Compare commits

...
10 Commits
Author SHA1 Message Date
mesh-admin 3cfc121670 Merge pull request 'Compose each read-write access's grant for the node-engine (hq ADR 0312, issue 537)' (#256) from feat/537-a-read-write-access-grants-the-paths-group into main 2026-10-12 14:20:27 +00:00
mesh-admin db7e4869ae Merge pull request 'Raise a machine that refuses or only partly applies its declarations past one apply interval (issue 233)' (#263) from fix/233-a-failed-probe-fails-only-what-needs-the-tool-2 into main 2026-10-12 14:11:10 +00:00
mesh-admin 9eddf36503 Merge pull request 'A directory's owner may follow the account its module runs as (issue 279)' (#260) from fix/279-owner-follows-a-setting into main 2026-10-12 14:00:23 +00:00
jochen d4de4543cc Judge S21 by time since the first failing report while the machine is heard, not by how many reports said it (issue 233 review)
mesh/merge-gate pass: builds mesh-controller → novox; no bus step; every machine composes with the change as it did without (4 of 4 compose)
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery delivered
mesh/delivery-group group fix/233-a-failed-probe-fails-only-what-needs-the-tool-2 delivered: every member is delivered
2026-10-12 15:39:08 +02:00
jochen a371de37fe Keep a directory owner's settings at the controller's terminal (issue 279)
mesh/merge-gate pass: builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without …
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery delivered
Root chowns a directory to the number its owner's setting names, so a key
an owner asks for must not change through the settings verb: any caller
could have set puid to 0 or to its own uid. TerminalKeys now takes every
${setting:…} in a directory's owner, whatever else asks for it, and the
owner is written as the parsed number so "0001" never reaches the
node-engine as text (review 60).
2026-10-12 15:37:29 +02:00
jochen 77acb40092 Let a directory's owner follow the account its module runs as (issue 279)
A media manager runs as the puid/pgid its assignment sets per node, but a
declared directory's owner was literal, so a folder the app must write was
refused on a machine whose identity was not the image's default. A
directory's owner now takes ${setting:<key>} (user, or user:group), filled
at composition and refused unless the value is a whole number, and the
module check refuses any other shape or a key the module asks for nowhere
else.
2026-10-12 15:37:23 +02:00
jochen 3e10ad1535 Raise a machine that refuses or only partly applies its declarations past one apply interval, naming the package manager and its words, so the mesh says it rather than only status (issue 233)
mesh/delivery-group group fix/233-a-failed-probe-fails-only-what-needs-the-tool-2 ready: every member ready, and composed together they pass
mesh/merge-gate pass: builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without …
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery superseded: a newer head of the same pull request
2026-10-12 14:49:33 +02:00
jschoubben 74bd90df7e Pin the node-engine at its merge of the access grant, and capture it beside (hq ADR 0312, issue 537)
mesh/merge-gate pass: builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step; every machine composes with the change as it…
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery failed: its walk failed, carrying its own commit: a gate on a first machine (what it carried put back), a build, a machine
2026-10-12 13:47:28 +02:00
jschoubben 30599c1858 Pin the node-engine at its reviewed grant, and hold the mount match to the path itself (hq ADR 0312, issue 537 review)
A path beginning with the same letters is another path, and a read-only mount
of a read-write access is still granted: the grant follows what mounts it.
2026-10-12 13:47:17 +02:00
jschoubben 10d4712f82 Name who a read-write access grants its path's group to, so modules write shared data through the group the operator gave it (hq ADR 0312, issue 537)
Each read-write access is composed with a grant: the module's containers that
mount its path and the account its tools run as (the operator's, or the module's
own under ADR 0259 §8); that runtime restarts on the access, so it holds a group
the engine put its account in. A read access grants nothing. Sent only to an
engine whose statement says GrantContract (4), since an older one refuses the
field. Pins the node-engine at the branch that reads it.
2026-10-12 13:47:17 +02:00
24 changed files with 1089 additions and 12 deletions
+2
View File
@@ -6,3 +6,5 @@
/mesh-controller
/postgres-provisioner
/redis-provisioner
# And where `go build` in cmd/mesh-controller leaves it (novox/hq issue 233).
/cmd/mesh-controller/mesh-controller
+118
View File
@@ -0,0 +1,118 @@
package main
import (
"fmt"
"sort"
"strings"
"time"
"github.com/novox/mesh-controller/internal/conditions"
"github.com/novox/mesh-controller/internal/inventory"
)
// kindApplyFailing is S21's kind: a machine whose declarations have been refused, or applied only in part,
// for longer than one apply interval (novox/hq issue 233).
const kindApplyFailing = "apply-failing"
// applyEvery is how often a node-engine applies what it was last told when nothing new arrives: its
// ReconcileEvery. A failure still there after one more of them is not one the next apply clears.
const applyEvery = 5 * time.Minute
// packageManagerKey is how a node-engine names a package manager that does not answer among what failed
// (mesh-host's probe.go): "the package manager (pacman)".
const packageManagerKey = "the package manager ("
// watchApplyFailing is S21: **a machine that does not apply what it was sent is said by the mesh, not only
// to someone who asks for `status`** (novox/hq issue 233, issue 187's family). A workstation refused every
// declaration for an hour and a half because its package manager did not answer, and the mesh's only word
// on it was a line in `status`. Raised once the machine's first report of this failure is older than one apply
// interval **while the machine is heard** — judged by time, not by how many reports said it: a node-engine
// says a re-apply only when how it went changed, so the same failure is one report however long it lasts.
// Names the refusal's or the failure's words; cleared by the clean apply the node-engine then says. Not while
// the machine is silent, asleep or not known: S1 says silent, and asleep is not lost.
func watchApplyFailing(f *signalFacts) []conditions.Observation {
heard := map[string]bool{}
for _, m := range f.machines {
if !m.asleep() && !m.lastHeard.IsZero() && f.now.Sub(m.lastHeard) <= heartbeatBound(m.every) {
heard[m.name] = true
}
}
var out []conditions.Observation
for _, d := range f.doing {
if !d.Wrong() || d.Since == nil || f.now.Sub(*d.Since) <= applyEvery || !heard[d.Node] {
continue
}
out = append(out, applyFailing(d, f.now))
}
return out
}
// applyFailing is one machine's S21 condition.
func applyFailing(d inventory.Doing, now time.Time) conditions.Observation {
m := d.Node
since := d.Since.UTC().Format("2006-01-02 15:04 MST")
o := conditions.Observation{Scope: conditions.ScopeMachine, ID: m, Kind: kindApplyFailing, Machine: m,
Severity: conditions.Warning, Said: doingWords(d),
Resolved: m + " applies what it is sent again"}
how := "refused its declaration whole"
if d.Outcome != inventory.OutcomeRefused {
how = "applied its declaration only in part"
}
o.Summary = fmt.Sprintf("%s has %s since %s (bound %s): %s", m, how, since, ago(applyEvery), failedNames(d))
o.Headline = m + " is not applying its update"
o.Explanation = fmt.Sprintf("%s has %s for %s, and trying again has not helped. The details say what fails.",
m, plainHow(d), humanDuration(now.Sub(*d.Since)))
if tool := unansweringTool(d); tool != "" {
o.Summary = fmt.Sprintf("%s: %s does not answer, so no package reaches it; everything else it was sent "+
"is applied, since %s (bound %s)", m, tool, since, ago(applyEvery))
o.Headline = fmt.Sprintf("%s cannot install packages", m)
o.Explanation = fmt.Sprintf("The package manager on %s (%s) has not answered for %s, so no package reaches "+
"it. Everything else is still applied, so an update that repairs it can land; if none does, it needs a "+
"look on the machine.", m, tool, humanDuration(now.Sub(*d.Since)))
}
return o
}
// plainHow is how a machine failed, as the operator reads it.
func plainHow(d inventory.Doing) string {
if d.Outcome == inventory.OutcomeRefused {
return "refused every update"
}
return "applied its updates only in part"
}
// unansweringTool is the package manager a machine said does not answer, or "".
func unansweringTool(d inventory.Doing) string {
for _, r := range d.Failed {
if rest, ok := strings.CutPrefix(r.ID, packageManagerKey); ok {
return strings.TrimSuffix(rest, ")")
}
}
return ""
}
// failedNames is what failed, by name: the refusal, or the failed resources' ids.
