Watch
0
0
Fork
You've already forked hyperhive
0

nix trusts the swarm's root through its own CA bundle, so root's nix clients reach the cache

This commit is contained in:
atlas 2026-11-08 18:32:37 +01:00
commit 9d62ef44b5
5 changed files with 299 additions and 79 deletions

View file

@ -12,20 +12,18 @@
# - the cache's public key and the pull token are read from the store by a root
# oneshot, `hive-priv-cache-facts`, under this hive's own store identity. The
# key lands in `/run/hyperhive-priv/attic/public-key` (directory `0700`, file
# `0600`).
# The cache's address, the key and a netrc also land in a nix.conf fragment,
# `/run/hyperhive-nix/cache.conf`, as `extra-substituters`,
# `extra-trusted-public-keys` and `netrc-file`, which the host's nix.conf
# includes (`!include`, so a host whose unit has not run yet still starts).
# Every client of the daemon (hive-c0re, agent containers on `nixBuilds =
# host`) can have a swarm closure substituted; closures hold no secret;
# - that netrc, `/run/hyperhive-nix/netrc` (`root:root 0600`), is the pull
# token's entry plus the entries of the host's own netrc, which
# `netrc-file` would otherwise replace. The host's netrc is
# `nix.settings.netrc-file` when the host sets it, else nix's default
# `/etc/nix/netrc`. The unit merges again on every run, so an edit reaches the
# daemon within the timer's ten minutes. A netrc named only in
# `nix.extraOptions` is not seen.
# `0600`). The address, the key, a netrc and a CA bundle also land in a
# nix.conf fragment, `/run/hyperhive-nix/cache.conf`, as `extra-substituters`,
# `extra-trusted-public-keys`, `netrc-file` and `ssl-cert-file`, which the
# host's nix.conf includes (`!include`, so a host whose unit has not run yet
# still starts). Every nix client of the host can have a swarm closure
# substituted; the fragment holds no secret;
# - `netrc-file` and `ssl-cert-file` take one path and would replace the host's
# own, so the files they name merge it in: the pull token's entry plus the
# host's netrc entries, and the host's CA bundle plus the swarm's root, which
# the cache's certificate chains to. The host's are `nix.settings.netrc-file`
# else `/etc/nix/netrc`, and `nix.settings.ssl-cert-file` else
# `security.pki.caBundle`; one named only in `nix.extraOptions` is not seen.
#
# hive-c0re keeps its own copies under `/run/hyperhive/attic/` for the probe.
{
@ -60,6 +58,13 @@ let
netrcFile = "${confDir}/netrc";
# `netrc-file` is not additive: the fragment's setting replaces the host's.
hostNetrc = toString (config.nix.settings.netrc-file or "/etc/nix/netrc");
# `ssl-cert-file` is not additive either. The host's bundle is the one nix
# would use without the fragment: `nix.settings.ssl-cert-file` when the host
# sets it, else the system bundle the nix daemon's `CURL_CA_BUNDLE` names.
caBundleFile = "${confDir}/ca-bundle.crt";
baseBundle = toString (config.nix.settings.ssl-cert-file or config.security.pki.caBundle);
# Written by `swarm-services-cert` on a host whose gateway is self-signed.
swarmCa = "${tls.stateDir}/swarm-services-root.pem";
# Where the store runs on this host, the first login waits for its container.
# No `wants`: a pull would start the container mid-switch on the old closure.
@ -118,6 +123,9 @@ in
confFile
netrcFile
hostNetrc
caBundleFile
baseBundle
swarmCa
;
url = cacheUrl;
};

