libcurl contains a heap buffer overflow in the function (tftp_receive_packet()) that receives data from a TFTP server. It can call recvfrom() with the default size for the buffer rather than with the size that was used to allocate it. Thus, the content that might overwrite the heap memory is controlled by the server. This flaw is only triggered if the TFTP server sends an OACK without the BLKSIZE option, when a BLKSIZE smaller than 512 bytes was requested by the TFTP client. OACK is a TFTP extension and is not used by all TFTP servers.
libcurl can be told to use kerberos over FTP to a server, as set with the CURLOPT_KRBLEVEL option. During such kerberos FTP data transfer, the server sends data to curl in blocks with the 32 bit size of each block first and then that amount of data immediately following. A malicious or just broken server can claim to send a very large block and if by doing that it makes curl's subsequent call to realloc() to fail, curl would then misbehave in the exit path and double-free the memory. In practical terms, an up to 4 GB memory area may very well be fine to allocate on a modern 64 bit system but on 32 bit systems it will fail.
In situations where an attacker receives automated notification of the success or failure of a decryption attempt an attacker, after sending a very large number of messages to be decrypted, can recover a CMS/PKCS7 transported encryption key or decrypt any RSA encrypted message that was encrypted with the public RSA key, using a Bleichenbacher padding oracle attack. Applications are not affected if they use a certificate together with the private RSA key to the CMS_decrypt or PKCS7_decrypt functions to select the correct recipient info to decrypt.
OpenSSL 1.1.1 introduced a rewritten random number generator (RNG). This was intended to include protection in the event of a fork() system call in order to ensure that the parent and child processes did not share the same RNG state. However this protection was not being used in the default case. A partial mitigation for this issue is that the output from a high precision timer is mixed into the RNG state so the likelihood of a parent and child process sharing state is significantly reduced. If an application already calls OPENSSL_init_crypto() explicitly using OPENSSL_INIT_ATFORK then this problem does not occur at all.
Normally in OpenSSL EC groups always have a co-factor present and this is used in side channel resistant code paths. However, in some cases, it is possible to construct a group using explicit parameters (instead of using a named curve). In those cases it is possible that such a group does not have the cofactor present. This can occur even where all the parameters match a known named curve. If such a curve is used then OpenSSL falls back to non-side channel resistant code paths which may result in full key recovery during an ECDSA signature operation. In order to be vulnerable an attacker would have to have the ability to time the creation of a large number of signatures where explicit parameters with no co-factor present are in use by an application using libcrypto.