270 lines
13 KiB
Bash
270 lines
13 KiB
Bash
#!/bin/sh
|
||
# MoShiro 容器节点（Agent、Probe）的宿主机更新器。用法：moshiro-update <check|apply> <组件> 组件 = agent 或 probe
|
||
#
|
||
# 信任模型（防篡改、防中间人）：
|
||
# - 只接受用发布私钥签名的清单（ECDSA P-256 / SHA-256，对清单文件的原始字节签名）；公钥放在本机，面板和 Registry 都改不了它。
|
||
# - 清单里写的是镜像的 digest：拉取时用 repo@sha256:…，内容寻址，Registry 或链路上的人换不出另一个镜像。
|
||
# - 镜像仓库、组件、服务名都来自本机的配置（/etc/moshiro-update/<组件>.conf），清单里写什么都不会改变拉哪个仓库。
|
||
# - 清单的 seq 必须比已安装的大（防止旧清单被重放回滚）。
|
||
# - 业务进程（Agent、Probe）只能往数据目录里写「清单」和「请求」文件，更新由这个宿主机脚本完成，业务进程不执行任何命令。
|
||
# Shell snippets below execute inside helper containers.
|
||
# shellcheck disable=SC2016
|
||
set -eu
|
||
umask 077
|
||
|
||
CMD="${1:-}"
|
||
COMPONENT="${2:-}"
|
||
case "$CMD" in check|apply) ;; *) echo "usage: moshiro-update <check|apply> <agent|probe|backend>" >&2; exit 2 ;; esac
|
||
case "$COMPONENT" in agent|probe|backend) ;; *) echo "component must be agent, probe or backend" >&2; exit 2 ;; esac
|
||
|
||
if [ "$(id -u)" -eq 0 ]; then
|
||
DEFAULT_CONF_DIR=/etc/moshiro-update
|
||
else
|
||
DEFAULT_CONF_DIR="${XDG_CONFIG_HOME:-$HOME/.config}/moshiro"
|
||
fi
|
||
CONF="${MOSHIRO_UPDATE_CONF_DIR:-$DEFAULT_CONF_DIR}/$COMPONENT.conf"
|
||
[ -r "$CONF" ] || { echo "missing $CONF" >&2; exit 2; }
|
||
# 配置：DATA_DIR（容器里 /etc/moshiro 对应的宿主机目录）、IMAGE_REPO、UNIT、PUBKEY、POLICY（auto|manual）、AUTHFILE（可选）、
|
||
# STATE_DIR（已安装版本的记录）、ENV_FILE（容器单元读取的 IMAGE=… 文件）
|
||
# shellcheck disable=SC1090
|
||
. "$CONF"
|
||
: "${IMAGE_REPO:?}" "${UNIT:?}" "${PUBKEY:?}" "${ENV_FILE:?}"
|
||
POLICY="${POLICY:-manual}"
|
||
STATE_DIR="${STATE_DIR:-/var/lib/moshiro-update}"
|
||
ENGINE="${ENGINE:-podman}"
|
||
SERVICE_SCOPE="${SERVICE_SCOPE:-system}"
|
||
DATA_VOLUME="${DATA_VOLUME:-}"
|
||
STATUS_VOLUME="${STATUS_VOLUME:-moshiro-$COMPONENT-status}"
|
||
CHANNEL="${CHANNEL:-}"
|
||
HEALTH_URL="${HEALTH_URL:-}"
|
||
if [ "$COMPONENT" = backend ]; then
|
||
[ "$POLICY" = manual ] || { echo "Backend updates require manual confirmation" >&2; exit 2; }
|
||
case "$CHANNEL" in dev|release) ;; *) echo "Backend needs CHANNEL=dev or release" >&2; exit 2 ;; esac
|
||
[ -n "$HEALTH_URL" ] || { echo "Backend needs a local HEALTH_URL" >&2; exit 2; }
|
||
[ -n "$DATA_VOLUME" ] || { echo "Backend backup needs DATA_VOLUME" >&2; exit 2; }
|
||
fi
|
||
case "$ENGINE" in podman|docker) ;; *) echo "unsupported engine: $ENGINE" >&2; exit 2 ;; esac
|
||
case "$SERVICE_SCOPE" in system|user) ;; *) echo "unsupported service scope: $SERVICE_SCOPE" >&2; exit 2 ;; esac
|
||
