BSD Commit Log Search

   fq_codel: fix CoDel drop scheduling

   Use drops - 1 as the codel_intervals index; the table starts at interval
   / sqrt(1) while cd->drops is one-based. Assert non-zero.

   Exclude the candidate packet from the non-starvation backlog check; the
   decision is about the backlog remaining after dequeue.

   Enter dropping when now equals cd->start, not only when it exceeds it.

   OK sashan@

   Add some notes about Pi 5 and the debug port, tweak suggestions about
   preferred storage, stop pointing at pftf UEFI files. Feedback kettenis@

   revert 'drm/amd/display: Fix backlight max_brightness to match exported range'

   jmc@ reports this 6.18.42 change caused a very dim display on boot and with
   display.brightness values of 97-100 on a Dell Inspiron 5505

   it looks like this patch will be reverted in linux stable branches
   https://gitlab.freedesktop.org/drm/amd/-/work_items/5562

   relayd: remove dead backup flag and token

   The idea behind it was good, but it was never fully and properly implemented.
   It's better to remove it before reimplementing the concept from scratch
   and doing it right. Reading the man page makes it a little easier.

   OK kirill@

   relayd: fix deferring connections log

   OK kirill@

   relayd: check imsg_compose_event return values

   relayd: check proc_compose return values

   Fatal on config, setup, and startup imsg send failures. Inconsistent
   state means we can't recover.

   Warn on per-request responses and sync messages. A single dropped
   message shouldn't kill the whole process.

   OK kirill@

   Protect the IPsec TDB replay counter with mutex.

   Put all tdb_rpl read, write and increment under tdb_mtx.  Although
   shared net lock and kernel lock should be enough for now, we want
   to move towards fine grained locking.  Mark tdb_rpl and tdb_seen
   as MP safe.
   The wrap-around check in ah_output() is not atomic with the increment
   code and would not work.  As tdb_rpl is 64 bit nowadays, it cannot
   warp.  Better remove the useless tdb_rpl == 0 check than making it
   MP safe.

   OK deraadt@

   drop casts now that quotactl() takes a void* argument

   also clean up 0 vs. NULL while here, suggested by tb@

   ok tb@ deraadt@

   sync

   change the fourth argument of quotactl() from char* to void*

   Depending on the quota control command, the fourth argument is a
   filename, a pointer to struct, or it is ignored.  This change brings
   us back in sync with FreeBSD and NetBSD, who already changed the
   type in the previous century.

   ok deraadt@ tb@

   uvm: avoid reading am_hashshift out of a small amap

   UVM_AMAP_BUCKET() violates memory correctness if UVM_AMAP_SMALL().
   It reads am_hashshift in the union's ami_normal arm at offset 80,
   past the end of an 80-byte amappl1 item.

   Found by KASAN once free pool items were poisoned end to end.

   ok kettenis@

   Fix three instances of a grammar error where "practise" (the British
   English verb) was used incorrectly in place of "practice" (the noun).
   Reported by Nathan <nluuirl at gmail.com>.

   sndiod: If a MIDI port is gone, disconnect its control clients

   Update libexpat to version 2.8.3.

   Relevant for OpenBSD are bug fixes #1297 #1300 #1286 #1305 #1306,
   other changes #1303.  Library bump is not necessary.  CVE-2026-72522

   OK deraadt@ tb@

   sndiod: Drop unused numbers of opt structures

   sndiod: Move the -s option checks near the command-line args parsing

   libsndio: On the wire, identify MIDI ports by names instead of indices

   This makes handling of all type of MIDI ports and audio devices
   roughtly the same, which is simpler. The old scheme is still accepted by
   the server for compatibility with old libraries.

   libsndio: Delete support of "snd/0.foo" device syntax

   Support for such syntax was dropped around 5 years ago, but the
   corresponding code was forgotten.

   sndiod: Associate a midithru structure to MIDI ports

   Allows "midi/X" and "midithru/X" ports to be handled by the same code,
   which is simpler than the current approach.

   sndiod: In the CTL_SW case, the third ctl_new arg must be NULL

   sndiod: Make midithru_ref() return 1 on success

   Currently midithru_ref() never fails, but this is what all other
   xxx_ref() functions do and may ease future changes.

   Fix various context switch fallout in lpt(4).

   Joint work with Johann Hoepfner. OK deraadt.

   Unpublish the segment before entering the potentially sleeping
   shm_deallocate_segment() in the sys_shmat() and sys_shmctl() paths. The
   remaining shm_delete_mapping() already has the right order.

   From Acts1631

   wycheproof: recognize Cobblestone-{128,256}

   Put back the whitelist approach previous to 1.18, because without a
   specific whitelist, arbitrary *TINY* requests can flow through here
   which lack the required structure header.  Issue found by gnezdo.
   ok gnezdo jmatthew

   SIOCGIFDATA (on most architectures) leaked 4 bytes of kernel stack.
   from Acts1631

   Reassemble IP fragments in correct routing domain.

   Fragment reassembly in pf(4) and inet(4) and inet6(4) did not respect
   the routing domain.  Fragments from different routing domains could
   be merged together.  Keep track of rdomain in pf and network stack.
   Convert routing table to routing domain where necessary.  Sort
   struct fields by size.

   report and diff from Acts1631; fixes by bluhm@; OK sashan@

   remove a stale .Sx internal link;
   similar to a patch sent in by <hi at pgray.dev>

   iked: remove pointless call to X509_get_subject_name()

   A deserialized certificate always has a subject, so this check doesn't
   do anything useful. This was added in r1.7 (2010) as a mostly cosmetic
   return check, and it has been useless since r1.30 (2014) when the subject
   variable became unused. A subsequent removal of subject in r1.66 (2020)
   made this look weird (why would you check subject but not issuer, SPKI,
   ...?), so garbage collect it.

   ok hshoexer