box/drill/doctor.sh

365 lines
18 KiB
Bash
Raw Normal View History

#!/usr/bin/env bash
# doctor.sh — is this host fit to mint boxes (and to drill), and if not, what
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
# is wrong? Users reach it as 'box doctor'; the drill runs it directly.
#
# bash drill/doctor.sh # report
# bash drill/doctor.sh --fix # report, then revert what the drill left behind
#
# The drill MUTATES the host in phase D (dns.mode, NIC filtering, ACL rules) to
# rehearse the #16 hardening. If a run aborts before it reverts them, those
# mutations outlive it — and the next run mints boxes on a broken network. That
# is not a hypothetical: it is how a box came up with no DNS at all
# ("Temporary failure resolving deb.debian.org" in cloud-init), and how a false
# design veto against #16 got posted from a poisoned baseline.
#
# This script is the answer to "what state is the host actually in?" — the
# question that kept getting answered by hand.
#
# ok/no/inf/head_ all return 0, so the 'A && ok "…" || no "…"' idiom this file
# is built on cannot hit the C-may-run-when-A-is-true trap SC2015 warns about
# (same reasoning as drill.sh's ok/no).
# shellcheck disable=SC2015
set -u
FIX=0; PIN=0
case "${1:-}" in
--fix) FIX=1 ;;
--pin-dns) PIN=1 ;;
"") ;;
*) echo "usage: doctor.sh [--fix | --pin-dns]"; exit 2 ;;
esac
bad=0
ok() { printf ' \033[32mOK\033[0m %s\n' "$*"; }
no() { printf ' \033[31mDIRTY\033[0m %s\n' "$*"; bad=$((bad + 1)); }
inf() { printf ' %s\n' "$*"; }
head_() { printf '\n\033[1m%s\033[0m\n' "$*"; }
command -v incus >/dev/null || { echo "doctor: incus is not installed on this host."; exit 1; }
timeout 10 incus list >/dev/null 2>&1 || {
echo "doctor: the incus daemon is not answering (see issue #26 for recovery):"
echo " sudo pkill -9 -f 'incusd shutdown'"
echo " sudo systemctl stop incus.service incus.socket"
echo " sudo systemctl reset-failed incus.service incus.socket"
echo " sudo systemctl start incus.socket incus.service"
exit 1
}
feat: the restricted tier — box grant/revoke converge users onto the hardened boxnet (#74) incus-user confines an incus-group user to their own project, but its defaults miss box's contract three measured ways (Debian 13 / Incus 6.0.4): a private UNHARDENED NAT bridge per user (ipv6.nat=true, no ACL, no DNS isolation), snapshots blocked, and the box-net profile invisible to their project. So the tier is an admin-run idempotent convergence: box grant <user> # incus group; touch incus-user (the project is lazy); # drop the private-bridge eth0 from their default # profile; restricted.networks.access=boxnet — and ONLY # boxnet, or the unhardened bridge stays one --network # flag away; restricted.snapshots=allow; install the # shipped box-net profile into their project box revoke <user> # group removal closes the socket, boxes keep running --purge # ...or delete their world, and assert the absence box_tier() (live credentials, argless id -nG; byte-identical copy in setup-host.sh) drives the tier-aware surface: new pre-flights the profile and names the right fix per tier, expose refuses before any daemon call (without the guard the failure is a lie — restricted certs cannot read boxnet's redacted config, so box_net_ip claims a running box has no address), setup-host exits 0 with the honest note, doctor judges only what the caller can see. Also fixed while the rehearsal exercised the lifecycle: box restore dispatched 'incus restore', which does not exist in Incus 6 (it is 'incus snapshot restore') — the verb had never worked. Fixed for every tier. Convergence survives incus-user restarts by that tool's own design (it configures a project only at creation) — read in its source, then measured. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-18 04:09:13 +00:00
# The tier changes what this doctor can SEE, so it changes what it may JUDGE.
# bin/box exports BOX_TIER; unset means a hand-run, which was always admin.
# A restricted (incus-group) user cannot read the nft tables, the kernel's
# bridge ports, or boxnet's (redacted) config — reporting those as DIRTY
# would blame the host for the reader's own, correct, confinement. They get
# the checks that are theirs: is the tier granted, is the contract in their
# project, do their boxes actually resolve and route.
TIER="${BOX_TIER:-admin}"
if [ "$TIER" = restricted ]; then
head_ "Access tier — restricted (the incus group: your own boxes, nothing else)"
inf "the host stack (network, ACL, firewall, kernel state) is admin-owned;"
inf "this doctor judges only what is yours to see"
if [ "$FIX" = 1 ] || [ "$PIN" = 1 ]; then
inf "--fix / --pin-dns are admin levers — ignored on this tier"
FIX=0; PIN=0
fi
head_ "Your project — is the tier granted?"
