@@ -245,6 +245,17 @@ read-only root disk does not need modification at boot. Rebuild the runner
 image to pick up this layout change. A missing resolver handoff fails startup
 rather than leaving the guest with an unrelated public DNS server.
 
+The guest kernel keeps loopback traffic on its own loopback device instead of
+proxying it through TSI, so it cannot reach a loopback resolver such as Docker's
+embedded `127.0.0.11`. The launcher entrypoint relays the first loopback
+nameserver (`127.0.0.0/8`, `::1` or IPv4-mapped `127.0.0.0/8`, in any spelling
+glibc accepts) over UDP and TCP port 53 from the runner's
+own IPv4 address with `socat`, and forwards that address to the guest in its
+place; further loopback nameservers are dropped, and startup fails if the relay
+cannot listen. Relay children exit when idle, and the TCP relay caps its
+concurrent children. Routable nameservers, including Kubernetes cluster DNS, are forwarded
+unchanged.
+
 libkrun delivers the guest environment on the kernel command line, which only
 carries single-line printable ASCII and is capped at 2048 bytes by the guest
 kernel. The launcher entrypoint therefore forwards only the `nameserver`,
 
@@ -12,15 +12,129 @@ set -e
 # joined by a separator that api/src/guest-dns.sh expands back into lines.
 RESOLV_FIELD_SEPARATOR='|'
 
+# The guest kernel keeps loopback traffic on its own loopback device instead of
+# proxying it through TSI, so a loopback resolver such as Docker's embedded
+# 127.0.0.11 never answers inside the guest. Relay it from this container's own
+# address, which the guest reaches through TSI, and hand the guest that address.
+# The relay forks per datagram and per connection, so bound how long an idle
+# child may live, and how many TCP children may run at once. socat 1.8 fails
+# every UDP-RECVFROM child given max-children, so the UDP relay relies on its
+# short idle timeout. Only the runner network reaches the relay: NsJail jobs in
+# the guest run in their own network namespace.
+RESOLVER_RELAY_READY_ATTEMPTS=50
+RESOLVER_RELAY_MAX_CHILDREN=64
+RESOLVER_RELAY_UDP_IDLE_SECONDS=5
+RESOLVER_RELAY_TCP_IDLE_SECONDS=30
+
+# One inet_aton(3) part: hex with 0x, octal with a leading 0, else decimal.
+inet_aton_part() {
+    case "$1" in
+        0x*) [[ "$1" =~ ^0x[0-9a-f]+$ ]] && echo $((16#${1#0x})) ;;
+        0*) [[ "$1" =~ ^0[0-7]*$ ]] && echo $((8#$1)) ;;
+        *) [[ "$1" =~ ^[0-9]+$ ]] && echo $((10#$1)) ;;
+    esac
+}
+
+# Numeric value of an IPv4 address in the inet_aton(3) forms glibc accepts in
+# resolv.conf: one to four parts, the last filling the remaining bytes.
+ipv4_value() {
+    local parts part value=0 index last
+    IFS=. read -ra parts <<< "$1"
+    [[ "$1" != *. ]] && (( ${#parts[@]} >= 1 && ${#parts[@]} <= 4 )) || return 1
+    last=$(( ${#parts[@]} - 1 ))
+    for ((index = 0; index < last; index++)); do
+        part="$(inet_aton_part "${parts[index]}")" || return 1
+        (( part <= 255 )) || return 1
+        value=$(( value | part << (24 - 8 * index) ))
+    done
+    part="$(inet_aton_part "${parts[last]}")" || return 1
+    (( part < 1 << (32 - 8 * last) )) || return 1
+    echo $(( value | part ))
+}
+
+# The eight 16-bit groups of an IPv6 address in decimal, accepting compressed
+# zeros and a dotted IPv4 tail.