service_ctl() {
|
||
if [ "$SERVICE_SCOPE" = user ]; then systemctl --user "$@"; else systemctl "$@"; fi
|
||
}
|
||
pull_image() {
|
||
if [ -n "${AUTHFILE:-}" ]; then
|
||
if [ "$ENGINE" = podman ]; then
|
||
"$ENGINE" pull --quiet --authfile "$AUTHFILE" "$1"
|
||
else
|
||
mkdir -m 700 "$WORK/docker-auth"
|
||
cp "$AUTHFILE" "$WORK/docker-auth/config.json"
|
||
DOCKER_CONFIG="$WORK/docker-auth" "$ENGINE" pull --quiet "$1"
|
||
fi
|
||
else
|
||
"$ENGINE" pull --quiet "$1"
|
||
fi
|
||
}
|
||
# 业务进程（容器里的 Agent、Probe）可写的目录只有 update/：它在里面放 offer.manifest、offer.sig、apply.request。
|
||
# 更新器在容器外运行：只读取普通文件（不跟符号链接）、读取之后先复制到自己的临时目录再验证和使用（防止验证之后被换掉），
|
||
# 状态目录在 rootless 容器里另行只读挂载；更新器配置和已安装版本状态不挂入容器。
|
||
if [ -n "$DATA_VOLUME" ]; then
|
||
# Read/write volume contents via short-lived engine-managed helpers, without host mountpoint access.
|
||
: "${HELPER_IMAGE:?}"
|
||
"$ENGINE" volume inspect "$DATA_VOLUME" >/dev/null
|
||
"$ENGINE" volume inspect "$STATUS_VOLUME" >/dev/null
|
||
else
|
||
: "${DATA_DIR:?}"
|
||
UPDATE_DIR="$DATA_DIR/update"
|
||
STATUS_DIR="$DATA_DIR/update-status"
|
||
STATUS="$STATUS_DIR/status.json"
|
||
mkdir -p "$STATUS_DIR"
|
||
chmod 755 "$STATUS_DIR"
|
||
fi
|
||
INSTALLED="$STATE_DIR/$COMPONENT.installed" # 两行：seq、digest
|
||
mkdir -p "$STATE_DIR"
|
||
# Serialize check/apply path events for this component.
|
||
exec 9>"$STATE_DIR/$COMPONENT.lock"
|
||
flock -w 290 9 || { echo "another update is still running" >&2; exit 1; }
|
||
WORK="$(mktemp -d)"
|
||
trap 'rm -rf "$WORK"' EXIT
|
||
OFFER_MANIFEST="$WORK/offer.manifest"
|
||
OFFER_SIG="$WORK/offer.sig"
|
||
|
||
# 把业务进程放的清单和签名复制到临时目录；不是普通文件（缺失、符号链接）就当作没有
|
||
fetch_offer() {
|
||
rm -f "$OFFER_MANIFEST" "$OFFER_SIG"
|
||
for name in offer.manifest offer.sig; do
|
||
if [ -n "$DATA_VOLUME" ]; then
|
||
"$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||
-v "$DATA_VOLUME:/etc/moshiro:ro" --entrypoint /bin/sh "$HELPER_IMAGE" \
|
||
-c 'f="/etc/moshiro/update/$1"; if [ -f "$f" ] && [ ! -L "$f" ]; then head -c 8192 "$f"; fi' sh "$name" > "$WORK/$name"
|
||
continue
|
||
fi
|
||
f="$UPDATE_DIR/$name"
|
||
if [ -f "$f" ] && [ ! -L "$f" ]; then head -c 8192 "$f" > "$WORK/$name"; fi
|
||
done
|
||
}
|
||
|
||
now() { date +%s; }
|
||
|
||
# 状态文件：更新器写，Agent / Probe 从只读挂载中读取并随心跳上报。
|
||
write_status() { # state available_json error
|
||
tmp="$WORK/status.json"
|
||
printf '{"state":"%s","current":"%s","available":%s,"error":"%s","at":%s}\n' \
|
||