func failedNames(d inventory.Doing) string {
if d.Outcome == inventory.OutcomeRefused {
return "refused"
}
ids := make([]string, 0, len(d.Failed))
for _, r := range d.Failed {
ids = append(ids, r.ID)
}
sort.Strings(ids)
return "failed " + strings.Join(ids, ", ")
}
// doingWords is the machine's own words on what failed: the evidence, kept inside the mesh.
func doingWords(d inventory.Doing) string {
if d.Outcome == inventory.OutcomeRefused {
return "refused: " + oneLine(d.Refused)
}
var parts []string
for _, r := range d.Failed {
parts = append(parts, r.ID+": "+oneLine(r.Error))
}
sort.Strings(parts)
return strings.Join(parts, "; ")
}
+123
View File
@@ -0,0 +1,123 @@
package main
import (
"context"
"strings"
"testing"
"time"
"github.com/novox/mesh-controller/internal/conditions"
"github.com/novox/mesh-controller/internal/inventory"
)
// unansweringPackageManager is anchor's last report as the workstation's read on 2026-10-04, once its
// node-engine applies the rest (novox/hq issue 233): the package failed and the package manager did not answer,
// the same failure on times reports since for before now.
func unansweringPackageManager(now time.Time, forHowLong time.Duration, times int) inventory.Doing {
since := now.Add(-forHowLong)
return inventory.Doing{Node: "anchor", Outcome: inventory.OutcomeFailed, At: now.Add(-time.Second),
Since: &since, Times: times, Applied: 41, Failed: []inventory.FailedResource{
{ID: "apply", Error: `applying "pacman.base": nothing can be said about "base": this is the arch host and ` +
"pacman does not answer here. Either this machine is not Arch, or its package database is broken: " +
"pacman exited 1: error: config file /etc/pacman.conf could not be read: No such file or directory"},
{ID: "the package manager (pacman)", Error: "pacman does not answer here, so every package in this " +
"declaration fails and the rest of it is applied (novox/hq issue 233): pacman exited 1: error: config " +
"file /etc/pacman.conf could not be read: No such file or directory"}}}
}
// **A machine whose package manager does not answer is raised by the mesh itself, naming the tool and the
// probe's words** (novox/hq issue 233): the words in the evidence, the path kept out of the summary.
func TestAPackageManagerThatDoesNotAnswerIsRaisedNamingItsWords(t *testing.T) {
now := time.Date(2026, 10, 4, 14, 0, 0, 0, time.UTC)
f := calm(now)
f.doing = []inventory.Doing{unansweringPackageManager(now, 90*time.Minute, 18)}
got := watchApplyFailing(f)
if len(got) != 1 {
t.Fatalf("raised %d conditions, want one: %+v", len(got), got)
}
o := got[0]
if o.Key() != "machine.anchor.apply-failing" || o.Machine != "anchor" {
t.Errorf("the condition is not the machine's: %s", o.Key())
}
if !strings.Contains(o.Summary, "pacman does not answer") || !strings.Contains(o.Summary, "since 2026-10-04 12:30") {
t.Errorf("the summary does not name the tool and since when: %s", o.Summary)
}
if strings.Contains(o.Summary, "/etc/") {
t.Errorf("the summary carries a path, which belongs in the evidence: %s", o.Summary)
}
if !strings.Contains(o.Said, "could not be read") {
t.Errorf("the evidence does not carry the probe's words: %s", o.Said)
}
if !strings.Contains(o.Headline, "cannot install packages") || o.Explanation == "" || o.Resolved == "" {
t.Errorf("the plain words do not say it: %+v", o)
}
}
// **A whole refusal is raised too**, with the node-engine's words: what the workstation's report said before the
// node-engine applied the rest.
func TestARefusalThatGoesOnIsRaised(t *testing.T) {
now := time.Date(2026, 10, 4, 14, 0, 0, 0, time.UTC)
since := now.Add(-time.Hour)
f := calm(now)
f.doing = []inventory.Doing{{Node: "anchor", Outcome: inventory.OutcomeRefused, Since: &since, Times: 12,
Refused: "this is the arch host and pacman does not answer here", At: now.Add(-time.Minute)}}
got := watchApplyFailing(f)
if len(got) != 1 || !strings.Contains(got[0].Summary, "refused its declaration whole") ||
!strings.Contains(got[0].Said, "pacman does not answer") {
t.Fatalf("a refusal that went on for an hour was not raised with its words: %+v", got)
}
}
// **Judged by time while the machine is heard, not by how many reports said it** (review of novox/hq issue 233):
// a node-engine says the same failure once, so one report an apply interval old is a fault that goes on; one
// younger is not yet; and a silent or sleeping machine is S1's.
func TestOneReportOlderThanAnApplyIsRaisedAndAQuietMachineIsNot(t *testing.T) {
now := time.Date(2026, 10, 4, 14, 0, 0, 0, time.UTC)
f := calm(now)
f.doing = []inventory.Doing{unansweringPackageManager(now, 6*time.Minute, 1)}
if got := watchApplyFailing(f); len(got) != 1 {
t.Errorf("one report of a failure six minutes old was not raised: %+v", got)
}
f.doing = []inventory.Doing{unansweringPackageManager(now, 4*time.Minute, 1)}
if got := watchApplyFailing(f); len(got) != 0 {
t.Errorf("a failure four minutes old was raised before one apply could clear it: %+v", got)
}
f.doing = []inventory.Doing{unansweringPackageManager(now, time.Hour, 1)}
f.machines[0].lastHeard = now.Add(-time.Hour)
if got := watchApplyFailing(f); len(got) != 0 {
t.Errorf("a silent machine was raised for its last report: %+v", got)
}
f = calm(now)
f.doing = []inventory.Doing{unansweringPackageManager(now, time.Hour, 1)}
f.doing[0].Node = "a-machine-the-mesh-does-not-know"
if got := watchApplyFailing(f); len(got) != 0 {
t.Errorf("a machine not heard of was raised: %+v", got)
}
}
// **A failed re-apply followed by a clean one clears S21** (review of novox/hq issue 233): the node-engine says
// the clean re-apply, the store keeps it as applied with no failure since, and the condition the failure raised
// is cleared through the keeper on the next tick.
func TestACleanReapplyAfterAFailedOneClearsTheCondition(t *testing.T) {
now := time.Date(2026, 10, 4, 14, 0, 0, 0, time.UTC)
store := conditions.NewInMemory()
k := conditions.NewKeeper(t.Context(), conditions.Options{Store: store, History: store,
Teller: &conditions.Told{}, Now: func() time.Time { return now }})
defer k.Close(context.Background())
w := &watchdogs{keeper: k, started: now.Add(-time.Hour)}
failing := calm(now)
failing.doing = []inventory.Doing{unansweringPackageManager(now, 20*time.Minute, 1)}
w.see(t.Context(), failing)
open, err := k.Open(t.Context())
if err != nil || len(open) != 1 || open[0].Kind != kindApplyFailing {
t.Fatalf("the failed re-apply raised %+v (%v), want apply-failing", open, err)
}
clean := calm(now.Add(5 * time.Minute))
clean.doing = nil // the clean re-apply recorded: NotDoingWhatTheyWereTold no longer lists the machine
w.see(t.Context(), clean)
if open, _ := k.Open(t.Context()); len(open) != 0 {
t.Fatalf("the condition stayed open after the clean re-apply: %+v", open)
}
}
+7
View File
@@ -81,6 +81,13 @@ var plainWordings = map[string]func(conditions.Observation) words{
"It usually catches up by itself.", m, m),
Resolved: m + " applied the change"}
}),
kindApplyFailing: worded(func(o conditions.Observation) words {
m := machineOr(o, "a machine")
return words{Headline: m + " is not applying its update",
Explanation: fmt.Sprintf("%s has refused or only partly applied what the mesh sent it, again and again. "+
"The details say what fails.", conditions.Capital(m)),
Resolved: m + " applies what it is sent again"}
}),
"tools-silent": worded(func(o conditions.Observation) words {
m := machineOr(o, "a machine")
return words{Headline: "Tools on " + m + " are not answering",
+15
View File
@@ -947,9 +947,14 @@ func renderingFor(ctx context.Context, open *stores, node string,
if err != nil {
return catalogue.Rendering{}, inventory.Node{}, err
}
grantsGroups, err := engineGrantsGroups(ctx, inv, node)
if err != nil {
return catalogue.Rendering{}, inventory.Node{}, err
}
return catalogue.Rendering{
ReadsHealth: readsHealth,
JudgesRoot: judgesRoot,
GrantsGroups: grantsGroups,
BusMembership: memberships[node],
Settings: settings, Generators: gens, Grants: grants, Needed: needed, Foreseen: foreseen, Ports: ports,
Certificate: certificate, Authority: authority, Mesh: private, Names: names,
@@ -982,6 +987,16 @@ func engineJudgesRoot(ctx context.Context, inv *inventory.Inventory, node string
return had && stated.Contract >= link.RootContract, nil
}
// engineGrantsGroups says whether a machine's node-engine reads an access's `grant` (novox/hq ADR 0312), by its own
// newest statement, for the same reason as engineReadsHealth: an older engine parses strictly.
func engineGrantsGroups(ctx context.Context, inv *inventory.Inventory, node string) (bool, error) {
stated, had, err := inv.HealthOf(ctx, node)
if err != nil {
return false, err
}
return had && stated.Contract >= link.GrantContract, nil
}
// zonesInTheMesh is every zone a module in the mesh declares, where the mesh placed it (novox/hq ADR
// 0199): the zone settled from that node's settings, the node's private address, the port the
// answering listen is published on there.