+ipv6_groups() {
+    local address="$1" head tail octets octet group groups=() tail_groups=() fill
+    if [[ "$address" =~ ^(.*:)([0-9]+\.[0-9]+\.[0-9]+\.[0-9]+)$ ]]; then
+        address="${BASH_REMATCH[1]}"
+        IFS=. read -ra octets <<< "${BASH_REMATCH[2]}"
+        for octet in "${octets[@]}"; do
+            [[ "$octet" =~ ^(0|[1-9][0-9]{0,2})$ ]] && (( octet <= 255 )) || return 1
+        done
+        address+="$(printf '%x:%x' $(( octets[0] << 8 | octets[1] )) $(( octets[2] << 8 | octets[3] )))"
+    fi
+    [[ "$address" =~ ^[0-9a-f:]+$ && "$address" != :[!:]* && "$address" != *[!:]: && "$address" != *:::* ]] || return 1
+    if [[ "$address" == *::* ]]; then
+        head="${address%%::*}"
+        tail="${address#*::}"
+        [[ "$tail" != *::* ]] || return 1
+        [ -z "$head" ] || IFS=: read -ra groups <<< "$head"
+        [ -z "$tail" ] || IFS=: read -ra tail_groups <<< "$tail"
+        fill=$(( 8 - ${#groups[@]} - ${#tail_groups[@]} ))
+        (( fill >= 1 )) || return 1
+        for ((; fill > 0; fill--)); do groups+=(0); done
+        groups+=("${tail_groups[@]}")
+    else
+        IFS=: read -ra groups <<< "$address"
+    fi
+    (( ${#groups[@]} == 8 )) || return 1
+    for group in "${groups[@]}"; do
+        [[ "$group" =~ ^[0-9a-f]{1,4}$ ]] || return 1
+        printf '%d ' $((16#$group))
+    done
+}
+
+# Canonical form of a loopback nameserver in any spelling glibc accepts: IPv4
+# 127.0.0.0/8, IPv6 ::1, or IPv4-mapped 127.0.0.0/8 (relayed as IPv4). Fails for
+# every other address.
+canonical_loopback() {
+    local address="${1,,}" text groups value
+    address="${address%%\%*}"
+    if [[ "$address" == *:* ]]; then
+        text="$(ipv6_groups "$address")" || return 1
+        read -ra groups <<< "$text"
+        if [ "${groups[*]}" = '0 0 0 0 0 0 0 1' ]; then
+            echo ::1
+            return 0
+        fi
+        [ "${groups[*]:0:6}" = '0 0 0 0 0 65535' ] || return 1
+        value=$(( groups[6] << 16 | groups[7] ))
+    else
+        value="$(ipv4_value "$address")" || return 1
+    fi
+    (( value >> 24 == 127 )) || return 1
+    printf '%d.%d.%d.%d' $(( value >> 24 )) $(( value >> 16 & 255 )) $(( value >> 8 & 255 )) $(( value & 255 ))
+}
+
+loopback_nameserver() {
+    local line words
+    while IFS= read -r line || [ -n "$line" ]; do
+        read -ra words <<< "${line%$'\r'}"
+        if [ "${words[0]:-}" = nameserver ] && canonical_loopback "${words[1]:-}"; then
+            return 0
+        fi
+    done < "$1"
+}
+
+runner_address() {
+    getent ahostsv4 "$(cat /proc/sys/kernel/hostname)" | awk '$1 !~ /^127\./ { print $1; exit }'
+}
+
 encode_resolv_conf() {
+    local loopback="${1:-}" relay="${2:-}"
     local LC_ALL=C
-    local line words encoded=''
+    local line words canonical encoded=''
     while IFS= read -r line || [ -n "$line" ]; do
         line="${line%$'\r'}"
         if [[ ! "$line" =~ ^[[:space:]]*(nameserver|search|domain|options|sortlist)[[:space:]] ]]; then
             continue
         fi
         read -ra words <<< "$line"
+        if [ "${words[0]}" = nameserver ] && canonical="$(canonical_loopback "${words[1]}")"; then
+            [ "$canonical" = "$loopback" ] || continue
+            words[1]="$relay"
+        fi
         line="${words[*]}"
         if [[ "$line" == *[!' '-'~']* || "$line" == *[\"$RESOLV_FIELD_SEPARATOR]* ]]; then
             echo "ERROR: runner /etc/resolv.conf line cannot cross the kernel command line: $line" >&2
@@ -31,13 +145,62 @@ encode_resolv_conf() {
     printf '%s' "$encoded"
 }
 
-SANDBOX_RESOLV_CONF="$(encode_resolv_conf < /etc/resolv.conf)"
+proc_net_address() {
+    local a b c d
+    IFS=. read -r a b c d <<< "$1"
+    printf '%02X%02X%02X%02X:0035' "$d" "$c" "$b" "$a"
+}
+
+start_resolver_relay() {
+    local nameserver="$1" address="$2" local_address udp_pid tcp_pid attempt
+    local udp_target="UDP4-SENDTO:$nameserver:53" tcp_target="TCP4:$nameserver:53"
+    if [ "$nameserver" = ::1 ]; then
+        udp_target='UDP6-SENDTO:[::1]:53'
+        tcp_target='TCP6:[::1]:53'
+    fi
+    socat -T "$RESOLVER_RELAY_UDP_IDLE_SECONDS" \
+        "UDP4-RECVFROM:53,bind=$address,fork" "$udp_target" &
+    udp_pid=$!