if incus profile show box-net >/dev/null 2>&1 </dev/null; then
ok "the box-net profile is in your project — 'box new' lands on the hardened boxnet"
iso="$(incus profile device get box-net eth0 security.port_isolation </dev/null 2>/dev/null)"
[ "$iso" = "true" ] \
&& ok "security.port_isolation = true (as shipped)" \
|| no "security.port_isolation is NOT set in your box-net profile — re-grant refreshes it: ask an admin to re-run 'box grant $(id -un)'"
else
no "no box-net profile in your project — the restricted tier is granted per user"
inf "fix: an admin runs: box grant $(id -un)"
fi
if incus network show boxnet >/dev/null 2>&1 </dev/null; then
ok "boxnet is reachable from your project"
else
no "boxnet is not visible from your project — ask an admin to re-run 'box grant $(id -un)'"
fi
head_ "Can one of your boxes actually resolve DNS?"
probe="$({ incus list "user.box=1" --format csv --columns ns 2>/dev/null
incus list "user.claudebox=1" --format csv --columns ns 2>/dev/null; } \
| awk -F, '$2 == "RUNNING" { print $1; exit }')"
if [ -n "$probe" ]; then
inf "probing inside '$probe':"
timeout -k 5 25 incus exec "$probe" -- curl -sS -m 10 -o /dev/null https://1.1.1.1 </dev/null 2>/dev/null \
&& ok "reaches 1.1.1.1 by address — egress routing is fine" \
|| no "cannot reach 1.1.1.1 by address — egress routing is broken (an admin problem: box doctor as admin)"
timeout -k 5 25 incus exec "$probe" -- getent hosts deb.debian.org </dev/null >/dev/null 2>&1 \
&& ok "resolves deb.debian.org — DNS works" \
|| no "CANNOT resolve deb.debian.org — an admin problem (the resolver pin lives on the host): box doctor as admin"
else
inf "no running box to probe with (mint one: box new --name work)"
fi
head_ "Verdict"
if [ "$bad" -eq 0 ]; then
printf ' \033[32mclean\033[0m — your tier is granted and your boxes are fit.\n\n'
exit 0
fi
printf ' \033[31m%s problem(s)\033[0m — see the fixes above (most need an admin).\n\n' "$bad"
exit 1
fi
# A FRESH host (no boxnet) is not a DIRTY one. Everything below that would
# scream about a missing piece must first ask: missing from a stack, or never
# set up? setup-host creates all of it, and the drill runs setup-host itself —
# a bare host went 84/84 green minutes after this doctor called it unfit.
FRESH=0
head_ "Network — boxnet"
if incus network show boxnet >/dev/null 2>&1; then
# dns.mode=none is SHIPPED — it is what stops a box enumerating its siblings
# through the gateway's dnsmasq. Its ABSENCE is the problem, not its presence.
dns="$(incus network get boxnet dns.mode 2>/dev/null)"
if [ "$dns" = none ]; then
ok "dns.mode = none — a box cannot enumerate its siblings by name"
else
no "dns.mode = ${dns:-<unset>} — a box can RESOLVE its siblings' names and addresses"
inf "fix: re-run: box setup-host"
[ "$FIX" = 1 ] && { incus network set boxnet dns.mode=none && inf "set: dns.mode=none"; }
fi
inf "ipv4.address = $(incus network get boxnet ipv4.address 2>/dev/null)"
