rpm-ostree/tests/common/libvm.sh
Colin Walters 21bdb54423 ci: Pull Fedora from api.ci
Currently we pull from quay.io and api.ci, and the Fedora registry
is being very flaky.  I set up a mirror on api.ci for now.
2020-02-27 19:24:31 +01:00

714 lines
20 KiB
Bash

# Source library for installed virtualized shell script tests
#
# Copyright (C) 2016 Jonathan Lebon <jlebon@redhat.com>
#
# This library is free software; you can redistribute it and/or
# modify it under the terms of the GNU Lesser General Public
# License as published by the Free Software Foundation; either
# version 2 of the License, or (at your option) any later version.
#
# This library is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# Lesser General Public License for more details.
#
# You should have received a copy of the GNU Lesser General Public
# License along with this library; if not, write to the
# Free Software Foundation, Inc., 59 Temple Place - Suite 330,
# Boston, MA 02111-1307, USA.
if test -z "${LIBTEST_SH:-}"; then
. ${commondir}/libtest.sh
fi
# prepares the VM and library for action
vm_setup() {
export VM=${VM:-vmcheck}
export SSH_CONFIG=${SSH_CONFIG:-${topsrcdir}/ssh-config}
SSHOPTS="-o User=root -o ControlMaster=auto \
-o ControlPath=/dev/shm/ssh-$VM-$(date +%s%N).sock \
-o ControlPersist=yes"
# If we're provided with an ssh-config, make sure we tell
# ssh to pick it up.
if [ -f "${SSH_CONFIG}" ]; then
SSHOPTS="${SSHOPTS} -F ${SSH_CONFIG}"
fi
export SSHOPTS
export SSH="ssh ${SSHOPTS} $VM"
export SCP="scp ${SSHOPTS}"
}
# prepares a fresh VM for action via `kola spawn`
vm_kola_spawn() {
local outputdir=$1; shift
exec 4> info.json
mkdir kola-ssh
setpriv --pdeathsig SIGKILL -- \
env MANTLE_SSH_DIR="$PWD/kola-ssh" kola spawn -p qemu-unpriv \
--qemu-image "${topsrcdir}/tests/vmcheck/image.qcow2" -v --idle \
--json-info-fd 4 --output-dir "$outputdir" &
# hack; need cleaner API for async kola spawn
while [ ! -s info.json ]; do sleep 1; done
local ssh_ip_port ssh_ip ssh_port
ssh_ip_port=$(jq -r .public_ip info.json)
ssh_ip=${ssh_ip_port%:*}
ssh_port=${ssh_ip_port#*:}
cat > ssh-config <<EOF
Host vmcheck
HostName ${ssh_ip}
Port ${ssh_port}
StrictHostKeyChecking no
UserKnownHostsFile /dev/null
EOF
SSH_CONFIG=$PWD/ssh-config
# XXX: should just have kola output the path to the socket
SSH_AUTH_SOCK=$(ls kola-ssh/agent.*)