View file

@ -1,32 +1,32 @@
# The root-side copy of the cache facts the host's nix daemon and hive-priv
# use: the key a downloaded closure must be signed by and the netrc the pull
# token goes out in. The script runs as root on the hive host under the hive's
# own store identity and writes:
# The root-side copy of the cache facts the host's nix daemon and hive-priv use:
# the key a closure must be signed by, the netrc the pull token goes out in and
# the CA bundle that trusts the cache. The script runs as root on the hive host
# under the hive's own store identity and writes:
#
# - the key into `dir`, which only root can read;
# - the netrc at `netrcFile`, root 0600: the pull token's entry for `host`
# followed by the entries of the host's own netrc (`hostNetrc`) for other
# machines. `netrc-file` is one path, so a fragment that names this file must
# carry what the host already keeps there;
# - the netrc at `netrcFile`, root 0600: the pull token's entry for `host` then
# the entries of the host's own netrc (`hostNetrc`), as `netrc-file` is one path;
# - the CA bundle at `caBundleFile`, root 0644: the host's own bundle
# (`baseBundle`) then the swarm's root (`swarmCa`), when the host holds it,
# which the cache's certificate chains to; `ssl-cert-file` is one path too;
# - a nix.conf fragment at `confFile` that the host's nix.conf includes, with
# `extra-substituters` (the cache's address), `extra-trusted-public-keys` and
# `netrc-file` (that merged netrc), so the daemon
# substitutes from the cache and its other substituters. It holds no secret.
# `extra-substituters` (the cache's address), `extra-trusted-public-keys`,
# `netrc-file` and `ssl-cert-file`. It holds no secret.
#
# Nothing is read from a file hive-c0re (unprivileged) writes: the key and the
# token come from the store.
# token come from the store. A store that holds no facts yet, or cannot be
# reached, leaves what is in `dir`, `netrcFile`, `caBundleFile` and `confFile`
# as they are. Nothing here removes a file.
#
# A store that holds no facts yet, or cannot be reached, leaves what is in
# `dir`, `netrcFile` and `confFile` as they are. Nothing here removes a file.
#
# Pure function — NOT a NixOS module. Call it from a unit's `script`, with
# `bao`, `coreutils`, `gnugrep` and `systemd` on its `path` and the `BAO_*`
# client environment set:
# Pure function — NOT a NixOS module. Call it from a unit's `script`, with `bao`,
# `coreutils`, `gnugrep`, `systemd` on its `path` and the `BAO_*` client set:
#
# script = import ./lib/attic-root-facts.nix { inherit lib; } {
# hive = "alpha"; host = "attic.example.org"; dir = "/run/hyperhive-priv/attic";
# url = "https://attic.example.org/swarm"; confFile = "/run/hyperhive-nix/cache.conf";
# netrcFile = "/run/hyperhive-nix/netrc"; hostNetrc = "/etc/nix/netrc";
# caBundleFile = "/run/hyperhive-nix/ca-bundle.crt"; baseBundle = "/etc/ssl/certs/ca-certificates.crt";
# swarmCa = "/var/lib/hive-tls/swarm-services-root.pem";
# };
{ lib }:
{
@ -45,6 +45,15 @@
# module, which is merged in and never written.
netrcFile,
hostNetrc,
# Where the CA bundle goes.
caBundleFile,
# The CA bundle nix uses without this module, copied in and never written:
# `ssl-cert-file` is one path, so a fragment that names the new bundle must
# carry the host's CAs.
baseBundle,
# The swarm's root certificate. The file may be absent: the bundle then holds
# the host's CAs alone until a later run finds it.
swarmCa,
}:
''
set -euo pipefail
@ -56,8 +65,11 @@
conf=${lib.escapeShellArg confFile}
netrc_file=${lib.escapeShellArg netrcFile}
host_netrc=${lib.escapeShellArg hostNetrc}
ca_bundle_file=${lib.escapeShellArg caBundleFile}
base_bundle=${lib.escapeShellArg baseBundle}
swarm_ca=${lib.escapeShellArg swarmCa}
install -d -m 0700 "$dir"
install -d -m 0755 "$(dirname -- "$conf")" "$(dirname -- "$netrc_file")"
install -d -m 0755 "$(dirname -- "$conf")" "$(dirname -- "$netrc_file")" "$(dirname -- "$ca_bundle_file")"
store_init
store_login
@ -128,14 +140,33 @@
atomic_write_secret 0600 "" "$netrc_file" "$netrc"
fi
# Last, so a daemon that reads the fragment finds the netrc in place. The
# mtime moves only with the content. Lix re-reads nix.conf for every
# connection, CppNix only at start, so a CppNix daemon that started before
# this write is restarted.
printf -v fragment '%s\n%s\n%s' \
# The host's CAs, then the swarm's root. Nix reads the file for each
# download, so a root that lands after the fragment reaches the next one
# without a restart. A file without a certificate is not appended: the bundle
# serves every https fetch of every nix client.
if [ ! -r "$base_bundle" ]; then
echo "cannot read the host's CA bundle $base_bundle; $ca_bundle_file stays as it is." >&2
exit 1
fi
bundle="$(< "$base_bundle")"
if [ -s "$swarm_ca" ] && grep -q -- '-----BEGIN CERTIFICATE-----' "$swarm_ca"; then
bundle+=$'\n'"$(< "$swarm_ca")"
else
echo "the swarm's root $swarm_ca is not on this host yet; $ca_bundle_file holds the host's CAs only."
fi
if secret_differs "$ca_bundle_file" "$bundle" || [ "$(stat -c %a "$ca_bundle_file")" != 644 ]; then
atomic_write_secret 0644 "" "$ca_bundle_file" "$bundle"
fi
# Last, so a daemon that reads the fragment finds the netrc and the bundle in
# place. The mtime moves only with the content. Lix re-reads nix.conf for
# every connection, CppNix only at start, so a CppNix daemon that started
# before this write is restarted.
printf -v fragment '%s\n%s\n%s\n%s' \
${lib.escapeShellArg "extra-substituters = ${url}"} \
"extra-trusted-public-keys = $key" \
"netrc-file = $netrc_file"
"netrc-file = $netrc_file" \
"ssl-cert-file = $ca_bundle_file"
if secret_differs "$conf" "$fragment"; then
atomic_write_secret 0644 "" "$conf" "$fragment"
fi

