Merge pull request 'Create the nftables table in the configured address family' (#32) from benvin/v4-only-family into main

Reviewed-on: #32

.woodpecker/build.yaml

@@ -3,9 +3,21 @@ when:
 
 steps:
   - name: build
-    image: golang:1.23
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/gobuilder:0.1.2-alma9
     commands:
       - go build ./...
+    environment:
+      GOCACHE_S3_BUCKET: gocache
+      GOCACHE_S3_REGION: us-east-1
+      GOCACHE_S3_ENDPOINT_URL: https://s3.ceph.unkin.net
+      GOCACHE_S3_PATH_STYLE: "true"
+      GOCACHE_KEY_PREFIX: ci-tomswall
+      GOCACHE_METRICS: "true"
+      GOCACHEPROG: go-cache-plugin --cache-dir=/tmp/gocache
+      AWS_ACCESS_KEY_ID:
+        from_secret: GOCACHE_AWS_ACCESS_KEY_ID
+      AWS_SECRET_ACCESS_KEY:
+        from_secret: GOCACHE_AWS_SECRET_ACCESS_KEY
     backend_options:
       kubernetes:
         serviceAccountName: default

.woodpecker/pre-commit.yaml

@@ -3,9 +3,21 @@ when:
 
 steps:
   - name: pre-commit
-    image: git.unkin.net/unkin/almalinux9-gobuilder:20260606
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/gobuilder:0.1.2-alma9
     commands:
       - uvx pre-commit run --all-files
+    environment:
+      GOCACHE_S3_BUCKET: gocache
+      GOCACHE_S3_REGION: us-east-1
+      GOCACHE_S3_ENDPOINT_URL: https://s3.ceph.unkin.net
+      GOCACHE_S3_PATH_STYLE: "true"
+      GOCACHE_KEY_PREFIX: ci-tomswall
+      GOCACHE_METRICS: "true"
+      GOCACHEPROG: go-cache-plugin --cache-dir=/tmp/gocache
+      AWS_ACCESS_KEY_ID:
+        from_secret: GOCACHE_AWS_ACCESS_KEY_ID
+      AWS_SECRET_ACCESS_KEY:
+        from_secret: GOCACHE_AWS_SECRET_ACCESS_KEY
     backend_options:
       kubernetes:
         serviceAccountName: default

.woodpecker/release.yaml

@@ -4,9 +4,21 @@ when:
 
 steps:
   - name: test
-    image: golang:1.23
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/gobuilder:0.1.2-alma9
     commands:
       - go test ./...
+    environment:
+      GOCACHE_S3_BUCKET: gocache
+      GOCACHE_S3_REGION: us-east-1
+      GOCACHE_S3_ENDPOINT_URL: https://s3.ceph.unkin.net
+      GOCACHE_S3_PATH_STYLE: "true"
+      GOCACHE_KEY_PREFIX: ci-tomswall
+      GOCACHE_METRICS: "true"
+      GOCACHEPROG: go-cache-plugin --cache-dir=/tmp/gocache
+      AWS_ACCESS_KEY_ID:
+        from_secret: GOCACHE_AWS_ACCESS_KEY_ID
+      AWS_SECRET_ACCESS_KEY:
+        from_secret: GOCACHE_AWS_SECRET_ACCESS_KEY
     backend_options:
       kubernetes:
         serviceAccountName: default
@@ -19,10 +31,22 @@ steps:
             cpu: 2
 
   - name: build
-    image: git.unkin.net/unkin/almalinux9-gobuilder:20260606
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/gobuilder:0.1.2-alma9
     commands:
       - make dist-build VERSION=${CI_COMMIT_TAG}
     depends_on: [test]
+    environment:
+      GOCACHE_S3_BUCKET: gocache
+      GOCACHE_S3_REGION: us-east-1
+      GOCACHE_S3_ENDPOINT_URL: https://s3.ceph.unkin.net
+      GOCACHE_S3_PATH_STYLE: "true"
+      GOCACHE_KEY_PREFIX: ci-tomswall
+      GOCACHE_METRICS: "true"
+      GOCACHEPROG: go-cache-plugin --cache-dir=/tmp/gocache
+      AWS_ACCESS_KEY_ID:
+        from_secret: GOCACHE_AWS_ACCESS_KEY_ID
+      AWS_SECRET_ACCESS_KEY:
+        from_secret: GOCACHE_AWS_SECRET_ACCESS_KEY
     backend_options:
       kubernetes:
         serviceAccountName: default
@@ -80,3 +104,45 @@ steps:
           limits:
             memory: 512Mi
             cpu: 500m
+
+  - name: release
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/tea:0.1.0
+    environment:
+      RELEASER_TOKEN:
+        from_secret: RELEASER_TOKEN
+    commands:
+      - |
+        tea logins add --name gitea --url https://git.unkin.net --token "$${RELEASER_TOKEN}" --no-version-check
+        CUR_SHA=$$(git rev-list -n1 "${CI_COMMIT_TAG}")
+        PREV_TAG=""
+        for t in $$(git tag --sort=-v:refname); do
+          [ "$$t" = "${CI_COMMIT_TAG}" ] && continue
+          [ "$$(git rev-list -n1 "$$t")" = "$$CUR_SHA" ] && continue
+          if git merge-base --is-ancestor "$$t" "${CI_COMMIT_TAG}" 2>/dev/null; then
+            PREV_TAG="$$t"; break
+          fi
+        done
+        if [ -n "$$PREV_TAG" ]; then
+          NOTES=$$(git log "$${PREV_TAG}..${CI_COMMIT_TAG}" --pretty=format:"- %s")
+        else
+          NOTES=$$(git log --pretty=format:"- %s")
+        fi
+        tea releases create --tag "${CI_COMMIT_TAG}" --title "${CI_COMMIT_TAG}" --note "$${NOTES}" --login gitea --repo "${CI_REPO}"
+        cp dist/tomswall tomswall-linux-amd64
+        ASSETS="tomswall-linux-amd64"
+        RPM=$$(ls dist/*.rpm 2>/dev/null | head -1)
+        [ -n "$$RPM" ] && ASSETS="$$ASSETS $$RPM"
+        sha256sum $$ASSETS > sha256sums.txt
+        tea releases assets create "${CI_COMMIT_TAG}" $$ASSETS sha256sums.txt \
+          --login gitea --repo "${CI_REPO}"
+    depends_on: [upload-rpm]
+    backend_options:
+      kubernetes:
+        serviceAccountName: default
+        resources:
+          requests:
+            memory: 128Mi
+            cpu: 100m
+          limits:
+            memory: 512Mi
+            cpu: 500m

.woodpecker/test.yaml

@@ -3,10 +3,22 @@ when:
 
 steps:
   - name: test
-    image: golang:1.23
+    image: artifactapi.k8s.syd1.au.unkin.net/docker-internal/gobuilder:0.1.2-alma9
     commands:
       - go vet ./...
       - go test ./...
+    environment:
+      GOCACHE_S3_BUCKET: gocache
+      GOCACHE_S3_REGION: us-east-1
+      GOCACHE_S3_ENDPOINT_URL: https://s3.ceph.unkin.net
+      GOCACHE_S3_PATH_STYLE: "true"
+      GOCACHE_KEY_PREFIX: ci-tomswall
+      GOCACHE_METRICS: "true"
+      GOCACHEPROG: go-cache-plugin --cache-dir=/tmp/gocache
+      AWS_ACCESS_KEY_ID:
+        from_secret: GOCACHE_AWS_ACCESS_KEY_ID
+      AWS_SECRET_ACCESS_KEY:
+        from_secret: GOCACHE_AWS_SECRET_ACCESS_KEY
     backend_options:
       kubernetes:
         serviceAccountName: default

cmd/tomswall/guard_test.go

@@ -0,0 +1,30 @@
+package main
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/spf13/cobra"
+
+	"git.unkin.net/unkin/tomswall/internal/tryapply"
+)
+
+func TestMutatingCommandsRefuseWhileTryPending(t *testing.T) {
+	orig := tryapply.Dir
+	tryapply.Dir = t.TempDir()
+	t.Cleanup(func() { tryapply.Dir = orig })
+	if err := os.WriteFile(filepath.Join(tryapply.Dir, "try-snapshot.json"), []byte("{}"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	configPath = "../../tomswall.example.yaml"
+
+	for _, cmd := range []*cobra.Command{applyCmd(), flushCmd(), purgeCmd()} {
+		t.Run(cmd.Use, func(t *testing.T) {
+			if err := cmd.RunE(cmd, nil); !errors.Is(err, tryapply.ErrPending) {
+				t.Errorf("got %v, want ErrPending", err)
+			}
+		})
+	}
+}

cmd/tomswall/main.go

@@ -12,6 +12,7 @@ import (
 	"git.unkin.net/unkin/tomswall/internal/config"
 	"git.unkin.net/unkin/tomswall/internal/nftables"
 	"git.unkin.net/unkin/tomswall/internal/shorewall"
+	"git.unkin.net/unkin/tomswall/internal/tryapply"
 )
 
 var configPath string
@@ -33,6 +34,9 @@ Use 'tomswall migrate' to convert a shorewall config to YAML.`,
 
 	root.AddCommand(
 		applyCmd(),
+		tryCmd(),
+		confirmCmd(),
+		revertCmd(),
 		planCmd(),
 		validateCmd(),
 		statusCmd(),
@@ -86,6 +90,12 @@ The firewall is never torn down — existing connections are preserved.`,
 				return err
 			}
 
