Md at debian.org

tales of a debian maintainer

Extending access to the systemd RuntimeDirectory with a POSIX ACL

inn2 uses ephemeral UNIX domain sockets in /run/news/ to communicate with the ctlinnd program. Since the directory is only writeable by the "news" user, other unprivileged users are not able to use the command.

I solved this by extending the inn2.service systemd unit with a drop-in file which uses setfacl to give access to my user "md" to the RuntimeDirectory created by systemd. This is the content of /etc/systemd/system/inn2.service.d/md-ctlinnd.conf:

[Service]
# innd will change the permissions of /run/news/ when started: without
# creating it now with mode 0775 then that will change the ACL mask.
RuntimeDirectoryMode=0775
# allow user md to run ctlinnd(8), which creates sockets in /run/news/
ExecStartPost=/usr/bin/setfacl --modify user:md:rwx $RUNTIME_DIRECTORY

The non-obvious issue here is that the innd daemon on startup will change the directory permissions in a way which sets a more restrictive (non group-writeable) ACL mask, and this would make the newly created user ACL ineffective. The solution is to create the directory group-writeable from start.

(Beware: this creates a trivial privileges escalation from md to news.)

On having a track record in operating systems development

Now that Debian 12 has been released with proprietary firmwares on the official media, non-optional merged-/usr and systemd adopted by everybody, I want to take a moment to list, not without some pride, a few things that I was right about over the last 20 years:

Accepting the obvious solution about firmwares took 18 years. My work on the merged-/usr transition started in 2014, and the first discussions about replacing sysvinit are from 2011. The general adoption of udev (and dynamic device names, and persistent network interface names...) took less time in comparison and no large-scale flame wars, since people could enable it at their own pace. But it required countless little debates in the Debian Bug Tracking System: I still remember the people insisting that they would never use this newfangled dynamic /dev/, or complaining about their beloved /dev/cdrom symbolic link and persistent network interface names.

So follow me for more rants about inevitable technologies.

Installing Debian 12 on a Banana Pi M5

I recently bought a Banana Pi BPI-M5, which uses the Amlogic S905X3 SoC: these are my notes about installing Debian on it.

While this SoC is supported by the upstream U-Boot it is not supported by the Debian U-Boot package, so debian-installer does not work. Do not be fooled by seeing the DTB file for this exact board being distributed with debian-installer: all DTB files are, and it does not mean that the board is supposed to work.

As I documented in #1033504, the Debian kernels are currently missing some patches needed to support the SD card reader.

I started by downloading an Armbian Banana Pi image and booted it from an SD card. From there I partitioned the eMMC, which always appears as /dev/mmcblk1:

parted /dev/mmcblk1
(parted) mklabel msdos
(parted) mkpart primary ext4 4194304B -1
(parted) align-check optimal 1
mkfs.ext4 /dev/mmcblk1p1

Make sure to leave enough space before the first partition, or else U-Boot will overwrite it: as it is common for many ARM SoCs, U-Boot lives somewhere in the gap between the MBR and the first partition.

I looked at Armbian's /usr/lib/u-boot/platform_install.sh and installed U-Boot by manually copying it to the eMMC:

dd if=/usr/lib/linux-u-boot-edge-bananapim5_22.08.6_arm64/u-boot.bin of=/dev/mmcblk1 bs=1 count=442
dd if=/usr/lib/linux-u-boot-edge-bananapim5_22.08.6_arm64/u-boot.bin of=/dev/mmcblk1 bs=512 skip=1 seek=1

Beware: Armbian's U-Boot 2022.10 is buggy, so I had to use an older image.

I did not want to install a new system, so I copied over my old Cubieboard install:

mount /dev/mmcblk1p1 /mnt/
rsync -xaHSAX --delete --numeric-ids root@old-server:/ /mnt/ --exclude='/tmp/*' --exclude='/var/tmp/*'

Since the Cubieboard has a 32 bit CPU and the Banana Pi requires an arm64 kernel I enabled the architecture and installed a new kernel:

dpkg --add-architecture arm64
apt update
apt install linux-image-arm64
apt purge linux-image-6.1.0-6-armmp linux-image-armmp

At some point I will cross-grade the entire system.