export SSH_CONFIG SSH_AUTH_SOCK
# Hack around kola's Ignition config only setting up the core user; but we
# want to be able to ssh directly as root. We still want all the other goodies
# that kola injects in its Ignition config though, so we don't want to
# override it. `cosa run`'s merge semantics would do nicely.
ssh -o User=core -F "${SSH_CONFIG}" vmcheck 'sudo cp -RT {/home/core,/root}/.ssh'
vm_setup
# XXX: hack around https://github.com/systemd/systemd/issues/14328
vm_cmd systemctl mask --now systemd-logind
# Some tests expect the ref to be on `vmcheck`. We should drop that
# requirement, but for now let's just mangle the origin
local deployment_root
vm_cmd ostree refs --create vmcheck "$(vm_get_booted_csum)"
deployment_root=$(vm_get_deployment_root 0)
vm_cmd "sed -ie '/^refspec=/ s/=.*/=vmcheck/' ${deployment_root}.origin"
vm_cmd "sed -ie '/^baserefspec=/ s/=.*/=vmcheck/' ${deployment_root}.origin"
vm_cmd systemctl try-restart rpm-ostreed
# also move the default yum repos, we don't want em
vm_cmd mv /etc/yum.repos.d{,.bak}
vm_cmd mkdir /etc/yum.repos.d
}
# $1 - file to send
# $2 - destination path
vm_send() {
$SCP ${1} ${VM}:${2}
}
# $1 - destination path
vm_send_inline() {
f=$(mktemp -p $PWD)
cat > ${f}
vm_send ${f} ${1}
rm -f ${f}
}
vm_shell_inline() {
script=$(mktemp -p $PWD)
echo "set -xeuo pipefail" > ${script}
cat >> ${script}
vm_send ${script} /tmp/$(basename ${script})
rm -f ${script}
vm_cmd bash /tmp/$(basename ${script})
}
# rsync wrapper that sets up authentication
vm_raw_rsync() {
local rsyncopts="ssh -o User=root"
if [ -f "${SSH_CONFIG}" ]; then
rsyncopts="$rsyncopts -F '${SSH_CONFIG}'"
fi
rsync -az --no-owner --no-group -e "$rsyncopts" "$@"
}
vm_rsync() {
if ! test -f .vagrant/using_sshfs; then
pushd ${topsrcdir}
vm_raw_rsync --delete --exclude target/ --exclude bindgen-target/ --exclude .git/ . $VM:/var/roothome/sync
popd
fi
}
# run command in vm as user
# - $1 username
# - $@ command to run
vm_cmd_as() {
local user=$1; shift
# don't reuse root's ControlPath
local sshopts="-o User=$user"
if [ -f "${SSH_CONFIG}" ]; then
sshopts="$sshopts -F ${SSH_CONFIG}"
fi
ssh $sshopts $VM "$@"
}
# run command in vm
# - $@ command to run
vm_cmd() {
$SSH "$@"
}
# Delete anything which we might change between runs
vm_clean_caches() {
vm_cmd rm /ostree/repo/refs/heads/rpmostree/pkg/* -rf
}
# run rpm-ostree in vm
# - $@ args
vm_rpmostree() {
vm_cmd env ASAN_OPTIONS=detect_leaks=false rpm-ostree "$@"
}
# copy the test repo to the vm
# $1 - repo file mode: nogpgcheck (default), gpgcheck, skip (don't send)
vm_send_test_repo() {
mode=${1:-nogpgcheck}
# note we use -c here because we might be called twice within a second
vm_raw_rsync -c --delete ${test_tmpdir}/yumrepo $VM:/var/tmp/vmcheck
if [[ $mode == skip ]]; then
return
fi
cat > vmcheck.repo << EOF
[test-repo]
name=test-repo
baseurl=file:///var/tmp/vmcheck/yumrepo
EOF
if [[ $mode == gpgcheck ]]; then
cat >> vmcheck.repo <<EOF
gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-fedora-25-primary
EOF
else
assert_streq "$mode" nogpgcheck
echo "Enabling vmcheck.repo without GPG"
echo 'gpgcheck=0' >> vmcheck.repo
fi
vm_send vmcheck.repo /etc/yum.repos.d
}
# wait until ssh is available on the vm
# - $1 timeout in second (optional)
# - $2 previous bootid (optional)
vm_ssh_wait() {
local timeout=${1:-0}; shift
local old_bootid=${1:-}; shift
if ! vm_cmd true; then
echo "Failed to log into VM, retrying with debug:"
$SSH -o LogLevel=debug true || true
fi
while [ $timeout -gt 0 ]; do
if bootid=$(vm_get_boot_id 2>/dev/null); then
if [[ $bootid != $old_bootid ]]; then
# if this is a reboot, display some info about new boot
if [ -n "$old_bootid" ]; then
vm_rpmostree status
vm_rpmostree --version
fi
return 0
fi
fi
if test $(($timeout % 5)) == 0; then
echo "Still failed to log into VM, retrying for $timeout seconds"
fi
timeout=$((timeout - 1))
sleep 1
done
false "Timed out while waiting for SSH."
}
vm_get_boot_id() {
vm_cmd cat /proc/sys/kernel/random/boot_id
}
# Run a command in the VM that will cause a reboot
vm_reboot_cmd() {
vm_cmd sync
local bootid=$(vm_get_boot_id 2>/dev/null)
vm_cmd "$@" || :
vm_ssh_wait 120 $bootid
}
# reboot the vm
vm_reboot() {
vm_reboot_cmd systemctl reboot
}
# check that the given files/dirs exist on the VM
# - $@ files/dirs to check for
vm_has_files() {
for file in "$@"; do
if ! vm_cmd test -e $file; then
return 1
fi
done
}
# check that the packages are installed
# - $@ packages to check for
vm_has_packages() {
for pkg in "$@"; do
if ! vm_cmd rpm -q $pkg; then
return 1
fi
done
}
# retrieve info from a deployment
# - $1 index of deployment (or -1 for booted)
# - $2 key to retrieve
vm_get_deployment_info() {
local idx=$1
local key=$2
vm_rpmostree status --json | \
python3 -c "
import sys, json
deployments = json.load(sys.stdin)[\"deployments\"]
idx = $idx
if idx < 0:
for i, depl in enumerate(deployments):
if depl[\"booted\"]:
idx = i
if idx < 0:
print(\"Failed to determine currently booted deployment\")
exit(1)
if idx >= len(deployments):
print(\"Deployment index $idx is out of range\")
exit(1)
depl = deployments[idx]
if \"$key\" in depl:
data = depl[\"$key\"]
if type(data) is list:
print(\" \".join(data))
else:
print(data)
"
}
# retrieve the deployment root
# - $1 index of deployment
vm_get_deployment_root() {
local idx=$1
local csum=$(vm_get_deployment_info $idx checksum)
local serial=$(vm_get_deployment_info $idx serial)
local osname=$(vm_get_deployment_info $idx osname)
echo /ostree/deploy/$osname/deploy/$csum.$serial
}
# retrieve info from the booted deployment
# - $1 key to retrieve
vm_get_booted_deployment_info() {
vm_get_deployment_info -1 $1
}
# print the layered packages
vm_get_layered_packages() {
vm_get_booted_deployment_info packages
}
# print the requested packages
vm_get_requested_packages() {
vm_get_booted_deployment_info requested-packages
}
vm_get_local_packages() {
vm_get_booted_deployment_info requested-local-packages
}
# check that the packages are currently layered
# - $@ packages to check for
vm_has_layered_packages() {
local pkgs=$(vm_get_layered_packages)
for pkg in "$@"; do
if [[ " $pkgs " != *$pkg* ]]; then
return 1
fi
done
}
# check that the packages are currently requested
# - $@ packages to check for
vm_has_requested_packages() {
local pkgs=$(vm_get_requested_packages)
for pkg in "$@"; do
if [[ " $pkgs " != *$pkg* ]]; then
return 1
fi
done
}
vm_has_local_packages() {
local pkgs=$(vm_get_local_packages)
for pkg in "$@"; do
if [[ " $pkgs " != *$pkg* ]]; then
return 1
fi
done
}
vm_has_dormant_packages() {
vm_has_requested_packages "$@" && \
! vm_has_layered_packages "$@"
}
vm_get_booted_stateroot() {
vm_get_booted_deployment_info osname
}
# retrieve the checksum of the currently booted deployment
vm_get_booted_csum() {
vm_get_booted_deployment_info checksum
}
# retrieve the checksum of the pending deployment
vm_get_pending_csum() {
vm_get_deployment_info 0 checksum
}
# make multiple consistency checks on a test pkg
# - $1 package to check for
# - $2 either "present" or "absent"
vm_assert_layered_pkg() {
local pkg=$1; shift
local policy=$1; shift
set +e
vm_has_packages $pkg; pkg_in_rpmdb=$?
vm_has_layered_packages $pkg; pkg_is_layered=$?
vm_has_local_packages $pkg; pkg_is_layered_local=$?
vm_has_requested_packages $pkg; pkg_is_requested=$?
[ $pkg_in_rpmdb == 0 ] && \
( ( [ $pkg_is_layered == 0 ] &&
[ $pkg_is_requested == 0 ] ) ||
[ $pkg_is_layered_local == 0 ] ); pkg_present=$?
[ $pkg_in_rpmdb != 0 ] && \
[ $pkg_is_layered != 0 ] && \
[ $pkg_is_layered_local != 0 ] && \
[ $pkg_is_requested != 0 ]; pkg_absent=$?
set -e
if [ $policy == present ] && [ $pkg_present != 0 ]; then
vm_cmd rpm-ostree status
assert_not_reached "pkg $pkg is not present"
fi
if [ $policy == absent ] && [ $pkg_absent != 0 ]; then
vm_cmd rpm-ostree status
assert_not_reached "pkg $pkg is not absent"
fi
}
# Takes a list of `jq` expressions, each of which should evaluate to a boolean,
# and asserts that they are true.
vm_assert_status_jq() {
vm_rpmostree status --json > status.json
assert_jq status.json "$@"
}
vm_pending_is_staged() {
vm_rpmostree status --json > status-staged.json
local rc=1
if jq -e ".deployments[0][\"staged\"]" < status-staged.json; then
rc=0
fi
rm -f status-staged.json
return $rc
}
# Like build_rpm, but also sends it to the VM
vm_build_rpm() {
build_rpm "$@"
vm_send_test_repo
}
# Like uinfo_cmd, but also sends it to the VM
vm_uinfo() {
uinfo_cmd "$@"
vm_send_test_repo
}
# Like vm_build_rpm but takes a yumrepo mode
vm_build_rpm_repo_mode() {
mode=$1; shift
build_rpm "$@"
vm_send_test_repo $mode
}
vm_build_selinux_rpm() {
build_selinux_rpm "$@"
vm_send_test_repo
}
vm_get_journal_cursor() {
vm_cmd journalctl -o json -n 1 | jq -r '.["__CURSOR"]'
}
# Wait for a message logged after $cursor matching a regexp to appear
# $1 - cursor
# $2 - regex to wait for
vm_wait_content_after_cursor() {
from_cursor=$1; shift
regex=$1; shift
vm_shell_inline <<EOF
tmpf=\$(mktemp /var/tmp/journal.XXXXXX)
for x in \$(seq 60); do
journalctl -u rpm-ostreed --after-cursor "${from_cursor}" > \${tmpf}
if grep -q -e "${regex}" \${tmpf}; then
exit 0
else
cat \${tmpf}
sleep 1
fi
done
echo "timed out after 60s" 1>&2
journalctl -u rpm-ostreed --after-cursor "${from_cursor}" | tail -100
exit 1
EOF
}
# Minor helper that makes sure to get quoting right
vm_get_journal_after_cursor() {
from_cursor=$1; shift
to_file=$1; shift
# add an extra helping of quotes for hungry ssh
vm_cmd journalctl --after-cursor "'$from_cursor'" > $to_file
}
vm_assert_journal_has_content() {
from_cursor=$1; shift
vm_get_journal_after_cursor $from_cursor tmp-journal.txt
assert_file_has_content tmp-journal.txt "$@"
rm -f tmp-journal.txt
}
# usage: <podman args> -- <container args>
vm_run_container() {
local podman_args=
while [ $# -ne 0 ]; do
local arg=$1; shift
if [[ $arg == -- ]]; then
break
fi
podman_args="$podman_args $arg"
done
[ $# -ne 0 ] || fatal "No container args provided"
# just automatically always share dnf cache so we don't redownload each time
# (use -n so this ssh invocation doesn't consume stdin)
vm_cmd -n mkdir -p /var/cache/dnf
# This uses a mirror since the Fedora registry is very flaky
# https://pagure.io/releng/issue/9282
vm_cmd podman run --rm -v /var/cache/dnf:/var/cache/dnf:z $podman_args \
registry.svc.ci.openshift.org/coreos/fedora:31 "$@"
}
# $1 - service name
# $2 - dir to serve
# $3 - port to serve on
vm_start_httpd() {
local name=$1; shift
local dir=$1; shift
local port=$1; shift
vm_cmd podman rm -f $name || true
vm_run_container --net=host -d --name $name --privileged \
-v $dir:/srv --workdir /srv -- \
python3 -m http.server $port
# NB: the EXIT trap is used by libtest, but not the ERR trap
trap "vm_stop_httpd $name" ERR
set -E # inherit trap
# Ideally systemd-run would support .socket units or something
vm_cmd 'while ! curl --head http://127.0.0.1:8888 &>/dev/null; do sleep 1; done'