View file

@ -24,6 +24,7 @@ let
;
})
hive
hiveWith
hiveWithController
runGroup
;
@ -41,16 +42,18 @@ let
# Fronting none of the swarm's names, reading a store it does not run,
# with the reader identity, the services-issuer pair and the hive's own
# identity placed by hand.
noService = hiveWithController {
deploy.bao.clientCertFile = "/etc/pki/bao-client.pem";
deploy.bao.clientKeyFile = "/etc/pki/bao-client-key.pem";
deploy.bao.serverCaFile = "/etc/pki/bao-ca.pem";
deploy.bao.servicesIssuerClientCertFile = "/etc/pki/bao-services-issuer.pem";
deploy.bao.servicesIssuerClientKeyFile = "/etc/pki/bao-services-issuer-key.pem";
deploy.hive-controller.tls.baoHiveIdentityCertFile = "/etc/pki/bao-hive.pem";
deploy.hive-controller.tls.baoHiveIdentityKeyFile = "/etc/pki/bao-hive-key.pem";
};
# identity placed by hand, and its own `ssl-cert-file` in nix.conf.
noService = hiveWith [ { nix.settings.ssl-cert-file = "/etc/pki/host-ca.pem"; } ] (
lib.recursiveUpdate { deploy.hive-controller.enable = true; } {
deploy.bao.clientCertFile = "/etc/pki/bao-client.pem";
deploy.bao.clientKeyFile = "/etc/pki/bao-client-key.pem";
deploy.bao.serverCaFile = "/etc/pki/bao-ca.pem";
deploy.bao.servicesIssuerClientCertFile = "/etc/pki/bao-services-issuer.pem";
deploy.bao.servicesIssuerClientKeyFile = "/etc/pki/bao-services-issuer-key.pem";
deploy.hive-controller.tls.baoHiveIdentityCertFile = "/etc/pki/bao-hive.pem";
deploy.hive-controller.tls.baoHiveIdentityKeyFile = "/etc/pki/bao-hive-key.pem";
}
);
noSvcFetch = noService.systemd.services.swarm-services-cert.script;
# The same hive holding its own identity and no services-issuer pair: the
@ -898,6 +901,15 @@ let
&& lib.hasInfix ''atomic_write_secret 0644 "" "$conf"'' u.script
# The host's own netrc is merged in: nix's default path when the host sets none.
&& lib.hasInfix "host_netrc=/etc/nix/netrc" u.script
# The fragment names the bundle: the host's own CAs plus the swarm's root, in a file of its own.
&& lib.hasInfix "ssl-cert-file = $ca_bundle_file" u.script
&& lib.hasInfix "ca_bundle_file=/run/hyperhive-nix/ca-bundle.crt" u.script
&& lib.hasInfix "swarm_ca=/var/lib/hive-tls/swarm-services-root.pem" u.script
&& lib.hasInfix ''atomic_write_secret 0644 "" "$ca_bundle_file"'' u.script
# No ssl-cert-file on the host: the system bundle is the base. A host that sets one is merged, not replaced.
&& lib.hasInfix "base_bundle=${allLocal.security.pki.caBundle}" u.script
&& lib.hasInfix "base_bundle=/etc/pki/host-ca.pem" noService.systemd.services.hive-priv-cache-facts.script
&& !(lib.hasInfix "base_bundle=${allLocal.security.pki.caBundle}" noService.systemd.services.hive-priv-cache-facts.script)
# No client copy of the token for hive-core, and hive-c0re is not told the cache.
&& !(lib.hasInfix "client_netrc" u.script)
&& !(c0re ? HIVE_C0RE_CACHE_URL)