Even if ttyS0 exists it is not the serial console, which appears as ttyAML0 instead. Nowadays systemd automatically starts a getty if the serial console is enabled on the kernel command line, so I just had to disable the old manually-configured getty:

systemctl disable serial-getty@ttyS0.service

I wanted to have a fully working flash-kernel, so I used Armbian's boot.scr as a template to create /etc/flash-kernel/bootscript/bootscr.meson and then added a custom entry for the Banana Pi to /etc/flash-kernel/db:

Machine: Banana Pi BPI-M5
Kernel-Flavors: arm64
DTB-Id: amlogic/meson-sm1-bananapi-m5.dtb
U-Boot-Initrd-Address: 0x0
Boot-Initrd-Path: /boot/uInitrd
Boot-Initrd-Path-Version: yes
Boot-Script-Path: /boot/boot.scr
U-Boot-Script-Name: bootscr.meson
Required-Packages: u-boot-tools

All things considered I do not think that I would recommend to Debian users to buy Amlogic-based boards since there are many other better supported SoCs.

I replaced grub with systemd-boot

To be able to investigate and work on the the measured boot features I have switched from grub to systemd-boot (sd-boot).

This initial step is optional, but it is useful because this way /etc/kernel/cmdline will become the new place where the kernel command line can be configured:

. /etc/default/grub
echo "root=/dev/mapper/root $GRUB_CMDLINE_LINUX $GRUB_CMDLINE_LINUX_DEFAULT" > /etc/kernel/cmdline

Do not forget to set the correct root file system there, because initramfs-tools does not support discovering it at boot time using the Discoverable Partitions Specification.

The installation has been automated since systemd version 252.6-1, so installing the package has the effect of installing sd-boot in the ESP, enabling it in the UEFI boot sequence and then creating boot loader entries for the kernels already installed on the system:

apt install systemd-boot

If needed, it could be manually installed again just by running bootctl install.

I like to show the boot menu by default, at least until I will be more familiar with sd-boot:

bootctl set-timeout 4

Since other UEFI binaries can be easily chainloaded, I am also going to keep around grub for a while, just to be sure:

cat <<END > /boot/efi/loader/entries/grub.conf
title Grub
linux /EFI/debian/grubx64.efi
END

At this point sd-boot works, but I still had to enable secure boot. So far sd-boot has not been signed with a Debian key known to the shim bootloader, so I needed to create a Machine Owner Key (MOK), enroll it in UEFI and then use it to sign everything.

I dislike the complexity of mokutil and the other related programs, so after removing it and the boot shim I have decided to use sbctl instead. With it I easily created new keys, enrolled them in the EFI key store and then signed everything:

sbctl create-keys
sbctl enroll-keys
for file in /boot/efi/*/*/linux /boot/efi/EFI/*/*.efi; do
  sbctl sign -s $file
done

Since there is no sbctl package yet I need to make sure that also the kernels installed in the future will be automatically signed, so I have created a trivial script in /etc/kernel/install.d/ which automatically runs sbctl sign -s or sbctl remove-file.

The Debian wiki SecureBoot page documents how do do this with mokutil and sbsigntool, but I think that sbctl is much friendlier.

Since I am not using the boot shim, I also had to set DisableShimForSecureBoot=true in /etc/fwupd/uefi_capsule.conf to make firmware updates work automatically.

