sys/qwz: fix REO queue lifetime
Track REO completions before publication and wait for peer unmap,
deletion, and cache flushes before reusing queue DMA. Submission errors
and timeouts retain ownership; hardware failures block reuse until cold
cleanup. HAL error conventions and flush semantics follow ath12k;
tracking and the reuse barrier adapt qwz's retained pool.
sys/qwz: fix WCN7850 REO layout
Use WCN7850 REO tags and 64-bit TLV headers so commands and completions
use the correct offsets. Correct the status ring size and clear command
payloads before reuse.
The layout follows Linux ath12k's WCN7850 definitions.
OK: stsp@
Update to 2026egtz from https://github.com/JodaOrg/global-tz
o Manitoba moves to permanent -05 on 2026-10-31.
o In 1925 Ireland fell back on 09-20 not 10-04.
rpki-client: do not fatal after RB_INSERT() into the NCA trees
rpki-client is generally a bit too quick to error out and a repeated source
of problems has been errx after RB_INSERT() (one fixed just yesterday).
The first of these is probably not reachable but do that for good measure.
The other one was shown to be reachable in somewhat contrived setups by
eur1ka, which means rpki-client would refuse to start.
ok claudio
Track the nofetch variable by TAL.
This matches better with the MAX_REPO_PER_TAL limit which is already tracked
by TAL and with that a TAL hitting the limit will not affect the other TALs.
Reported by eur1ka
OK tb@
sys/qwz: preserve decoded radiotap frequency
Management RX parameters already contain a hostorder chanel
frequency. Avoid decoding it again before writing the little endian
radiotap field.
OK: stsp@
sys/qwz: report radiotap channels and rates
Based on sys/dev/ic/qwx.c,v 1.93 and sys/dev/ic/qwxvar.h,v 1.31
Populate radiotap channel and rate fields with WCN7850 RX rate decoding.
Use QWZ presence masks and omit unavailable timestamps, noise and
signal strength for data frames.
Correct 54 Mb/s encoding from 104 to 108 in 500 kb/s.
OK: stsp@
sys/qwz: drain REO RX exceptions
Based on sys/dev/ic/qwx.c,v 1.33
Drain REO RX exceptions using descriptor layouts and qwz cookie.
Reclaim packet buffers, return link descriptors and replenish RX,
checking bank bounds and release ring space.
OK: stsp@
sys/qwz: handle WBM RX errors
Based on sys/dev/ic/qwx.c,v 1.35 and sys/dev/ic/qwxvar.h,v 1.18 ,
sys/dev/ic/qwx.c,v 1.89 , sys/dev/ic/qwx.c,v 1.121 and sys/dev/ic/qwxvar.h,v 1.36
Process WBM RX releases using WCN7850 descriptor and cookie formats.
Deliver valid null queue frames through existing RX processing, clear
mbuf pointers after delivery, and then replenish descriptors
OK: stsp@
sys/qwz: report hardware RX aggregation
Based on sys/dev/ic/qwx.c,v 1.85 and sys/dev/ic/qwx.c,v 1.90
Report hardware deaggregation and reordering after successful RX
reconstruction; allow repaeted sequence numbers for later A-MSDU
subframes and clear the AMSDU QoS bit.
OK: stsp@
start process of deprecating the -R flag. This was the old way of
performing a remote-to-remote copy that was basically executed scp on
the remote host. It barely worked (needing agent forwarding enabled or
usable credentials on the remote host) and has largely been replaced
by a better SFTP-protocol remote-to-remote copy that runs through the
host performing the copy.
We'll disable this option in a release or two; ok dtucker@
Implement a maximum number of KDF rounds that will be accepted when
writing an OpenSSH-format private key or when loading one. This limit
is set pretty high (1<<20), but ensures that a service that is passed a
bad key with an ridiculously high number of rounds will _eventually_
complete parsing it.
Also bump the default number of KDF rounds from 24 to 32 (this is a
linear increase, not like bcrypt(3) which is exponential).
Pointed out by Aris Adamantiadis
fix the bit length of ML-DSA 44/Ed25519 keys that was being
incorrectly reported as 256. The private key length for these
composite keys is 512 bits. This value is only used for display.
Spotted by Yiyue Wang
sftp: be stricter in accepting paths returned by the server for
SSH_FXP_REALPATH or SSH2_FXP_READDIR replies, as these can be
used in some situations to decide the destination path for recursive
transfers.
Report and patch from Junghoon Cho
handle max-pk-ok path identically when the incoming user is invalid;
avoids max-pk-ok feature presenting a username validity oracle
analysis and patch from Chris Rohlf in collaboration with Claude and
Anthropic Research