+    socat -T "$RESOLVER_RELAY_TCP_IDLE_SECONDS" \
+        "TCP4-LISTEN:53,bind=$address,reuseaddr,fork,max-children=$RESOLVER_RELAY_MAX_CHILDREN" "$tcp_target" &
+    tcp_pid=$!
+    local_address="$(proc_net_address "$address")"
+    for ((attempt = 0; attempt < RESOLVER_RELAY_READY_ATTEMPTS; attempt++)); do
+        if grep -q " $local_address " /proc/net/udp && grep -q " $local_address 00000000:0000 0A " /proc/net/tcp; then
+            echo "Relaying guest DNS from $address:53 to loopback resolver $nameserver"
+            return 0
+        fi
+        if ! kill -0 "$udp_pid" "$tcp_pid" 2>/dev/null; then
+            break
+        fi
+        sleep 0.1
+    done
+    kill "$udp_pid" "$tcp_pid" 2>/dev/null || true
+    echo "ERROR: guest DNS relay could not listen on $address:53 for loopback resolver $nameserver" >&2
+    return 1
+}
+
+LOOPBACK_NAMESERVER="$(loopback_nameserver /etc/resolv.conf)"
+RESOLVER_RELAY_ADDRESS=''
+if [ -n "$LOOPBACK_NAMESERVER" ]; then
+    RESOLVER_RELAY_ADDRESS="$(runner_address)"
+    if [ -z "$RESOLVER_RELAY_ADDRESS" ]; then
+        echo "ERROR: runner has no non-loopback IPv4 address to relay loopback resolver $LOOPBACK_NAMESERVER" >&2
+        exit 1
+    fi
+fi
+
+SANDBOX_RESOLV_CONF="$(encode_resolv_conf "$LOOPBACK_NAMESERVER" "$RESOLVER_RELAY_ADDRESS" < /etc/resolv.conf)"
 export SANDBOX_RESOLV_CONF
 if [[ "$RESOLV_FIELD_SEPARATOR$SANDBOX_RESOLV_CONF" != *"${RESOLV_FIELD_SEPARATOR}nameserver "[!\#]* ]]; then
     echo 'ERROR: runner /etc/resolv.conf has no nameserver' >&2
     exit 1
 fi
 
+if [ -n "$LOOPBACK_NAMESERVER" ]; then
+    start_resolver_relay "$LOOPBACK_NAMESERVER" "$RESOLVER_RELAY_ADDRESS"
+fi
+
 if [ "${LAUNCHER_FILTER_VSOCK_ENOTCONN:-true}" = "true" ]; then
     # libkrun can emit this benign TSI/vsock teardown line after the guest has
     # already closed its side of the socket. It contains the word "error", so
 
@@ -80,31 +80,60 @@ printf '%s\n' "$SANDBOX_RESOLV_CONF" > "$TEST_RESOLVER_OUTPUT"
 STUB
 cat > "$TEST_DIR/bin/getent" <<'STUB'
 #!/usr/bin/env bash
-printf '192.0.2.99 stale-address\n'
+[[ -z "${TEST_RUNNER_ADDRESS-192.0.2.99}" ]] || printf '%s stale-address\n' "${TEST_RUNNER_ADDRESS-192.0.2.99}"
 STUB
-chmod +x "$TEST_DIR/bin/launcher" "$TEST_DIR/bin/getent"
+# The relay stub records its endpoints; readiness comes from the fake /proc/net.