+10
View File
@@ -254,6 +254,16 @@ var signalsTable = []signalRow{
newest: func(f *signalFacts) time.Time {
return newestOf(f.bus.waits, func(w busWaitFacts) time.Time { return w.since })
}},
{Row: "S21", Signal: "a clean apply", Emitter: "node-engine", Trigger: "each report of a declaration refused or applied only in part (novox/hq issue 233)",
Bound: "one apply interval (5 min) past the first report of the same failure, judged by time while the machine " +
"is heard — the node-engine says a re-apply only when how it went changed; not while it is silent or " +
"asleep; naming the refusal, or the package manager that does not answer with its words; cleared by the " +
"clean apply it then says",
Kind: kindApplyFailing, Severity: conditions.Warning, Phase: 2,
needs: func(f *signalFacts) error { return f.doingErr }, watch: watchApplyFailing,
newest: func(f *signalFacts) time.Time {
return newestOf(f.doing, func(d inventory.Doing) time.Time { return d.At })
}},
}
// watchFacts is S14: the snapshot a merge check is fed is older than its bound, or none was kept since
+8
View File
@@ -186,6 +186,14 @@ var suppressions = map[string]suppression{
},
past: func(f *signalFacts) { f.refusedSends = []inventory.Send{refusedSend(f.now.Add(-time.Second))} },
},
// A machine whose package manager does not answer (novox/hq issue 233): said once its first report of the
// failure is older than one apply interval while it is heard. Inside: exactly five minutes old.
"S21": {
inside: func(f *signalFacts) { f.doing = []inventory.Doing{unansweringPackageManager(f.now, 5*time.Minute, 1)} },
past: func(f *signalFacts) {
f.doing = []inventory.Doing{unansweringPackageManager(f.now, 5*time.Minute+time.Second, 1)}
},
},
// Twice by hand within a fortnight is a healer wanted; once, or the first of two a day too old, is not.
"S15": {
inside: func(f *signalFacts) {
@@ -344,3 +344,58 @@ func TestAnyKeyOfAModuleWithATrustedMergeableFileIsRefusedThroughAVerb(t *testin
t.Fatalf("a file marked untrusted through the verb: %v", err)
}
}
// Who owns a module's directory is said at the terminal alone (novox/hq issue 279): a directory's owner may name
// the settings the module runs as, root chowns the folder to them at the next push, and through the settings verb
// any caller could have set puid to 0 or to its own uid. The key is refused through every verb route even where a
// file nothing trusts also asks for it, and taken at the terminal; the layer a refused call meets is unchanged.
func TestADirectoryOwnersKeyIsRefusedThroughAVerb(t *testing.T) {
open := aMesh(t)
ctx := t.Context()
register(t, open, catalogue.Manifest{Module: "sonarr", Version: "1",
Resources: []map[string]any{
{"id": "recycle-bin", "type": "directory", "mode": "0755", "owner": "${setting:puid}:${setting:pgid}"},
{"id": "identity", "type": "file", "path": "/etc/sonarr.env", "mode": "0644", "trusted": false,
"content": "PUID=${setting:puid}\nPGID=${setting:pgid}\nX=${setting:x}\n"}}})
if err := atTheTerminal(t, "settings", "set", "sonarr", `{"puid":1001,"pgid":2000,"x":0}`,
"--node", "anchor"); err != nil {
t.Fatalf("the owner at the terminal: %v", err)
}
if _, err := assign(ctx, open, "anchor", "sonarr"); err != nil {
t.Fatal(err)
}
layer := func() string {
t.Helper()
values, _, err := open.inventory.Layer(ctx, "anchor", "sonarr")
if err != nil {
t.Fatal(err)
}
raw, _ := json.Marshal(values)
return string(raw)
}
kept := layer()
refused := func(what string, err error) {
t.Helper()
if err == nil || !strings.Contains(err.Error(), "controller's terminal") {
t.Fatalf("%s: %v", what, err)
}
}
refused("puid 0 through the verb", throughVerb(t, "settings", map[string]any{"module": "sonarr",
"node": "anchor", "values": `{"puid":0,"pgid":2000,"x":0}`}))
refused("pgid through the verb, mesh-wide", throughVerb(t, "settings", map[string]any{"module": "sonarr",
"values": `{"pgid":0}`}))
refused("cleared through the verb", throughVerb(t, "settings", map[string]any{"module": "sonarr",
"node": "anchor", "clear": "true"}))
if err := throughVerb(t, "command", map[string]any{
"command": `settings set sonarr '{"puid":0,"pgid":2000,"x":0}' --node anchor`}); err == nil {
t.Fatal("the command verb set the owner")
}
if got := layer(); got != kept {
t.Fatalf("a refused call changed the layer: %s, was %s", got, kept)
}
// Another key, the owner kept, is still the verb's.
if err := throughVerb(t, "settings", map[string]any{"module": "sonarr", "node": "anchor",
"values": `{"puid":1001,"pgid":2000,"x":1}`}); err != nil {
t.Fatalf("another key through the verb, the owner kept: %v", err)
}
}
+5
View File
@@ -59,6 +59,10 @@ type signalFacts struct {
// from, each naming its sender (S20, novox/hq issue 234).
refusedSends []inventory.Send
refusedSendsErr error
// doing is every machine whose last report was not a clean apply, with since when it has said the same
// failure (S21, novox/hq issue 233).
doing []inventory.Doing
doingErr error
loop loopFacts
loopErr error
@@ -369,6 +373,7 @@ func (w *watchdogs) gather(ctx context.Context) *signalFacts {
}
f.handActs, f.handActsErr = w.gatherHandActs(ctx, now)
f.refusedSends, f.refusedSendsErr = inv.RefusedSendsSince(ctx, now.Add(-generationRefusalsSaid))
f.doing, f.doingErr = inv.NotDoingWhatTheyWereTold(ctx)
f.facts.taken, _, f.facts.began, f.facts.err = exportedFacts.last()
return f
}
+1 -1
View File
@@ -35,4 +35,4 @@ require (
// committed. Every build (the build agent's `go build`, the Dockerfile) compiles from vendor/ and
// fetches nothing; go refuses to build when vendor/ and this file disagree, so a pin moved without
// `go mod vendor` fails loudly, at once, everywhere.
replace github.com/novox/mesh-host => git.novox.be/novox/mesh-host v0.0.0-20261011092230-20d5af9b2d5f
replace github.com/novox/mesh-host => git.novox.be/novox/mesh-host v0.0.0-20261012112256-3f7f797168d3
+2 -2
View File
@@ -1,5 +1,5 @@
git.novox.be/novox/mesh-host v0.0.0-20261011092230-20d5af9b2d5f h1:8N5OW2mdTNIck2pe4EciTYX5NsrPsHrTLENGNIWYNTU=
git.novox.be/novox/mesh-host v0.0.0-20261011092230-20d5af9b2d5f/go.mod h1:tcTK4LMs1d6JpZUwy3hSHMFG/MIuDr/XFs7/nbeaW/c=
git.novox.be/novox/mesh-host v0.0.0-20261012112256-3f7f797168d3 h1:VpH8CesuBOo7k8GcXZq1eRUfbvUoktfmcUynI5N3IFQ=
git.novox.be/novox/mesh-host v0.0.0-20261012112256-3f7f797168d3/go.mod h1:tcTK4LMs1d6JpZUwy3hSHMFG/MIuDr/XFs7/nbeaW/c=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261010191001-33917f91ac3c h1:l5onJwoIeH8yE/PjVlEeoPyXBsHp2NQiWLllz2fwX7s=
git.novox.be/novox/mesh-sdk/go v0.1.14-0.20261010191001-33917f91ac3c/go.mod h1:GFuZUElBZ9A++mxgIKo97aXXo+kV0uJ/UkbhQPPIbrY=
github.com/antithesishq/antithesis-sdk-go v0.7.0-default-no-op h1:Z/MZK75wC/NSrkgqeNIa7jexam9uWzhLmFTSCPI/kn0=
+22
View File
@@ -246,6 +246,11 @@ type Rendering struct {
// statement's contract is link.RootContract or later). To an older, strict engine the field is not
// sent, and the account it names is not judged — which the self-check says, as not judged.
