1
0
mirror of https://passt.top/passt synced 2024-12-22 13:45:32 +00:00
passt/doc/demo.sh

262 lines
7.7 KiB
Bash
Raw Normal View History

passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#!/bin/sh -e
#
# SPDX-License-Identifier: GPL-2.0-or-later
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#
# PASST - Plug A Simple Socket Transport
#
# demo.sh - Set up namespace with pasta, start qemu and passt, step by step
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#
# Copyright (c) 2020-2022 Red Hat GmbH
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
# Author: Stefano Brivio <sbrivio@redhat.com>
# mbuto_profile() - Profile for https://mbuto.sh/, sourced, return after setting
mbuto_profile() {
PROGS="${PROGS:-ash,dash,bash ip mount ls ln chmod insmod mkdir sleep
lsmod modprobe find grep mknod mv rm umount iperf3 dhclient cat
hostname chown socat dd strace ping killall sysctl wget,curl}"
KMODS="${KMODS:- virtio_net virtio_pci}"
LINKS="${LINKS:-
ash,dash,bash /init
ash,dash,bash /bin/sh}"
DIRS="${DIRS} /tmp /sbin /var/log /var/run /var/lib"
# shellcheck disable=SC2016
FIXUP="${FIXUP}"'
cat > /sbin/dhclient-script << EOF
#!/bin/sh
[ -n "\${new_interface_mtu}" ] && ip link set dev \${interface} mtu \${new_interface_mtu}
[ -n "\${new_ip_address}" ] && ip addr add \${new_ip_address}/\${new_subnet_mask} dev \${interface}
[ -n "\${new_routers}" ] && for r in \${new_routers}; do ip route add default via \${r} dev \${interface}; done
[ -n "\${new_domain_name_servers}" ] && for d in \${new_domain_name_servers}; do echo "nameserver \${d}" >> /etc/resolv.conf; done
[ -n "\${new_domain_name}" ] && echo "search \${new_domain_name}" >> /etc/resolf.conf
[ -n "\${new_domain_search}" ] && (printf "search"; for d in \${new_domain_search}; do printf " %s" "\${d}"; done; printf "\n") >> /etc/resolv.conf
[ -n "\${new_ip6_address}" ] && ip addr add \${new_ip6_address}/\${new_ip6_prefixlen} dev \${interface}
[ -n "\${new_dhcp6_name_servers}" ] && for d in \${new_dhcp6_name_servers}; do echo "nameserver \${d}%\${interface}" >> /etc/resolv.conf; done
[ -n "\${new_dhcp6_domain_search}" ] && (printf "search"; for d in \${new_dhcp6_domain_search}; do printf " %s" "\${d}"; done; printf "\n") >> /etc/resolv.conf
[ -n "\${new_host_name}" ] && hostname "\${new_host_name}"
exit 0
EOF
chmod 755 /sbin/dhclient-script
mkdir -p /etc/dhcp
echo "timeout 3;" > /etc/dhcp/dhclient.conf
ln -s /sbin /usr/sbin
:> /etc/fstab
echo
echo "The guest is up and running. Networking is not configured yet:"
echo
echo "$ ip address show"
echo
ip address show
echo
echo "...the next step will take care of that."
read x
echo "$ ip link set dev eth0 up"
ip link set dev eth0 up
sleep 3
echo "$ dhclient -4 -1 -sf /sbin/dhclient-script"
dhclient -4 -1 -sf /sbin/dhclient-script
sleep 2
echo "$ dhclient -6 -1 -sf /sbin/dhclient-script"
dhclient -6 -1 -sf /sbin/dhclient-script
sleep 2
echo
echo "$ ip address show"
ip address show
echo
echo "$ ip route show"
ip route show
echo
echo "...done."
read x
echo "Checking connectivity..."
echo
echo "$ wget --no-check-certificate https://passt.top/ || curl -k https://passt.top/"
wget --no-check-certificate https://passt.top/ || curl -k https://passt.top/
echo "...done."
read x
echo "An interactive shell will start now. When you are done,"
echo "use ^C to terminate the guest and exit the demo."
echo
sh +m
'
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
}
[ "${0##*/}" = "mbuto" ] && mbuto_profile && return 0
# cmd() - Show command being executed, then run it
# $@: Command and arguments
cmd() {
echo "$" "$@"
"$@"
}
# next() - Go to next step once a key is pressed, sets $KEY
next() {
KEY="$(dd ibs=1 count=1 2>/dev/null)"
echo
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
}
# cleanup() - Terminate pasta and passt, clean up, restore TTY settings
# shellcheck disable=SC2317
cleanup() {
[ -f "${DEMO_DIR}/pasta.pid" ] && kill "$(cat "${DEMO_DIR}/pasta.pid")"
[ -f "${DEMO_DIR}/passt.pid" ] && kill "$(cat "${DEMO_DIR}/passt.pid")"
rm -rf "${DEMO_DIR}" 2>/dev/null
[ -n "${STTY_BACKUP}" ] && stty "${STTY_BACKUP}"
}
# start_pasta_delayed() - Start pasta once $DEMO_DIR/pasta.wait is gone
start_pasta_delayed() {
trap '' EXIT
while [ -d "${DEMO_DIR}/pasta.wait" ]; do sleep 1; done
cmd pasta --config-net -P "${DEMO_DIR}/pasta.pid" \
"$(cat "${DEMO_DIR}/shell.pid")"
echo
echo "...pasta is running."