+cat > "$TEST_DIR/bin/socat" <<'STUB'
+#!/usr/bin/env bash
+printf '%s\n' "$*" >> "$TEST_SOCAT_CALLS"
+STUB
+chmod +x "$TEST_DIR/bin/launcher" "$TEST_DIR/bin/getent" "$TEST_DIR/bin/socat"
+# 192.0.2.99:53 bound for UDP and listening for TCP, in /proc/net byte order.
+mkdir "$TEST_DIR/proc-net-bound" "$TEST_DIR/proc-net-unbound"
+printf '  sl  local_address rem_address   st\n   0: 630200C0:0035 00000000:0000 07 00000000:00000000\n' > "$TEST_DIR/proc-net-bound/udp"
+printf '  sl  local_address rem_address   st\n   0: 630200C0:0035 00000000:0000 0A 00000000:00000000\n' > "$TEST_DIR/proc-net-bound/tcp"
+printf '  sl  local_address rem_address   st\n' | tee "$TEST_DIR/proc-net-unbound/udp" > "$TEST_DIR/proc-net-unbound/tcp"
 # libkrun appends every guest environment entry to the kernel command line,
 # which linux-loader limits to single-line printable ASCII. Docker's generated
 # resolv.conf is multi-line with comments, so the entrypoint must forward only
 # the resolver directives, joined by the separator guest-dns.sh expands.
 run_entrypoint() {
-    local resolv_conf="$1"
+    local resolv_conf="$1" proc_net="${2:-$TEST_DIR/proc-net-bound}"
     sed -e "s|/usr/local/bin/launcher|$TEST_DIR/bin/launcher|g" \
         -e "s|/etc/resolv.conf|$resolv_conf|g" \
+        -e "s|/proc/net/|$proc_net/|g" \
+        -e 's|RESOLVER_RELAY_READY_ATTEMPTS=50|RESOLVER_RELAY_READY_ATTEMPTS=3|' \
         "$ROOT/launcher/entrypoint.sh" > "$TEST_DIR/entrypoint.sh"
-    rm -f "$TEST_DIR/forwarded"
+    rm -f "$TEST_DIR/forwarded" "$TEST_DIR/socat-calls"
     PATH="$TEST_DIR/bin:$PATH" \
     EGRESS_GATEWAY_URL='https://egress_gateway:3190/base' \
     FILE_SERVER_URL='http://[::1]:3000/base' \
     SANDBOX_FORWARD_TARGET='tool_call_server:3033' \
     LAUNCHER_FILTER_VSOCK_ENOTCONN=false \
     TEST_RESOLVER_OUTPUT="$TEST_DIR/forwarded" \
+    TEST_SOCAT_CALLS="$TEST_DIR/socat-calls" \
     bash "$TEST_DIR/entrypoint.sh"
 }
 # Independent reference for the expected handoff: directives only, whitespace
-# collapsed, joined by the separator.
+# collapsed, joined by the separator. The first loopback nameserver becomes the
+# relay address; other loopback nameservers are unreachable and dropped.
 encoded_reference() {
-    LC_ALL=C awk '/^[[:space:]]*(nameserver|search|domain|options|sortlist)[[:space:]]/ { sub(/\r$/, ""); $1 = $1; print }' "$1" | paste -sd '|' -
+    LC_ALL=C awk -v relay=192.0.2.99 '/^[[:space:]]*(nameserver|search|domain|options|sortlist)[[:space:]]/ {
+        sub(/\r$/, ""); $1 = $1
+        address = tolower($2)
+        if ($1 == "nameserver" && (address ~ /^127\./ || address ~ /^[0:]*:0*1$/ || address ~ /^::ffff:127\./)) { if (loop == "") loop = $2; if ($2 != loop) next; $2 = relay }
+        print
+    }' "$1" | paste -sd '|' -
+}
+# The relays start in the background, so wait for both before comparing. Both
+# bound idle children; the TCP listener also caps concurrent children.