+			unlock, err := tryapply.Acquire()
+			if err != nil {
+				return err
+			}
+			defer unlock()
+
 			engine, err := nftables.NewEngine(cfg)
 			if err != nil {
 				return fmt.Errorf("initializing nftables: %w", err)
@@ -227,6 +237,14 @@ func purgeCmd() *cobra.Command {
 				return err
 			}
 
+			if !dryRun {
+				unlock, err := tryapply.Acquire()
+				if err != nil {
+					return err
+				}
+				defer unlock()
+			}
+
 			engine, err := nftables.NewEngine(cfg)
 			if err != nil {
 				return fmt.Errorf("initializing nftables: %w", err)
@@ -272,6 +290,12 @@ func flushCmd() *cobra.Command {
 				return err
 			}
 
+			unlock, err := tryapply.Acquire()
+			if err != nil {
+				return err
+			}
+			defer unlock()
+
 			engine, err := nftables.NewEngine(cfg)
 			if err != nil {
 				return fmt.Errorf("initializing nftables: %w", err)

cmd/tomswall/try.go

@@ -0,0 +1,168 @@
+package main
+
+import (
+	"fmt"
+	"os"
+	"os/signal"
+	"strings"
+	"syscall"
+	"time"
+
+	"github.com/spf13/cobra"
+
+	"git.unkin.net/unkin/tomswall/internal/config"
+	"git.unkin.net/unkin/tomswall/internal/nftables"
+	"git.unkin.net/unkin/tomswall/internal/tryapply"
+)
+
+// revertGrace lets the in-process revert win before the systemd fallback fires.
+const revertGrace = 30 * time.Second
+
+func tryCmd() *cobra.Command {
+	var timeout time.Duration
+	cmd := &cobra.Command{
+		Use:   "try",
+		Short: "Apply configuration and revert unless confirmed within a timeout",
+		Long: `Try snapshots the live tomswall table to disk, applies the configuration, then
+waits for 'tomswall confirm'. On timeout, interrupt or hangup the snapshot is
+restored atomically. A transient systemd timer restores it too if this process
+dies. Confirm from a new session to prove new connections still work.`,
+		RunE: func(cmd *cobra.Command, args []string) error {
+			cfg, err := loadConfig()
+			if err != nil {
+				return err
+			}
+
+			// Registered before apply so a confirm or hangup cannot be missed.
+			signal.Ignore(syscall.SIGPIPE)
+			confirm := make(chan os.Signal, 1)
+			signal.Notify(confirm, syscall.SIGUSR1)
+			abort := make(chan os.Signal, 1)
+			signal.Notify(abort, syscall.SIGHUP, syscall.SIGINT, syscall.SIGTERM)
+
+			id, err := tryApply(cfg, timeout+revertGrace)
+			if err != nil || id == "" {
+				return err
+			}
+			fmt.Printf("Applied. Run 'tomswall confirm' within %s or the previous ruleset is restored.\n", timeout)
+			msg, err := confirmOrRevert(confirm, abort, timeout, func() (bool, error) { return tryapply.Revert(id) })
+			if err != nil {
+				return err
+			}
+			fmt.Println(msg)
+			return nil
+		},
+	}
+	cmd.Flags().DurationVar(&timeout, "timeout", 60*time.Second, "time to wait for confirmation before reverting")
+	return cmd
+}
+
+// tryApply applies cfg under a pending try and returns its ID, or "" when
+// there was nothing to change.
+func tryApply(cfg *config.Config, fallback time.Duration) (string, error) {
+	unlock, err := tryapply.Acquire()
+	if err != nil {
+		return "", err
+	}
+	defer unlock()
+
+	engine, err := nftables.NewEngine(cfg)
+	if err != nil {
+		return "", fmt.Errorf("initializing nftables: %w", err)
+	}
+	changes, err := engine.Plan()
+	if err != nil {
+		return "", fmt.Errorf("computing changes: %w", err)
+	}
+	if changes.Empty() {
+		fmt.Println("No changes needed — firewall is up to date.")
+		return "", nil
+	}
+	fmt.Println(changes.Summary())
+
+	snap, err := engine.Snapshot()
+	if err != nil {
+		return "", fmt.Errorf("snapshotting ruleset: %w", err)
+	}
+	id, err := tryapply.Arm(snap, os.Getpid(), fallback)
+	if err != nil {
+		return "", err
+	}
+	if err := engine.Apply(changes); err != nil {
+		if aerr := tryapply.Abort(); aerr != nil {
+			return "", fmt.Errorf("applying changes: %w; %v; the revert timer restores the previous ruleset within %s", err, aerr, fallback)
+		}
+		return "", fmt.Errorf("applying changes: %w: previous ruleset restored", err)
+	}
+	return id, nil
+}
+
+// confirmOrRevert waits for confirm; on abort or timeout it runs revert.
+// A confirm already delivered wins over a simultaneous abort or timeout.
+func confirmOrRevert(confirm, abort <-chan os.Signal, timeout time.Duration, revert func() (bool, error)) (string, error) {
+	reason := "not confirmed within " + timeout.String()
+	select {
+	case <-confirm:
+		return "Confirmed.", nil
+	case s := <-abort:
+		reason = "interrupted by " + s.String()
+	case <-time.After(timeout):
+	}
+	select {
+	case <-confirm:
+		return "Confirmed.", nil
+	default:
+	}
+	reverted, err := revert()
+	if err != nil {
+		return "", fmt.Errorf("%s; revert failed: %w", reason, err)
+	}
+	if !reverted {
+		return "Already resolved by 'tomswall confirm' or 'tomswall revert'.", nil
+	}
+	return "", fmt.Errorf("%s: previous ruleset restored", reason)
+}
+
+func confirmCmd() *cobra.Command {
+	return &cobra.Command{
+		Use:   "confirm",
+		Short: "Keep the configuration applied by a pending 'tomswall try'",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			pid, ok, err := tryapply.Confirm()
+			if err != nil {
+				return err
+			}
+			if !ok {
+				return fmt.Errorf("no pending 'tomswall try': it was already reverted or confirmed")
+			}
+			// A recycled PID must not be signalled.
+			if comm, err := os.ReadFile(fmt.Sprintf("/proc/%d/comm", pid)); err == nil && strings.TrimSpace(string(comm)) == "tomswall" {
+				_ = syscall.Kill(pid, syscall.SIGUSR1)
+			}
+			fmt.Println("Confirmed.")
+			return nil
+		},
+	}
+}
+
+func revertCmd() *cobra.Command {
+	var id string
+	cmd := &cobra.Command{
+		Use:   "revert",
+		Short: "Restore the ruleset saved by a pending 'tomswall try'",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			reverted, err := tryapply.Revert(id)
+			if err != nil {
+				return err
+			}
+			if !reverted {
+				fmt.Println("No pending 'tomswall try'.")
+				return nil
+			}
+			fmt.Println("Previous ruleset restored.")
+			return nil
+		},
+	}
+	cmd.Flags().StringVar(&id, "id", "", "only revert the try with this ID (used by the revert timer)")
+	return cmd
+}

cmd/tomswall/try_test.go

@@ -0,0 +1,63 @@
+package main
+
+import (
+	"errors"
+	"os"
+	"syscall"
+	"testing"
+	"time"
+)
+
+func TestConfirmOrRevert(t *testing.T) {
+	tests := []struct {
+		name         string
+		confirm      bool
+		abort        bool
+		resolved     bool
+		revertErr    error
+		wantRevert   bool
+		wantErr      bool
+		wantResolved bool
+	}{
+		{name: "confirmed keeps ruleset", confirm: true},
+		{name: "confirm wins over simultaneous abort", confirm: true, abort: true},
+		{name: "timeout reverts", wantRevert: true, wantErr: true},
+		{name: "hangup reverts", abort: true, wantRevert: true, wantErr: true},
+		{name: "revert failure surfaces", revertErr: errors.New("boom"), wantRevert: true, wantErr: true},
+		{name: "resolved elsewhere", resolved: true, wantRevert: true, wantResolved: true},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			for i := 0; i < 20; i++ { // select between ready channels is random
+				confirm := make(chan os.Signal, 1)
+				abort := make(chan os.Signal, 1)
+				if tt.confirm {
+					confirm <- syscall.SIGUSR1
+				}
+				if tt.abort {
+					abort <- syscall.SIGHUP
+				}
+				called := false
+				msg, err := confirmOrRevert(confirm, abort, 10*time.Millisecond, func() (bool, error) {
+					called = true
+					return !tt.resolved, tt.revertErr
+				})
+				if called != tt.wantRevert {
+					t.Fatalf("revert called = %v, want %v", called, tt.wantRevert)
+				}
+				if (err != nil) != tt.wantErr {
+					t.Fatalf("err = %v, wantErr %v", err, tt.wantErr)
+				}
+				if tt.revertErr != nil && !errors.Is(err, tt.revertErr) {
+					t.Fatalf("revert error not wrapped: %v", err)
+				}
+				if tt.wantResolved && msg == "" {
+					t.Fatal("expected already-resolved message")
+				}
+				if !(tt.confirm && tt.abort) {
+					break
+				}
+			}
+		})
+	}
+}