JudgesRoot bool
// GrantsGroups says this machine's node-engine reads an access's `grant` (novox/hq ADR 0312: its statement's
// contract is link.GrantContract or later). To an older, strict engine the field is not sent, nor the
// runtime's restarts on it, and a read-write access is reached as before: by whoever owns the path.
GrantsGroups bool
}
// RootField is a user resource's field saying the account must never become root without a person
@@ -543,6 +548,9 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
}
var out []map[string]any
// The read-write accesses each module grants its tools' account, by module (novox/hq ADR 0312): the runtime
// that runs as that account is restarted on them.
grantedToTools := map[string][]string{}
for _, m := range r.Modules {
if with.Adopted && m.Filtering != nil {
// Nothing of a module that loads a filter, on an adopted node: its table would drop
@@ -1115,6 +1123,13 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
out = append(out, copied)
}
// **A read-write access grants its path's group to what runs this module's code** (novox/hq ADR 0312):
// the containers composed above that mount it, and the account its tools run as. Only to an engine that
// reads the field.
if with.GrantsGroups {
grantedToTools[m.Module] = grantAccesses(out, owner, m.Module, r.toolsAccount(m, with))
}
// **What only the mesh knows, where this module asked for it.** The graph is the control
// plane's; making a name resolve is the module's software. Emitted as ordinary files under
// this module's name, so they are applied, reported and removed exactly as anything else
@@ -1141,6 +1156,12 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
}
owner[fmt.Sprint(process["id"])] = RuntimeModule
out = append(out, process)
// Restarted when an access moved its account's groups, so its process holds them (novox/hq ADR 0312 §6).
for _, m := range r.Modules {
if m.RunsAs == "" {
restartOnGrants(process, grantedToTools[m.Module])
}
}
// And a runtime of its own for each module of its own account, after it (novox/hq ADR 0259 §8).
owns, err := r.ownRuntimes(with)
if err != nil {
@@ -1149,6 +1170,7 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
for _, p := range owns {
id := fmt.Sprint(p["id"])
owner[id] = strings.TrimSuffix(id, "."+OwnRuntimeID())
restartOnGrants(p, grantedToTools[owner[id]])
out = append(out, p)
}
// What each module's bundles are given is read as the account the runtime runs as — not what a
+134
View File
@@ -0,0 +1,134 @@
package catalogue
import (
"fmt"
"path/filepath"
"sort"
"strings"
)
// A read-write access grants the path's own group to what runs the module's code (novox/hq ADR 0312, issue 537).
//
// **The controller says who; the node-engine reads which group.** Which group owns the operator's path is a fact
// of the machine the controller cannot know, and the path is never changed (§2): so each read-write access
// reaches a machine naming the identities its module's code runs as, and the node-engine reads the path's group
// at apply and grants it to them. Two kinds of identity run a module's code here:
//
// - each of its containers that mounts the path — the source of one of its volumes is the path, or below it —
// which gains the group as a supplementary group of its process and is recreated with it;
// - the account its tools run as, where it has a tools bundle the machine's runtime loads: the operator's
// account the runtime runs as, or the module's own account for a module served by a runtime of its own
// (ADR 0259 §8). That runtime names the access under its `restart-on`, so it is restarted when the access
// moved the account's groups, and its new process holds the group with no new login (§6).
//
// A read access grants nothing (§3): it is sent as before. A read-write access whose module runs nothing here is
// sent an empty grant, so the node-engine still states it unhealthy while its group cannot write (§2).
//
// **Only to an engine that reads it.** An older node-engine parses strictly and refuses a whole declaration for a
// field it does not know; the field and the restarts it brings are composed only where the engine's statement
// says link.GrantContract or later (Rendering.GrantsGroups).
// GrantField is an access's field naming who its path's group is granted to (novox/hq ADR 0312).
const GrantField = "grant"
// toolsAccount is the account module m's tools run as on this machine, or empty where they run nowhere as one:
// no runtime here, no bundle it loads, a module left out of an adopted machine's declaration, or a runtime
// running as root on a machine with no account.
func (r Resolution) toolsAccount(m Manifest, with Rendering) string {
if !r.runtimeHere() || (with.Adopted && m.Filtering != nil) {
return ""
}
loads := false
for _, b := range m.Bundles {
if len(b.Loads) > 0 {
loads = true
}
}
if !loads {
return ""
}
if m.RunsAs != "" {
return m.RunsAs
}
return r.Account
}
// grantAccesses writes the grant of every read-write access of module, among the resources composed so far: the
// module's containers that mount the access's path, and account when it is not empty. The ids of the accesses
// that grant to account, for the runtime that runs as it to be restarted on.
func grantAccesses(out []map[string]any, owner map[string]string, module, account string) []string {
var containers []map[string]any
for _, res := range out {
if res["type"] == "container" && owner[fmt.Sprint(res["id"])] == module {
containers = append(containers, res)
}
}
var toAccount []string
for _, res := range out {
if res["type"] != "access" || owner[fmt.Sprint(res["id"])] != module || res["mode"] != AccessReadWrite {
continue
}
path, _ := res["path"].(string)
grant := map[string]any{}
var mounting []any
for _, c := range containers {
if mounts(c, path) {
mounting = append(mounting, fmt.Sprint(c["id"]))
}
}
if len(mounting) > 0 {
grant["containers"] = mounting
}
if account != "" {
grant["account"] = account
toAccount = append(toAccount, fmt.Sprint(res["id"]))
}
res[GrantField] = grant
}
sort.Strings(toAccount)
return toAccount
}
// mounts says whether a container mounts path, or something below it, from the machine. A path that merely
// begins with the same letters — /storage/media2 beside /storage/media — is another path. A read-only mount of a
// read-write access still counts: the module declared the access read-write, the engine grants the group to what
// mounts it, and a read-only mount only means that container chose not to write through it.
func mounts(container map[string]any, path string) bool {
volumes, _ := container["volumes"].([]any)
path = filepath.Clean(path)
for _, v := range volumes {
s, ok := v.(string)
if !ok {
continue
}
source, _, _ := strings.Cut(s, ":")
if !strings.HasPrefix(source, "/") {
continue // a named volume: the runtime's, not the operator's path
}
source = filepath.Clean(source)
if source == path || strings.HasPrefix(source, path+"/") {
return true
}
}
return false
}
// restartOnGrants adds to a runtime process's restart-on the accesses that grant to the account it runs as, so it
// is restarted to hold a group the account was put in or taken out of (novox/hq ADR 0312 §6).
func restartOnGrants(process map[string]any, ids []string) {
if process == nil || len(ids) == 0 {
return
}
restarts, _ := process["restart-on"].([]any)
seen := map[string]bool{}
for _, id := range restarts {
seen[fmt.Sprint(id)] = true
}
for _, id := range ids {
if !seen[id] {
restarts = append(restarts, id)
seen[id] = true
}
}
process["restart-on"] = restarts
}
+174
View File
@@ -0,0 +1,174 @@
package catalogue
import (
"encoding/json"
"fmt"
"testing"
"github.com/novox/mesh-host/validate"
)
// Defends novox/hq ADR 0312 and issue 537 on the controller's side: each read-write access reaches the machine
// naming who its path's group is granted to — the module's containers that mount it, and the account its tools
// run as — and the tool runner is restarted on it; a read access grants nothing; and none of it is sent to a
// node-engine older than the field, which would refuse the whole declaration.
const anImage = "x@sha256:1111111111111111111111111111111111111111111111111111111111111111"
// anImporter files into one library it writes and reads another: a container mounting both, a step mounting
// neither, and tools served by the machine's runtime.
