While ld.so cannot lock mappings in the normal dlopen() case, we can apply
the same immutable protection as load-time shared libraries for dlopen'd
libraries mapped with RTLD_NODELETE.
With refactoring and comments by deraadt@, thanks!
Factor the code used by ucom(4) to forcibly detach processes from the tty
about to get removed upon detach to a ttydetach() routine, and make use of it
also when deleting wsdisplay screens, as well as com at pcmcia.
Bug report for wsdisplay by OpenAI.
Work by deraadt@, polishing by me.
rpki-client: and another time BGPsec Router Keys are special
This time the story is that proc_parser_cert() and proc_parser_root_cert()
hang the cert they return off a tree in the parser, but proc_parser_cert()
does not take ownership of a BRK. This is yet another wart coming from the
fact that you have no clue what a .cer is going to be.
Pointed out by Ties de Kock
ok claudio
Ignore invalid hostnames and domainnames.
Nothing in base looks at these but they still end up in the lease file
where unprepared scripts might pick them up.
Pointed out by Acts1631.
OK krw, kn
error if get_unused_fd_flags() is called without O_CLOEXEC set
Previously this was a KASSERT. landry saw it trigger on inteldrm
(Kaby Lake). Caller is unclear, likely DRM_IOCTL_PRIME_HANDLE_TO_FD.
rpki-client: adjust comment regarding TA tie breakers
Tie breaking is now in an RFC, so remove claim to the contrary and our
logic is now the logic of said RFC.
ok job
Initialize m_len after MGET() and properly enforce the MRU limit.
Previously, the uninitialized m_len could cause a KASSERT().
reported Acts1631, input mvs
ok mvs
Update RPKI Trust Anchor Tiebreaker reference
As luck would have it, on the very day this spec got published as RFC,
one of the RIRs helped demonstrate the usefulness of the concept in the
production environment.
https://www.rfc-editor.org/rfc/rfc10097.html with Theo Buehler & Ties de Kock
Rpki-client instances globally did not suffer from the below RPKI
certification service outage.
"""
RPKI repositories unavailable - Incident Report for RIPE NCC
Future Dated Trust Anchor Certificate temporarily in the RPKI Repository
The RPKI Trust Anchor (TA) Certificate is signed using an offline
machine that's disconnected from the Internet. As the clock may drift
on this machine while it's turned off, the normal procedure is to
[15 lines not shown]
Update RPKI Trust Anchor Tiebreaker reference
As luck would have it, on the very day this spec got published as RFC,
one of the RIRs helped demonstrate the usefulness of the concept in the
production environment. Rpki-client instances did not suffer from the
below certification service outage.
with tb@
"""
RPKI repositories unavailable - Incident Report for RIPE NCC
Future Dated Trust Anchor Certificate temporarily in the RPKI Repository
The RPKI Trust Anchor (TA) Certificate is signed using an offline
machine that's disconnected from the Internet. As the clock may drift
on this machine while it's turned off, the normal procedure is to
recalibrate it manually before signing a new certificate. During the
re-signing on 18 November 2026, the clock on the TA machine was set
[13 lines not shown]
rpki-client: reject CustomerASID 0 in CCR as well
Add a check to CCR's parse_aspa_payloads() analogous to the one added to
aspa_parse_econtent() in aspa.c 1.50.
ok job
Update to zlib 1.4
ok deraadt
The security fixes are already in tree, the remaining relevant fixes
for OpenBSD include these:
- Deliver all gzip data before reporting error
- Fix truncated stream detection in gzread()
- Fix pending skip handling in gzread()
- Fix loss of zero fill on repeated gzseek()'s when writing
- Fix inaccurate total_in after Z_NEED_DICT returned
- Detect and report bad deflateSetDictionary() calls
- Avoid undefined behaviors in C language usage
- Don't assume NULL is defined in compress.c and uncompr.c
- Clarify documentation
Bring back the special case where CRLs are not accounted for in the entity
count.
CRL are checked as part of the MFT and are therefor no real work items.
Especially there is a case where an MFT is invalid but the CRL was still
processed and a RTYPE_CRL entity was sent back to the parent. In that
case the count would be off since a failed MFT has no work items.
So excluding CRLs from work_items is the best solution here.
With and OK tb@
drm/amd/pm:restate all customized OD features on every table upload for smu 13.0.0/smu 14.0.2
From Kenneth Feng
64e8107a8c8b9dbe9d5dea9a249fc69d1503dd32 in linux-6.18.y/6.18.64
44d0d3031e47973ed65fd327669d5e2eaaf6c565 in mainline linux
drm/dp: keep MOT set between messages when no_zero_sized is set
From Finn Zeruhn
c34605fd0725dedbfcf3b941b32d1d97f7cb352a in linux-6.18.y/6.18.64
06bb2e5c71e9983519929a8482bea89da6d80ad8 in mainline linux
drm/amdgpu: don't drop the gang dependency fence twice
From Jesse Zhang
162879c7258267c094f541314fac4133fa222f2d in linux-6.18.y/6.18.64
6d1f901a83603a7d314e4a756b416b11d8dc7e35 in mainline linux