fts_read: fix potential double free on FCHDIR() failure
Part of the contract for fts_read() is that we must always reset
sp->fts_cur immediately before return. There was one instance where
a failed FCHDIR() was not setting sp->fts_cur to 'p' before return,
leaving sp->fts_cur pointing to a freed value.
From Acts1631
OK tb@
SEV-SNP: Disable ipmi(4) on confidential VMs
The CD_COCOVM filter already keeps ipmi(4) from attaching on a
confidential VM, but it runs too late: ipmi_probe() is called directly
from mainbus_attach() and scans the BIOS ROM window at 0xc0000-0xfffff
for an IPMI signature before config_found() is reached. Skip the probe
as well.
ok mlarkin@
disable squid's use of nettle, it has a fallback internal copy of an old
version of nettle's code for the things it uses it for (base64 and md5).
the base64 API was changed in nettle 4.0 breaking squid's use of it (e.g.
decoding HTTP headers used for proxy authentication).
big thanks to Fabrice LEVIEUX for tracking it down.
Add ML-DSA 44 support with tests
This adds ML-DSA 44 support based on BoringSSL's changes to
add support for the 44 rank.
It also adds the NIST ACVP test vectors and unit tests for
MLDSA 44 to the existing regress tests.
ok tb@
SEV-SNP: Add memory clobber to vmgexit()
The hypervisor reads and writes the GHCB while handling a VMGEXIT.
Without a memory clobber the compiler may cache GHCB contents across
the instruction. "asm volatile" alone does not prevent that.
No change in generated code.
ok mlarkin@
Add ML-DSA known answer tests against the NIST ACVP vectors
This uses a file-test driver that runs the NIST keygen
and siggen vectors for MLDSA-65 and MLDSA-87. It re-uses
the mlkem utilities written by tb@ from the mlkem test
harness, calling the internal deteministic functions
with known entropy values so the fixed vectors are reproducible.
parse_test_file for nist should probably move to the
generic test harness we use, but that can happen later.
ok tb@ kenjiro@
Convert the BoringSSL c++ code to C
This mirrors the approach taken with ML-KEM to encode the rank
in the key, instead of BoringSSL's templated C++ approach.
This is still not hooked up to build, that will follow
ok tb@ kenjiro@
Avoid use-after-free in IPv6 duplicate address detection.
An expired DAD timer waiting for netlock could find the dadq of a
new DAD for the same address and destroy it while its timer was
still pending. Pass struct dadq to the timer and check under
exclusive netlock that it is still the active one.
from OpenAI; OK deraadt@ mvs@