func anImporter(t *testing.T) Manifest {
t.Helper()
m := aToolsModule(t, "importer", "tools/index.js")
m.Accesses = []Access{
{ID: "music", Path: "/storage/media/music", Mode: AccessReadWrite},
{ID: "movies", Path: "/storage/media/movies", Mode: AccessRead},
}
m.Resources = []map[string]any{
{"id": "app", "type": "container", "name": "importer", "image": anImage,
"volumes": []any{"${access:music}/incoming:/music", "${access:movies}:/movies:ro"}},
{"id": "sidecar", "type": "container", "name": "importer-sidecar", "image": anImage},
}
return m
}
func composeImporter(t *testing.T, r Resolution, with Rendering) map[string]map[string]any {
t.Helper()
with.Needed = map[string]map[string]string{RuntimeModule: {"broker": "sealed-credential"}}
with.ArtifactStore = "anchor.internal:5101"
out, err := r.Declaration(with)
if err != nil {
t.Fatal(err)
}
// What the node-engine this controller pins would make of it: nothing refused.
body, err := json.Marshal(map[string]any{"declaration": validate.Version, "resources": out})
if err != nil {
t.Fatal(err)
}
if problems := validate.Declaration(body); len(problems) > 0 {
t.Fatalf("composed into something the node-engine refuses: %v", problems)
}
return byID(out)
}
func TestAReadWriteAccessNamesTheContainersThatMountItAndTheToolsAccount(t *testing.T) {
r := Resolution{Node: "ace", Account: "jochen", Modules: []Manifest{anImporter(t), theRuntime(t)}}
got := composeImporter(t, r, Rendering{GrantsGroups: true})
music := got["importer."+AccessID("/storage/media/music")]
if music == nil {
t.Fatalf("no access resource for the library: %v", keysOf(got))
}
grant, _ := music[GrantField].(map[string]any)
if fmt.Sprint(grant["containers"]) != "[importer.app]" || grant["account"] != "jochen" {
t.Errorf("the read-write access grants %v; want the container that mounts it and the tools' account", grant)
}
// A read access grants nothing (§3).
if movies := got["importer."+AccessID("/storage/media/movies")]; movies[GrantField] != nil {
t.Errorf("a read access names a grant: %v", movies[GrantField])
}
// The tool runner is restarted when the access moved its account's groups (§6).
runtime := got[RuntimeModule+"."+RuntimeProcessID()]
found := false
for _, id := range runtime["restart-on"].([]any) {
if id == "importer."+AccessID("/storage/media/music") {
found = true
}
}
if !found {
t.Errorf("the tool runner is not restarted on the access that grants its account: %v", runtime["restart-on"])
}
}
func TestAModuleOfItsOwnAccountIsGrantedAsThatAccount(t *testing.T) {
m := anImporter(t)
m.RunsAs = "importer"
m.OwnSecrets = OwnSecrets{"broker": {Path: "/var/lib/mesh/importer/broker"}}
runtime := theRuntime(t)
runtime.Bundles[0].Binary = "node-tools"
r := Resolution{Node: "ace", Account: "jochen", Modules: []Manifest{m, runtime}}
with := Rendering{GrantsGroups: true}
with.Needed = map[string]map[string]string{RuntimeModule: {"broker": "sealed"}, "importer": {"broker": "sealed"}}
with.ArtifactStore = "anchor.internal:5101"
out, err := r.Declaration(with)
if err != nil {
t.Fatal(err)
}
got := byID(out)
grant, _ := got["importer."+AccessID("/storage/media/music")][GrantField].(map[string]any)
if grant["account"] != "importer" {
t.Errorf("the access grants %v; want the module's own account", grant)
}
own := got["importer."+OwnRuntimeID()]
if fmt.Sprint(own["restart-on"]) == "" || !restartsOnIt(own["restart-on"], "importer."+AccessID("/storage/media/music")) {
t.Errorf("the module's own runtime is not restarted on its access: %v", own["restart-on"])
}
if restartsOnIt(got[RuntimeModule+"."+RuntimeProcessID()]["restart-on"], "importer."+AccessID("/storage/media/music")) {
t.Error("the machine's runtime is restarted on an access of a module it does not serve")
}
}
func TestAModuleThatRunsNothingHereSendsAnEmptyGrant(t *testing.T) {
m := mod("spool", nil, nil, nil)
m.Accesses = []Access{{Path: "/storage/downloads", Mode: AccessReadWrite}}
out, err := Resolution{Node: "ace", Account: "jochen", Modules: []Manifest{m}}.Declaration(Rendering{GrantsGroups: true})
if err != nil {
t.Fatal(err)
}
grant, has := byID(out)["spool."+AccessID("/storage/downloads")][GrantField].(map[string]any)
if !has || len(grant) != 0 {
t.Errorf("a read-write access of a module running nothing here is sent %v (sent %v); want an empty grant, "+
"so its group is still judged", grant, has)
}
}
func TestNoGrantIsSentToAnEngineOlderThanTheField(t *testing.T) {
r := Resolution{Node: "ace", Account: "jochen", Modules: []Manifest{anImporter(t), theRuntime(t)}}
got := composeImporter(t, r, Rendering{})
for id, res := range got {
if _, has := res[GrantField]; has {
t.Errorf("%s names a grant to an engine that does not read it", id)
}
}
if restartsOnIt(got[RuntimeModule+"."+RuntimeProcessID()]["restart-on"], "importer."+AccessID("/storage/media/music")) {
t.Error("the tool runner is restarted on an access an older engine grants nothing for")
}
}
func restartsOnIt(list any, want string) bool {
items, _ := list.([]any)
for _, i := range items {
if fmt.Sprint(i) == want {
return true
}
}
return false
}
func keysOf(m map[string]map[string]any) []string {
var out []string
for k := range m {
out = append(out, k)
}
return out
}
// Only a container that mounts the access's path, or below it, is named: a path that begins with the same letters is
// another path. A read-only mount of a read-write access is still named — the access is read-write, and the grant
// follows what mounts it, not how.
func TestOnlyAContainerMountingThePathIsNamedReadOnlyIncluded(t *testing.T) {
m := mod("library", nil, nil, nil)
m.Accesses = []Access{{ID: "media", Path: "/storage/media", Mode: AccessReadWrite}}
m.Resources = []map[string]any{
{"id": "beside", "type": "container", "name": "beside", "image": anImage, "volumes": []any{"/storage/media2:/x"}},
{"id": "reader", "type": "container", "name": "reader", "image": anImage, "volumes": []any{"${access:media}:/media:ro"}},
}
out, err := Resolution{Node: "ace", Modules: []Manifest{m}}.Declaration(Rendering{GrantsGroups: true})
if err != nil {
t.Fatal(err)
}
grant, _ := byID(out)["library."+AccessID("/storage/media")][GrantField].(map[string]any)
if fmt.Sprint(grant["containers"]) != "[library.reader]" {
t.Errorf("the access grants %v; want the read-only mount and not /storage/media2", grant)
}
}
+1
View File
@@ -1859,6 +1859,7 @@ func ParseManifest(raw []byte) (Manifest, error) {
problems = append(problems, invokeProblems(m)...)
problems = append(problems, replacesProblems(m)...)
problems = append(problems, UnitSettingProblems(m)...)
problems = append(problems, OwnerSettingProblems(m)...)
problems = append(problems, endpointNameProblems(m)...)
problems = append(problems, RouteProblems(m)...)
for _, port := range m.Guards {
+181
View File
@@ -0,0 +1,181 @@
package catalogue
import (
"strings"
"testing"
)
// A declared directory may be owned by the account its module runs as (novox/hq issue 279): a media
// manager runs as the puid and pgid its assignment sets per node, and a folder it must write was owned
// by the manifest's literal, so the app refused it on a machine whose identity was not the image's.
func recycler() Manifest {
return Manifest{Module: "sonarr",
Resources: []map[string]any{
{"id": "recycle-bin", "type": "directory", "path": "/srv/sonarr/recycle", "mode": "0755",
"owner": "${setting:puid}:${setting:pgid}"},
{"id": "identity", "type": "file", "path": "/srv/sonarr/identity.env", "mode": "0644",
"content": "PUID=${setting:puid}\nPGID=${setting:pgid}\n"},
},
}
}
func copied(r map[string]any) map[string]any {
out := map[string]any{}
for k, v := range r {
out[k] = v
}
return out
}
// The node's identity reaches the directory's owner, the node's layer over the mesh's, a number given
// as a string or as a JSON number alike.