# Incus reports the network as "Created" whether or not anything is actually
# SERVING it. Kill the daemon uncleanly (a wedge, an OOM, a SIGKILL) and it
# can come back without respawning this network's dnsmasq — the bridge is up,
# the config is perfect, and every box minted afterwards gets NO DHCP lease,
# so it dies deep inside cloud-init with "Temporary failure resolving
# deb.debian.org". Two cold mints and an hour of hunting went into learning
# that Incus's own status does not cover this. Ask the process table instead.
if pgrep -af 'dnsmasq.*--interface=boxnet' >/dev/null 2>&1; then
ok "a dnsmasq is serving boxnet (DHCP + DNS)"
else
no "NO dnsmasq is serving boxnet — the bridge is up and incus says 'Created', but nothing hands out leases"
inf "every box minted now gets no address, no DNS, and dies in cloud-init"
inf "fix: timeout 60 incus delete -f <any boxes>; sudo systemctl restart incus"
inf " (if it does not come back: teardown-host.sh, then re-run the drill)"
[ "$FIX" = 1 ] && {
inf "restarting incus to respawn it…"
sudo systemctl restart incus && sleep 5
pgrep -af 'dnsmasq.*--interface=boxnet' >/dev/null 2>&1 \
&& inf "reverted: dnsmasq is serving boxnet again" \
|| inf "STILL missing — run teardown-host.sh and let the drill rebuild the network"
}
fi
ipv6="$(incus network get boxnet ipv6.address 2>/dev/null)"
[ "$ipv6" = none ] && ok "ipv6.address = none (the isolation contract — every ACL rule is IPv4-only)" \
|| no "ipv6.address = $ipv6 — IPv6 is on and NOT covered by any ACL rule"
else
FRESH=1
inf "boxnet does not exist (a fresh host — setup-host.sh will create it)"
fi
head_ "Firewall — the box-to-box drop"
if sudo nft list table bridge box >/dev/null 2>&1; then
ok "nft bridge table 'box' is present — boxes cannot reach each other"
elif [ "$FRESH" = 1 ]; then
inf "not installed yet (a fresh host — setup-host.sh installs it)"
else
no "the box-to-box drop is MISSING — boxes can reach each other"
inf "an L3 ACL never sees frames switched between two ports of one bridge;"
inf "the drop is an nft BRIDGE-family rule, and without it siblings are wide open."
inf "fix: sudo /usr/local/sbin/box-firewall"
inf " (or: sudo systemctl restart box-firewall.service)"
fi
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
# box-net is the placement contract since the 0.4.0 rename; claude-dev is its
# pre-rename ancestor and may linger while legacy boxes still reference it.
# Check whichever exist — an unisolated NIC is a fault on either.
PROFILES=""
incus profile show box-net >/dev/null 2>&1 && PROFILES="box-net"
incus profile show claude-dev >/dev/null 2>&1 && PROFILES="$PROFILES claude-dev"
head_ "Profile — the NIC is the isolation contract"
if [ -n "$PROFILES" ]; then
for p in $PROFILES; do
[ "$p" = claude-dev ] && inf "claude-dev is legacy (pre-rename boxes still reference it)"
iso="$(incus profile device get "$p" eth0 security.port_isolation 2>/dev/null)"
if [ "$iso" = "true" ]; then
ok "$p: security.port_isolation = true — boxes cannot reach each other at L2"
else
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
no "$p: security.port_isolation is NOT set — BOXES CAN REACH EACH OTHER"
inf "an L3 ACL cannot do this: two boxes on one bridge are on the same L2"
inf "segment, so their frames are switched, never routed past the ACL."
inf "fix: re-run: box setup-host"
fi
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
for k in security.mac_filtering security.ipv4_filtering; do
v="$(incus profile device get "$p" eth0 "$k" 2>/dev/null)"
if [ -z "$v" ]; then
ok "$p: $k unset (as shipped)"
else
no "$p: $k = $v ← phase D left this behind. A box can fail to get on the network at all."
[ "$FIX" = 1 ] && { incus profile device unset "$p" eth0 "$k" && inf "reverted: $k unset"; }
fi
done
done
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
inf "resources are per-box since 0.4.0 (stamped from the template at mint; BOX_CPU/BOX_MEMORY override)"
else
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
inf "box-net does not exist (a fresh host — setup-host.sh will create it)"
fi
head_ "ACL — box-isolate"
if incus network acl show box-isolate >/dev/null 2>&1; then
n="$(incus network acl show box-isolate | grep -c 'action:' || true)"
inf "$n rules"
incus network acl show box-isolate | grep -E 'action:|destination:' | sed 's/^/ /'
if incus network acl show box-isolate | grep -q '@internal'; then
no "an @internal rule survived phase D"
[ "$FIX" = 1 ] && { incus network acl rule remove box-isolate egress action=drop destination=@internal && inf "reverted: @internal rule removed"; }
fi
else
inf "box-isolate does not exist (a fresh host)"