+assert_relayed() {
+    local udp_target="$1" tcp_target="$2" attempt
+    for ((attempt = 0; attempt < 50; attempt++)); do
+        [[ -e "$TEST_DIR/socat-calls" && "$(wc -l < "$TEST_DIR/socat-calls")" -ge 2 ]] && break
+        sleep 0.1
+    done
+    [[ "$(LC_ALL=C sort "$TEST_DIR/socat-calls")" == "$(printf -- '-T 30 TCP4-LISTEN:53,bind=192.0.2.99,reuseaddr,fork,max-children=64 %s\n-T 5 UDP4-RECVFROM:53,bind=192.0.2.99,fork %s' "$tcp_target" "$udp_target")" ]]
 }
 cat > "$TEST_DIR/docker-resolv.conf" <<'RESOLV'
 # Generated by Docker Engine.
@@ -119,24 +148,92 @@ options ndots:0
 # Overrides: []
 # Option ndots from: internal
 RESOLV
+# The guest keeps loopback traffic local, so Docker's embedded resolver is
+# relayed from the runner address and the guest gets that address instead.
 run_entrypoint "$TEST_DIR/docker-resolv.conf"
-[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 127.0.0.11|options ndots:0' ]]
+[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 192.0.2.99|options ndots:0' ]]
 [[ "$(cat "$TEST_DIR/forwarded")" == "$(encoded_reference "$TEST_DIR/docker-resolv.conf")" ]]
+assert_relayed UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
 [[ "$(wc -l < "$TEST_DIR/forwarded")" == 1 ]]
 if LC_ALL=C grep -q '[^ -~]' "$TEST_DIR/forwarded"; then
     echo 'forwarded resolver contains bytes the kernel command line rejects' >&2; exit 1
 fi
-# The guest restores exactly the directives Docker supplied.
+# The guest restores the Docker directives with the relayed nameserver.
 rm -rf "$TEST_DIR/guest/run/codeapi-resolver"
 SANDBOX_RESOLV_CONF="$(cat "$TEST_DIR/forwarded")" configure_guest_dns "$TEST_DIR/guest"
-printf 'nameserver 127.0.0.11\noptions ndots:0\n' > "$TEST_DIR/expected"
+printf 'nameserver 192.0.2.99\noptions ndots:0\n' > "$TEST_DIR/expected"
 cmp "$TEST_DIR/expected" "$TEST_DIR/guest/etc/resolv.conf"
 
+# Only one loopback resolver can be relayed; other loopback entries would time
+# out in the guest, while routable nameservers keep their order.
+printf 'nameserver 127.0.0.53\r\nnameserver 10.0.0.2\nnameserver\t127.0.0.11\nsearch example.internal\n' > "$TEST_DIR/multi-loopback-resolv.conf"
+run_entrypoint "$TEST_DIR/multi-loopback-resolv.conf"
+[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 192.0.2.99|nameserver 10.0.0.2|search example.internal' ]]
+[[ "$(cat "$TEST_DIR/forwarded")" == "$(encoded_reference "$TEST_DIR/multi-loopback-resolv.conf")" ]]
+assert_relayed UDP4-SENDTO:127.0.0.53:53 TCP4:127.0.0.53:53
+
+# An IPv6 loopback resolver is just as unreachable from the guest; the IPv4
+# relay forwards to it over IPv6.
+printf 'nameserver ::1\noptions ndots:0\n' > "$TEST_DIR/ipv6-loopback-resolv.conf"
+run_entrypoint "$TEST_DIR/ipv6-loopback-resolv.conf"
+[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 192.0.2.99|options ndots:0' ]]
+[[ "$(cat "$TEST_DIR/forwarded")" == "$(encoded_reference "$TEST_DIR/ipv6-loopback-resolv.conf")" ]]
+assert_relayed 'UDP6-SENDTO:[::1]:53' 'TCP6:[::1]:53'
+printf 'nameserver 127.0.0.11\nnameserver ::1\n' > "$TEST_DIR/mixed-loopback-resolv.conf"
+run_entrypoint "$TEST_DIR/mixed-loopback-resolv.conf"
+[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 192.0.2.99' ]]
+assert_relayed UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+# Every spelling glibc accepts for a loopback resolver is relayed to its
+# canonical address: compressed or zero-padded ::1, IPv4-mapped 127/8 in any
+# IPv6 form, and inet_aton(3) IPv4 forms.