go.mod

@@ -3,19 +3,18 @@ module git.unkin.net/unkin/tomswall
 go 1.23
 
 require (
-	github.com/google/nftables v0.2.0
+	github.com/google/nftables v0.3.0
+	github.com/mdlayher/netlink v1.7.3-0.20250113171957-fbb4dce95f42
 	github.com/spf13/cobra v1.8.1
-	golang.org/x/sys v0.18.0
+	golang.org/x/sys v0.28.0
 	gopkg.in/yaml.v3 v3.0.1
 )
 
 require (
 	github.com/google/go-cmp v0.6.0 // indirect
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
-	github.com/josharian/native v1.1.0 // indirect
-	github.com/mdlayher/netlink v1.7.2 // indirect
 	github.com/mdlayher/socket v0.5.1 // indirect
 	github.com/spf13/pflag v1.0.5 // indirect
-	golang.org/x/net v0.23.0 // indirect
+	golang.org/x/net v0.33.0 // indirect
 	golang.org/x/sync v0.6.0 // indirect
 )

go.sum

@@ -1,14 +1,12 @@
 github.com/cpuguy83/go-md2man/v2 v2.0.4/go.mod h1:tgQtvFlXSQOSOSIRvRPT7W67SCa46tRHOmNcaadrF8o=
 github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
 github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
-github.com/google/nftables v0.2.0 h1:PbJwaBmbVLzpeldoeUKGkE2RjstrjPKMl6oLrfEJ6/8=
+github.com/google/nftables v0.3.0 h1:bkyZ0cbpVeMHXOrtlFc8ISmfVqq5gPJukoYieyVmITg=
-github.com/google/nftables v0.2.0/go.mod h1:Beg6V6zZ3oEn0JuiUQ4wqwuyqqzasOltcoXPtgLbFp4=
+github.com/google/nftables v0.3.0/go.mod h1:BCp9FsrbF1Fn/Yu6CLUc9GGZFw/+hsxfluNXXmxBfRM=
 github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
 github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLfsEA9PFc4w1p2J65bw=
-github.com/josharian/native v1.1.0 h1:uuaP0hAbW7Y4l0ZRQ6C9zfb7Mg1mbFKry/xzDAfmtLA=
+github.com/mdlayher/netlink v1.7.3-0.20250113171957-fbb4dce95f42 h1:A1Cq6Ysb0GM0tpKMbdCXCIfBclan4oHk1Jb+Hrejirg=
-github.com/josharian/native v1.1.0/go.mod h1:7X/raswPFr05uY3HiLlYeyQntB6OO7E/d2Cu7qoaN2w=
+github.com/mdlayher/netlink v1.7.3-0.20250113171957-fbb4dce95f42/go.mod h1:BB4YCPDOzfy7FniQ/lxuYQ3dgmM2cZumHbK8RpTjN2o=
-github.com/mdlayher/netlink v1.7.2 h1:/UtM3ofJap7Vl4QWCPDGXY8d3GIY2UGSDbK+QWmY8/g=
-github.com/mdlayher/netlink v1.7.2/go.mod h1:xraEF7uJbxLhc5fpHL4cPe221LI2bdttWlU+ZGLfQSw=
 github.com/mdlayher/socket v0.5.1 h1:VZaqt6RkGkt2OE9l3GcC6nZkqD3xKeQLyfleW/uBcos=
 github.com/mdlayher/socket v0.5.1/go.mod h1:TjPLHI1UgwEv5J1B5q0zTZq12A/6H7nKmtTanQE37IQ=
 github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
@@ -16,14 +14,14 @@ github.com/spf13/cobra v1.8.1 h1:e5/vxKd/rZsfSJMUX1agtjeTDf+qv1/JdBF8gg5k9ZM=
 github.com/spf13/cobra v1.8.1/go.mod h1:wHxEcudfqmLYa8iTfL+OuZPbBZkmvliBWKIezN3kD9Y=
 github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
 github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
-github.com/vishvananda/netns v0.0.0-20180720170159-13995c7128cc h1:R83G5ikgLMxrBvLh22JhdfI8K6YXEPHx5P03Uu3DRs4=
+github.com/vishvananda/netns v0.0.4 h1:Oeaw1EM2JMxD51g9uhtC0D7erkIjgmj8+JZc26m1YX8=
-github.com/vishvananda/netns v0.0.0-20180720170159-13995c7128cc/go.mod h1:ZjcWmFBXmLKZu9Nxj3WKYEafiSqer2rnvPr0en9UNpI=
+github.com/vishvananda/netns v0.0.4/go.mod h1:SpkAiCQRtJ6TvvxPnOSyH3BMl6unz3xZlaprSwhNNJM=
-golang.org/x/net v0.23.0 h1:7EYJ93RZ9vYSZAIb2x3lnuvqO5zneoD6IvWjuhfxjTs=
+golang.org/x/net v0.33.0 h1:74SYHlV8BIgHIFC/LrYkOGIwL19eTYXQ5wc6TBuO36I=
-golang.org/x/net v0.23.0/go.mod h1:JKghWKKOSdJwpW2GEx0Ja7fmaKnMsbu+MWVZTokSYmg=
+golang.org/x/net v0.33.0/go.mod h1:HXLR5J+9DxmrqMwG9qjGCxZ+zKXxBru04zlTvWlWuN4=
 golang.org/x/sync v0.6.0 h1:5BMeUDZ7vkXGfEr1x9B4bRcTH4lpkTkpdh0T/J+qjbQ=
 golang.org/x/sync v0.6.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
-golang.org/x/sys v0.18.0 h1:DBdB3niSjOA/O0blCZBqDefyWNYveAYMNF1Wum0DYQ4=
+golang.org/x/sys v0.28.0 h1:Fksou7UEQUWlKvIdsqzJmUmCX3cZuD2+P3XyyzwMhlA=
-golang.org/x/sys v0.18.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/sys v0.28.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=

internal/agent/agent.go

@@ -8,6 +8,7 @@ import (
 
 	"git.unkin.net/unkin/tomswall/internal/config"
 	"git.unkin.net/unkin/tomswall/internal/nftables"
+	"git.unkin.net/unkin/tomswall/internal/tryapply"
 )
 
 // Applier applies a translated config to the firewall. Abstracted so the run
@@ -103,8 +104,14 @@ func (a *Agent) applyConfig(ctx context.Context, rc *RenderedConfig, report bool
 // EngineApplier applies via the real nftables differential engine.
 type EngineApplier struct{}
 
-// Apply computes and applies the differential change set for cfg.
+// Apply computes and applies the differential change set for cfg. It refuses
+// while a 'tomswall try' awaits confirmation.
 func (EngineApplier) Apply(_ context.Context, cfg *config.Config) error {
+	unlock, err := tryapply.Acquire()
+	if err != nil {
+		return err
+	}
+	defer unlock()
 	engine, err := nftables.NewEngine(cfg)
 	if err != nil {
 		return fmt.Errorf("initializing nftables: %w", err)

internal/config/blrules.go

@@ -55,20 +55,11 @@ func (c *Config) validateBlrules() error {
 			return fmt.Errorf("blrules[%d]: dest required", i)
 		}
 
-		if r.Source != "all" && r.Source != "any" && r.Source != "none" &&
+		if err := c.validateZoneRef(r.Source); err != nil {
-			!hasPrefix(r.Source, "all!") && !hasPrefix(r.Source, "any!") {
+			return fmt.Errorf("blrules[%d]: source %w", i, err)
-			srcZone := zoneFromSpec(r.Source)
-			if _, ok := c.Zones[srcZone]; !ok {
-				return fmt.Errorf("blrules[%d]: source zone %q not defined", i, srcZone)
-			}
 		}
-
+		if err := c.validateZoneRef(r.Dest); err != nil {
-		if r.Dest != "all" && r.Dest != "any" && r.Dest != "none" &&
+			return fmt.Errorf("blrules[%d]: dest %w", i, err)
-			!hasPrefix(r.Dest, "all!") && !hasPrefix(r.Dest, "any!") {
-			dstZone := zoneFromSpec(r.Dest)
-			if _, ok := c.Zones[dstZone]; !ok {
-				return fmt.Errorf("blrules[%d]: dest zone %q not defined", i, dstZone)
-			}
 		}
 	}
 	return nil

internal/config/config.go

@@ -59,6 +59,11 @@ type Settings struct {
 
 	// When true, auto-generate CONTINUE policies for sub-zones to their parent zones.
 	ImplicitContinue bool `yaml:"implicit_continue,omitempty"`
+
+	// Verdict for ct state invalid/untracked packets; continue passes them to the rules.
+	// Unset: invalid drops, untracked continues.
+	InvalidDisposition   PolicyAction `yaml:"invalid_disposition,omitempty"`
+	UntrackedDisposition PolicyAction `yaml:"untracked_disposition,omitempty"`
 }
 