fi
# Config is a claim; the bridge port is the fact. Incus can accept
# security.port_isolation and the kernel can still have 'isolated off' on the
# tap — and then boxes reach each other while every config says they cannot.
# Ask the kernel.
head_ "Bridge ports — the KERNEL's view (config is a claim; this is the fact)"
BRIDGE=""
for c in bridge /usr/sbin/bridge /sbin/bridge; do
sudo "$c" -V >/dev/null 2>&1 && { BRIDGE="$c"; break; }
done
if [ -n "$BRIDGE" ]; then
ports="$(sudo "$BRIDGE" -d link show 2>/dev/null | grep -A1 'master boxnet')"
if [ -z "$ports" ]; then
inf "no instance is attached to boxnet right now (mint a box to check the taps)"
else
printf '%s\n' "$ports" | sed 's/^/ /'
if printf '%s' "$ports" | grep -q 'isolated on'; then
ok "the bridge ports are ISOLATED — boxes cannot exchange frames at L2"
else
no "the bridge ports are NOT isolated ('isolated off') — BOXES CAN REACH EACH OTHER"
inf "security.port_isolation in the profile is a claim; this line is the fact."
inf "if the profile says true and the kernel says off, the flag is not being"
inf "applied to VM taps and the isolation needs a different mechanism."
fi
fi
else
inf "'bridge' (iproute2) not found — cannot read the kernel's view"
fi
head_ "Instances"
left="$(incus list --format csv --columns ns 2>/dev/null)"
[ -z "$left" ] && inf "(none)" || printf ' %s\n' "$left"
for b in drill clone archive peer payroll cbprobe cbcopy cbnotours; do
if incus config show "$b" >/dev/null 2>&1; then
no "leftover drill box: $b"
[ "$FIX" = 1 ] && { timeout 60 incus delete -f "$b" >/dev/null 2>&1 && inf "reverted: deleted $b"; }
fi
done
# An interrupted multiuser.sh leaves its users (and their projects) behind —
# and nothing else on this host will ever mention them. Its own cleanup is
# 'box revoke --purge + userdel'; say so rather than absorbing them silently.
grant/rehearsal: the codex round — verified rollback, loud partial states, and the raw-attach guarantee measured (#75) Review 4727756972 (A2): the backout no longer trusts gpasswd — it re-reads the live group database after removal; verified-absent gets the safe message, anything else screams ROLLBACK INCOMPLETE, exits nonzero, and names the exact remediation. The concurrent-login window (a session begun between usermod and backout keeps the group) is CLOSED to the extent the database can't reach: the backout detects live processes and names loginctl terminate-user, and the success wording claims only what was verified. Review 4727641752 (A1): a failed grant for a user whose membership predates the run (the hand-added-user scenario) now fails LOUDLY — they retain socket access on part-converged policy, and the message says so with both remediations (box revoke now, or fix and re-run). Their membership is not stripped: breaking a working user over a failed re-grant is its own hazard. The default-profile eth0 removal is deliberately not restored on failure — that mutation only reduces capability, and restoring it would move the failure state AWAY from fail-closed. Injected-failure coverage is criterion (n), both flavors: fresh-user backout (fault at the LAST mutation, so the rollback runs after every earlier one) with the group's absence verified and a converging re-run; blocked narrowing staged for real with an instance-local NIC parked on the private bridge. Review A3, resolution 3 with the measurement demanded: criterion (m) launches exactly 'incus launch --network boxnet' as the restricted user and probes