuTLS before 3.3.26 and 3.5.x before 3.5.8 allow remote attackers to have unspecified impact via a crafted OpenPGP certificate.
Multiple unspecified vulnerabilities in the DER decoder in GNU Libtasn1 before 3.6, as used in GnuTLS, allow remote attackers to cause a denial of service (out-of-bounds read) via crafted ASN.1 data.
The asn1_get_bit_der function in GNU Libtasn1 before 3.6 does not properly report an error when a negative bit length is identified, which allows context-dependent attackers to cause out-of-bounds access via crafted ASN.1 data.
The (1) asn1_read_value_type and (2) asn1_read_value functions in GNU Libtasn1 before 3.6 allows context-dependent attackers to cause a denial of service (NULL pointer dereference and crash) via a NULL value in an ivalue argument.
Mutt 1.5.19, when linked against (1) OpenSSL (mutt_ssl.c) or (2) GnuTLS (mutt_ssl_gnutls.c), allows connections when only one TLS certificate in the chain is accepted instead of verifying the entire chain, which allows remote attackers to spoof trusted servers via a man-in-the-middle attack.