As a bonus, I have also added to the boot menu the excellent Debian-based GRML live distribution. Since sd-boot is not capable of loopback-mounting CD-ROM images like grub, I first had to extract the kernel and initramfs and copy them to the ESP:

mount -o loop /boot/grml/grml64-full_2022.11.iso /mnt/
mkdir /boot/efi/grml/
cp /mnt/boot/grml64full/* /boot/efi/grml/
umount /mnt/

cat <<END > /boot/efi/loader/entries/grml.conf
title GRML
linux /grml/vmlinuz
initrd /grml/initrd.img
options boot=live bootid=grml64full202211 findiso=/grml/grml64-full_2022.11.iso live-media-path=/live/grml64-full net.ifnames=0 
END

As expected, after a reboot bootctl reports the new security features:

System:
      Firmware: UEFI 2.70 (Lenovo 0.4496)
 Firmware Arch: x64
   Secure Boot: enabled (user)
  TPM2 Support: yes
  Boot into FW: supported

Current Boot Loader:
      Product: systemd-boot 252.5-2
     Features:  Boot counting
                Menu timeout control
                One-shot menu timeout control
                Default entry control
                One-shot entry control
                Support for XBOOTLDR partition
                Support for passing random seed to OS
                Load drop-in drivers
                Support Type #1 sort-key field
                Support @saved pseudo-entry
                Support Type #1 devicetree field
                Boot loader sets ESP information
          ESP: /dev/disk/by-partuuid/1b767f8e-70fa-5a48-b444-cfe5c272d66e
         File: └─/EFI/systemd/systemd-bootx64.efi

...

Relevant documentation:

Using a custom domain as the Mastodon identity

I just did again the usual web search, and I have verified that Mastodon still does not support managing multiple domains on the same instance, and that there is still no way to migrate an account to a different instance without losing all posts (and more?).

As much as I like the idea of a federated social network, open standards and so on, I do not think that it would be wise for me to spend time developing a social network identity on somebody else's instance which could disappear at any time.

I have managed my own email server since the '90s, but I do not feel that the system administration effort required to maintain a private Mastodon instance would be justified at this point: there is not even a Debian package! Mastodon either needs to become much simpler to maintain or become much more socially important, and so far it is neither. Also, it would be wasteful to use so many computing resources for a single-user instance.

While it is not ideal, for the time being I compromised by redirecting WebFinger requests for md@linux.it using this Apache configuration:

<Location /.well-known/host-meta>
  Header set Access-Control-Allow-Origin: "*"
  Header set Content-Type: "application/xrd+xml; charset=utf-8"
  Header set Cache-Control: "max-age=86400"
</Location>

<Location /.well-known/webfinger>
  Header set Access-Control-Allow-Origin: "*"
  Header set Content-Type: "application/jrd+json; charset=utf-8"
  Header set Cache-Control: "max-age=86400"
</Location>

# WebFinger (https://www.rfc-editor.org/rfc/rfc7033)
RewriteMap lc int:tolower
RewriteMap unescape int:unescape
RewriteCond %{REQUEST_URI} ^/\.well-known/webfinger$
RewriteCond ${lc:${unescape:%{QUERY_STRING}}} (?:^|&)resource=acct:([^&]+)@linux\.it(?:$|&)
RewriteRule .+ /home/soci/%1/public_html/webfinger.json [L,QSD]
# answer 404 to requests missing "acct:" or for domains != linux.it
RewriteCond %{REQUEST_URI} ^/\.well-known/webfinger$
RewriteCond ${unescape:%{QUERY_STRING}} (?:^|&)resource=
RewriteRule .+ - [L,R=404]
# answer 400 to requests without the resource parameter
RewriteCond %{REQUEST_URI} ^/\.well-known/webfinger$
RewriteRule .+ - [L,R=400]

This is the equivalent lighttpd configuration:

  $HTTP["url"] =~ "^/\.well-known/" {
    setenv.add-response-header += (
      "Access-Control-Allow-Origin" => "*",
      "Cache-Control" => "max-age=86400",
    )
  }

  $HTTP["url"] == "/.well-known/host-meta" {
    mimetype.assign = ("" => "application/xrd+xml; charset=utf-8")
  }

  $HTTP["url"] == "/.well-known/webfinger" {
    mimetype.assign = ("" => "application/jrd+json; charset=utf-8")
    $HTTP["querystring"] =~ "(?:^|&)resource=acct:([^&]+)@linux\.it(?:$|&)" {
      alias.url = ( "" => "/home/soci/$1/public_html/webfinger.json" )
    } else {
      alias.url = ( "" => "/does-not-exist" )
    }
  }

Debian bookworm on a Lenovo T14s Gen3 AMD

I recently upgraded my laptop to a Lenovo T14s Gen3 AMD and I am happy to report that it works just fine with Debian/unstable using a 5.19 kernel.