the raw NIC from inside — egress works, RFC1918 dropped (the ACL is the network's), sibling probes dropped BOTH directions (the nft drop is the host's), name enumeration blocked. The scoped guarantee is now stated in box-design.md and measured on every run: box-minted instances carry per-NIC port_isolation; raw attachments keep every network- and host-owned control, losing only that redundant L2 layer. Instrument lesson kept as MU-5: the probe's first cut minted the non-cloud image — no DHCP client, no lease, and a dead NIC passes every negative probe vacuously; it now requires the lease before believing its own answers. Rehearsal: 54/54 (containers). test/cli.sh: 82 checks. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-18 06:41:05 +00:00
for u in boxdrill1 boxdrill2 boxdrill3 boxdrill4; do
if getent passwd "$u" >/dev/null 2>&1; then
no "leftover rehearsal user: $u (an interrupted drill/multiuser.sh run)"
inf "fix: sudo BOX_YES=1 box revoke $u --purge && sudo userdel -r $u"
fi
done
# --- the box's DNS comes from the HOST's resolver. See issue #33. -----------
head_ "Host resolver — a box's DNS is forwarded through this"
hostns="$(grep -E '^nameserver' /etc/resolv.conf 2>/dev/null | awk '{print $2}' | tr '\n' ' ')"
inf "/etc/resolv.conf: ${hostns:-<none>}"
raw="$(incus network get boxnet raw.dnsmasq 2>/dev/null | tr '\n' ';')"
if [ -n "$raw" ]; then
ok "boxnet has a pinned resolver (raw.dnsmasq: $raw)"
inf "boxes do NOT inherit the host's resolver — good (issue #33)"
elif [ "$FRESH" = 1 ]; then
# No boxnet, so there is nothing to pin YET. A VPN resolver on the host is
# worth naming, but it is a fact about the host, not a fault in a stack
# that does not exist — setup-host pins around it at creation.
if printf '%s' "$hostns" | grep -qE '(^| )100\.(6[4-9]|[7-9][0-9]|1[01][0-9]|12[0-7])\.'; then
inf "the host resolves via a CGNAT/Tailscale resolver ($hostns) — noted, not a fault:"
inf "setup-host pins the box resolver to public upstreams, so boxes will not inherit it (issue #33)"
else
inf "no boxnet yet — setup-host.sh pins its resolver at creation"
fi
else
# 100.64.0.0/10 is CGNAT — which is exactly Tailscale's range.
if printf '%s' "$hostns" | grep -qE '(^| )100\.(6[4-9]|[7-9][0-9]|1[01][0-9]|12[0-7])\.'; then
no "the host resolves via a CGNAT/Tailscale resolver ($hostns), and boxes INHERIT it — see issue #33"
inf "· box DNS breaks whenever the tailnet's resolver does (this is what kills cold mints)"
inf "· and tailnet names RESOLVE from inside a box, though its ACL blocks connecting to them"
inf "fix: re-run: box setup-host (it pins the resolver)"
inf " or quick-test the pin alone: bash drill/doctor.sh --pin-dns"
else
inf "boxes inherit the host's resolver (unpinned — setup-host.sh pins this now; re-run it)"
fi
fi
if [ "$PIN" = 1 ]; then
head_ "Pinning boxnet's resolver (issue #33)"
if incus network set boxnet raw.dnsmasq "$(printf 'no-resolv\nserver=1.1.1.1\nserver=8.8.8.8\n')" 2>/tmp/pin.err; then
ok "set raw.dnsmasq: no-resolv + 1.1.1.1 + 8.8.8.8 — dnsmasq now ignores /etc/resolv.conf"
inf "a box's DNS no longer depends on the host's VPN state, and MagicDNS is out of the path"
inf "re-run the drill; if the cold mint now succeeds, issue #33 is confirmed and the fix belongs in setup-host.sh"
else
no "incus rejected raw.dnsmasq: $(head -1 /tmp/pin.err 2>/dev/null)"
inf "then raw.dnsmasq is the wrong lever and #33 needs a different mechanism — say so on the issue"
fi
fi
head_ "Can a box actually resolve DNS?"