+while read -r spelling udp_target tcp_target; do
+    printf 'nameserver %s\nnameserver 10.0.0.2\n' "$spelling" > "$TEST_DIR/spelled-loopback-resolv.conf"
+    run_entrypoint "$TEST_DIR/spelled-loopback-resolv.conf"
+    [[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 192.0.2.99|nameserver 10.0.0.2' ]]
+    assert_relayed "$udp_target" "$tcp_target"
+done <<'SPELLINGS'
+0:0:0:0:0:0:0:1 UDP6-SENDTO:[::1]:53 TCP6:[::1]:53
+0000::0001 UDP6-SENDTO:[::1]:53 TCP6:[::1]:53
+::1%lo UDP6-SENDTO:[::1]:53 TCP6:[::1]:53
+::FFFF:127.0.0.11 UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+0:0:0:0:0:ffff:127.0.0.11 UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+::ffff:7f00:b UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+127.1 UDP4-SENDTO:127.0.0.1:53 TCP4:127.0.0.1:53
+2130706443 UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+0x7f.0.0.0xb UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+0177.0.0.013 UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+SPELLINGS
+# Spellings of the same loopback resolver collapse into one relayed entry.
+printf 'nameserver 127.0.0.11\nnameserver ::ffff:7f00:b\nnameserver 127.0.0.53\n' > "$TEST_DIR/same-loopback-resolv.conf"
+run_entrypoint "$TEST_DIR/same-loopback-resolv.conf"
+[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 192.0.2.99|nameserver 192.0.2.99' ]]
+assert_relayed UDP4-SENDTO:127.0.0.11:53 TCP4:127.0.0.11:53
+# Routable resolvers and addresses glibc rejects are forwarded unchanged, without
+# a relay: 0127 is octal 87, and ::7f00:1 is not IPv4-mapped.
+printf 'nameserver 2001:db8::1\nnameserver fd00::11\nnameserver ::ffff:10.0.0.2\nnameserver ::7f00:1\nnameserver 0127.0.0.1\nnameserver 127.0.0.256\nnameserver 1:2:3:4:5:6:7:8:\nnameserver 127..1\n' > "$TEST_DIR/routable-resolv.conf"
+run_entrypoint "$TEST_DIR/routable-resolv.conf"
+[[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 2001:db8::1|nameserver fd00::11|nameserver ::ffff:10.0.0.2|nameserver ::7f00:1|nameserver 0127.0.0.1|nameserver 127.0.0.256|nameserver 1:2:3:4:5:6:7:8:|nameserver 127..1' ]]
+[[ ! -e "$TEST_DIR/socat-calls" ]]
+
+# A relay that never listens, or a runner without a routable address, fails
+# startup instead of booting a guest that cannot resolve service names.
+if run_entrypoint "$TEST_DIR/docker-resolv.conf" "$TEST_DIR/proc-net-unbound" 2> "$TEST_DIR/entrypoint-error"; then
+    echo 'started launcher without a listening resolver relay' >&2; exit 1
+fi
+grep -q 'guest DNS relay could not listen on 192.0.2.99:53' "$TEST_DIR/entrypoint-error"
+[[ ! -e "$TEST_DIR/forwarded" ]]
+if TEST_RUNNER_ADDRESS='' run_entrypoint "$TEST_DIR/docker-resolv.conf" 2> "$TEST_DIR/entrypoint-error"; then
+    echo 'started launcher without a resolver relay address' >&2; exit 1
+fi
+grep -q 'no non-loopback IPv4 address to relay loopback resolver 127.0.0.11' "$TEST_DIR/entrypoint-error"
+[[ ! -e "$TEST_DIR/forwarded" && ! -e "$TEST_DIR/socat-calls" ]]
+
 # Kubernetes resolvers: tabs, CRLF, trailing spaces, and unknown keywords are
 # normalized away; search domains and options survive intact.
 printf 'nameserver\t10.96.0.10  \r\nsearch   tenant.svc.cluster.local svc.cluster.local cluster.local\n; resolver comment\nlookup file bind\noptions ndots:5\n' > "$TEST_DIR/k8s-resolv.conf" # leak-check:allow
 run_entrypoint "$TEST_DIR/k8s-resolv.conf"
 [[ "$(cat "$TEST_DIR/forwarded")" == 'nameserver 10.96.0.10|search tenant.svc.cluster.local svc.cluster.local cluster.local|options ndots:5' ]] # leak-check:allow
+[[ ! -e "$TEST_DIR/socat-calls" ]]
 
 # The runner's own resolver must round-trip through the same reference.
 run_entrypoint /etc/resolv.conf