OpenSSL Valhalla Rampage

Tearing apart OpenSSL, one arcane VMS hack at a time. Like what OpenBSD is doing to OpenSSL? Donate here.

[opensslrampage]

LibreSSL now turns off SSL 3 by default as a fix for POODLE. Thanks to the awesome people that have made this possible. Remember to donate!

[opensslrampage]

Many thanks to all of the awesome hackers that have made this release possible. Again, if you like the work that OpenBSD is doing, please donate here

[opensslrampage]

Add ressl_config_clear_keys()

Provide a ressl config function that explicitly clears keys. Now that ressl config takes copies of the keys passed to it, the keys need to be explicitly cleared. While this can be done by calling the appropriate functions with a NULL pointer, it is simpler and more obvious to call one function that does this for you.

— jsing

[opensslrampage]

Add SSL_CTX_use_certificate_chain()

Add a new API function SSL_CTX_use_certificate_chain() that allows to read the PEM-encoded certificate chain from memory instead of a file. This idea is derived from an older implementation in relayd that was needed to use the function with a privep'ed process in a chroot. Now it is time to get it into LibreSSL to make the API more privsep- friendly and to make it available for other programs and the ressl library.

— reyk

[opensslrampage]

Don't overzealously free(3) what you did not allocate

X509v3_add_ext(): do not free stuff we did not allocate in the error path.

— miod

[opensslrampage]

Don't dereference when NULL, also don't leak memory

X509_TRUST_add(): check X509_TRUST_get0() return value before dereferencing it, for it may be NULL. Do not leak memory upon error.

— miod

[opensslrampage]

Don't avoid your own API for the sake of misguided cleverness

Someone (TM) thought it was smart to save memory by using malloc(1) and manual field fiddling to create an ASN1_INTEGER object, instead of using M_ASN1_INTEGER_new() which will allocate sizeof(long) bytes. That person had probably never looked into malloc(3) and never heard of allocation size rounding.

Thus, replace the obfuscated code with M_ASN1_INTEGER_new() followed by ASN1_INTEGER_set(), to achieve a similar result, without the need for /* version == 0 */ comments.

— miod

[opensslrampage]

Config interface should own its own memory

revamp the config interface to own memory. easier to use correctly without caller worrying about leaks or lifetimes.

— tedu

[opensslrampage]

Re-enable ghash assembly on hppa

Revert r1.5 and reenable assembler version of ghash now that it has been fixed.

— miod

[opensslrampage]

Disable more ghash hppa assembler

Doh, rev 1.4 had left out one routine with both 32-bit and 64-bit code, where the 64-bit code has to be disabled under OpenBSD/hppa.

— miod

[opensslrampage]

Disable assembly code for ghash on hppa

Disable assembler code for ghash on hppa, causes wrong computations in some cases and breaks TLS 1.2; crank libcrypto.so minor version out of safety and to be able to tell broken versions apart easily.

— miod

[opensslrampage]

Don't double NULL check ecdhp

There is not much point checking ecdhp is not NULL... twice.

— jsing

[opensslrampage]

Ensure that client actually wants the curve that the server has chosen

Check that the specified curve is one of the client preferences. Based on OpenSSL.

— jsing

[opensslrampage]

Only check mmap(2) return value for MAP_FAILED

Fix mmap() calls that check for a result other than MAP_FAILED.

— doug

[opensslrampage]

Another memory leak in an error-handling codepath

X509_STORE_new(): do not leak memory upon error. X509_STORE_get1_certs(), X509_STORE_get1_crls(): check the result of allocations.

[opensslrampage]

Don't leak memory in error-handling codepaths

X509_issuer_and_serial_hash(): do not leak memory if an error occurs during the first EVP block.

— miod

[opensslrampage]

Don't free what you didn't allocate

X509at_add1_attr(): do not free stuff we did not allocate in the error path.

— miod