feat!: claudebox becomes box — the Claude box is one template among several The tool underneath was already generic: a thin, honest wrapper over Incus. What was Claude-specific was welded on — one image, one profile, one cloud-init file, one hardcoded 'sudo -u claude'. The weld is now a template. The mechanic: 'box new' stamps the template's identity onto the instance (user.box=1, user.box.template, user.box.user); shell/exec/ tmux read the user back off the instance, and 'incus copy' carries user.* keys (audit B2), so a clone knows what it is without consulting the template. Templates are box.env (parsed against a strict allowlist, never sourced — no key for a network exists, on purpose) plus a verbatim cloud-init. Every template launches with the shared box-net profile: the isolated NIC and root disk, nothing template-controlled — resources land per-instance from box.env, overridable via BOX_CPU/ BOX_MEMORY/BOX_DISK (which is also how the drill shrinks boxes on a small host now that profile edits can't). The three open calls, taken as recommended: clean cut at 0.4.0 (no claudebox shim; the installer retires the old symlink); default template = claude (muscle memory survives); repo stays heavy-duty/ claudebox, binary is box. Compat is the tag, not the name: resolve_box and list honor the legacy user.claudebox=1 forever, and the legacy tag maps to the claude user — a pre-rename box lists, shells, clones, unchanged. Deliberate divergence from #17's table: the host-stack resource names (claudenet, claude-isolate, nft tables, claudebox-firewall.*) are NOT renamed — they are host-internal, invisible to users, and renaming them breaks every provisioned host for zero user-visible gain. claude-dev is no longer created; setup-host creates box-net, teardown removes both. Closes #17
2026-07-14 14:22:50 +00:00
# Any box will do — the drill's names are not the only boxes on a host, and
# a pre-rename box (legacy tag) is as good a probe as a new one.
probe="$({ incus list "user.box=1" --format csv --columns ns 2>/dev/null
incus list "user.claudebox=1" --format csv --columns ns 2>/dev/null; } \
| awk -F, '$2 == "RUNNING" { print $1; exit }')"
if [ -n "$probe" ] && [ "$FIX" != 1 ]; then
# Stdin MUST be pinned to /dev/null: with a TTY on stdin, 'incus exec' goes
# interactive and puts the terminal in raw mode — the probe hangs forever,
# timeout's TERM never takes (hence -k), and ^C is forwarded INTO the box
# instead of killing the script. The drill learned this in #22; same rule here.
inf "probing inside '$probe' — this separates DNS from routing, which is the whole question:"
inf "its resolv.conf: $(timeout -k 5 20 incus exec "$probe" -- sh -c 'grep -m2 nameserver /etc/resolv.conf' </dev/null 2>/dev/null | tr '\n' ' ')"
# Routing is probed by ADDRESS against the public internet, NOT by pinging
# the gateway: box-firewall.sh drops everything from a box to the host
# except DNS/DHCP, so ICMP to 10.88.0.1 fails BY DESIGN on a healthy host.
# A gateway ping here is a check that can only ever lie.
if timeout -k 5 25 incus exec "$probe" -- curl -sS -m 10 -o /dev/null https://1.1.1.1 </dev/null 2>/dev/null; then
routing=1; ok "reaches 1.1.1.1 by address — egress routing is fine"
else
routing=0; no "cannot reach 1.1.1.1 by address — egress routing is broken (this is not DNS)"
fi
if timeout -k 5 25 incus exec "$probe" -- getent hosts deb.debian.org </dev/null >/dev/null 2>&1; then
ok "resolves deb.debian.org — DNS works"
else
no "CANNOT resolve deb.debian.org — this is exactly what kills cloud-init on every cold mint"
# Egress by address was probed above. If it worked, the fault is purely
# name resolution — i.e. the forwarder, i.e. issue #33.
if [ "$routing" = 1 ]; then
inf "…but it CAN reach 1.1.1.1 by address. So egress works and only NAME RESOLUTION is broken:"
inf "the fault is the forwarder the box inherits from the host — issue #33."
inf "test the fix: bash drill/doctor.sh --pin-dns then re-run the drill"
else
inf "…and it cannot reach 1.1.1.1 by address either — so egress itself is broken, not just DNS."
fi
fi
else
inf "no box to probe with (mint one, or run without --fix after a run)"
fi
head_ "Verdict"
if [ "$bad" -eq 0 ]; then
if [ "$FRESH" = 1 ]; then
printf ' \033[32mfresh\033[0m — no box stack on this host yet, and nothing dirty either.\n'
printf ' run: box setup-host (or the drill — it sets the host up itself)\n\n'
else
printf ' \033[32mclean\033[0m — this host is fit to mint boxes (and to drill).\n\n'
fi
exit 0
fi
printf ' \033[31m%s problem(s)\033[0m — this host is NOT fit to mint boxes (or to drill).\n' "$bad"
if [ "$FIX" = 1 ]; then
printf ' reverted what could be reverted; re-run doctor to confirm.\n\n'
else
printf ' run: bash drill/doctor.sh --fix\n\n'
fi
exit 1