The only issue is that some firmware files are still missing and I had to install them manually.

Updates are needed for the firmware-amd-graphics package (#1019847) for the Radeon 680M GPU (AMD Rembrandt) and for the firmware-atheros package (#1021157) for the Qualcomm NFA725A Wi-Fi card (which is actually reported as a NFA765).

s2idle (AKA "modern suspend") works too, and a ~10 seconds delay on resume has been removed by setting iommu=pt on the kernel command line.

For improved energy efficiency it is recommended to switch from the acpi_cpufreq CPU frequency scaling driver to amd_pstate. Please note that so far it is not loaded automatically.

As expected, fwupdmgr can update the system BIOS and the firmware of the NVMe device. Everybody should do it immediately, because there are major suspend bugs with BIOS releases earlier than 1.25.

Run an Ansible playbook in a remote chroot

Running a playbook in a remote chroot or container is not supported by Ansible, but I have invented a good workaround to do it anyway.

The first step is to install Mitogen for Ansible (ansible-mitogen in Debian) and then configure ansible.cfg to use it:

[defaults]
strategy = mitogen_linear

But everybody should use Mitogen anyway, because it makes Ansible much faster.

The trick to have Ansible operate in a chroot is to make it call a wrapper script instead of Python. The wrapper can be created manually or by another playbook, e.g.:

  vars:
  - fsroot: /mnt

  tasks:
  - name: Create the chroot wrapper
    copy:
      dest: "/usr/local/sbin/chroot_{{inventory_hostname_short}}"
      mode: 0755
      content: |
        #!/bin/sh -e
        exec chroot {{fsroot}} /usr/bin/python3 "$@"

  - name: Continue with stage 2 inside the chroot
    debug:
      msg:
        - "Please run:"
        - "ansible-playbook therealplaybook.yaml -l {{inventory_hostname}} -e ansible_python_interpreter=/usr/local/sbin/chroot_{{inventory_hostname_short}}"

This works thanks to Mitogen, which funnels all remote tasks inside that single call to Python. It would not work with standard Ansible, because it copies files to the remote system with SFTP and would do it outside of the chroot.

The same principle can also be applied to containers by changing the wrapper script, e.g:

#!/bin/sh -e
exec systemd-run --quiet --pipe --machine={{container_name}} --service-type=exec /usr/bin/python3 "$@"

After the wrapper will have been installed then you can run the real playbook by setting the ansible_python_interpreter variable, either on the command line, in the inventory or anywhere else that variables can be defined:

ansible-playbook therealplaybook.yaml -l {{inventory_hostname}} -e ansible_python_interpreter=/usr/local/sbin/chroot_{{inventory_hostname_short}}

My resignation from freenode

As it is now known, the freenode IRC network has been taken over by a Trumpian wannabe korean royalty bitcoins millionaire. To make a long story short, the former freenode head of staff secretly "sold" the network to this person even if it was not hers to sell, and our lawyers have advised us that there is not much that we can do about it without some of us risking financial ruin. Fuck you Christel, lilo's life work did not deserve this.

What you knew as freenode after 12:00 UTC of May 19 will be managed by different people.

As I have no desire to volunteer under the new regime, this marks the end of my involvement with freenode. It had started in 1999 when I encouraged the good parts of #linux-it to leave ircnet, and soon after I became senior staff. Even if I have not been very active recently, at this point I was the longest-serving freenode staff member and now I expect that I will hold this record forever.

The people that I have met on IRC, on freenode and other networks, have been and still are a very important part of my life, second only to the ones that I have known thanks to Usenet. I am not fine, but I know that the communities which I have been a part of are not defined by a domain name and will regroup somewhere else.

The current freenode staff members have resigned with me, these are some of their farewell messages:

Together we have created Libera.Chat, a new IRC network based on the same principles of the old freenode.

RPKI validation with FORT Validator

This article documents how to install FORT Validator (an RPKI relying party software which also implements the RPKI to Router protocol in a single daemon) on Debian 10 to provide RPKI validation to routers. If you are using testing or unstable then you can just skip the part about apt pinnings.

The packages in bullseye (Debian testing) can be installed as is on Debian stable with no need to rebuild them, by configuring an appropriate pinning for apt:

cat <<END > /etc/apt/sources.list.d/bullseye.list
deb http://deb.debian.org/debian/ bullseye main
END

cat <<END > /etc/apt/preferences.d/pin-rpki
# by default do not install anything from bullseye
Package: *
Pin: release bullseye
Pin-Priority: 100

Package: fort-validator rpki-trust-anchors
Pin: release bullseye
Pin-Priority: 990
END

apt update

Before starting, make sure that curl (or wget) and the web PKI certificates are installed:

apt install curl ca-certificates

If you already know about the legal issues related to the ARIN TAL then you may instruct the package to automatically install it. If you skip this step then you will be asked at installation time about it, either way is fine.

echo 'rpki-trust-anchors rpki-trust-anchors/get_arin_tal boolean true' \
  | debconf-set-selections

Install the package as usual:

apt install fort-validator

You may also install rpki-client and gortr on Debian 10, or maybe cfrpki and gortr. I have also tried packaging Routinator 3000 for Debian, but this effort is currently on hold because the Rust ecosystem is broken and hostile to the good packaging practices of Linux distributions.

RPKI validation with OpenBSD's rpki-client and Cloudflare's gortr

This article documents how to install rpki-client (an RPKI relying party software, the actual validator) and gortr (which implements the RPKI to Router protocol) on Debian 10 to provide RPKI validation to routers. If you are using testing or unstable then you can just skip the part about apt pinnings.

The packages in bullseye (Debian testing) can be installed as is on Debian stable with no need to rebuild them, by configuring an appropriate pinning for apt:

cat <<END > /etc/apt/sources.list.d/bullseye.list
deb http://deb.debian.org/debian/ bullseye main
END

cat <<END > /etc/apt/preferences.d/pin-rpki
# by default do not install anything from bullseye
Package: *
Pin: release bullseye
Pin-Priority: 100

Package: gortr rpki-client rpki-trust-anchors
Pin: release bullseye
Pin-Priority: 990
END

apt update

Before starting, make sure that curl (or wget) and the web PKI certificates are installed:

apt install curl ca-certificates

If you already know about the legal issues related to the ARIN TAL then you may instruct the package to automatically install it. If you skip this step then you will be asked at installation time about it, either way is fine.

echo 'rpki-trust-anchors rpki-trust-anchors/get_arin_tal boolean true' \
  | debconf-set-selections

Install the packages as usual:

apt install rpki-client gortr

And then configure rpki-client to generate its output in the the JSON format needed by gortr:

echo 'OPTIONS=-j' > /etc/default/rpki-client

You may manually start the service unit to immediately generate the data instead of waiting for the next timer run:

systemctl start rpki-client &

gortr too needs to be configured to use the JSON data generated by rpki-client:

echo 'GORTR_ARGS=-bind :323 -verify=false -checktime=false -cache /var/lib/rpki-client/json' > /etc/default/gortr

And then it needs to be restarted to use the new configuration:

systemctl restart gortr

You may also install FORT Validator on Debian 10, or maybe cfrpki with gortr. I have also tried packaging Routinator 3000 for Debian, but this effort is currently on hold because the Rust ecosystem is broken and hostile to the packaging practices of Linux distributions.

About

This is the blog of Marco d'Itri.

S M T W T F S
      1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29      

See also:

My blogroll:


W3C HTML 4.01
W3C CSS 2.0     

Powered by Bryar.pm.