exit 0
}
# start_mbuto_delayed() - Run mbuto once, and if, $DEMO_DIR/mbuto.wait is gone
start_mbuto_delayed() {
trap '' EXIT
while [ -d "${DEMO_DIR}/mbuto.wait" ]; do sleep 1; done
cmd git -C "${DEMO_DIR}" clone git://mbuto.sh/mbuto
echo
cmd "${DEMO_DIR}/mbuto/mbuto" \
-p "$(realpath "${0}")" -f "${DEMO_DIR}/demo.img"
mkdir "${DEMO_DIR}/mbuto.done"
exit 0
}
# into_ns() - Entry point and demo script to run inside new namespace
into_ns() {
echo "We're in the new namespace now."
next
echo "Networking is not configured yet:"
echo
cmd ip link show
echo
cmd ip address show
next
echo "Let's run pasta(1) to configure networking and connect this"
echo "namespace. Note that we'll run pasta(1) from outside this"
echo "namespace, because it needs to implement the connection between"
echo "this namespace and the initial (\"outer\") one."
next
echo "$$" > "${DEMO_DIR}/shell.pid"
rmdir "${DEMO_DIR}/pasta.wait"
next
echo "Back to the new namespace, networking is configured:"
echo
cmd ip link show
echo
cmd ip address show
next
echo "and we can now start passt(1), to connect this namespace to a"
echo "virtual machine. If you want to start a shell in this namespace,"
echo "press 's' now. Exiting the shell will resume the script."
next
[ "${KEY}" = "s" ] && ${SHELL}
cmd passt -P "${DEMO_DIR}/passt.pid"
echo
echo "...passt is running."
next
__arch="$(uname -m)"
case ${__arch} in
x86_64)
__arch_supported=1
__qemu_arch="qemu-system-x86_64 -M pc,accel=kvm:tcg"
;;
*)
__arch_supported=0
;;
esac
if [ "${__arch_supported}" -eq 1 ]; then
echo "We're ready to start a virtual machine now. This script"
echo "can download and use mbuto (https://mbuto.sh/) to build a"
echo "basic initramfs image. Otherwise, press 's' to skip this"
echo "step, and start an existing virtual machine yourself."
echo "You'll need to use the qrap(1) wrapper, with qemu options"
echo "as reported above."
next
else
echo "This script doesn't know, yet, how to run a virtual"
echo "machine on your architecture (${__arch}). Please start an"
echo "existing virtual machine yourself, using the qrap(1)"
echo "wrapper, with qemu options as reported above."
echo
fi
if [ "${__arch_supported}" -eq 0 ] || [ "${KEY}" = "s" ]; then
echo "Start a virtual machine now. Pressing any key here will"
echo "terminate passt and pasta, and clean up."
next
exit 0
fi
rmdir "${DEMO_DIR}/mbuto.wait"
while [ ! -d "${DEMO_DIR}/mbuto.done" ]; do sleep 1; done
echo "The guest image is ready. The next step will start the guest."
echo "Use ^C to terminate it."
next
# shellcheck disable=SC2086
cmd qrap 5 ${__qemu_arch} \
-smp "$(nproc)" -m 1024 \
-nographic -serial stdio -nodefaults -no-reboot -vga none \
-initrd "${DEMO_DIR}/demo.img" \
-kernel "/boot/vmlinuz-$(uname -r)" -append "console=ttyS0" \
-net socket,fd=5 -net nic,model=virtio || :
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
}
STTY_BACKUP="$(stty -g)"
stty -icanon
trap cleanup EXIT INT
[ "${1}" = "into_ns" ] && into_ns && exit 0
DEMO_DIR="$(mktemp -d)"
mkdir "${DEMO_DIR}/pasta.wait"
mkdir "${DEMO_DIR}/mbuto.wait"
echo "This script sets up a network and user namespace using pasta(1), then"
echo "starts a virtual machine in it, connected via passt(1), pausing at every"
echo "step. Press any key to go to the next step."
next
echo "Let's create the network and user namespace, first. This could be done"
echo "with pasta(1) itself (just issue \`pasta\`), but for the sake of this"
echo "script we'll create it first with unshare(1), and run the next steps"
echo "of this script from there."
next
start_pasta_delayed &
start_mbuto_delayed &
DEMO_DIR="${DEMO_DIR}" cmd unshare -rUn "${0}" into_ns
exit 0