"$1" "$(current_digest)" "${2:-null}" "$(printf '%s' "${3:-}" | tr -d '"\\\n' | cut -c1-200)" "$(now)" > "$tmp"
|
||
if [ -n "$DATA_VOLUME" ]; then
|
||
# 状态卷的属主可能是容器里的 app（1654）；root 丢掉全部能力后写不进去，所以只加回 DAC_OVERRIDE
|
||
"$ENGINE" run --pull=never --rm -i --user 0 --network none --read-only --cap-drop ALL --cap-add DAC_OVERRIDE --security-opt no-new-privileges \
|
||
-v "$STATUS_VOLUME:/status" --entrypoint /bin/sh "$HELPER_IMAGE" \
|
||
-c 'tmp=$(mktemp /status/.status.XXXXXX); cat > "$tmp"; chmod 644 "$tmp"; mv -f "$tmp" /status/status.json' < "$tmp"
|
||
else
|
||
chmod 644 "$tmp"
|
||
destination="$(mktemp "$STATUS_DIR/.status.XXXXXX")"
|
||
cp "$tmp" "$destination"
|
||
chmod 644 "$destination"
|
||
mv -f "$destination" "$STATUS"
|
||
fi
|
||
}
|
||
|
||
current_digest() { [ -r "$INSTALLED" ] && sed -n 2p "$INSTALLED" || echo ""; }
|
||
installed_seq() { [ -r "$INSTALLED" ] && sed -n 1p "$INSTALLED" || echo 0; }
|
||
|
||
# Backend 的请求带用户确认的 seq；复制以后再验清单，避免后来到达的 tag 被自动套用。
|
||
fetch_request() {
|
||
if [ -n "$DATA_VOLUME" ]; then
|
||
"$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||
-v "$DATA_VOLUME:/etc/moshiro:ro" --entrypoint /bin/sh "$HELPER_IMAGE" \
|
||
-c 'f=/etc/moshiro/update/apply.request; if [ -f "$f" ] && [ ! -L "$f" ]; then if [ "$1" = backend ]; then head -c 64 "$f"; else printf yes; fi; fi' sh "$COMPONENT"
|
||
elif [ -f "$UPDATE_DIR/apply.request" ] && [ ! -L "$UPDATE_DIR/apply.request" ]; then
|
||
if [ "$COMPONENT" = backend ]; then head -c 64 "$UPDATE_DIR/apply.request"; else printf yes; fi
|
||
fi
|
||
}
|
||
|
||
backend_backup() {
|
||
BACKUP="$STATE_DIR/backend-before-$M_SEQ.tar.gz"
|
||
service_ctl stop "$UNIT"
|
||
# 容器的标准输出就是备份本身：不能再让引擎把它当日志写进 journal（每次更新会多灌一份备份大小的日志）
|
||
if ! "$ENGINE" run --pull=never --rm --log-driver=none --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||
-v "$DATA_VOLUME:/data:ro" --entrypoint /bin/sh "$HELPER_IMAGE" -c 'tar -czf - -C /data .' > "$BACKUP.tmp"; then
|
||
rm -f "$BACKUP.tmp"
|
||
service_ctl start "$UNIT" || true
|
||
write_status failed "$avail" "database backup failed"
|
||
return 1
|
||
fi
|
||
chmod 600 "$BACKUP.tmp"
|
||
mv -f "$BACKUP.tmp" "$BACKUP"
|
||
}
|
||
|
||
backend_restore() {
|
||
service_ctl stop "$UNIT" || true
|
||
"$ENGINE" run --pull=never --rm -i --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||
-v "$DATA_VOLUME:/data" --entrypoint /bin/sh "$HELPER_IMAGE" -c 'tar -xzf - --no-same-owner -C /data' < "$BACKUP"
|
||
}
|
||
|
||
# 从扁平的清单里取字段（清单由 CI 生成，只有字符串和整数，没有嵌套）
|
||
field() { sed -n 's/.*"'"$1"'":"\([^"]*\)".*/\1/p' "$OFFER_MANIFEST" | head -n 1; }
|
||
num_field() { sed -n 's/.*"'"$1"'":\([0-9][0-9]*\).*/\1/p' "$OFFER_MANIFEST" | head -n 1; }
|
||
|
||
# 验签 + 字段校验。成功时设置 M_DIGEST、M_TAG、M_VERSION、M_SEQ；失败返回非 0，原因在 ERR。
|
||
verify_offer() {
|
||
ERR=""
|
||
[ -s "$OFFER_MANIFEST" ] && [ -s "$OFFER_SIG" ] || { ERR="no offer"; return 1; }
|
||
[ "$(wc -c < "$OFFER_MANIFEST")" -le 4096 ] || { ERR="manifest too large"; return 1; }
|
||
sigbin="$WORK/sig.bin"
|
||
base64 -d "$OFFER_SIG" > "$sigbin" 2>/dev/null || { ERR="bad signature encoding"; return 1; }
|
||
openssl dgst -sha256 -verify "$PUBKEY" -signature "$sigbin" "$OFFER_MANIFEST" >/dev/null 2>&1 || { ERR="signature check failed"; return 1; }
|
||
[ "$(field component)" = "$COMPONENT" ] || { ERR="manifest is for another component"; return 1; }
|
||
[ "$(field image)" = "$IMAGE_REPO" ] || { ERR="manifest names another repository"; return 1; }
|
||
if [ -n "$CHANNEL" ]; then
|
||
[ "$(field channel)" = "$CHANNEL" ] || { ERR="manifest is for another channel"; return 1; }
|
||
fi
|
||
M_DIGEST="$(field digest)"
|
||
case "$M_DIGEST" in sha256:*) ;; *) ERR="bad digest"; return 1 ;; esac
|
||
[ "${#M_DIGEST}" -eq 71 ] && [ -z "$(printf '%s' "${M_DIGEST#sha256:}" | tr -d '0-9a-f')" ] || { ERR="bad digest"; return 1; }
|
||
M_TAG="$(field tag)"; M_VERSION="$(field version)"; M_SEQ="$(num_field seq)"
|
||
[ -n "$M_SEQ" ] || { ERR="no seq"; return 1; }
|
||
[ "$M_SEQ" -gt "$(installed_seq)" ] || { ERR="stale manifest (seq $M_SEQ <= $(installed_seq))"; return 1; }
|
||
}
|
||
|
||
available_json() { printf '{"version":"%s","tag":"%s","digest":"%s","seq":%s}' "$(printf '%s' "$M_VERSION" | tr -d '\042\134')" "$(printf '%s' "$M_TAG" | tr -d '\042\134')" "$M_DIGEST" "$M_SEQ"; }
|
||
|
||
do_check() {
|
||
fetch_offer
|
||
if ! verify_offer; then
|
||
# 没有新清单，或者清单已经安装过：不算错误，只是没有更新
|
||
case "$ERR" in
|
||
"no offer"|"stale manifest"*) write_status up_to_date null "" ;;
|
||
*) write_status invalid null "$ERR"; echo "offer rejected: $ERR" >&2 ;;
|
||
esac
|
||
return 0
|
||
fi
|
||
if [ "$M_DIGEST" = "$(current_digest)" ]; then
|
||
write_status up_to_date null ""
|
||
return 0
|
||
fi
|
||
write_status available "$(available_json)" ""
|
||
echo "update available: $M_TAG ($M_DIGEST)"
|
||
if [ "$POLICY" = auto ]; then do_apply; fi
|
||
return 0
|
||
}
|
||
|
||
do_apply() {
|
||
requested_seq="$(fetch_request)"
|
||
if [ -n "$DATA_VOLUME" ]; then
|
||
"$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
|
||
-v "$DATA_VOLUME:/etc/moshiro" --entrypoint /bin/sh "$HELPER_IMAGE" -c 'rm -f /etc/moshiro/update/apply.request'
|
||
else
|
||
rm -f "$UPDATE_DIR/apply.request"
|
||
fi
|
||
fetch_offer
|
||
verify_offer || { write_status invalid null "$ERR"; echo "refusing to apply: $ERR" >&2; return 1; }
|
||
if [ "$COMPONENT" = backend ]; then
|
||
case "$requested_seq" in ''|*[!0-9]*) write_status failed "$(available_json)" "a confirmed release sequence is required"; return 1 ;; esac
|
||
[ "$requested_seq" = "$M_SEQ" ] || { write_status available "$(available_json)" "release changed; confirm again"; return 1; }
|
||
fi
|
||
avail="$(available_json)"
|
||
write_status applying "$avail" ""
|
||
ref="$IMAGE_REPO@$M_DIGEST"
|
||
if ! pull_image "$ref" >/dev/null 2>&1; then
|
||
write_status failed "$avail" "pull failed"; echo "pull failed: $ref" >&2; return 1
|
||
fi
|
||
prev="$(cat "$ENV_FILE" 2>/dev/null || true)"
|
||
if [ "$COMPONENT" = backend ]; then backend_backup || return 1; fi
|
||
printf 'IMAGE=%s\n' "$ref" > "$ENV_FILE.new" && mv -f "$ENV_FILE.new" "$ENV_FILE"
|
||
service_ctl restart "$UNIT" || true
|
||
# 健康确认：服务在跑，并且 20 秒内没有再被重启（崩溃循环算失败）
|
||
sleep 5
|
||
before="$(service_ctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
|
||
sleep 15
|
||
after="$(service_ctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
|
||
healthy=true
|
||
if [ -n "$HEALTH_URL" ]; then
|
||
if ! curl --fail --silent --show-error --max-time 5 "$HEALTH_URL" | grep -q '"status":"ok"'; then healthy=false; fi
|
||
fi
|
||
if [ "$healthy" = true ] && service_ctl is-active --quiet "$UNIT" && [ "$before" = "$after" ]; then
|
||
printf '%s\n%s\n' "$M_SEQ" "$M_DIGEST" > "$INSTALLED"
|
||
write_status up_to_date null ""
|
||
echo "updated to $M_TAG"
|
||
return 0
|
||
fi
|
||
echo "new version did not become healthy, rolling back" >&2
|
||
if [ "$COMPONENT" = backend ]; then
|
||
if ! backend_restore; then
|
||
write_status failed "$avail" "database restore failed; backup retained"
|
||
return 1
|
||
fi
|
||
fi
|
||
if [ -n "$prev" ]; then printf '%s\n' "$prev" > "$ENV_FILE"; fi
|
||
service_ctl restart "$UNIT" || true
|
||
write_status rolled_back "$avail" "new version did not become healthy"
|
||
return 1
|
||
}
|
||
|
||
case "$CMD" in
|
||
check)
|
||
if [ -n "$DATA_VOLUME" ] && [ "$POLICY" != auto ]; then
|
||
requested="$(fetch_request)"
|
||
if [ -n "$requested" ]; then do_apply; else do_check; fi
|
||
else
|
||
do_check
|
||
fi ;;
|
||
apply) do_apply ;;
|
||
esac
|