View file

@ -4,30 +4,39 @@
# ./attic-root-facts.sh.
#
# What runs is the rendered script, with the directory it writes moved into the
# test's scratch space.
# test's scratch space. `SCRIPT_HOSTCA` is the same script for a host whose nix
# sets its own `ssl-cert-file`, which becomes the base of the CA bundle.
{
pkgs,
...
}:
let
script = pkgs.writeText "attic-root-facts.sh" (
import ../host-modules/lib/attic-root-facts.nix { inherit (pkgs) lib; } {
hive = "alpha";
host = "cache.example";
dir = "@DIR@/attic";
url = "https://cache.example/swarm";
confFile = "@DIR@/nix/cache.conf";
netrcFile = "@DIR@/nix/netrc";
hostNetrc = "@DIR@/host-netrc";
}
);
render =
baseBundle:
pkgs.writeText "attic-root-facts.sh" (
import ../host-modules/lib/attic-root-facts.nix { inherit (pkgs) lib; } {
hive = "alpha";
host = "cache.example";
dir = "@DIR@/attic";
url = "https://cache.example/swarm";
confFile = "@DIR@/nix/cache.conf";
netrcFile = "@DIR@/nix/netrc";
hostNetrc = "@DIR@/host-netrc";
caBundleFile = "@DIR@/nix/ca-bundle.crt";
inherit baseBundle;
swarmCa = "@DIR@/swarm-root.pem";
}
);
in
pkgs.runCommand "hyperhive-script-test-attic-root-facts"
{
SCRIPT = script;
SCRIPT = render "@DIR@/system-bundle.crt";
SCRIPT_HOSTCA = render "@DIR@/host-ca.pem";
nativeBuildInputs = [
pkgs.coreutils
pkgs.curl
pkgs.gnugrep
pkgs.openssl
];
}
''

