Skip to content

Ngfw trixie changes - #460

Open
singhrohit23 wants to merge 24 commits into
masterfrom
ngfw-trixie
Open

Ngfw trixie changes#460
singhrohit23 wants to merge 24 commits into
masterfrom
ngfw-trixie

Conversation

@singhrohit23

Copy link
Copy Markdown
Contributor

No description provided.

singhrohit23 and others added 24 commits August 26, 2026 17:09
…okworm

fbpanel was removed from Debian after bullseye and is not available in
bookworm. Replace with xfce4-panel which is already part of the xfce4
stack (xfce4-session/xfwm4 are existing dependencies).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The postinst was looking for xorg safe config at
/home/kiosk/xorg-confs/safe which never existed. The file is
actually installed at /etc/X11/xorg-kiosk-safe.conf by the package.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- untangle-postgresql-config: PG 13 to PG 15 migration
- untangle-spamassassin-update: spamd service + symlink + /etc/default/spamd
- untangle-suricata-config: pin suricata (<< 1:7) to avoid DPDK chain
- untangle-kiosk: panel alignment fix (center)
- untangle-firmware-common/05modules: modprobe br_netfilter + iptable_tune || true
- untangle-hardware-config: PG 15 pg_hba.conf scram-sha-256 to trust
- untangle-clamav-config: removed obsolete clamd.conf.buster
- untangle-development-build: removed lintian dep
- untangle-development-kernel: qemu to qemu-system-x86,qemu-utils
- Various debian/control: bookworm dependency updates
- build-order.txt: updated for bookworm
- .travis.yml: REPOSITORY=bookworm, amd64 only
- docker-compose.build.yml: REPOSITORY and TRAVIS_BRANCH defaults to bookworm
- docker-compose-dev.yml: new local dev build file with host networking
- position: p=12 is bottom-center in xfce4-panel 4.18 (was p=6)
- length: type changed from uint to double (4.18 rejects uint 0)
- size: 100px to match fbpanel's original icon height
- .xsession: start xfwm4 before xfce4-panel so panel can position
  correctly via window manager dock hints

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…p_mark

Bump linux-image/linux-headers dep to 6.1.0-44-untangle-amd64
(>= 6.1.159-1+untangle3bookworm) so regular builds pull the kernel
with the NGFW-15626 NF_DROP conntrack-confirm patch.

Add /etc/modprobe.d/untangle-nfqueue.conf to set
nfqueue_confirm_drop_mark=255 at module load, activating the patch
on every NGFW without manual configuration.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…lpers

Round 7-13 dep-resolution work + scaffolding to make ngfw_pkgs build
under trixie. Companion to ngfw_imgtools start_new_disc/late_command
and ngfw_upstream mod_python 3.5.0.7 commits.

debian/control trixie alternatives (avoids hard-coded bookworm-only deps):
- untangle-development-runtime: mime-support → media-types | mime-support;
  ntp → ntpsec | ntp; openjdk-17-jre-headless → openjdk-21-jre-headless |
  openjdk-17-jre-headless; dropped python-unittest2 alternative.
- untangle-postgresql-config: postgresql-15 → postgresql-17 | postgresql-15.
- untangle-suricata-config: drop suricata (<< 1:7) ceiling — trixie ships 7.x
  by default and we validated end-to-end.
- untangle-lxc-client: drop cgroupfs-mount (deprecated in trixie).
- untangle-extra-utils: drop latencytop (gone from trixie).
- untangle-kiosk: drop gvidm (gone from trixie).
- untangle-linux-config: trixie kernel deps refresh.

NEW package — untangle-kernel-modules-trixie-amd64:
  Trixie equivalent of untangle-kernel-modules-bookworm-amd64. Provides:
  untangle-kernel-modules-trixie virtual package consumed by
  untangle-virus-blocker and other apps that need raw module access.

untangle-vue-ui/Makefile:
  Trixie's apt yarnpkg is Yarn Berry (4.x) which rejects --force and
  removed the `clean` subcommand. Install Yarn Classic 1.22.22 via npm
  on first run; keeps yarn.lock and existing build scripts unchanged.