func TestASettingOwnsADirectory(t *testing.T) {
m := recycler()
layers := []Layer{
{From: "the mesh", Values: map[string]any{"puid": "1000", "pgid": "1000"}},
{From: "this node", Values: map[string]any{"puid": float64(1001), "pgid": "2000"}},
}
r := copied(m.Resources[0])
if err := settingInto(r, layers, m.Module); err != nil {
t.Fatal(err)
}
if r["owner"] != "1001:2000" {
t.Fatalf("owned by %q, not 1001:2000", r["owner"])
}
user := copied(m.Resources[0])
user["owner"] = "${setting:puid}"
if err := settingInto(user, layers, m.Module); err != nil || user["owner"] != "1001" {
t.Fatalf("user alone: %v, %v", user["owner"], err)
}
}
// Composed for a machine, the way a push writes it: the directory reaches the node-engine owned by the
// number the node's settings name, never by the placeholder.
func TestAComposedMachineGetsTheOwnerTheSettingNames(t *testing.T) {
r := anAdoptedAnchor()
r.Modules = append(r.Modules, recycler())
with := anchorRendering(false)
with.Settings["sonarr"] = []Layer{{From: "anchor", Values: map[string]any{"puid": "1001", "pgid": "2000"}}}
composed, err := r.Compose(with)
if err != nil {
t.Fatal(err)
}
if why, left := composed.LeftOut["sonarr"]; left {
t.Fatalf("left out: %s", why)
}
got := byID(composed.Resources)["sonarr.recycle-bin"]
if got["owner"] != "1001:2000" {
t.Fatalf("composed owned by %v, not 1001:2000", got["owner"])
}
}
// Only a whole number owns a directory: an account's name, text, a sign, a fraction or a value past the
// kernel's ids is refused by name, and the directory is left as it was. So is a key nothing sets.
func TestAnOwnerSettingThatIsNoNumberIsRefused(t *testing.T) {
m := recycler()
for _, bad := range []any{"abc", "root", "", "-1", "+5", "1000:1000", "10 00", "1000\nx", float64(1.5),
float64(-3), "4294967295", "99999999999", true} {
r := copied(m.Resources[0])
err := settingInto(r, []Layer{{From: "ace", Values: map[string]any{"puid": bad, "pgid": "1000"}}}, m.Module)
if err == nil || !strings.Contains(err.Error(), `"puid"`) || !strings.Contains(err.Error(), "whole number") {
t.Errorf("%v: %v", bad, err)
}
if r["owner"] != m.Resources[0]["owner"] {
t.Errorf("%v: the owner was changed on refusal: %v", bad, r["owner"])
}
}
r := copied(m.Resources[0])
if err := settingInto(r, nil, m.Module); err == nil || !strings.Contains(err.Error(), "${setting:puid}") {
t.Errorf("nothing set: %v", err)
}
if err := JudgeSettings(m, []Layer{{From: "ace", Values: map[string]any{"puid": "media", "pgid": "1000"}}}, false); err == nil ||
!strings.Contains(err.Error(), "whole number") {
t.Errorf("a stored name is not refused when judged: %v", err)
}
}
// Where the manifest is read: an owner holding a setting is a whole placeholder or a whole number on
// each side of the colon, and every key is one the module asks for elsewhere or declares a default for.
func TestAnOwnerSettingIsJudgedWhereTheManifestIsRead(t *testing.T) {
ok := []string{"${setting:puid}:${setting:pgid}", "${setting:puid}", "${setting:puid}:1000", "1000:${setting:pgid}",
"1000:1000", "media:media"}
bad := []string{
"${setting:puid}x",
"x${setting:puid}",
"${setting:puid}:${setting:pgid}:1",
"${setting:puid}${setting:pgid}",
"media:${setting:pgid}",
"${setting:puid}:",
"${setting:other}:${setting:pgid}",
}
for _, owner := range append(ok, bad...) {
m := recycler()
m.Resources[0] = copied(m.Resources[0])
m.Resources[0]["owner"] = owner
err := parsed(t, m)
refused := err != nil && strings.Contains(err.Error(), "owner")
want := false
for _, b := range bad {
want = want || b == owner
}
if refused != want {
t.Errorf("%s: refused %v, want %v (%v)", owner, refused, want, err)
}
}
// A key with a default is declared, though nothing else asks for it.
m := recycler()
m.Resources = m.Resources[:1]
m.Resources[0] = copied(m.Resources[0])
m.Resources[0]["owner"] = "${setting:workers}"
m.Settings = map[string]SettingDeclaration{"workers": {Kind: KindPreference, Default: float64(4), Why: "a test"}}
if err := parsed(t, m); err != nil && strings.Contains(err.Error(), "owner") {
t.Errorf("a declared key was refused: %v", err)
}
// A file's owner takes no setting: the sweep says so.
f := recycler()
f.Resources[1] = copied(f.Resources[1])
f.Resources[1]["owner"] = "${setting:puid}"
if err := parsed(t, f); err == nil || !strings.Contains(err.Error(), "not filled in this field") {
t.Errorf("a file's owner setting passed: %v", err)
}
}
// A key a directory's owner asks for is a destination, so setting it is not called stray.
func TestASettingAnOwnerAsksForIsNotStray(t *testing.T) {
m := recycler()
m.Resources = m.Resources[:1]
stray := strings.Join(UnusedSettings(m, []Layer{{From: "ace", Values: map[string]any{"puid": "1001", "pgid": "2000"}}}), "; ")
if strings.Contains(stray, "puid") || strings.Contains(stray, "pgid") {
t.Errorf("called stray: %s", stray)
}
}
// An owner is written as the number parsed: "0001" and "00" reach the node-engine as 1 and 0, never as text.
// Uid 0 is allowed, because a key an owner asks for is the terminal's alone.
func TestAnOwnerSettingIsWrittenAsTheNumber(t *testing.T) {
m := recycler()
r := copied(m.Resources[0])
if err := settingInto(r, []Layer{{From: "ace", Values: map[string]any{"puid": "0001", "pgid": "00"}}}, m.Module); err != nil {
t.Fatal(err)
}
if r["owner"] != "1:0" {
t.Fatalf("owned by %q, not 1:0", r["owner"])
}
}
// Every key a directory's owner asks for is the controller's terminal's alone (novox/hq issue 279, issue 339):
// root chowns the folder to it, so it is terminal-only whatever else asks for it — a file nothing trusts, a
// default, or nothing at all.
func TestAnOwnersKeyIsTheTerminals(t *testing.T) {
m := recycler()
m.Resources[1] = copied(m.Resources[1])
m.Resources[1]["trusted"] = false
m.Settings = map[string]SettingDeclaration{"pgid": {Kind: KindPreference, Default: "1000", Why: "a test"}}
keys := strings.Join(TerminalKeys(m), ",")
for _, k := range []string{"puid", "pgid"} {
if !strings.Contains(","+keys+",", ","+k+",") {
t.Errorf("%s is not the terminal's: %s", k, keys)
}
}
}
+125 -3
View File
@@ -4,6 +4,7 @@ import (
"fmt"
"regexp"
"sort"
"strconv"
"strings"
)
@@ -15,7 +16,8 @@ import (
// (novox/hq issues 122, 134). ADR 0112 names the operator as one of the four providers; this is the
// operator answering.
//
// `${setting:<key>}` in a file's content, and in a service's unit name, is filled from the module's settings layers — the mesh's,
// `${setting:<key>}` in a file's content, in a service's unit name and in a directory's owner (a uid or
// gid only, issue 279) is filled from the module's settings layers — the mesh's,
// then this node's — the same layers a mergeable JSON file and a contribution already take, so
// `settings set <module>` is the one place a person's values go. **Refused when no layer sets it**,
// naming the key and the remedy: a definition that carried a default for a mail domain would be
@@ -59,6 +61,9 @@ func settingInto(resource map[string]any, layers []Layer, module string) error {
if fmt.Sprint(resource["type"]) == "service" {
return settingIntoUnit(resource, layers, module)
}
if fmt.Sprint(resource["type"]) == "directory" {
return settingIntoOwner(resource, layers, module)
}
if fmt.Sprint(resource["type"]) != "file" {
return nil
}
@@ -143,6 +148,107 @@ func settingIntoUnit(resource map[string]any, layers []Layer, module string) err
return nil
}
// A declared directory may be owned by the account its module runs as (novox/hq issue 279).