View file

@ -62,11 +62,39 @@ STUB
chmod +x "$dir/bin/systemctl"
export PATH="$dir/bin:$PATH"
# Certificates made here, with fixed names and subjects: `system-ca` stands for
# a CA the host's own bundle holds, `swarm-ca` for the swarm's root, `host-ca`
# for a bundle the host's nix.conf names. `swarm-leaf` and `system-leaf` are
# servers for localhost signed by the first two.
pki=$dir/pki
mkdir -p "$pki"
newca() { # <name>
openssl req -x509 -newkey ec -pkeyopt ec_paramgen_curve:prime256v1 -nodes \
-keyout "$pki/$1.key" -out "$pki/$1.pem" -subj "/CN=$1" -days 2 \
-addext basicConstraints=critical,CA:TRUE -addext keyUsage=keyCertSign 2>/dev/null
}
newleaf() { # <ca> <name>
openssl req -newkey ec -pkeyopt ec_paramgen_curve:prime256v1 -nodes \
-keyout "$pki/$2.key" -out "$pki/$2.csr" -subj "/CN=localhost" 2>/dev/null
printf 'subjectAltName=DNS:localhost\nbasicConstraints=CA:FALSE\n' >"$pki/$2.ext"
openssl x509 -req -in "$pki/$2.csr" -CA "$pki/$1.pem" -CAkey "$pki/$1.key" \
-CAcreateserial -out "$pki/$2.pem" -days 2 -extfile "$pki/$2.ext" 2>/dev/null
}
newca system-ca
newca swarm-ca
newca host-ca
newleaf swarm-ca swarm-leaf
newleaf system-ca system-leaf
certs() { grep -c -- '-----BEGIN CERTIFICATE-----' "$1" 2>/dev/null || echo 0; }
KEYLEAF=secret_swarm_hives_alpha_attic_public-key
TOKENLEAF=secret_swarm_hives_alpha_attic_pull-token
reset() {
rm -rf "$dir/store" "$dir/attic" "$dir/nix" "$dir/host-netrc" "$calls" "$dir/systemctl"
cp "$pki/system-ca.pem" "$dir/system-bundle.crt"
cp "$pki/swarm-ca.pem" "$dir/swarm-root.pem"
cp "$pki/host-ca.pem" "$dir/host-ca.pem"
mkdir -p "$dir/store"
: >"$calls"
: >"$dir/systemctl"
@ -97,9 +125,10 @@ expect "netrc" "$(body "$dir/nix/netrc")" "machine cache.example password tok-1"
expect "key mode" "$(mode "$dir/attic/public-key")" 600
expect "netrc mode" "$(mode "$dir/nix/netrc")" 600
expect "dir mode" "$(mode "$dir/attic")" 700
expect "fragment names the cache, the key and the root-only netrc" "$(body "$dir/nix/cache.conf")" "extra-substituters = https://cache.example/swarm
expect "fragment names the cache, the key, the root-only netrc and the CA bundle" "$(body "$dir/nix/cache.conf")" "extra-substituters = https://cache.example/swarm
extra-trusted-public-keys = swarm:S0VZ
netrc-file = $dir/nix/netrc"
netrc-file = $dir/nix/netrc
ssl-cert-file = $dir/nix/ca-bundle.crt"
expect "fragment mode" "$(mode "$dir/nix/cache.conf")" 644
expect "fragment dir mode" "$(mode "$dir/nix")" 755
expect "no secret in the fragment" "$(grep -c tok-1 "$dir/nix/cache.conf")" 0
@ -314,6 +343,123 @@ run
expect "exit" $? 1
expect "no fragment" "$(body "$dir/nix/cache.conf")" missing
# The CA bundle the fragment names: the host's own CAs, then the swarm's root.
bundle=$dir/nix/ca-bundle.crt
case="the bundle holds the host's CAs and the swarm's root, 0644, and the fragment names it"
reset
seed "swarm:S0VZ" "tok-1"
run
expect "exit" $? 0
expect "bundle" "$(body "$bundle")" "$(cat "$pki/system-ca.pem")
$(cat "$pki/swarm-ca.pem")"
expect "two certificates" "$(certs "$bundle")" 2
expect "mode" "$(mode "$bundle")" 644
expect "fragment names it" "$(grep -c "^ssl-cert-file = $bundle\$" "$dir/nix/cache.conf")" 1
expect "control: the host's bundle held one certificate" "$(certs "$dir/system-bundle.crt")" 1
expect "the host's bundle is not written" "$(body "$dir/system-bundle.crt")" "$(cat "$pki/system-ca.pem")"
case="the swarm and host servers verify against the bundle, and the swarm one does not against the host CAs alone"
expect "swarm leaf, bundle" "$(openssl verify -CAfile "$bundle" "$pki/swarm-leaf.pem" >/dev/null 2>&1 && echo ok || echo bad)" ok
expect "system leaf, bundle" "$(openssl verify -CAfile "$bundle" "$pki/system-leaf.pem" >/dev/null 2>&1 && echo ok || echo bad)" ok
expect "control: swarm leaf, host's CAs alone" "$(openssl verify -CAfile "$dir/system-bundle.crt" "$pki/swarm-leaf.pem" >/dev/null 2>&1 && echo ok || echo bad)" bad
# curl takes the file as nix does. A TLS server for localhost with the swarm's
# leaf, on a port picked at random until one binds.
case="curl trusts the swarm's server with the bundle and refuses it with the host's CAs alone"
port=
for _ in 1 2 3 4 5 6 7 8; do
p=$((20000 + RANDOM % 20000))
openssl s_server -www -accept "$p" -cert "$pki/swarm-leaf.pem" -key "$pki/swarm-leaf.key" \
>"$dir/s_server.log" 2>&1 &
server=$!
for _ in 1 2 3 4 5 6 7 8 9 10; do
if curl -s --cacert "$pki/swarm-ca.pem" "https://localhost:$p/" >/dev/null 2>&1; then port=$p; break; fi
kill -0 "$server" 2>/dev/null || break
sleep 0.3
done
[ -n "$port" ] && break
kill "$server" 2>/dev/null
done
if [ -z "$port" ]; then
fail "no TLS server for the curl check came up: $(cat "$dir/s_server.log")"
else
curl -s --cacert "$bundle" "https://localhost:$port/" >/dev/null 2>&1
expect "with the bundle" $? 0
curl -s --cacert "$dir/system-bundle.crt" "https://localhost:$port/" >/dev/null 2>&1
expect "with the host's CAs alone" $? 60
kill "$server" 2>/dev/null
fi
case="a second run with the same CAs rewrites nothing"
b=$(ino "$bundle")
run
expect "exit" $? 0
expect "bundle untouched" "$(ino "$bundle")" "$b"
case="a bundle left 0640 is rewritten 0644 though its content is current"
chmod 640 "$bundle"
run
expect "exit" $? 0
expect "mode" "$(mode "$bundle")" 644
case="a host that sets its own ssl-cert-file has that file merged in, not replaced"
reset
seed "swarm:S0VZ" "tok-1"
sed -e "s|@DIR@|$dir|g" "$SCRIPT_HOSTCA" >"$dir/script-hostca.sh"
cp "$dir/script.sh" "$dir/script-default.sh"
cp "$dir/script-hostca.sh" "$dir/script.sh"
run
expect "exit" $? 0
expect "the host's file and the swarm's root" "$(body "$bundle")" "$(cat "$pki/host-ca.pem")
$(cat "$pki/swarm-ca.pem")"
line=$(sed -n 2p "$pki/system-ca.pem")
expect "the default bundle's CA is not in it" "$(grep -c -x -F -- "$line" "$bundle")" 0
expect "control: the same search finds it in its own file" "$(grep -c -x -F -- "$line" "$pki/system-ca.pem")" 1
expect "the host's file is not written" "$(body "$dir/host-ca.pem")" "$(cat "$pki/host-ca.pem")"
cat "$pki/system-ca.pem" >>"$dir/host-ca.pem"
run
expect "an edit to the host's file reaches the bundle" "$(certs "$bundle")" 3
cp "$dir/script-default.sh" "$dir/script.sh"
case="a swarm root that is not on the host yet leaves the host's CAs alone, and a later run adds it"
reset
seed "swarm:S0VZ" "tok-1"
rm "$dir/swarm-root.pem"
run DAEMON_STARTED="Sat 2000-01-01 00:00:00.000000 UTC"
expect "exit" $? 0
expect "bundle" "$(body "$bundle")" "$(cat "$pki/system-ca.pem")"
expect "the fragment still names it" "$(grep -c "^ssl-cert-file = $bundle\$" "$dir/nix/cache.conf")" 1
expect "said so" "$(grep -c "swarm-root.pem is not on this host yet" "$dir/out")" 1
started=$(now)
c=$(ino "$dir/nix/cache.conf")
: >"$dir/systemctl"
cp "$pki/swarm-ca.pem" "$dir/swarm-root.pem"
run DAEMON_STARTED="$started"
expect "exit" $? 0
expect "root added" "$(certs "$bundle")" 2
expect "fragment untouched" "$(ino "$dir/nix/cache.conf")" "$c"
expect "no restart: nix reads the bundle per download" "$(restarts)" 0
case="a swarm root file holding no certificate is not appended"
reset
seed "swarm:S0VZ" "tok-1"
echo "garbage" >"$dir/swarm-root.pem"
run
expect "exit" $? 0
expect "bundle" "$(body "$bundle")" "$(cat "$pki/system-ca.pem")"
case="a host bundle that cannot be read fails the run and leaves the fragment and bundle"
good
cp "$bundle" "$dir/bundle-before"
rm "$dir/system-bundle.crt"
seed "swarm:NEW" "tok-new"
run
expect "exit" $? 1
expect "says why" "$(grep -c "cannot read the host.s CA bundle" "$dir/err")" 1
expect "bundle kept" "$(body "$bundle")" "$(body "$dir/bundle-before")"
expect "fragment kept" "$(grep -c 'extra-trusted-public-keys = swarm:S0VZ' "$dir/nix/cache.conf")" 1
if [ "$failures" -ne 0 ]; then
echo "$failures of $count checks failed" >&2
exit 1