Dockerfile.bookworm-current (NEW):
  Adds nvm/Node-on-bookworm-current image used by the vue-ui builder.
  Apparently was needed once vue-ui's yarn handling diverged from the
  base bookworm-build-multiarch image's package set.

untangle-archive-keyring trixie-automatic .asc (NEW):
  Trixie's apt uses sqv (Sequoia OpenPGP) which rejects keybox-format
  .gpg keyrings. Ship the ASCII-armored key alongside the existing .gpg
  so InRelease verifies under both bookworm and trixie. Memory pointer:
  trixie-keyring-sqv-fix.md.

build-order.txt:
  All 85 entries flipped bookworm → trixie. Adds the new
  untangle-kernel-modules-trixie-amd64 package. waagent removed (handled
  in ngfw_upstream commit). Existing bookworm metadata still buildable
  via parallel branch.

Validated: round-15 trixie ISO bundles all of the above; install on
192.168.56.137 succeeds; PG17 + ntpsec + suricata 7.x + media-types
all install correctly without dep conflicts.
…ld entries

Three "auto build" stub entries with version "18.0.0..-1trixie" (note
the empty middle between dots) leaked into the seed changelog from
local set-version.sh runs that hit a git error and emitted placeholder
versions. Drop them; keep only the (0-0) initial entry, matching the
pattern used by untangle-kernel-modules-bookworm-amd64.

The set-version.sh + dch pipeline computes the real version
(18.0.0.<TIMESTAMP>.<COMMIT>-1trixie) from this baseline at build time,
so the seed-only state is the correct on-disk source.
pkgtools build.sh runs do-build in the background and redirects its
stdout/stderr to a per-package logfile, which is invisible to the Travis
console. With VERBOSE=1, do-build is foregrounded with `2>&1 | tee
$logfile` so the actual command output streams live. Needed to diagnose
the apt-get build-dep hang on untangle-cloud-config under ngfw-trixie
(suspected root cause: sqv keyring rejection of the unsigned
package-server trixie/current InRelease).

Revert this once the keyring issue is fixed at the Docker-image layer.
Previous VERBOSE=1 run on Travis got past untangle-cloud-config build +
dput (both succeeded) but then hung silently in do-build's post-upload
cleanup block — most likely fakeroot debian/rules clean or the find/
quilt cleanup. Travis killed the job after 10 min no-output.

pkgtools build.sh already has a DEBUG env var that sets -x and forces
VERBOSE=1 (lines 36-39). Adding it surfaces the exact wedged command.

Revert once root cause is identified + fixed.
Root cause of trixie CI hang identified — fakeroot SysV backend wedged
under Travis Docker IPC namespace. Fixed at the image layer by
switching to fakeroot-tcp in ngfw_src Dockerfile.trixie-build (commit
cd7d1eac29). Diagnostic flags no longer needed; revert to silent CI
output matching bookworm pattern.
Image rebuild with update-alternatives fakeroot->fakeroot-tcp confirmed
deployed (digest f057bcf2fc...). Forced fresh pull locally shows
fakeroot-tcp active. Travis still hangs at untangle-cloud-config, but
without VERBOSE we can't tell whether:
- fakeroot-tcp also hangs under Travis IPC (would need different fix)
- Hang has moved to a different command (different root cause)
- docker-compose.build.yml uses a different image path

Re-enable trace flags to see exact wedge point. Revert again once
diagnosed.
…trixie

fakeroot-tcp backend (deployed in trixie-build-multiarch image
cd7d1eac29) does NOT resolve the hang — Travis Docker IPC + socket
restrictions appear to wedge BOTH fakeroot SysV and TCP variants on
post-build cleanup.

Post-build cleanup is non-essential for CI: dpkg-buildpackage does its
own pre-build clean for each package, and Travis runners are ephemeral
(no artifact buildup across runs). NO_CLEAN=1 skips the post-build
fakeroot/find/quilt block entirely, unblocking the trixie matrix.

Also drops VERBOSE+DEBUG diagnostic flags — no longer needed once
cleanup is bypassed.

Long-term: investigate fakeroot-tcp + Travis IPC interaction OR move
the cleanup outside fakeroot wrapper (clean doesn't need root). Both
deferred.
On trixie (Python 3.13), python3-gdbm is not installed by default.
Without it, mod_python's DbmSession falls back to dbm.sqlite3 which
lacks the gdbm-specific db.first()/next() API used by session cleanup,
breaking HTTP session persistence for remote clients. This caused
cert upload tests (test_021-023) to fail with 60s timeouts in ATS.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant