Patrick, this is fantastic. These guys have learned to run Qubes in live mode! I checked, and indeed the data is not saved. This allows the use of Whonix and Kicksecure in Qubes in amnesia mode. I would like to draw the attention of developers and enthusiasts to get involved in this topic and develop this wonderful project. Perhaps it should be added to the Whonix wiki. Let more users learn about it
so, you recommend using whonix exactly like this for maximum protection and anonymity?
Yes, this is the most effective solution. Patrick has been suggesting adding a live mode to Qubes for a long time implement live boot by porting grub-live to Qubes - amnesia / non-persistent boot / anti-forensics · Issue #4982 · QubesOS/qubes-issues · GitHub, and these guys implemented it several years ago. This solution has excellent reviews. I checked it and amnesia works. Therefore, the best options for using Whonix:
Live Kicksecure host + Whonix and Live Qubes with Whonix and Kicksecure VMs
I am not so excited yet.
- Lengthy instructed hosted in a forum. I am not a fan of lengthy documentation in forums as it tends bit rot over time, cannot be updated collaboratively as if it was in a wiki or source code and requires reading the whole forum discussion.
- Not officially supported by Qubes.
- Anti-forensic Claims
I published a comment that they updated. I agree that it’s not very convenient - perhaps it’s worth suggesting to them to create a repository on GitHub-GitLab.
I don’t know why the Qubes team doesn’t officially support this - the project has been around for 3 years. Many users have long been asking for an amnesia mode for Qubes - many topics have been created about it. Perhaps it hasn’t been included in the official support because it’s just dracut settings. Dracut is very flexible and can be configured in various ways.
I think we can write in the wiki: “Some Qubes users create their own solutions for running live mode, which can enhance the protection of Whonix VMs” or something like that. I believe these users deserve attention. Their solution has been working on several versions of Qubes since 2022. This could attract other developers to the very important topic of anti-forensic protection and it will enhance the security of Whonix users in Qubes
Hello. I tested it. It works great! but you need to specify the correct configuration for maximum protection against forensics:
- Follow this instruction and make dom0 run in memory Qubes in tmpfs 🤫 - #30 by qstateless - Community Guides - Qubes OS Forum
This will make only dom0 amnesic! - Then it is desirable to increase the size of dom0 (I highly doubt that 20 GB will be enough after the 3 point):
sudo lvresize --size 30G /dev/mapper/qubes_dom0-root
sudo resize2fs /dev/mapper/qubes_dom0-root
- Follow this instruction and recreate the qubes so that all this is stored exactly in dom0 Qubes in tmpfs 🤫 - #56 by qstateless - Community Guides - Qubes OS Forum
a. Instead of using AppQubes based on TemplateQubes, you could alternatively create StandaloneQubes in the ‘
varlibqubes
’ storage pool, which appear to store ALL data in traditional image files within Dom0’s ‘/var/lib/qubes
’ directory. This is very costly in Dom0 RAM, as it copies your entire TemplateQube’s OS into Dom0 RAM space for each StandaloneQube you make, whether it is actively running or not, which is usually multiple extra GBs per qube, in addition to the RAM it takes to store any user files and the RAM it takes to run and operate the qube’s OS & apps.b. You could create a new TemplateQube from the previous TemplateQube you want to use, but store that new TemplateQube in the ‘
varlibqubes
’ storage pool. Then create a new AppQubes based on this new TemplateQube, and store this new AppQube in the ‘varlibqubes
’ storage pool too. Now, when you use this new AppQube, it appears to store ALL data in traditional image files within Dom0’s ‘/var/lib/qubes
’ directory. This is as costly as the other method for the first AppQube, but you do not have to copy & store the entire TemplateQube OS root filesystem for every AppQube you want to make with it, so this saves a lot of RAM space for using more than one qube.c. Like “b”, you could create a new TemplateQube from the previous TemplateQube you want to use, but store that new TemplateQube in the ‘
varlibqubes
’ storage pool. Then create a Disposable Template by creating new AppQubes based on this new TemplateQube, and store this new AppQube in the ‘varlibqubes
’ storage pool too. After creation, in the settings of this AppQube, under the “Advanced” tab, you can check to turn on “Disposable template” and after applying also select “Default disposable template” to either be “(none)” or that very same AppQube itself. Now, you can use both this TemplateQube and Disposable Template AppQube to create new AppQubes and DisposableQubes fully within the ‘varlibqubes
’ storage pool. This is likely to generally be the most desirable approach for most people.Here is an example implementation of approach “c”:
Let’s say you want to base some of your fully stateless qubes on the ‘
debian-12-xfce
’ persistent template.
- In Persistent mode: Create & Configure a new TemplateQube named ‘
debian-12-xfce-stateless
’ based on ‘debian-12-xfce
’ and choose to store it in storage pool ‘varlibqubes
’ (Advanced tab).
- In Persistent mode: Create & Configure a new AppQube named ‘
debian-12-xfce-stateless-dvm
’ based on ‘debian-12-xfce-stateless
’ and choose to store it in storage pool ‘varlibqubes
’ (Advanced tab).
- In Persistent mode: After creation, for the AppQube ‘
debian-12-xfce-stateless-dvm
’, change the ‘Advanced’ tab setting ‘Disposable template’ to be checked as turned on (click Apply), then the ‘Default disposable template’ to either be ‘(none)’ or ‘debian-12-xfce-stateless-dvm
’ itself.
- In Persistent mode: Create & Configure any new AppQubes based on ‘
debian-12-xfce-stateless
’ that you want to exist across multiple stateless boot sessions and choose to store them in storage pool ‘varlibqubes
’ (Advanced tab).
- In Stateless mode: You are free to now use any AppQubes based on ‘
debian-12-xfce-stateless
’ and DisposableQubes based on ‘debian-12-xfce-stateless-dvm
’, which appear to remain fully stateless by storing ALL data in traditional image files within Dom0’s ‘/var/lib/qubes
’ directory (that directory gets wiped and reset back to match the state of your last persistent session once your stateless session is powered down).
I made an amnesic template Kicksecure and created all secure qubes based on it. I also changed all Whonix templates. I created a Fedora-live template for cubes with gimp, strawberry, vlc. In the end, it works great. There is no saving in either appvm or dvm qubes. For good performance, I recommend having at least 32 GB of memory (I recommend the same amount for running Whonix from the Kicksecure-live host). 32 GB in Qubes allows you to run many virtual machines and it is convenient. If you have 16 GB of memory, only dvm is used in amnesic mode. Otherwise, you will quickly run out of free space. By the way, some Whonix forum moderators are impressed with this solution. It is really amazing, but the devs write that it can be improved and they are happy with new ideas and solutions. It’s a shame that this is published on the forum and not on GitHub. You have to read all the messages to not miss the relevant information. But I am very satisfied with this solution. I hope it will continue to develop.
Your instruction is very good!
I think this solution will work for a long time. It’s independent of Qubes, so it should work with other distros as well. This solution is also more efficient in terms of memory usage for Qubes compared to grub-live. I have a laptop with 62GB of memory. In Kicksecure live, I have 31GB of free disk space. In Qubes live, I allocated 50GB for dom0, and all of it is available for running VMs. That’s really cool. Grub-live would require 128GB of memory to work normally with Qubes live. Maybe someone will come up with updates to this solution to require even less memory for Qubes, but I’m very satisfied with it now.
But if you only use whonix-workstation, whonix-workstation-dvm, and, for example, mullvad-dvm, it won’t require a lot of memory. Computers with 16GB of memory will be sufficient for such amnesic sessions
If you are copying your amnesic VMs, always choose ‘varlibqubes’! Otherwise, it will lose amnesia! I conducted a test and copied an amnesic AppVM based on an amnesic template, but chose ‘vm-pool’. This almost completely removed amnesia - only newly added programs in the Qubes Manager were not saved.
Patrick, maybe we should edit this text:
At time of writing, there is no shortcut, substitute or workaround available that users can easily use to get Qubes Anti-Forensics capabilities. This is elaborated in developers chapter Forensic Considerations and Anti-forensics Claims.
Disposables are not amnesic. In practice this means traces of their activity can be left on storage or in memory, making them vulnerable to forensic operations. [20]
I read this a long time ago and it upset me then. I think it’s worth mentioning this solution. For example, “Experienced Qubes users have created a workaround for anti-forensic protection…”
Did anyone consider Forensic Considerations and Anti-forensics Claims?
If you have a lot of memory and plan to create only amnesic varlibqubes, then during the Qubes installation, do not select the installation of templates. You can choose only sys-net and sys-firewall. The installation will proceed very quickly.
After logging into the system, open the dom0 terminal and change the default Storage pool:
qubes-prefs default_pool varlibqubes
After that, install the Whonix templates How-to: Install Qubes-Whonix It will be installed directly in dom0. Then install other templates.
After that, proceed with this instruction:
Qubes in tmpfs 🤫 - #30 by qstateless - Community Guides - Qubes OS Forum
In Qubes OS, you cannot perform a TRIM operation on varlibqubes
to increase free disk space. This is very important for live mode. You can easily perform a TRIM operation as follows: create a copy of the templates in the vm-pool
→ start a terminal in the whonix-gateway-clone-1
(vm-pool) → clear cache and unnecessary packages → then enter sudo fstrim -av
→ create a copy of the templates in the varlibqubes
. To perform a TRIM operation in a workstation, do the following: open the terminal in dom0 and enter qvm-console-dispvm whonix-workstation-17-clone-1
(vm-pool) → then log in as root
→ clear cache and unnecessary packages → write sudo fstrim -av
→ create a copy of the templates in the varlibqubes
. With these actions, I was able to remove more than 3 GB of trash!
PS: You may encounter error copy from a dirty volume
when testing it in live mode. But in persistent mode, everything should work fine. I made a trim in Debian, Fedora, and Whonix templates without any problems.
[quote=“Patrick, post:10, topic:21912”]
Patrick, I have studied Qubes live and grub-live a bit. It seems that no one has been able to port grub-live to Qubes due to very high memory requirements. These guys changed the approach to this solution - you can implement live only for dom0 storage. This saves a lot of memory. I think we can try to implement grub-live only for dom0. But it should work differently than in Kicksecure. I have half of my memory on disk in Kicksecure Live. If I have 16 GB of memory, I won’t be able to run Whonix from KVM/VirtualBox in live mode due to quick memory exhaustion. But this solution allows running sys-whonix, anon-whonix, and whonix-dvm in memory - I asked a friend to check it on a laptop with 16 GB of memory. One of the authors of this project thoroughly studied the amnesic properties of this solution. No data remains if both the template and appvm are created in varlibqubes. I decided to check this myself and also did not find any traces of the session - snapshots, images, and metadata are stored in dom0, so they disappear. Is this an ideal solution for forensic protection? We don’t know. Is grub-live ideal for it? We also don’t know. You say you are not a forensic expert, but you implemented live mode to protect your users. You know that this will greatly enhance their protection. You are a very kind person and I admire you. I think this solution for Whonix on Qubes is very good and also greatly enhances forensic protection. But I also think it can be significantly improved. Essentially, live Qubes works according to this script:
mkdir /mnt
umount /sysroot
mount /dev/mapper/qubes_dom0-root /mnt
modprobe zram
echo 10G > /sys/block/zram0/disksize
/mnt/usr/sbin/mkfs.ext2 /dev/zram0
mount /dev/zram0 /sysroot
cp -a /mnt/* /sysroot
exit 0
It just copies the dom0 partition to zram. Perhaps this can be improved with overlayfs. It seems you are familiar with this (or someone from the Whonix team). I think it would be great to improve this and recommend it to Whonix and Kicksecure on Whonix users. Maybe you will immediately see where improvements can be made in this small code. I am ready to test any new ideas about it. Maybe it can be implemented through overlayfs. But only so that memory is not halved, as in Kicksecure
A post was split to a new topic: Grub-live improvement - overlay-mount.sh - add, increase size= mount command parameter
This needs to be contributed upstream. Qubes amnesia development is fully independent from Whonix. This here is the wrong place to discuss this.
Unfortunately, not. Only Qubes developers can comment on that.
Perhaps size=100% parameter will help implement this suggest (implement live boot by porting grub-live to Qubes - amnesia / non-persistent boot / anti-forensics · Issue #4982 · QubesOS/qubes-issues · GitHub). For example, my laptop has 32 GB of memory. I had 15 GB in the default live mode in Kicksecure host. Now I have 31 GB in live mode. It allows me to run Whonix gateway and Whonix workstation, Kicksecure VM, Ubuntu KDE, and Fedora KDE simultaneously in KVM (provided I use fstrim -av
in the virtual machines). These are full-fledged distributions, not lightweight Qubes templates. We will always have only half of the memory in overlayfs-tmpfs if we do not specify the size.
Copying dom0 to zram is a very simple solution. It is efficient, but the boot time increases. Perhaps this can be implemented using overlayfs for dom0. I’m not an expert in this, but I would be interested in trying it next week
Wow, size=100% works! Thank you! And I don’t notice any issues with RAM shortage - I opened 10 tabs in browser and ran KeepassXC, Thunar, VLC, and everything works without problems
Whonix and Kicksecure in Live Mode with OverlayFS in Qubes OS
Qubes in tmpfs 🤫 - #30 by qstateless - Community Guides - Qubes OS Forum - Method of copying to zram is very slow, less secure and has more limitations on memory usage in live mode than OverlayFS. With OverlayFS /
filesystem will be read-only and launching in live mode will be very fast. and you will also have more free disk space. The original live mode in Whonix and Kicksecure works on OverlayFS.
Instructions ( do it in dom0):
-
Create a folder
90overlay-root
in/usr/lib/dracut/modules
.
sudo mkdir /usr/lib/dracut/modules.d/90overlay-root
-
Create two files
module-setup.sh
andoverlay-mount.sh
in this folder and make it executable.
sudo touch /usr/lib/dracut/modules.d/90overlay-root/module-setup.sh
sudo touch /usr/lib/dracut/modules.d/90overlay-root/overlay-mount.sh
sudo chmod 755 /usr/lib/dracut/modules.d/90overlay-root/module-setup.sh
sudo chmod 755 /usr/lib/dracut/modules.d/90overlay-root/overlay-mount.sh
- Open
module-setup.sh
sudo nano /usr/lib/dracut/modules.d/90overlay-root/module-setup.sh
Add this code:
#!/bin/bash
check() {
# do not add modules if the kernel does not have overlayfs support
[ -d /lib/modules/$kernel/kernel/fs/overlayfs ] || return 1
}
depends() {
# We do not depend on any modules - just some root
return 0
}
# called by dracut
installkernel() {
hostonly='' instmods overlay
}
install() {
inst_hook pre-pivot 10 "$moddir/overlay-mount.sh"
}
Press Ctrl + O
and Enter
Press Ctrl + X
to exit nano editor.
- Open
overlay-mount.sh
sudo nano /usr/lib/dracut/modules.d/90overlay-root/overlay-mount.sh
Add this code:
#!/bin/sh
# make a read-only nfsroot writeable by using overlayfs
# the nfsroot is already mounted to $NEWROOT
# add the parameter rootovl to the kernel, to activate this feature
. /lib/dracut-lib.sh
if ! getargbool 0 rootovl ; then
return
fi
modprobe overlay
# a little bit tuning
mount -o remount,nolock,noatime $NEWROOT
# Move root
# --move does not always work. Google >mount move "wrong fs"< for
# details
mkdir -p /live/image
mount --bind $NEWROOT /live/image
umount $NEWROOT
# Create tmpfs
mkdir /cow
mount -n -t tmpfs -o mode=0755,size=100% tmpfs /cow
mkdir /cow/work /cow/rw
# Merge both to new Filesystem
mount -t overlay -o noatime,lowerdir=/live/image,upperdir=/cow/rw,workdir=/cow/work,default_permissions overlay $NEWROOT
# Let filesystems survive pivot
mkdir -p $NEWROOT/live/cow
mkdir -p $NEWROOT/live/image
mount --bind /cow/rw $NEWROOT/live/cow
umount /cow
mount --bind /live/image $NEWROOT/live/image
umount /live/image
Press Ctrl + O
and Enter
Press Ctrl + X
to exit nano editor.
- Create a file
dracut-lib.sh
inlib
and make it executable
sudo touch lib/dracut-lib.sh
sudo chmod 755 lib/dracut-lib.sh
And add this code:
#!/bin/sh
type wait_for_dev > /dev/null 2>&1 || . /lib/dracut-dev-lib.sh
export DRACUT_SYSTEMD
export NEWROOT
if [ -n "$NEWROOT" ]; then
[ -d "$NEWROOT" ] || mkdir -p -m 0755 "$NEWROOT"
fi
# shellcheck disable=SC2153
if [ -z "$PREFIX" ]; then
if ! [ -d /run/initramfs ]; then
mkdir -p -m 0755 /run/initramfs/log
ln -sfn /run/initramfs/log /var/log
fi
[ -d /run/lock ] || mkdir -p -m 0755 /run/lock
[ -d /run/log ] || mkdir -p -m 0755 /run/log
fi
debug_off() {
set +x
}
debug_on() {
[ "$RD_DEBUG" = "yes" ] && set -x
}
# returns OK if $1 contains literal string $2 (and isn't empty)
strstr() {
[ "${1##*"$2"*}" != "$1" ]
}
# returns OK if $1 matches (completely) glob pattern $2
# An empty $1 will not be considered matched, even if $2 is * which technically
# matches; as it would match anything, it's not an interesting case.
strglob() {
[ -n "$1" -a -z "${1##$2}" ]
}
# returns OK if $1 contains (anywhere) a match of glob pattern $2
# An empty $1 will not be considered matched, even if $2 is * which technically
# matches; as it would match anything, it's not an interesting case.
strglobin() {
[ -n "$1" -a -z "${1##*$2*}" ]
}
# returns OK if $1 contains literal string $2 at the beginning, and isn't empty
str_starts() {
[ "${1#"$2"*}" != "$1" ]
}
# returns OK if $1 contains literal string $2 at the end, and isn't empty
str_ends() {
[ "${1%*"$2"}" != "$1" ]
}
trim() {
local var="$*"
var="${var#"${var%%[![:space:]]*}"}" # remove leading whitespace characters
var="${var%"${var##*[![:space:]]}"}" # remove trailing whitespace characters
printf "%s" "$var"
}
if [ -z "$DRACUT_SYSTEMD" ]; then
warn() {
check_quiet
echo "<28>dracut Warning: $*" > /dev/kmsg
echo "dracut Warning: $*" >&2
}
info() {
check_quiet
echo "<30>dracut: $*" > /dev/kmsg
if [ "$DRACUT_QUIET" != "yes" ]; then
echo "dracut: $*" >&2
fi
}
else
warn() {
echo "Warning: $*" >&2
}
info() {
echo "$*"
}
fi
vwarn() {
while read -r line || [ -n "$line" ]; do
warn "$line"
done
}
vinfo() {
while read -r line || [ -n "$line" ]; do
info "$line"
done
}
killall_proc_mountpoint() {
local _pid
local _killed=0
for _pid in /proc/*; do
_pid=${_pid##/proc/}
case $_pid in
*[!0-9]*) continue ;;
esac
[ -e "/proc/$_pid/exe" ] || continue
[ -e "/proc/$_pid/root" ] || continue
if strstr "$(ls -l -- "/proc/$_pid" "/proc/$_pid/fd" 2> /dev/null)" "$1"; then
kill -9 "$_pid"
_killed=1
fi
done
return $_killed
}
getcmdline() {
local _line
local _i
local CMDLINE_ETC_D
local CMDLINE_ETC
local CMDLINE_PROC
unset _line
if [ -e /etc/cmdline ]; then
while read -r _line || [ -n "$_line" ]; do
CMDLINE_ETC="$CMDLINE_ETC $_line"
done < /etc/cmdline
fi
for _i in /etc/cmdline.d/*.conf; do
[ -e "$_i" ] || continue
while read -r _line || [ -n "$_line" ]; do
CMDLINE_ETC_D="$CMDLINE_ETC_D $_line"
done < "$_i"
done
if [ -e /proc/cmdline ]; then
while read -r _line || [ -n "$_line" ]; do
CMDLINE_PROC="$CMDLINE_PROC $_line"
done < /proc/cmdline
fi
CMDLINE="$CMDLINE_ETC_D $CMDLINE_ETC $CMDLINE_PROC"
printf "%s" "$CMDLINE"
}
getarg() {
debug_off
local _deprecated _newoption
CMDLINE=$(getcmdline)
export CMDLINE
while [ $# -gt 0 ]; do
case $1 in
-d)
_deprecated=1
shift
;;
-y)
if dracut-getarg "$2" > /dev/null; then
if [ "$_deprecated" = "1" ]; then
if [ -n "$_newoption" ]; then
warn "Kernel command line option '$2' is deprecated, use '$_newoption' instead."
else
warn "Option '$2' is deprecated."
fi
fi
echo 1
debug_on
return 0
fi
_deprecated=0
shift 2
;;
-n)
if dracut-getarg "$2" > /dev/null; then
echo 0
if [ "$_deprecated" = "1" ]; then
if [ -n "$_newoption" ]; then
warn "Kernel command line option '$2' is deprecated, use '$_newoption=0' instead."
else
warn "Option '$2' is deprecated."
fi
fi
debug_on
return 1
fi
_deprecated=0
shift 2
;;
*)
if [ -z "$_newoption" ]; then
_newoption="$1"
fi
if dracut-getarg "$1"; then
if [ "$_deprecated" = "1" ]; then
if [ -n "$_newoption" ]; then
warn "Kernel command line option '$1' is deprecated, use '$_newoption' instead."
else
warn "Option '$1' is deprecated."
fi
fi
debug_on
return 0
fi
_deprecated=0
shift
;;
esac
done
debug_on
return 1
}
# getargbool <defaultval> <args...>
# False if "getarg <args...>" returns "0", "no", or "off".
# True if getarg returns any other non-empty string.
# If not found, assumes <defaultval> - usually 0 for false, 1 for true.
# example: getargbool 0 rd.info
# true: rd.info, rd.info=1, rd.info=xxx
# false: rd.info=0, rd.info=off, rd.info not present (default val is 0)
getargbool() {
local _b
unset _b
local _default
_default="$1"
shift
_b=$(getarg "$@") || _b=${_b:-"$_default"}
if [ -n "$_b" ]; then
[ "$_b" = "0" ] && return 1
[ "$_b" = "no" ] && return 1
[ "$_b" = "off" ] && return 1
fi
return 0
}
isdigit() {
case "$1" in
*[!0-9]* | "") return 1 ;;
esac
return 0
}
# getargnum <defaultval> <minval> <maxval> <arg>
# Will echo the arg if it's in range [minval - maxval].
# If it's not set or it's not valid, will set it <defaultval>.
# Note all values are required to be >= 0 here.
# <defaultval> should be with [minval -maxval].
getargnum() {
local _b
unset _b
local _default _min _max
_default="$1"
shift
_min="$1"
shift
_max="$1"
shift
_b=$(getarg "$1") || _b=${_b:-"$_default"}
if [ -n "$_b" ]; then
isdigit "$_b" && _b=$((_b)) \
&& [ $_b -ge "$_min" ] && [ $_b -le "$_max" ] && echo $_b && return
fi
echo "$_default"
}
getargs() {
debug_off
CMDLINE=$(getcmdline)
export CMDLINE
local _val _i _gfound _deprecated
unset _val
unset _gfound
_newoption="$1"
for _i in "$@"; do
if [ "$_i" = "-d" ]; then
_deprecated=1
continue
fi
if _val="$(dracut-getargs "$_i")"; then
if [ "$_deprecated" = "1" ]; then
if [ -n "$_newoption" ]; then
warn "Option '$_i' is deprecated, use '$_newoption' instead."
else
warn "Option $_i is deprecated!"
fi
fi
if [ -n "$_val" ]; then
printf '%s\n' "$_val"
fi
_gfound=1
fi
_deprecated=0
done
if [ -n "$_gfound" ]; then
debug_on
return 0
fi
debug_on
return 1
}
# Prints value of given option. If option is a flag and it's present,
# it just returns 0. Otherwise 1 is returned.
# $1 = options separated by commas
# $2 = option we are interested in
#
# Example:
# $1 = cipher=aes-cbc-essiv:sha256,hash=sha256,verify
# $2 = hash
# Output:
# sha256
getoptcomma() {
local line=",$1,"
local opt="$2"
local tmp
case "${line}" in
*,${opt}=*,*)
tmp="${line#*,${opt}=}"
echo "${tmp%%,*}"
return 0
;;
*,${opt},*) return 0 ;;
esac
return 1
}
# Splits given string 'str' with separator 'sep' into variables 'var1', 'var2',
# 'varN'. If number of fields is less than number of variables, remaining are
# not set. If number of fields is greater than number of variables, the last
# variable takes remaining fields. In short - it acts similary to 'read'.
#
# splitsep sep str var1 var2 varN
#
# example:
# splitsep ':' 'foo:bar:baz' v1 v2
# in result:
# v1='foo', v2='bar:baz'
#
# TODO: ':' inside fields.
splitsep() {
debug_off
local sep="$1"
local str="$2"
shift 2
local tmp
while [ -n "$str" -a "$#" -gt 1 ]; do
tmp="${str%%$sep*}"
eval "$1='${tmp}'"
str="${str#"$tmp"}"
str="${str#$sep}"
shift
done
[ -n "$str" -a -n "$1" ] && eval "$1='$str'"
debug_on
return 0
}
setdebug() {
[ -f /usr/lib/initrd-release ] || return
if [ -z "$RD_DEBUG" ]; then
if [ -e /proc/cmdline ]; then
RD_DEBUG=no
if getargbool 0 rd.debug -d -y rdinitdebug -d -y rdnetdebug; then
RD_DEBUG=yes
[ -n "$BASH" ] \
&& export PS4='${BASH_SOURCE}@${LINENO}(${FUNCNAME[0]-}): '
fi
fi
export RD_DEBUG
fi
debug_on
}
setdebug
source_all() {
local f
local _dir
_dir=$1
shift
[ "$_dir" ] && [ -d "/$_dir" ] || return
for f in "/$_dir"/*.sh; do
if [ -e "$f" ]; then
# shellcheck disable=SC1090
# shellcheck disable=SC2240
. "$f" "$@"
fi
done
}
hookdir=/lib/dracut/hooks
export hookdir
source_hook() {
local _dir
_dir=$1
shift
source_all "/lib/dracut/hooks/$_dir" "$@"
}
check_finished() {
local f
for f in "$hookdir"/initqueue/finished/*.sh; do
[ "$f" = "$hookdir/initqueue/finished/*.sh" ] && return 0
# shellcheck disable=SC1090
{ [ -e "$f" ] && (. "$f"); } || return 1
done
return 0
}
source_conf() {
local f
[ "$1" ] && [ -d "/$1" ] || return
# shellcheck disable=SC1090
for f in "/$1"/*.conf; do [ -e "$f" ] && . "$f"; done
}
die() {
{
echo "<24>dracut: FATAL: $*"
echo "<24>dracut: Refusing to continue"
} > /dev/kmsg
{
echo "warn dracut: FATAL: \"$*\""
echo "warn dracut: Refusing to continue"
} >> $hookdir/emergency/01-die.sh
[ -d /run/initramfs ] || mkdir -p -- /run/initramfs
: > /run/initramfs/.die
if getargbool 0 "rd.shell"; then
emergency_shell
else
source_hook "shutdown-emergency"
fi
if [ -n "$DRACUT_SYSTEMD" ]; then
systemctl --no-block --force halt
fi
exit 1
}
check_quiet() {
if [ -z "$DRACUT_QUIET" ]; then
DRACUT_QUIET="yes"
getargbool 0 rd.info -d -y rdinfo && DRACUT_QUIET="no"
getargbool 0 rd.debug -d -y rdinitdebug && DRACUT_QUIET="no"
getarg quiet || DRACUT_QUIET="yes"
a=$(getarg loglevel=)
[ -n "$a" ] && [ "$a" -ge 28 ] && DRACUT_QUIET="yes"
export DRACUT_QUIET
fi
}
check_occurances() {
# Count the number of times the character $ch occurs in $str
# Return 0 if the count matches the expected number, 1 otherwise
local str="$1"
local ch="$2"
local expected="$3"
local count=0
while [ "${str#*$ch}" != "${str}" ]; do
str="${str#*$ch}"
count=$((count + 1))
done
[ $count -eq "$expected" ]
}
incol2() {
debug_off
local check
local file="$1"
local str="$2"
[ -z "$file" ] && return 1
[ -z "$str" ] && return 1
while read -r _ check _ || [ -n "$check" ]; do
if [ "$check" = "$str" ]; then
debug_on
return 0
fi
done < "$file"
debug_on
return 1
}
udevsettle() {
# shellcheck disable=SC2086
udevadm settle --exit-if-exists=$hookdir/initqueue/work $settle_exit_if_exists
}
udevproperty() {
for i in "$@"; do
udevadm control --property="$i"
done
}
find_mount() {
local dev wanted_dev
wanted_dev="$(readlink -e -q "$1")"
while read -r dev _ || [ -n "$dev" ]; do
[ "$dev" = "$wanted_dev" ] && echo "$dev" && return 0
done < /proc/mounts
return 1
}
# usage: ismounted <mountpoint>
# usage: ismounted /dev/<device>
if command -v findmnt > /dev/null; then
ismounted() {
findmnt "$1" > /dev/null 2>&1
}
else
ismounted() {
if [ -b "$1" ]; then
find_mount "$1" > /dev/null && return 0
return 1
fi
while read -r _ m _ || [ -n "$m" ]; do
[ "$m" = "$1" ] && return 0
done < /proc/mounts
return 1
}
fi
# Create udev rule match for a device with its device name, or the udev property
# ID_FS_UUID or ID_FS_LABEL
#
# example:
# udevmatch LABEL=boot
# prints:
# ENV{ID_FS_LABEL}="boot"
#
# TODO: symlinks
udevmatch() {
case "$1" in
UUID=????????-????-????-????-???????????? | LABEL=* | PARTLABEL=* | PARTUUID=????????-????-????-????-????????????)
printf 'ENV{ID_FS_%s}=="%s"' "${1%%=*}" "${1#*=}"
;;
UUID=*)
printf 'ENV{ID_FS_UUID}=="%s*"' "${1#*=}"
;;
PARTUUID=*)
printf 'ENV{ID_FS_PARTUUID}=="%s*"' "${1#*=}"
;;
/dev/?*) printf -- 'KERNEL=="%s"' "${1#/dev/}" ;;
*) return 255 ;;
esac
}
label_uuid_to_dev() {
local _dev
_dev="${1#block:}"
case "$_dev" in
LABEL=*)
echo "/dev/disk/by-label/$(echo "${_dev#LABEL=}" | sed 's,/,\\x2f,g;s, ,\\x20,g')"
;;
PARTLABEL=*)
echo "/dev/disk/by-partlabel/$(echo "${_dev#PARTLABEL=}" | sed 's,/,\\x2f,g;s, ,\\x20,g')"
;;
UUID=*)
echo "/dev/disk/by-uuid/${_dev#UUID=}"
;;
PARTUUID=*)
echo "/dev/disk/by-partuuid/${_dev#PARTUUID=}"
;;
*)
echo "$_dev"
;;
esac
}
# Prints unique path for potential file inside specified directory. It consists
# of specified directory, prefix and number at the end which is incremented
# until non-existing file is found.
#
# funiq dir prefix
#
# example:
# # ls /mnt
# cdrom0 cdrom1
#
# # funiq /mnt cdrom
# /mnt/cdrom2
funiq() {
local dir="$1"
local prefix="$2"
local i=0
[ -d "${dir}" ] || return 1
while [ -e "${dir}/${prefix}$i" ]; do
i=$((i + 1)) || return 1
done
echo "${dir}/${prefix}$i"
}
# Creates unique directory and prints its path. It's using funiq to generate
# path.
#
# mkuniqdir subdir new_dir_name
mkuniqdir() {
local dir="$1"
local prefix="$2"
local retdir
local retdir_new
[ -d "${dir}" ] || mkdir -m 0755 -p "${dir}" || return 1
retdir=$(funiq "${dir}" "${prefix}") || return 1
until mkdir -m 0755 "${retdir}" 2> /dev/null; do
retdir_new=$(funiq "${dir}" "${prefix}") || return 1
[ "$retdir_new" = "$retdir" ] && return 1
retdir="$retdir_new"
done
echo "${retdir}"
}
# Copy the contents of SRC into DEST, merging the contents of existing
# directories (kinda like rsync, or cpio -p).
# Creates DEST if it doesn't exist. Overwrites files with the same names.
#
# copytree SRC DEST
copytree() {
local src="$1" dest="$2"
[ -d "$src" ] || return 1
mkdir -p "$dest" || return 1
dest=$(readlink -e -q "$dest") || return 1
(
cd "$src" || exit 1
cp -af . -t "$dest"
)
}
# Evaluates command for UUIDs either given as arguments for this function or all
# listed in /dev/disk/by-uuid. UUIDs doesn't have to be fully specified. If
# beginning is given it is expanded to all matching UUIDs. To pass full UUID to
# your command use '$___' as a place holder. Remember to escape '$'!
#
# foreach_uuid_until [ -p prefix ] command UUIDs
#
# prefix - string to put just before $___
# command - command to be evaluated
# UUIDs - list of UUIDs separated by space
#
# The function returns after *first successful evaluation* of the given command
# with status 0. If evaluation fails for every UUID function returns with
# status 1.
#
# Example:
# foreach_uuid_until "mount -U \$___ /mnt; echo OK; umount /mnt" \
# "01234 f512 a235567f-12a3-c123-a1b1-01234567abcb"
foreach_uuid_until() (
cd /dev/disk/by-uuid || return 1
[ "$1" = -p ] && local prefix="$2" && shift 2
local cmd="$1"
shift
local uuids_list="$*"
local uuid
local full_uuid
local ___
[ -n "${cmd}" ] || return 1
for uuid in ${uuids_list:-*}; do
for full_uuid in "${uuid}"*; do
[ -e "${full_uuid}" ] || continue
# shellcheck disable=SC2034
___="${prefix}${full_uuid}"
eval "${cmd}" && return 0
done
done
return 1
)
# Get kernel name for given device. Device may be the name too (then the same
# is returned), a symlink (full path), UUID (prefixed with "UUID=") or label
# (prefixed with "LABEL="). If just a beginning of the UUID is specified or
# even an empty, function prints all device names which UUIDs match - every in
# single line.
#
# NOTICE: The name starts with "/dev/".
#
# Example:
# devnames UUID=123
# May print:
# /dev/dm-1
# /dev/sdb1
# /dev/sdf3
devnames() {
local dev="$1"
local d
local names
case "$dev" in
UUID=*)
# shellcheck disable=SC2016
dev="$(foreach_uuid_until '! blkid -U $___' "${dev#UUID=}")" \
&& return 255
[ -z "$dev" ] && return 255
;;
LABEL=*) dev="$(blkid -L "${dev#LABEL=}")" || return 255 ;;
/dev/?*) ;;
*) return 255 ;;
esac
for d in $dev; do
names="$names
$(readlink -e -q "$d")" || return 255
done
echo "${names#
}"
}
usable_root() {
local _i
[ -d "$1" ] || return 1
for _i in "$1"/usr/lib*/ld-*.so "$1"/lib*/ld-*.so; do
[ -e "$_i" ] && return 0
done
for _i in proc sys dev; do
[ -e "$1"/$_i ] || return 1
done
return 0
}
inst_hook() {
local _hookname _unique _name _job _exe
while [ $# -gt 0 ]; do
case "$1" in
--hook)
_hookname="/$2"
shift
;;
--unique)
_unique="yes"
;;
--name)
_name="$2"
shift
;;
*)
break
;;
esac
shift
done
if [ -z "$_unique" ]; then
_job="${_name}$$"
else
_job="${_name:-$1}"
_job=${_job##*/}
fi
_exe=$1
shift
[ -x "$_exe" ] || _exe=$(command -v "$_exe")
if [ -n "$onetime" ]; then
{
# shellcheck disable=SC2016
echo '[ -e "$_job" ] && rm -f -- "$_job"'
echo "$_exe $*"
} > "/tmp/$$-${_job}.sh"
else
echo "$_exe $*" > "/tmp/$$-${_job}.sh"
fi
mv -f "/tmp/$$-${_job}.sh" "$hookdir/${_hookname}/${_job}.sh"
}
# inst_mount_hook <mountpoint> <prio> <name> <script>
#
# Install a mount hook with priority <prio>,
# which executes <script> as soon as <mountpoint> is mounted.
inst_mount_hook() {
local _prio="$2" _jobname="$3" _script="$4"
local _hookname
_hookname="mount-$(str_replace "$1" '/' '\\x2f')"
[ -d "$hookdir/${_hookname}" ] || mkdir -p "$hookdir/${_hookname}"
inst_hook --hook "$_hookname" --unique --name "${_prio}-${_jobname}" "$_script"
}
# wait_for_mount <mountpoint>
#
# Installs a initqueue-finished script,
# which will cause the main loop only to exit,
# if <mountpoint> is mounted.
wait_for_mount() {
local _name
_name="$(str_replace "$1" '/' '\\x2f')"
printf '. /lib/dracut-lib.sh\nismounted "%s"\n' "$1" \
>> "$hookdir/initqueue/finished/ismounted-${_name}.sh"
{
printf 'ismounted "%s" || ' "$1"
printf 'warn "\"%s\" is not mounted"\n' "$1"
} >> "$hookdir/emergency/90-${_name}.sh"
}
killproc() {
debug_off
local _exe
_exe="$(command -v "$1")"
local _sig=$2
local _i
[ -x "$_exe" ] || return 1
for _i in /proc/[0-9]*; do
[ "$_i" = "/proc/1" ] && continue
if [ -e "$_i"/_exe ] && [ "$_i/_exe" -ef "$_exe" ]; then
kill "$_sig" "${_i##*/}"
fi
done
debug_on
}
need_shutdown() {
: > /run/initramfs/.need_shutdown
}
wait_for_loginit() {
[ "$RD_DEBUG" = "yes" ] || return
[ -e /run/initramfs/loginit.pipe ] || return
debug_off
echo "DRACUT_LOG_END"
exec 0<> /dev/console 1<> /dev/console 2<> /dev/console
# wait for loginit
i=0
while [ $i -lt 10 ]; do
if [ ! -e /run/initramfs/loginit.pipe ]; then
j=$(jobs)
[ -z "$j" ] && break
[ -z "${j##*Running*}" ] || break
fi
sleep 0.1
i=$((i + 1))
done
if [ $i -eq 10 ]; then
kill %1 > /dev/null 2>&1
kill "$(while read -r line || [ -n "$line" ]; do echo "$line"; done < /run/initramfs/loginit.pid)"
fi
setdebug
rm -f -- /run/initramfs/loginit.pipe /run/initramfs/loginit.pid
}
# pidof version for root
if ! command -v pidof > /dev/null 2> /dev/null; then
pidof() {
debug_off
local _cmd
local _exe
local _rl
local _ret=1
local i
_cmd="$1"
if [ -z "$_cmd" ]; then
debug_on
return 1
fi
_exe=$(command -v "$1")
for i in /proc/*/exe; do
[ -e "$i" ] || continue
if [ -n "$_exe" ]; then
[ "$i" -ef "$_exe" ] || continue
else
_rl=$(readlink -f "$i")
[ "${_rl%/$_cmd}" != "$_rl" ] || continue
fi
i=${i%/exe}
echo "${i##/proc/}"
_ret=0
done
debug_on
return $_ret
}
fi
_emergency_shell() {
local _name="$1"
if [ -n "$DRACUT_SYSTEMD" ]; then
: > /.console_lock
echo "PS1=\"$_name:\\\${PWD}# \"" > /etc/profile
systemctl start dracut-emergency.service
rm -f -- /etc/profile
rm -f -- /.console_lock
else
debug_off
source_hook "$hook"
echo
/sbin/rdsosreport
echo 'You might want to save "/run/initramfs/rdsosreport.txt" to a USB stick or /boot'
echo 'after mounting them and attach it to a bug report.'
if ! RD_DEBUG='' getargbool 0 rd.debug -d -y rdinitdebug -d -y rdnetdebug; then
echo
echo 'To get more debug information in the report,'
echo 'reboot with "rd.debug" added to the kernel command line.'
fi
echo
echo 'Dropping to debug shell.'
echo
export PS1="$_name:\${PWD}# "
[ -e /.profile ] || : > /.profile
_ctty="$(RD_DEBUG='' getarg rd.ctty=)" && _ctty="/dev/${_ctty##*/}"
if [ -z "$_ctty" ]; then
_ctty=console
while [ -f /sys/class/tty/$_ctty/active ]; do
read -r _ctty < /sys/class/tty/$_ctty/active
_ctty=${_ctty##* } # last one in the list
done
_ctty=/dev/$_ctty
fi
[ -c "$_ctty" ] || _ctty=/dev/tty1
case "$(/usr/bin/setsid --help 2>&1)" in *--ctty*) CTTY="--ctty" ;; esac
setsid $CTTY /bin/sh -i -l 0<> $_ctty 1<> $_ctty 2<> $_ctty
fi
}
emergency_shell() {
local _ctty
set +e
local _rdshell_name="dracut" action="Boot" hook="emergency"
local _emergency_action
if [ "$1" = "-n" ]; then
_rdshell_name=$2
shift 2
elif [ "$1" = "--shutdown" ]; then
_rdshell_name=$2
action="Shutdown"
hook="shutdown-emergency"
shift 2
fi
echo
echo
warn "$*"
echo
_emergency_action=$(getarg rd.emergency)
[ -z "$_emergency_action" ] \
&& [ -e /run/initramfs/.die ] \
&& _emergency_action=halt
if getargbool 1 rd.shell -d -y rdshell || getarg rd.break -d rdbreak; then
_emergency_shell "$_rdshell_name"
else
source_hook "$hook"
warn "$action has failed. To debug this issue add \"rd.shell rd.debug\" to the kernel command line."
[ -z "$_emergency_action" ] && _emergency_action=halt
fi
case "$_emergency_action" in
reboot)
reboot || exit 1
;;
poweroff)
poweroff || exit 1
;;
halt)
halt || exit 1
;;
esac
}
# Retain the values of these variables but ensure that they are unexported
# This is a POSIX-compliant equivalent of bash's "export -n"
export_n() {
local var
local val
for var in "$@"; do
eval val=\$$var
unset $var
[ -n "$val" ] && eval "$var=\"$val\""
done
}
# returns OK if list1 contains all elements of list2, i.e. checks if list2 is a
# sublist of list1. An order and a duplication doesn't matter.
#
# $1 = separator
# $2 = list1
# $3 = list2
# $4 = ignore values, separated by $1
listlist() {
local _sep="$1"
local _list="${_sep}${2}${_sep}"
local _sublist="$3"
[ -n "$4" ] && local _iglist="${_sep}${4}${_sep}"
local IFS="$_sep"
local _v
[ "$_list" = "$_sublist" ] && return 0
for _v in $_sublist; do
if [ -n "$_v" ] && ! ([ -n "$_iglist" ] && strstr "$_iglist" "$_v"); then
strstr "$_list" "$_v" || return 1
fi
done
return 0
}
# returns OK if both lists contain the same values. An order and a duplication
# doesn't matter.
#
# $1 = separator
# $2 = list1
# $3 = list2
# $4 = ignore values, separated by $1
are_lists_eq() {
listlist "$1" "$2" "$3" "$4" && listlist "$1" "$3" "$2" "$4"
}
setmemdebug() {
if [ -z "$DEBUG_MEM_LEVEL" ]; then
DEBUG_MEM_LEVEL=$(getargnum 0 0 5 rd.memdebug)
export DEBUG_MEM_LEVEL
fi
}
setmemdebug
# parameters: func log_level prefix msg [trace_level:trace]...
make_trace_mem() {
local log_level prefix msg msg_printed
local trace trace_level trace_in_higher_levels insert_trace
msg=$1
shift
prefix='[debug_mem]'
log_level=$DEBUG_MEM_LEVEL
if [ -z "$log_level" ] || [ "$log_level" -le 0 ]; then
return
fi
# FIXME? useless echo?
# shellcheck disable=SC2116
msg=$(echo "$msg")
msg_printed=0
while [ $# -gt 0 ]; do
trace=${1%%:*}
trace_level=${trace%%+}
[ "$trace" != "$trace_level" ] && trace_in_higher_levels="yes"
trace=${1##*:}
if [ -z "$trace_level" ]; then
trace_level=0
fi
insert_trace=0
if [ -n "$trace_in_higher_levels" ]; then
if [ "$log_level" -ge "$trace_level" ]; then
insert_trace=1
fi
else
if [ "$log_level" -eq "$trace_level" ]; then
insert_trace=1
fi
fi
if [ $insert_trace -eq 1 ]; then
if [ $msg_printed -eq 0 ]; then
echo "$prefix $msg"
msg_printed=1
fi
show_memstats "$trace"
fi
shift
done
}
# parameters: type
show_memstats() {
case $1 in
shortmem)
while read -r line || [ -n "$line" ]; do
str_starts "$line" "MemFree" \
|| str_starts "$line" "Cached" \
|| str_starts "$line" "Slab" \
|| continue
echo "$line"
done < /proc/meminfo
;;
mem)
cat /proc/meminfo
;;
slab)
cat /proc/slabinfo
;;
iomem)
cat /proc/iomem
;;
esac
}
remove_hostonly_files() {
rm -fr /etc/cmdline /etc/cmdline.d/*.conf "$hookdir/initqueue/finished"
if [ -f /lib/dracut/hostonly-files ]; then
while read -r line || [ -n "$line" ]; do
[ -e "$line" ] || [ -h "$line" ] || continue
rm -f "$line"
done < /lib/dracut/hostonly-files
fi
}
# parameter: kernel_module [filesystem_name]
# returns OK if kernel_module is loaded
# modprobe fails if /lib/modules is not available (--no-kernel use case)
load_fstype() {
strstr "$(cat /proc/filesystems)" "${2:-$1}" || modprobe "$1"
}
-
Update
dracut
sudo dracut --force
-
Open
/etc/default/grub
sudo nano /etc/default/grub
addrootovl
toGRUB_CMDLINE_LINUX
, then changedom0_mem=max:
and add 90% of your available RAM (you can find this info with commandxl info | grep total_memory
in terminal). For example, if you have 15 GB of available RAM, writedom0_mem=max:13312
(if you have 64 GB of RAM, you can writedom0_mem=max:59392
).
Then updategrub
sudo grub2-mkconfig -o /boot/grub2/grub.cfg
You will now always launch dom0 in live mode!
Next, copy Whonix and Kicksecure templates and select varlibqubes
in the advanced settings. Then create new AppVMs based on the varlibqubes
templates and also select varlibqubes
in the advanced settings.
Templates in vm-pool will be in persistent mode (but metadata from dom0 will be deleted, so if you create a new qube in vm-pool in live mode, it will not be saved after rebooting).
To launch persistent mode for dom0 and perform updates or install new programs, do the following:
- Press
E
in grub menu and removerootovl
, then pressF10
. This will launch a persistent session for one time. - To completely remove live mode, remove
rootovl
in/etc/default/grub
and updategrub
in persistent mode.
You can choose not to create a vm-pool and only have dom0 and varlibqubes during Qubes installation, which will result in a complete disk amnesia in live mode. In this case, you will need to always launch persistent mode to save persistent files. This will require a significant amount of RAM for normal Qubes operation in live mode (at least 64 GB, but preferably 128 GB).
Now the Qubes team can easily port grub-live and add a live mode to the GRUB menu, just like it’s implemented in Whonix and Kicksecure! implement live boot by porting grub-live to Qubes - amnesia / non-persistent boot / anti-forensics · Issue #4982 · QubesOS/qubes-issues · GitHub You can write to the Qubes team on GitHub and suggest Patrick’s solution again (I’m not registered on GitHub)
So, I’ve conducted tests. Here are my results:
Overlayfs consumes significantly more memory when running virtual machines, but it provides more free space for downloading files.
Copying dom0 to zram run slowly and has less free disk space, but it hardly loads the memory - I couldn’t even get the memory to 100% load, even with 17 concurrently running varlibqubes qubes!! Overlayfs ran out of memory after loading 10 varlibqubes qubes.
These two modes can be used simultaneously, and you can choose one of them when launching the system, depending on your needs.