//
// A containerised module runs as the identity its assignment sets per node — its puid and pgid
// settings, written into the environment file its image reads (ADR 0112, issue 153) — and a directory
// its app must write, beyond the one its image re-owns at start, was owned by whatever the manifest
// spelt: right on a machine whose identity matched the image's default, refused by the app on one whose
// did not. So a directory's `owner` may name the settings that identity is: `${setting:<key>}` for the
// user alone, `${setting:<user>}:${setting:<group>}` for both, filled from the same layers at
// composition.
//
// **Whole numbers only.** An owner a setting fills is a uid or a gid — never an account's name, never
// text — so no value can put anything into the ownership the node-engine applies but a number; one that
// is not is refused at composition, naming the key, and the directory is left as it was. The shape is
// judged where the manifest is read (OwnerSettingProblems): each side of the colon a whole placeholder
// or a whole number, and every key one the module itself asks for elsewhere — in a file, a unit, a
// contribution or a served fact — or declares a default for, so the owner follows the identity its
// process is told it runs as, never a key that reaches nothing else.
// ownerSide is one side of an owner a setting may fill: a whole placeholder or a whole number.
var ownerSide = `(?:\$\{setting:[a-z0-9][a-z0-9_.-]*\}|[0-9]+)`
// ownerWithSetting is the only shape an owner holding a setting may have: user, or user:group.
var ownerWithSetting = regexp.MustCompile(`^` + ownerSide + `(?::` + ownerSide + `)?$`)
// wholeID is a value as a uid or a gid, and whether it is one: a whole number below the kernel's "no id"
// (2^32-1). Uid 0 is one: a key an owner asks for is the controller's terminal's alone (TerminalKeys).
func wholeID(v string) (uint64, bool) {
if v == "" || strings.TrimLeft(v, "0123456789") != "" {
return 0, false
}
n, err := strconv.ParseUint(v, 10, 32)
return n, err == nil && n < 1<<32-1
}
// OwnerSettingProblems refuses, where a manifest is read, a directory whose owner carries a setting
// in any shape but a whole placeholder or number on each side of the colon, or names a key the module
// asks for nowhere else and declares no default for.
func OwnerSettingProblems(m Manifest) []string {
var problems []string
asked := settingKeysAskedOutsideOwners(m)
for k := range m.Settings {
asked[k] = true
}
for _, r := range m.Resources {
owner, _ := r["owner"].(string)
if !strings.Contains(owner, "${setting:") {
continue
}
if fmt.Sprint(r["type"]) != "directory" {
// Said by the placeholder sweep: only a directory's owner takes a setting.
continue
}
if !ownerWithSetting.MatchString(owner) {
problems = append(problems, fmt.Sprintf("%s: the directory %v's owner %q carries a setting in a shape "+
"the mesh does not fill; an owner names the account by number only, each side of the colon a whole "+
"${setting:key} or a whole number (${setting:puid}:${setting:pgid}) (novox/hq issue 279)",
m.Module, r["id"], owner))
continue
}
for _, key := range settingsUsed(owner) {
if !asked[key] {
problems = append(problems, fmt.Sprintf("%s: the directory %v's owner says ${setting:%s}, and the "+
"module asks for %q nowhere else and declares no default for it: an owner follows the identity "+
"the module's process is told it runs as (its identity file's ${setting:%s}), never a key of its "+
"own (novox/hq issue 279)", m.Module, r["id"], key, key, key))
}
}
}
return problems
}
// settingIntoOwner fills ${setting:…} in a directory's owner, refusing — with the key and why, the
// resource left as it was — a key nothing sets and a value that is not a whole number.
func settingIntoOwner(resource map[string]any, layers []Layer, module string) error {
owner, ok := resource["owner"].(string)
if !ok {
return nil
}
for _, key := range settingsUsed(owner) {
value, set := settingValue(layers, key)
if !set {
return &UnsetSettingError{Module: module, Setting: key, said: fmt.Sprintf(
"%s has a directory %v whose owner says ${setting:%s}, and nothing sets %q for it — an operator's "+
"value is the assignment's, never the definition's (novox/hq ADR 0112): "+
"`settings set %s <file>` with {%q: …}%s",
module, resource["id"], key, key, module, key, orNoSettings(layers))}
}
v := plainly(value)
id, whole := wholeID(v)
if !whole {
return fmt.Errorf("%s: the setting %q is %q, which cannot own the directory %v: an owner a setting "+
"names is a uid or a gid, a whole number, never an account's name or other text (novox/hq issue 279)",
module, key, v, resource["id"])
}
// Written as the number parsed, so "0001" reaches the node-engine as 1, never as text.
owner = strings.ReplaceAll(owner, "${setting:"+key+"}", strconv.FormatUint(id, 10))
}
resource["owner"] = owner
return nil
}
func settingValue(layers []Layer, key string) (any, bool) {
var value any
set := false
@@ -171,9 +277,25 @@ func orNoSettings(layers []Layer) string {
return "; set today: " + strings.Join(keys, ", ")
}
// settingKeysUsedBy is every key a module's files, contributions and served facts ask for, so a
// setting that lands in one is not called stray.
// settingKeysUsedBy is every key a module's files, units, directory owners, contributions and served
// facts ask for, so a setting that lands in one is not called stray.
func settingKeysUsedBy(m Manifest) map[string]bool {
used := settingKeysAskedOutsideOwners(m)
for _, r := range m.Resources {
if fmt.Sprint(r["type"]) == "directory" {
if owner, ok := r["owner"].(string); ok {
for _, k := range settingsUsed(owner) {
used[k] = true
}
}
}
}
return used
}
// settingKeysAskedOutsideOwners is every key a module's files, units, contributions and served facts
// ask for: what a directory's owner may follow (OwnerSettingProblems).
func settingKeysAskedOutsideOwners(m Manifest) map[string]bool {
used := map[string]bool{}
note := func(s string) {
for _, k := range settingsUsed(s) {
+12 -2
View File
@@ -33,6 +33,9 @@ import (
// 4. **A module's whole layer, when it has a mergeable file not marked `"trusted": false`** (novox/hq issue 340,
// TrustedMergeable). A mergeable file takes any key a layer sets, not only those its content names, so no list
// of keys covers it: an empty runtime configuration a provider reads would take a `url` of the caller's.
// 5. **Every setting a directory's owner asks for** (novox/hq issue 279): root chowns the directory to the number
// at the next push, so through a verb any caller could hand itself — or uid 0 — a module's folder. Whatever
// else asks for the key, a file marked `"trusted": false` included, the owner makes it the terminal's.
//
// Derived from the manifest, never listed by hand, so a provider or a trusted file added tomorrow is covered.
@@ -42,8 +45,8 @@ import (
const TrustedField = "trusted"
// TerminalKeys are the settings keys of a module that are set at the controller's terminal alone: places and
// accesses, every key its provisions serve and every setting a served value asks for, and every setting a file
// asks for unless it says `"trusted": false`. Places and accesses first, then the rest sorted.
// accesses, every key its provisions serve and every setting a served value asks for, every setting a file
// asks for unless it says `"trusted": false`, and every setting a directory's owner asks for. Places and accesses first, then the rest sorted.
func TerminalKeys(m Manifest) []string {
keys := map[string]bool{}
for _, served := range m.Serves {
@@ -57,6 +60,13 @@ func TerminalKeys(m Manifest) []string {
}
}
for _, r := range m.Resources {
if fmt.Sprint(r["type"]) == "directory" {
owner, _ := r["owner"].(string)
for _, asked := range settingsUsed(owner) {
keys[asked] = true
}
continue
}
if fmt.Sprint(r["type"]) != "file" {
continue
}
+1 -1
View File
@@ -56,7 +56,7 @@ type filler struct {
// seatInto, machineInto and contributionsInto. A pass that comes to read another field adds it here,
// or the sweep refuses the placeholder it would have filled.