}
# $1 - service name
vm_stop_httpd() {
local name=$1; shift
vm_cmd podman rm -f $name
set +E
trap - ERR
}
# start up an ostree server to be used as an http remote
vm_ostreeupdate_prepare_repo() {
# Really testing this like a user requires a remote ostree server setup.
# Let's start by setting up the repo.
REMOTE_OSTREE=/ostree/repo/tmp/vmcheck-remote
vm_cmd mkdir -p $REMOTE_OSTREE
vm_cmd ostree init --repo=$REMOTE_OSTREE --mode=archive
vm_start_httpd ostree_server $REMOTE_OSTREE 8888
}
# this is split out for the sole purpose of making iterating easier when hacking
# (see below for more details)
_init_updated_rpmmd_repo() {
vm_build_rpm base-pkg-foo version 1.4 release 8 # upgraded
vm_build_rpm base-pkg-bar version 0.9 release 3 # downgraded
vm_build_rpm base-pkg-boo version 3.7 release 2.11 # added
vm_uinfo add VMCHECK-ENH enhancement
vm_uinfo add VMCHECK-SEC-NONE security none
vm_uinfo add VMCHECK-SEC-LOW security low
vm_uinfo add VMCHECK-SEC-CRIT security critical
vm_build_rpm base-pkg-enh version 2.0 uinfo VMCHECK-ENH
vm_build_rpm base-pkg-sec-none version 2.0 uinfo VMCHECK-SEC-NONE
vm_build_rpm base-pkg-sec-low version 2.0 uinfo VMCHECK-SEC-LOW
vm_build_rpm base-pkg-sec-crit version 2.0 uinfo VMCHECK-SEC-CRIT
vm_uinfo add-ref VMCHECK-SEC-LOW 1 http://example.com/vuln1 "CVE-12-34 vuln1"
vm_uinfo add-ref VMCHECK-SEC-LOW 2 http://example.com/vuln2 "CVE-12-34 vuln2"
vm_uinfo add-ref VMCHECK-SEC-LOW 3 http://example.com/vuln3 "CVE-56-78 CVE-90-12 vuln3"
vm_uinfo add-ref VMCHECK-SEC-LOW 4 http://example.com/vuln4 "CVE-12-JUNK CVE-JUNK vuln4"
}
# Start up a remote, and create two new commits (v1 and v2) which contain new
# pkgs. The 'vmcheck' ref on the remote is set at v1. You can then make a new
# update appear later using "vm_ostreeupdate_create v2".
vm_ostreeupdate_prepare() {
# first, let's make sure the timer is disabled so it doesn't mess up with our
# tests
vm_cmd systemctl disable --now rpm-ostreed-automatic.timer
# Prepare an OSTree repo with updates
vm_ostreeupdate_prepare_repo
# (delete ref but don't prune for easier debugging)
vm_cmd ostree refs --repo=$REMOTE_OSTREE vmcheck --delete
# now let's build some pkgs that we'll jury-rig into a base update
# this whole block can be commented out (except the init_updated_rpmmd_repo
# call) after the first run for a speed-up when iterating locally
vm_build_rpm base-pkg-foo version 1.4 release 7
vm_build_rpm base-pkg-bar
vm_build_rpm base-pkg-baz version 1.1 release 1
vm_build_rpm base-pkg-enh
vm_build_rpm base-pkg-sec-none
vm_build_rpm base-pkg-sec-low
vm_build_rpm base-pkg-sec-crit
vm_rpmostree install base-pkg-{foo,bar,baz,enh,sec-{none,low,crit}}
vm_ostreeupdate_lift_commit $(vm_get_pending_csum) v1
vm_rpmostree cleanup -p
# ok, we don't need those RPMs anymore since they're part of the base
rm -rf $test_tmpdir/yumrepo
# create new versions of those RPMs that we install in v2; we keep the repo
# around since that's where e.g. advisories are stored too when analyzing
# the v2 ostree update
_init_updated_rpmmd_repo
vm_rpmostree install base-pkg-{foo,bar,boo,enh,sec-{none,low,crit}}
vm_ostreeupdate_lift_commit $(vm_get_pending_csum) v2
vm_rpmostree cleanup -p
vm_ostreeupdate_create v1
vm_cmd ostree remote add vmcheckmote --no-gpg-verify http://localhost:8888/
vm_rpmostree reload
}
vm_ostreeupdate_prepare_reboot() {
vm_ostreeupdate_prepare
vm_rpmostree rebase vmcheckmote:vmcheck
vm_reboot
vm_rpmostree cleanup -pr
vm_assert_status_jq ".deployments[0][\"origin\"] == \"vmcheckmote:vmcheck\"" \
".deployments[0][\"booted\"]" \
".deployments[0][\"version\"] == \"v1\""
vm_rpmostree status > status.txt
assert_file_has_content_literal status.txt 'AutomaticUpdates: disabled'
# start it up again since we rebooted
vm_start_httpd ostree_server $REMOTE_OSTREE 8888
}
vm_change_update_policy() {
policy=$1; shift
vm_shell_inline <<EOF
cp /usr/etc/rpm-ostreed.conf /etc
echo -e "[Daemon]\nAutomaticUpdatePolicy=$policy" > /etc/rpm-ostreed.conf
rpm-ostree reload
EOF
}
# APIs to build up a history on the server. Rather than wasting time
# composing trees for real, we just use client package layering to create new
# trees that we then "lift" into the server before cleaning them up client-side.
# steal a commit from the system repo and tag it as a new version
vm_ostreeupdate_lift_commit() {
checksum=$1; shift
# ostree doesn't support tags, so just shove it in a branch
branch=vmcheck_tmp/$1; shift
vm_cmd ostree pull-local --repo=$REMOTE_OSTREE --disable-fsync \
/ostree/repo $checksum
vm_cmd ostree --repo=$REMOTE_OSTREE refs $branch --delete
vm_cmd ostree --repo=$REMOTE_OSTREE refs $checksum --create=$branch
}
_commit_and_inject_pkglist() {
local version=$1; shift
local src_ref=$1; shift
vm_cmd ostree commit --repo=$REMOTE_OSTREE -b vmcheck --fsync=no \
--tree=ref=$src_ref --add-metadata-string=version=$version
vm_rpmostree testutils inject-pkglist $REMOTE_OSTREE vmcheck
}
# use a previously stolen commit to create an update on our vmcheck branch,
# complete with version string and pkglist metadata
vm_ostreeupdate_create() {
version=$1; shift
_commit_and_inject_pkglist $version vmcheck_tmp/$version
}
# create a new no-op update with version metadata $1
vm_ostreeupdate_create_noop() {
version=$1; shift
_commit_and_inject_pkglist $version vmcheck
}
# takes a layered commit, and makes it into a base
vm_ostree_repo_commit_layered_as_base() {
local repo=$1; shift
local from_rev=$1; shift
local to_ref=$1; shift
local d=$repo/tmp/vmcheck_commit.tmp
rm -rf $d
vm_cmd ostree checkout --repo=$repo -H --fsync=no $from_rev $d
# need to update the base rpmdb
vm_cmd mkdir -p $d/usr/lib/sysimage/rpm-ostree-base-db
vm_cmd rsync -qa --delete $d/usr/share/rpm/ $d/usr/lib/sysimage/rpm-ostree-base-db
vm_cmd ostree commit --repo=$repo -b $to_ref --link-checkout-speedup --fsync=no --consume $d
# and inject pkglist metadata
vm_rpmostree testutils inject-pkglist $repo $to_ref >/dev/null
}
vm_ostree_commit_layered_as_base() {
vm_ostree_repo_commit_layered_as_base /ostree/repo "$@"
}
vm_status_watch_start() {
rm -rf status-watch.txt
while sleep 1; do
vm_rpmostree status >> status-watch.txt
done &
_status_watch_pid=$!
# NB: the EXIT trap is used by libtest, but not the ERR trap
trap "kill $_status_watch_pid" ERR
set -E # inherit trap
}
vm_status_watch_check() {
[ -n "${_status_watch_pid:-}" ]
kill $_status_watch_pid
_status_watch_pid=
set +E
[ -f status-watch.txt ]
assert_file_has_content_literal status-watch.txt "$@"
rm -rf status-watch.txt
}