 // Load reads a config file in YAML or JSON format (detected by extension).
@@ -115,6 +120,16 @@ func (c *Config) validateSettings() error {
 	if !validAddressFamilies[c.Settings.AddressFamily] {
 		return fmt.Errorf("unknown address_family %q (use inet, ip, or ip6)", c.Settings.AddressFamily)
 	}
+	for name, d := range map[string]PolicyAction{
+		"invalid_disposition":   c.Settings.InvalidDisposition,
+		"untracked_disposition": c.Settings.UntrackedDisposition,
+	} {
+		switch d {
+		case "", PolicyAccept, PolicyDrop, PolicyReject, PolicyContinue:
+		default:
+			return fmt.Errorf("unknown %s %q (use accept, drop, reject, or continue)", name, d)
+		}
+	}
 	return nil
 }
 

internal/config/config_test.go

@@ -3,6 +3,7 @@ package config
 import (
 	"os"
 	"path/filepath"
+	"reflect"
 	"strings"
 	"testing"
 )
@@ -854,6 +855,32 @@ func TestValidateRules(t *testing.T) {
 			},
 			wantErr: "source zone \"missing\" not defined",
 		},
+		{
+			name: "comma zone lists are valid",
+			rules: []Rule{
+				{Action: RuleAccept, Source: "fw,loc", Dest: "loc,net:192.0.2.1,198.51.100.1"},
+			},
+		},
+		{
+			name: "undefined zone in dest list",
+			rules: []Rule{
+				{Action: RuleAccept, Source: "loc", Dest: "net,missing"},
+			},
+			wantErr: "dest zone \"missing\" not defined",
+		},
+		{
+			name: "negation prefixing the whole address list is valid",
+			rules: []Rule{
+				{Action: RuleAccept, Source: "net:!192.0.2.1,198.51.100.1", Dest: "fw"},
+			},
+		},
+		{
+			name: "negation inside an address list",
+			rules: []Rule{
+				{Action: RuleAccept, Source: "net", Dest: "loc:192.0.2.1,!198.51.100.1"},
+			},
+			wantErr: "'!' may only prefix the whole address list",
+		},
 		{
 			name: "all keyword is valid source",
 			rules: []Rule{
@@ -1006,3 +1033,39 @@ func TestValidateSNAT(t *testing.T) {
 		})
 	}
 }
+
+func TestSplitZoneList(t *testing.T) {
+	tests := []struct {
+		in   string
+		want []ZoneSpec
+	}{
+		{"net", []ZoneSpec{{Zone: "net"}}},
+		{"fw,lan,svr", []ZoneSpec{{Zone: "fw"}, {Zone: "lan"}, {Zone: "svr"}}},
+		{"svr:192.0.2.17", []ZoneSpec{{Zone: "svr", Addr: "192.0.2.17"}}},
+		{"net:192.0.2.1,198.51.100.1", []ZoneSpec{{Zone: "net", Addr: "192.0.2.1,198.51.100.1"}}},
+		{"lan,svr:192.0.2.17", []ZoneSpec{{Zone: "lan"}, {Zone: "svr", Addr: "192.0.2.17"}}},
+		{"net:2001:db8::1", []ZoneSpec{{Zone: "net", Addr: "2001:db8::1"}}},
+	}
+	for _, tt := range tests {
+		if got := SplitZoneList(tt.in); !reflect.DeepEqual(got, tt.want) {
+			t.Errorf("SplitZoneList(%q) = %+v, want %+v", tt.in, got, tt.want)
+		}
+	}
+}
+
+func TestValidateDispositions(t *testing.T) {
+	for _, tc := range []struct {
+		invalid, untracked PolicyAction
+		wantErr            string
+	}{
+		{"", "", ""},
+		{PolicyContinue, PolicyDrop, ""},
+		{"bogus", "", `unknown invalid_disposition "bogus"`},
+		{PolicyAccept, "log", `unknown untracked_disposition "log"`},
+	} {
+		c := baseConfig()
+		c.Settings.InvalidDisposition = tc.invalid
+		c.Settings.UntrackedDisposition = tc.untracked
+		checkErr(t, c.Validate(), tc.wantErr)
+	}
+}

internal/config/conntrack.go

@@ -1,6 +1,9 @@
 package config
 
-import "fmt"
+import (
+	"fmt"
+	"strings"
+)
 
 type ConntrackAction string
 
@@ -68,8 +71,14 @@ func (c *Config) validateConntrack() error {
 			return fmt.Errorf("conntrack[%d]: helper name required for helper action", i)
 		}
 
-		if ct.Source == "" && ct.Dest == "" && ct.Action != ConntrackHelper {
+		if HasZoneExclusion(ct.Source) || HasZoneExclusion(ct.Dest) {
-			return fmt.Errorf("conntrack[%d]: source or dest required", i)
+			return fmt.Errorf("conntrack[%d]: zone exclusions are not supported in conntrack entries", i)
+		}
+		if err := c.validateZoneRef(ct.Source); err != nil {
+			return fmt.Errorf("conntrack[%d]: source %w", i, err)
+		}
+		if err := c.validateZoneRef(ct.Dest); err != nil {
+			return fmt.Errorf("conntrack[%d]: dest %w", i, err)
 		}
 
 		if ct.User != "" {
@@ -84,3 +93,9 @@ func (c *Config) validateConntrack() error {
 	}
 	return nil
 }
+
+// HasZoneExclusion reports an all/any zone ref with a "+" or "!" modifier (all+, all!x, any+!x, ...).
+func HasZoneExclusion(spec string) bool {
+	zones, _, _ := strings.Cut(spec, ":")
+	return (strings.HasPrefix(zones, "all") || strings.HasPrefix(zones, "any")) && strings.ContainsAny(zones[3:], "+!")
+}

internal/config/extras_test.go

@@ -53,11 +53,52 @@ func TestValidateConntrack(t *testing.T) {
 			},
 		},
 		{
-			name: "source or dest required for non-helper",
+			name:  "omitted source and dest is valid",
-			rules: []ConntrackRule{
+			rules: []ConntrackRule{{Action: ConntrackNoTrack, Proto: "udp", DPort: PortSpec{"53"}}},
-				{Action: ConntrackDrop},
+		},
-			},
+		{
-			wantErr: "source or dest required",
+			name:    "unknown source zone",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Source: "nte"}},
+			wantErr: `source zone "nte" not defined`,
+		},
+		{
+			name:    "unknown dest zone",
+			rules:   []ConntrackRule{{Action: ConntrackDrop, Source: "net", Dest: "nte:192.0.2.1"}},
+			wantErr: `dest zone "nte" not defined`,
+		},
+		{
+			name:  "all and plain zone forms are valid",
+			rules: []ConntrackRule{{Action: ConntrackNoTrack, Source: "net,fw", Dest: "all:192.0.2.1"}},
+		},
+		{
+			name:    "Source all!net rejected",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Source: "all!net"}},
+			wantErr: "zone exclusions are not supported in conntrack entries",
+		},
+		{
+			name:    "Dest all!net:192.0.2.1 rejected",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Dest: "all!net:192.0.2.1"}},
+			wantErr: "zone exclusions are not supported in conntrack entries",
+		},
+		{
+			name:    "Source all+ rejected",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Source: "all+"}},
+			wantErr: "zone exclusions are not supported in conntrack entries",
+		},
+		{
+			name:    "Dest all+!net rejected",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Dest: "all+!net"}},
+			wantErr: "zone exclusions are not supported in conntrack entries",
+		},
+		{
+			name:    "Source any!net rejected",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Source: "any!net"}},
+			wantErr: "zone exclusions are not supported in conntrack entries",
+		},
+		{
+			name:    "Dest any+ rejected",
+			rules:   []ConntrackRule{{Action: ConntrackNoTrack, Dest: "any+"}},
+			wantErr: "zone exclusions are not supported in conntrack entries",
 		},
 		{
 			name: "helper without source/dest is valid",

internal/config/rules.go

@@ -1,6 +1,9 @@
 package config
 
-import "fmt"
+import (
+	"fmt"
+	"strings"
+)
 
 type RuleAction string
 
@@ -170,25 +173,13 @@ func (c *Config) validateRules() error {
 			return fmt.Errorf("rule[%d]: dest required", i)
 		}
 
-		if r.Source != "all" && r.Source != "any" && r.Source != "none" &&
+		if err := c.validateZoneRef(r.Source); err != nil {
-			!hasPrefix(r.Source, "all+") && !hasPrefix(r.Source, "all!") && !hasPrefix(r.Source, "any!") {
+			return fmt.Errorf("rule[%d]: source %w", i, err)
-			for _, srcPart := range splitZones(r.Source) {
-				srcZone := zoneFromSpec(srcPart)
-				if _, ok := c.Zones[srcZone]; !ok {
-					return fmt.Errorf("rule[%d]: source zone %q not defined", i, srcZone)
-				}
-			}
 		}
 
 		if r.Action != RuleDNAT && r.Action != RuleRedirect && r.Action != RuleNoNAT {
-			if r.Dest != "all" && r.Dest != "any" && r.Dest != "none" &&
+			if err := c.validateZoneRef(r.Dest); err != nil {
-				!hasPrefix(r.Dest, "all+") && !hasPrefix(r.Dest, "all!") && !hasPrefix(r.Dest, "any!") {
+				return fmt.Errorf("rule[%d]: dest %w", i, err)
-				for _, dstPart := range splitZones(r.Dest) {
-					dstZone := zoneFromSpec(dstPart)
-					if _, ok := c.Zones[dstZone]; !ok {
-						return fmt.Errorf("rule[%d]: dest zone %q not defined", i, dstZone)
-					}
-				}
 			}
 		}
 