var fillers = []filler{
{namespace: "setting", patterns: []*regexp.Regexp{settingRef}, fields: map[string][]string{"file": {"content"}, "service": {"unit"}}},
{namespace: "setting", patterns: []*regexp.Regexp{settingRef}, fields: map[string][]string{"file": {"content"}, "service": {"unit"}, "directory": {"owner"}}},
{namespace: "dir", patterns: []*regexp.Regexp{dirRef}, fields: map[string][]string{"*": {"path", "content", "volumes", "env", "env-file"}}},
{namespace: "access", patterns: []*regexp.Regexp{accessRef}, fields: map[string][]string{"*": {"path", "content", "volumes", "env", "env-file"}}},
{namespace: "secret", patterns: []*regexp.Regexp{placeholder}, fields: map[string][]string{"file": {"content"}},
+11
View File
@@ -436,6 +436,12 @@ const ReadinessContract = 2
// parses strictly and would refuse the whole declaration for it.
const RootContract = 3
// GrantContract is the statement of an engine that also reads an access's `grant` (novox/hq ADR 0312): it grants a
// read-write access's path group to the containers and the account the grant names, takes it back, and states the
// access unhealthy while the group cannot write. Only to such an engine is the field sent: an older one parses
// strictly and would refuse the whole declaration for it.
const GrantContract = 4
// ReasonRoot starts the reason of an account verdict that found a way to root (ADR 0266); the node-engine's
// own words (mesh-host internal/accounts ReasonRoot).
const ReasonRoot = "can become root without a person"
@@ -487,6 +493,11 @@ const KindAccount = "account"
// ReasonUsedAsFound.
const KindDirectory = "directory"
// KindAccess is a read-write access that grants its path's group (novox/hq ADR 0312): healthy while the group may
// write the path, unhealthy while it cannot — the operator's to decide, since the mesh changes nothing on the path
// — or while the path is owned by a group that is never granted. The node-engine's own (mesh-host internal/link).
const KindAccess = "access"
// ReasonUsedAsFound starts the reason of a directory used as found.
const ReasonUsedAsFound = "used as found:"
+1 -1
View File
@@ -8,7 +8,7 @@ private addresses. That is acceptable while mesh-controller is private; if it ev
a second place to clean besides the catalogue itself.
mesh-catalog b9de001833b1b61b297182b8e3bcdae1cbdeace9 modules/*/module.json, modules/nats/Dockerfile
mesh-host a76baa4b47f07a8b2bab065603414530bd02b53a examples/foundation-first-node-nats.lock
mesh-host 3f7f797168d33ec22241bdd6ef2c64de86eea914 examples/foundation-first-node-nats.lock
mesh-sdk 33917f91ac3cc213b879b32f3c70e2aa70395f98 conformance/events/module-event.json
To move them, from this repository's root, with the two repositories checked out beside it:
+1
View File
@@ -26,3 +26,4 @@
./cmd/mesh-controller TestAnUnassignThatLeavesItsNodeUnresolvedSaysSoOnceInItsAnswer
./cmd/mesh-controller TestAnUnassignThatLeavesItsNodeResolvedSaysNothingOfIt
./cmd/mesh-controller TestAnAssignThatLeavesItsNodeUnresolvedSaysOnlyThePlansRefusal
./cmd/mesh-controller TestADirectoryOwnersKeyIsRefusedThroughAVerb
+78
View File
@@ -464,6 +464,35 @@ type Access struct {
Path string `json:"path"`
// Mode is how this module reaches the path: read or read-write. Absent narrows to read.
Mode string `json:"mode,omitempty"`
// Grant is who the path's own group is granted to, for a read-write access (novox/hq ADR 0312 §1): the
// identities the module's code runs as. The controller says who; the node-engine reads which group owns
// the path when it applies, because only the machine knows, and never changes the path (§2). Absent is a
// controller older than the grant, or an access that grants nothing — and a read access always is one (§3).
Grant *AccessGrant `json:"grant,omitempty"`
}
// AccessGrant is the identities a read-write access grants its path's group to (novox/hq ADR 0312 §1).
type AccessGrant struct {
// Containers are the module's containers that mount the path, by id: each is created with the group as a
// supplementary group of its process, and recreated when the group moves (§6).
Containers []string `json:"containers,omitempty"`
// Account is the account that runs the module's tools: put in the group as an account group, as ADR 0252
// §1 puts one, and taken out again as §3 takes back (§5). Empty for a module whose code runs nowhere as
// an account.
Account string `json:"account,omitempty"`
}
// UnmarshalJSON reads a grant as strictly as a resource is read: a field the engine does not know is a
// thing the controller believes it asked for, and is refused rather than ignored.
func (g *AccessGrant) UnmarshalJSON(raw []byte) error {
type plain AccessGrant
var p plain
if err := strictDecode(raw, &p); err != nil {
return fmt.Errorf("a grant: %w", err)
}
*g = AccessGrant(p)
return nil
}
func (a *Access) Identity() string { return a.ID }
@@ -482,6 +511,24 @@ func (a *Access) validate(where string, _ bool) []string {
problems = append(problems, fmt.Sprintf(
"%s: an access is %q or %q, not %q", where, AccessRead, AccessReadWrite, a.Mode))
}
if a.Grant != nil && a.Mode != AccessReadWrite {
// A read access reads as any account may, and gains no group (novox/hq ADR 0312 §3).
problems = append(problems, where+": a read access grants nothing, and this one names a grant; "+
"only a read-write access grants its path's group (novox/hq ADR 0312 §3)")
}
if a.Grant != nil {
for _, id := range a.Grant.Containers {
if id == "" {
problems = append(problems, where+": a grant names a container with no id")
}
}
if a.Grant.Account != "" && !accountName.MatchString(a.Grant.Account) {
// Handed to id, usermod and gpasswd as an argument: a name starting with a dash would be read as an
// option by each of them.
problems = append(problems, fmt.Sprintf("%s: a grant's account is an account's name — letters, digits, "+
"dots, underscores and dashes, not starting with a dash — and %q is not one", where, a.Grant.Account))
}
}
return problems
}
@@ -1668,6 +1715,7 @@ func parse(raw []byte, allowActions bool) (*Declaration, error) {
d.Resources = append(d.Resources, resource)
}
problems = append(problems, checkWhileStopped(d.Resources)...)
problems = append(problems, checkGrants(d.Resources)...)
problems = append(problems, checkAdoption(env.Adoption, d.Resources, allowActions)...)
problems = append(problems, checkReplaces(d.Resources)...)
if env.Adoption == nil {
@@ -1732,6 +1780,36 @@ func checkWhileStopped(resources []Resource) []string {
return problems
}
// accountName is what a grant's account may be: the portable account name the user tools take, optionally ending
// in `$`, never starting with a dash.
var accountName = regexp.MustCompile(`^[A-Za-z0-9_.][A-Za-z0-9_.-]*\$?$`)
// checkGrants judges each access's grant against the whole declaration (novox/hq ADR 0312): a container it
// names is a container of this declaration, as a maintenance window's are. A grant to a container the machine
// is not told about would be a group given to nothing, said as granted.
func checkGrants(resources []Resource) []string {
containers := map[string]bool{}
for _, r := range resources {
if r.Kind() == TypeContainer {
containers[r.Identity()] = true
}
}
var problems []string
for _, r := range resources {
a, ok := r.(*Access)
if !ok || a.Grant == nil {
continue
}
for _, id := range a.Grant.Containers {
if id != "" && !containers[id] {
problems = append(problems, fmt.Sprintf("resource %q: its grant names %q, and this declaration "+
"has no container by that id", a.ID, id))
}
}
}
return problems
}
func strictDecode(raw []byte, into any) error {
// DisallowUnknownFields is the whole point rather than strictness for its own sake: a
// field the host does not know is a thing the control plane believes it asked for.
+2 -2
View File
@@ -78,7 +78,7 @@ github.com/nats-io/nkeys
# github.com/nats-io/nuid v1.0.1
## explicit
github.com/nats-io/nuid
# github.com/novox/mesh-host v0.0.0 => git.novox.be/novox/mesh-host v0.0.0-20261011092230-20d5af9b2d5f
# github.com/novox/mesh-host v0.0.0 => git.novox.be/novox/mesh-host v0.0.0-20261012112256-3f7f797168d3
## explicit; go 1.26.0
github.com/novox/mesh-host/internal/declaration
github.com/novox/mesh-host/rootsearch
@@ -135,4 +135,4 @@ golang.org/x/text/width
# golang.org/x/time v0.15.0
## explicit; go 1.25.0
golang.org/x/time/rate
# github.com/novox/mesh-host => git.novox.be/novox/mesh-host v0.0.0-20261011092230-20d5af9b2d5f
# github.com/novox/mesh-host => git.novox.be/novox/mesh-host v0.0.0-20261012112256-3f7f797168d3