@@ -223,6 +214,30 @@ func (c *Config) validateRules() error {
 	return nil
 }
 
+// ZoneSpec is one zone of a SOURCE/DEST list; Addr is its comma-separated address list, if any.
+type ZoneSpec struct{ Zone, Addr string }
+
+// SplitZoneList parses "lan,svr:a,b": commas before the first colon separate zones,
+// commas after it separate addresses of the last zone (shorewall semantics).
+func SplitZoneList(spec string) []ZoneSpec {
+	zones, addr, _ := strings.Cut(spec, ":")
+	var out []ZoneSpec
+	for _, z := range strings.Split(zones, ",") {
+		if z = strings.TrimSpace(z); z != "" {
+			out = append(out, ZoneSpec{Zone: z})
+		}
+	}
+	if len(out) > 0 {
+		out[len(out)-1].Addr = addr
+	}
+	return out
+}
+
+// validAddrList reports whether '!' appears only at the start, negating the whole list.
+func validAddrList(addr string) bool {
+	return !strings.Contains(strings.TrimPrefix(addr, "!"), "!")
+}
+
 // zoneFromSpec extracts the zone name from a zone spec like "net" or "net:192.168.1.0/24".
 func zoneFromSpec(spec string) string {
 	for i, c := range spec {
@@ -233,6 +248,32 @@ func zoneFromSpec(spec string) string {
 	return spec
 }
 
-func hasPrefix(s, prefix string) bool {
+// validateZoneRef checks a SOURCE/DEST spec: all/any[+][!excluded,...][:addr], none, or a declared zone list.
-	return len(s) >= len(prefix) && s[:len(prefix)] == prefix
+func (c *Config) validateZoneRef(spec string) error {
+	zones, addr, _ := strings.Cut(spec, ":")
+	base, excl, isExcl := strings.Cut(zones, "!")
+	switch base {
+	case "", "none", "all", "all+", "any", "any+":
+		if base == "" && isExcl {
+			return fmt.Errorf("%q: exclusion needs all or any", spec)
+		}
+		for _, z := range strings.Split(excl, ",") {
+			if _, ok := c.Zones[strings.TrimSpace(z)]; isExcl && !ok {
+				return fmt.Errorf("excluded zone %q not defined", z)
+			}
+		}
+		if !validAddrList(addr) {
+			return fmt.Errorf("%q: '!' may only prefix the whole address list", addr)
+		}
+		return nil
+	}
+	for _, zs := range SplitZoneList(spec) {
+		if _, ok := c.Zones[zs.Zone]; !ok {
+			return fmt.Errorf("zone %q not defined", zs.Zone)
+		}
+		if !validAddrList(zs.Addr) {
+			return fmt.Errorf("%q: '!' may only prefix the whole address list", zs.Addr)
+		}
+	}
+	return nil
 }

internal/nftables/cleanup.go

@@ -28,7 +28,7 @@ func (e *Engine) FindForeignRules() ([]ForeignRule, error) {
 
 	var ourTable *nftables.Table
 	for _, t := range tables {
-		if t.Name == e.cfg.Settings.TableName && t.Family == nftables.TableFamilyINet {
+		if t.Name == e.cfg.Settings.TableName && t.Family == e.family() {
 			ourTable = t
 			break
 		}
@@ -52,7 +52,7 @@ func (e *Engine) FindForeignRules() ([]ForeignRule, error) {
 
 	var foreign []ForeignRule
 
-	chains, err := e.conn.ListChainsOfTableFamily(nftables.TableFamilyINet)
+	chains, err := e.conn.ListChainsOfTableFamily(e.family())
 	if err != nil {
 		return nil, fmt.Errorf("listing chains: %w", err)
 	}

internal/nftables/diff.go

@@ -2,6 +2,8 @@ package nftables
 
 import (
 	"fmt"
+	"reflect"
+	"sort"
 	"strings"
 
 	"github.com/google/nftables/expr"
@@ -12,19 +14,31 @@ type ManagedRule struct {
 	Handle uint64
 	Exprs  []expr.Any
 	Tag    string
+	// Before is the handle of the existing rule an added rule is inserted ahead of; 0 appends.
+	Before uint64
 }
 
 type FirewallState struct {
 	Rules map[string][]ManagedRule
+	// Helpers are the ct helper objects in kernel (insertion) order.
+	Helpers []Helper
+}
+
+// Helper is a named ct helper object.
+type Helper struct {
+	Name   string        `json:"name"`
+	Helper expr.CtHelper `json:"helper"`
 }
 
 type ChangeSet struct {
-	Add    []ManagedRule
+	Add           []ManagedRule
-	Remove []ManagedRule
+	Remove        []ManagedRule
+	AddHelpers    []Helper
+	RemoveHelpers []string
 }
 
 func (cs *ChangeSet) Empty() bool {
-	return len(cs.Add) == 0 && len(cs.Remove) == 0
+	return len(cs.Add) == 0 && len(cs.Remove) == 0 && len(cs.AddHelpers) == 0 && len(cs.RemoveHelpers) == 0
 }
 
 func (cs *ChangeSet) Summary() string {
@@ -32,7 +46,11 @@ func (cs *ChangeSet) Summary() string {
 	if len(cs.Add) > 0 {
 		fmt.Fprintf(&b, "  + %d rule(s) to add\n", len(cs.Add))
 		for _, r := range cs.Add {
-			fmt.Fprintf(&b, "    + [%s] %s\n", r.Chain, r.Tag)
+			if r.Before != 0 {
+				fmt.Fprintf(&b, "    + [%s] %s (before handle %d)\n", r.Chain, r.Tag, r.Before)
+			} else {
+				fmt.Fprintf(&b, "    + [%s] %s\n", r.Chain, r.Tag)
+			}
 		}
 	}
 	if len(cs.Remove) > 0 {
@@ -41,69 +59,120 @@ func (cs *ChangeSet) Summary() string {
 			fmt.Fprintf(&b, "    - [%s] %s (handle %d)\n", r.Chain, r.Tag, r.Handle)
 		}
 	}
+	for _, h := range cs.AddHelpers {
+		fmt.Fprintf(&b, "  + ct helper %q\n", h.Name)
+	}
+	for _, n := range cs.RemoveHelpers {
+		fmt.Fprintf(&b, "  - ct helper %q\n", n)
+	}
 	return b.String()
 }
 
+// Rules are first-match, so order matters: keep the common prefix and suffix of
+// each chain, replace the middle, and insert the new rules before the first kept
+// suffix rule (or append when there is none).
 func computeDiff(current, desired *FirewallState) *ChangeSet {
-	cs := &ChangeSet{}
+	cs := diffHelpers(current, desired)
 
-	currentByTag := make(map[string][]ManagedRule)
+	chains := make([]string, 0, len(current.Rules)+len(desired.Rules))
-	for _, rules := range current.Rules {
+	for c := range current.Rules {
-		for _, r := range rules {
+		chains = append(chains, c)
+	}
+	for c := range desired.Rules {
+		if _, ok := current.Rules[c]; !ok {
+			chains = append(chains, c)
+		}
+	}
+	sort.Strings(chains)
+
+	for _, chain := range chains {
+		var cur []ManagedRule
+		for _, r := range current.Rules[chain] {
 			if r.Tag != "" {
-				currentByTag[r.Tag] = append(currentByTag[r.Tag], r)
+				cur = append(cur, r)
 			}
 		}
-	}
+		want := desired.Rules[chain]
 
-	desiredByTag := make(map[string][]ManagedRule)
+		pre := 0
-	for _, rules := range desired.Rules {
+		for pre < len(cur) && pre < len(want) && ruleEqual(cur[pre], want[pre]) {
-		for _, r := range rules {
+			pre++
-			desiredByTag[r.Tag] = append(desiredByTag[r.Tag], r)
+		}
+		suf := 0
+		for suf < len(cur)-pre && suf < len(want)-pre &&
+			ruleEqual(cur[len(cur)-1-suf], want[len(want)-1-suf]) {
+			suf++
 		}
-	}
 
-	for tag, desiredRules := range desiredByTag {
+		var before uint64
-		currentRules, exists := currentByTag[tag]
+		if suf > 0 {
-		if !exists {
+			before = cur[len(cur)-suf].Handle
-			cs.Add = append(cs.Add, desiredRules...)
-			continue
 		}
-		if !rulesMatch(currentRules, desiredRules) {
+		cs.Remove = append(cs.Remove, cur[pre:len(cur)-suf]...)
-			cs.Remove = append(cs.Remove, currentRules...)
+		for _, r := range want[pre : len(want)-suf] {
-			cs.Add = append(cs.Add, desiredRules...)
+			r.Before = before
-		}
+			cs.Add = append(cs.Add, r)
-	}
-
-	for tag, currentRules := range currentByTag {
-		if _, exists := desiredByTag[tag]; !exists {
-			cs.Remove = append(cs.Remove, currentRules...)
 		}
 	}
 
 	return cs
 }
 
-func rulesMatch(a, b []ManagedRule) bool {
+func ruleEqual(a, b ManagedRule) bool {
-	if len(a) != len(b) {
+	return a.Chain == b.Chain && a.Tag == b.Tag && reflect.DeepEqual(a.Exprs, b.Exprs)
-		return false
-	}
-	for i := range a {
-		if a[i].Chain != b[i].Chain {
-			return false
-		}
-		if !exprsEqual(a[i].Exprs, b[i].Exprs) {
-			return false
-		}
-	}
-	return true
 }
 
-func exprsEqual(a, b []expr.Any) bool {
+// restoreChangeSet replaces every managed rule in current with the snapshot's,
-	if len(a) != len(b) {
+// in snapshot order, so a restore cannot reorder rules.
-		return false
+func restoreChangeSet(current, snap *FirewallState) *ChangeSet {
+	cs := &ChangeSet{AddHelpers: snap.Helpers}
+	for _, h := range current.Helpers {
+		cs.RemoveHelpers = append(cs.RemoveHelpers, h.Name)
 	}
-	as := fmt.Sprintf("%v", a)
+	for _, rules := range current.Rules {
-	bs := fmt.Sprintf("%v", b)
+		for _, r := range rules {
-	return as == bs
+			if r.Tag != "" {
+				cs.Remove = append(cs.Remove, r)
+			}
+		}
+	}
+	chains := make([]string, 0, len(snap.Rules))
+	for c := range snap.Rules {
+		chains = append(chains, c)
+	}
+	sort.Strings(chains)
+	for _, c := range chains {
+		for _, r := range snap.Rules[c] {
+			if r.Tag != "" {
+				cs.Add = append(cs.Add, r)
+			}
+		}
+	}
+	return cs
+}
+
+// diffHelpers replaces any ct helper object that is missing or differs.
+// L3Proto is ignored: the kernel narrows inet to ip/ip6 for single-family helpers such as pptp.
+func diffHelpers(current, desired *FirewallState) *ChangeSet {
+	cs := &ChangeSet{}
+	same := func(a, b expr.CtHelper) bool { return a.Name == b.Name && a.L4Proto == b.L4Proto }
+	find := func(hs []Helper, name string) (expr.CtHelper, bool) {
+		for _, h := range hs {
+			if h.Name == name {
+				return h.Helper, true
+			}
+		}
+		return expr.CtHelper{}, false
+	}
+	for _, h := range current.Helpers {
+		if want, ok := find(desired.Helpers, h.Name); !ok || !same(want, h.Helper) {
+			cs.RemoveHelpers = append(cs.RemoveHelpers, h.Name)
+		}
+	}
+	for _, h := range desired.Helpers {
+		if have, ok := find(current.Helpers, h.Name); !ok || !same(have, h.Helper) {
+			cs.AddHelpers = append(cs.AddHelpers, h)
+		}
+	}
+	return cs
 }

internal/nftables/diff_test.go

@@ -0,0 +1,106 @@
+package nftables
+
+import (
+	"reflect"
+	"testing"
+
+	"github.com/google/nftables/expr"
+	"golang.org/x/sys/unix"
+)
+
+func TestRestoreChangeSet(t *testing.T) {
+	accept := []expr.Any{&expr.Verdict{Kind: expr.VerdictAccept}}
+	drop := []expr.Any{&expr.Verdict{Kind: expr.VerdictDrop}}
+
+	snap := &FirewallState{Rules: map[string][]ManagedRule{
+		"input": {
+			{Chain: "input", Tag: "ssh", Exprs: accept, Handle: 4},
+			{Chain: "input", Tag: "web", Exprs: accept, Handle: 5},
+			{Chain: "input", Tag: "", Exprs: drop, Handle: 6},
+		},
+		"forward": {{Chain: "forward", Tag: "fwd", Exprs: accept, Handle: 7}},
+	}}
+	current := &FirewallState{Rules: map[string][]ManagedRule{
+		"input": {
+			{Chain: "input", Tag: "web", Exprs: accept, Handle: 10},
+			{Chain: "input", Tag: "ssh", Exprs: drop, Handle: 11},
+			{Chain: "input", Tag: "", Exprs: drop, Handle: 12},
+		},
+	}}
+
+	cs := restoreChangeSet(current, snap)
+
+	var removed []uint64
+	for _, r := range cs.Remove {
+		removed = append(removed, r.Handle)
+	}
+	if len(removed) != 2 || removed[0] != 10 || removed[1] != 11 {
+		t.Errorf("expected managed handles [10 11] removed, untagged kept; got %v", removed)
+	}
+
+	var added []string
+	for _, r := range cs.Add {
+		added = append(added, r.Tag)
+	}
+	want := []string{"fwd", "ssh", "web"}
+	if len(added) != len(want) {
+		t.Fatalf("added %v, want %v", added, want)
+	}
+	for i := range want {
+		if added[i] != want[i] {
+			t.Fatalf("added %v, want %v (snapshot order per chain)", added, want)
+		}
+	}
+	if !reflect.DeepEqual(cs.Add[1].Exprs, accept) {
+		t.Error("ssh not restored to its snapshot exprs")
+	}
+}
+
+func TestRestoreChangeSetEmptySnapshotRemovesAll(t *testing.T) {
+	current := &FirewallState{Rules: map[string][]ManagedRule{
+		"input": {{Chain: "input", Tag: "x", Handle: 1}},
+	}}
+	cs := restoreChangeSet(current, &FirewallState{Rules: map[string][]ManagedRule{}})
+	if len(cs.Remove) != 1 || len(cs.Add) != 0 {
+		t.Errorf("expected 1 remove 0 add, got %d/%d", len(cs.Remove), len(cs.Add))
+	}
+}
+
+func TestDiffHelpers(t *testing.T) {
+	h := func(name, typ string, l3 uint16, l4 uint8) Helper {
+		return Helper{Name: name, Helper: expr.CtHelper{Name: typ, L3Proto: l3, L4Proto: l4}}
+	}
+	ftp := h("ftp", "ftp", unix.NFPROTO_INET, unix.IPPROTO_TCP)
+	tftp := h("tftp", "tftp", unix.NFPROTO_INET, unix.IPPROTO_UDP)
+	sipUDP := h("sip", "sip", unix.NFPROTO_INET, unix.IPPROTO_UDP)
+	sipTCP := h("sip", "sip", unix.NFPROTO_INET, unix.IPPROTO_TCP)
+
+	current := &FirewallState{Helpers: []Helper{tftp, ftp, sipUDP}}
+	desired := &FirewallState{Helpers: []Helper{ftp, sipTCP}}
+
+	cs := computeDiff(current, desired)
+	if !reflect.DeepEqual(cs.RemoveHelpers, []string{"tftp", "sip"}) {
+		t.Errorf("remove = %v", cs.RemoveHelpers)
+	}
+	if !reflect.DeepEqual(cs.AddHelpers, []Helper{sipTCP}) {
+		t.Errorf("add = %v", cs.AddHelpers)
+	}
+
+	// Restore recreates every helper so kernel listing order matches the snapshot.
+	cs = restoreChangeSet(desired, current)
+	if !reflect.DeepEqual(cs.RemoveHelpers, []string{"ftp", "sip"}) || !reflect.DeepEqual(cs.AddHelpers, current.Helpers) {
+		t.Errorf("restore = -%v +%v", cs.RemoveHelpers, cs.AddHelpers)
+	}
+
+	live := &FirewallState{Helpers: []Helper{h("pptp", "pptp", unix.NFPROTO_IPV4, unix.IPPROTO_TCP)}}
+	want := &FirewallState{Helpers: []Helper{h("pptp", "pptp", unix.NFPROTO_INET, unix.IPPROTO_TCP)}}
+	if cs := computeDiff(live, want); !cs.Empty() {
+		t.Errorf("kernel-narrowed l3proto should not diff: %+v", cs)
+	}
+	if cs := computeDiff(desired, desired); !cs.Empty() {
+		t.Errorf("identical helpers should be empty: %+v", cs)
+	}
+	if cs := computeDiff(&FirewallState{}, desired); cs.Empty() {
+		t.Error("missing helpers should not be empty")
+	}
+}

internal/nftables/engine.go

@@ -4,6 +4,9 @@ import (
 	"fmt"
 
 	"github.com/google/nftables"
+	"github.com/google/nftables/expr"
+	"github.com/mdlayher/netlink"
+	"golang.org/x/sys/unix"
 
 	"git.unkin.net/unkin/tomswall/internal/config"
 )
@@ -14,21 +17,107 @@ type Engine struct {
 }
 
 func NewEngine(cfg *config.Config) (*Engine, error) {
-	conn, err := nftables.New()
+	conn, err := nftables.New(nftables.WithSockOptions(largeBuffers))
 	if err != nil {
 		return nil, fmt.Errorf("connecting to nftables: %w", err)
 	}
 	return &Engine{cfg: cfg, conn: conn}, nil
 }
 
+var tableFamilies = map[config.AddressFamily]nftables.TableFamily{
+	config.FamilyINET: nftables.TableFamilyINet,
+	config.FamilyIP:   nftables.TableFamilyIPv4,
+	config.FamilyIP6:  nftables.TableFamilyIPv6,
+}
+
+func (e *Engine) family() nftables.TableFamily {
+	if f, ok := tableFamilies[e.cfg.Settings.AddressFamily]; ok {
+		return f
+	}
+	return nftables.TableFamilyINet
+}
+
+func addressFamily(tf nftables.TableFamily) config.AddressFamily {
+	for f, t := range tableFamilies {
+		if t == tf {
+			return f
+		}
+	}
+	return config.FamilyINET
+}
+
+// withFamily is the engine for the same table name in another address family.
+func (e *Engine) withFamily(f config.AddressFamily) *Engine {
+	cfg := *e.cfg
+	cfg.Settings.AddressFamily = f
+	return &Engine{cfg: &cfg, conn: e.conn}
+}
+
+// overlaps reports whether tables of families a and b filter the same traffic:
+// inet covers both ip and ip6, which do not overlap each other.
+func overlaps(a, b nftables.TableFamily) bool {
+	if a == b {
+		return false
+	}
+	return (a == nftables.TableFamilyINet && (b == nftables.TableFamilyIPv4 || b == nftables.TableFamilyIPv6)) ||
+		(b == nftables.TableFamilyINet && (a == nftables.TableFamilyIPv4 || a == nftables.TableFamilyIPv6))
+}
+
+// staleTables are our-named tables left by a different address_family that
+// would still filter the traffic this family now owns.
+func (e *Engine) staleTables() ([]*nftables.Table, error) {
+	tables, err := e.conn.ListTables()
+	if err != nil {
+		return nil, fmt.Errorf("listing tables: %w", err)
+	}
+	var stale []*nftables.Table
+	for _, t := range tables {
+		if t.Name == e.cfg.Settings.TableName && overlaps(e.family(), t.Family) {
+			stale = append(stale, t)
+		}
+	}
+	return stale, nil
+}
+
+// batchBufSize bounds one batch: the kernel rejects a batch larger than the
+// send buffer (EMSGSIZE) and drops ACKs beyond the receive buffer (ENOBUFS)
+// after committing it.
+// ponytail: fixed cap of tens of thousands of rules; size per batch if exceeded.
+const batchBufSize = 64 << 20
+
+// largeBuffers raises both socket buffers, ignoring rmem_max/wmem_max when
+// CAP_NET_ADMIN allows it and falling back to the capped sizes otherwise.
+func largeBuffers(c *netlink.Conn) error {
+	rc, err := c.SyscallConn()
+	if err != nil {
+		return err
+	}
+	var serr error
+	err = rc.Control(func(fd uintptr) {
+		for _, o := range [][2]int{{unix.SO_SNDBUFFORCE, unix.SO_SNDBUF}, {unix.SO_RCVBUFFORCE, unix.SO_RCVBUF}} {
+			if unix.SetsockoptInt(int(fd), unix.SOL_SOCKET, o[0], batchBufSize) == nil {
+				continue
+			}
+			if serr = unix.SetsockoptInt(int(fd), unix.SOL_SOCKET, o[1], batchBufSize); serr != nil {
+				return
+			}
+		}
+	})
+	if err != nil {
+		return err
+	}
+	return serr
+}
+
 func (e *Engine) ensureTable() *nftables.Table {
 	return e.conn.AddTable(&nftables.Table{
-		Family: nftables.TableFamilyINet,
+		Family: e.family(),
 		Name:   e.cfg.Settings.TableName,
 	})
 }
 
-func (e *Engine) ensureChains(table *nftables.Table) map[string]*nftables.Chain {
+// ensureChains declares the base chains; policies overrides their default policy.
+func (e *Engine) ensureChains(table *nftables.Table, policies map[string]nftables.ChainPolicy) map[string]*nftables.Chain {
 	chains := map[string]*nftables.Chain{
 		"input": {
 			Name:     "input",
@@ -68,9 +157,42 @@ func (e *Engine) ensureChains(table *nftables.Table) map[string]*nftables.Chain
 			Hooknum:  nftables.ChainHookPrerouting,
 			Priority: nftables.ChainPriorityNATDest,
 		},
+		"helper_prerouting": {
+			Name:     "helper_prerouting",
+			Table:    table,
+			Type:     nftables.ChainTypeFilter,
+			Hooknum:  nftables.ChainHookPrerouting,
+			Priority: nftables.ChainPriorityMangle,
+		},
+		"helper_output": {
+			Name:     "helper_output",
+			Table:    table,
+			Type:     nftables.ChainTypeFilter,
+			Hooknum:  nftables.ChainHookOutput,
+			Priority: nftables.ChainPriorityMangle,
+		},
+		"raw_prerouting": {
+			Name:     "raw_prerouting",
+			Table:    table,
+			Type:     nftables.ChainTypeFilter,
+			Hooknum:  nftables.ChainHookPrerouting,
+			Priority: nftables.ChainPriorityRaw,
+			Policy:   policyPtr(nftables.ChainPolicyAccept),
+		},
+		"raw_output": {
+			Name:     "raw_output",
+			Table:    table,
+			Type:     nftables.ChainTypeFilter,
+			Hooknum:  nftables.ChainHookOutput,
+			Priority: nftables.ChainPriorityRaw,
+			Policy:   policyPtr(nftables.ChainPolicyAccept),
+		},
 	}
 
 	for name, chain := range chains {
+		if p, ok := policies[name]; ok {
+			chain.Policy = policyPtr(p)
+		}
 		chains[name] = e.conn.AddChain(chain)
 	}
 	return chains
@@ -93,8 +215,19 @@ func (e *Engine) Plan() (*ChangeSet, error) {
 }
 
 func (e *Engine) Apply(changes *ChangeSet) error {
+	return e.apply(changes, nil)
+}
+
+func (e *Engine) apply(changes *ChangeSet, policies map[string]nftables.ChainPolicy) error {
+	stale, err := e.staleTables()
+	if err != nil {
+		return err
+	}
+	for _, t := range stale {
+		e.conn.DelTable(t)
+	}
 	table := e.ensureTable()
-	chains := e.ensureChains(table)
+	chains := e.ensureChains(table, policies)
 
 	for _, r := range changes.Remove {
 		e.conn.DelRule(&nftables.Rule{
@@ -104,35 +237,67 @@ func (e *Engine) Apply(changes *ChangeSet) error {
 		})
 	}
 
+	for _, n := range changes.RemoveHelpers {
+		e.conn.DeleteObject(helperObj(table, n, expr.CtHelper{}))
+	}
+	for _, h := range changes.AddHelpers {
+		e.conn.AddObj(helperObj(table, h.Name, h.Helper))
+	}
+
 	for _, r := range changes.Add {
 		chain, ok := chains[r.Chain]
 		if !ok {
 			return fmt.Errorf("unknown chain %q", r.Chain)
 		}
-		e.conn.AddRule(&nftables.Rule{
+		rule := &nftables.Rule{
 			Table:    table,
 			Chain:    chain,
 			Exprs:    r.Exprs,
 			UserData: []byte(r.Tag),
-		})
+		}
+		if r.Before != 0 {
+			rule.Position = r.Before
+			e.conn.InsertRule(rule)
+		} else {
+			e.conn.AddRule(rule)
+		}
 	}
 
 	return e.conn.Flush()
 }
 
 func (e *Engine) Flush() error {
+	tables, err := e.staleTables()
+	if err != nil {
+		return err
+	}
+	own, err := e.findTable()
+	if err != nil {
+		return err
+	}
+	if own != nil {
+		tables = append(tables, own)
+	}
+	if len(tables) == 0 {
+		return nil
+	}
+	for _, t := range tables {
+		e.conn.DelTable(t)
+	}
+	return e.conn.Flush()
+}
+
+func (e *Engine) findTable() (*nftables.Table, error) {
 	tables, err := e.conn.ListTables()
 	if err != nil {
-		return fmt.Errorf("listing tables: %w", err)
+		return nil, fmt.Errorf("listing tables: %w", err)
 	}
-
 	for _, t := range tables {
-		if t.Name == e.cfg.Settings.TableName {
+		if t.Name == e.cfg.Settings.TableName && t.Family == e.family() {
-			e.conn.DelTable(t)
+			return t, nil
-			return e.conn.Flush()
 		}
 	}
-	return nil
+	return nil, nil
 }
 
 func (e *Engine) readCurrentState() (*FirewallState, error) {
@@ -140,26 +305,26 @@ func (e *Engine) readCurrentState() (*FirewallState, error) {
 		Rules: make(map[string][]ManagedRule),
 	}
 
-	tables, err := e.conn.ListTables()
+	ourTable, err := e.findTable()
-	if err != nil {
+	if err != nil || ourTable == nil {
-		return state, nil
+		return state, err
 	}
 
-	var ourTable *nftables.Table
+	objs, err := e.conn.GetNamedObjects(ourTable)
-	for _, t := range tables {
+	if err != nil {
-		if t.Name == e.cfg.Settings.TableName && t.Family == nftables.TableFamilyINet {
+		return nil, fmt.Errorf("listing objects: %w", err)
-			ourTable = t
+	}
-			break
+	for _, o := range objs {
+		if no, ok := o.(*nftables.NamedObj); ok && no.Type == nftables.ObjTypeCtHelper {
+			if h, ok := no.Obj.(*expr.CtHelper); ok {
+				state.Helpers = append(state.Helpers, Helper{Name: no.Name, Helper: *h})
+			}
 		}
 	}
 
-	if ourTable == nil {
+	chains, err := e.conn.ListChainsOfTableFamily(e.family())
-		return state, nil
-	}
-
-	chains, err := e.conn.ListChainsOfTableFamily(nftables.TableFamilyINet)
 	if err != nil {
-		return state, nil
+		return nil, fmt.Errorf("listing chains: %w", err)
 	}
 
 	for _, chain := range chains {
@@ -168,7 +333,7 @@ func (e *Engine) readCurrentState() (*FirewallState, error) {
 		}
 		rules, err := e.conn.GetRules(ourTable, chain)
 		if err != nil {
-			continue
+			return nil, fmt.Errorf("listing rules of %s: %w", chain.Name, err)
 		}
 		for _, rule := range rules {
 			state.Rules[chain.Name] = append(state.Rules[chain.Name], ManagedRule{
@@ -183,6 +348,98 @@ func (e *Engine) readCurrentState() (*FirewallState, error) {
 	return state, nil
 }
 
+// Snapshot is the tomswall table as captured live, serialisable so a revert
+// survives the process that took it.
+type Snapshot struct {
+	Table    string                          `json:"table"`
+	Family   config.AddressFamily            `json:"family,omitempty"`
+	Present  bool                            `json:"present"`
+	Policies map[string]nftables.ChainPolicy `json:"policies,omitempty"`
+	Rules    map[string][]SnapshotRule       `json:"rules,omitempty"`
+	Helpers  []Helper                        `json:"helpers,omitempty"`
+}
+
+// SnapshotRule is a managed rule with its expressions in netlink wire format.
+type SnapshotRule struct {
+	Tag   string   `json:"tag"`
+	Exprs [][]byte `json:"exprs"`
+}
+
+// Snapshot captures the live tomswall table so Restore can roll back to it,
+// falling back to the overlapping table of another family that apply replaces.
+func (e *Engine) Snapshot() (*Snapshot, error) {
+	t, err := e.findTable()
+	if err != nil {
+		return nil, err
+	}
+	if t == nil {
+		stale, err := e.staleTables()
+		if err != nil {
+			return nil, err
+		}
+		// ponytail: captures one stale table; an inet config replacing both ip and ip6 restores only the first.
+		if len(stale) > 0 {
+			return e.withFamily(addressFamily(stale[0].Family)).Snapshot()
+		}
+	}
+	snap := &Snapshot{Table: e.cfg.Settings.TableName, Family: addressFamily(e.family())}
+	if t == nil {
+		return snap, nil
+	}
+	snap.Present = true
+
+	chains, err := e.conn.ListChainsOfTableFamily(e.family())
+	if err != nil {
+		return nil, fmt.Errorf("listing chains: %w", err)
+	}
+	snap.Policies = make(map[string]nftables.ChainPolicy)
+	for _, c := range chains {
+		if c.Table.Name == snap.Table && c.Policy != nil {
+			snap.Policies[c.Name] = *c.Policy
+		}
+	}
+
+	state, err := e.readCurrentState()
+	if err != nil {
+		return nil, err
+	}
+	snap.Rules, err = encodeState(state, byte(e.family()))
+	if err != nil {
+		return nil, err
+	}
+	snap.Helpers = state.Helpers
+	return snap, nil
+}
+
+// Restore atomically returns the tomswall table to the snapshot: rule order
+// and chain policies included, or removed if it was absent.
+func (e *Engine) Restore(s *Snapshot) error {
+	if s.Table != e.cfg.Settings.TableName {
+		return fmt.Errorf("snapshot is of table %q, engine manages %q", s.Table, e.cfg.Settings.TableName)
+	}
+	// Snapshots predating the family field are of the inet table.
+	if f := addressFamily(tableFamilies[s.Family]); f != addressFamily(e.family()) {
+		return e.withFamily(f).Restore(s)
+	}
+	if !s.Present {
+		return e.Flush()
+	}
+	want, err := decodeState(s.Rules, byte(e.family()))
+	if err != nil {
+		return err
+	}
+	want.Helpers = s.Helpers
+	current, err := e.readCurrentState()
+	if err != nil {
+		return err
+	}
+	return e.apply(restoreChangeSet(current, want), s.Policies)
+}
+
+func helperObj(table *nftables.Table, name string, h expr.CtHelper) *nftables.NamedObj {
+	return &nftables.NamedObj{Table: table, Name: name, Type: nftables.ObjTypeCtHelper, Obj: &h}
+}
+
 func policyPtr(p nftables.ChainPolicy) *nftables.ChainPolicy {
 	return &p
 }

internal/nftables/engine_test.go

@@ -0,0 +1,59 @@
+package nftables
+
+import (
+	"os"
+	"strconv"
+	"strings"
+	"testing"
+
+	"github.com/mdlayher/netlink"
+	"golang.org/x/sys/unix"
+)
+
+func TestLargeBuffersRaisesSocketBuffers(t *testing.T) {
+	c, err := netlink.Dial(unix.NETLINK_NETFILTER, nil)
+	if err != nil {
+		t.Skipf("netlink unavailable: %v", err)
+	}
+	defer c.Close()
+	if err := largeBuffers(c); err != nil {
+		t.Fatal(err)
+	}
+	// Without CAP_NET_ADMIN the kernel caps at the sysctl max; it doubles either way.
+	for opt, sysctl := range map[int]string{unix.SO_RCVBUF: "rmem_max", unix.SO_SNDBUF: "wmem_max"} {
+		want := 2 * min(batchBufSize, procInt(t, "/proc/sys/net/core/"+sysctl))
+		if got := sockBuf(t, c, opt); got < want {
+			t.Errorf("%s-bounded buffer = %d, want >= %d", sysctl, got, want)
+		}
+	}
+}
+
+func procInt(t *testing.T, path string) int {
+	t.Helper()
+	b, err := os.ReadFile(path)
+	if err != nil {
+		t.Skipf("reading %s: %v", path, err)
+	}
+	v, err := strconv.Atoi(strings.TrimSpace(string(b)))
+	if err != nil {
+		t.Fatal(err)
+	}
+	return v
+}
+
+func sockBuf(t *testing.T, c *netlink.Conn, opt int) int {
+	t.Helper()
+	rc, err := c.SyscallConn()
+	if err != nil {
+		t.Fatal(err)
+	}
+	var v int
+	var serr error
+	if err := rc.Control(func(fd uintptr) { v, serr = unix.GetsockoptInt(int(fd), unix.SOL_SOCKET, opt) }); err != nil {
+		t.Fatal(err)
+	}
+	if serr != nil {
+		t.Fatal(serr)
+	}
+	return v
+}

internal/nftables/family_test.go

@@ -0,0 +1,113 @@
+package nftables
+
+import (
+	"testing"
+
+	"github.com/google/nftables"
+	"github.com/google/nftables/expr"
+	"github.com/mdlayher/netlink"
+	"golang.org/x/sys/unix"
+
+	"git.unkin.net/unkin/tomswall/internal/config"
+)
+
+type sentTable struct {
+	msg    int
+	family nftables.TableFamily
+}
+
+// familyEngine fakes a kernel holding tomswall tables of the given families
+// and records table creations/deletions.
+func familyEngine(t *testing.T, af config.AddressFamily, live ...nftables.TableFamily) (*Engine, *[]sentTable) {
+	var sent []sentTable
+	e := testEngine(t, func(req []netlink.Message) ([]netlink.Message, error) {
+		var out []netlink.Message
+		for _, m := range req {
+			switch m.Header.Type {
+			case nftType(unix.NFT_MSG_GETTABLE):
+				for _, f := range live {
+					attrs, _ := netlink.MarshalAttributes([]netlink.Attribute{{Type: unix.NFTA_TABLE_NAME, Data: []byte("tomswall\x00")}})
+					out = append(out, netlink.Message{Header: netlink.Header{Type: nftType(unix.NFT_MSG_NEWTABLE), Sequence: m.Header.Sequence}, Data: append([]byte{byte(f), 0, 0, 0}, attrs...)})
+				}
+			case nftType(unix.NFT_MSG_NEWTABLE):
+				sent = append(sent, sentTable{unix.NFT_MSG_NEWTABLE, nftables.TableFamily(m.Data[0])})
+			case nftType(unix.NFT_MSG_DELTABLE):
+				sent = append(sent, sentTable{unix.NFT_MSG_DELTABLE, nftables.TableFamily(m.Data[0])})
+			}
+		}
+		return out, nil
+	})
+	e.cfg.Settings.AddressFamily = af
+	return e, &sent
+}
+
+func TestApplyIPFamilyReplacesInetTable(t *testing.T) {
+	e, sent := familyEngine(t, config.FamilyIP, nftables.TableFamilyINet, nftables.TableFamilyIPv6)
+	if err := e.Apply(&ChangeSet{}); err != nil {
+		t.Fatal(err)
+	}
+	want := []sentTable{{unix.NFT_MSG_DELTABLE, nftables.TableFamilyINet}, {unix.NFT_MSG_NEWTABLE, nftables.TableFamilyIPv4}}
+	if len(*sent) != len(want) || (*sent)[0] != want[0] || (*sent)[1] != want[1] {
+		t.Errorf("got %+v, want %+v (the ip6 table must survive)", *sent, want)
+	}
+}
+
+func TestApplyInetFamilyReplacesIPTables(t *testing.T) {
+	e, sent := familyEngine(t, config.FamilyINET, nftables.TableFamilyIPv4, nftables.TableFamilyIPv6)
+	if err := e.Apply(&ChangeSet{}); err != nil {
+		t.Fatal(err)
+	}
+	var dels int
+	for _, s := range *sent {
+		if s.msg == unix.NFT_MSG_DELTABLE {
+			dels++
+		}
+	}
+	if dels != 2 {
+		t.Errorf("want ip and ip6 tables deleted, got %+v", *sent)
+	}
+}
+
+func TestFlushIPFamilyKeepsIP6Table(t *testing.T) {
+	e, sent := familyEngine(t, config.FamilyIP, nftables.TableFamilyIPv6, nftables.TableFamilyIPv4)
+	if err := e.Flush(); err != nil {
+		t.Fatal(err)
+	}
+	if len(*sent) != 1 || (*sent)[0] != (sentTable{unix.NFT_MSG_DELTABLE, nftables.TableFamilyIPv4}) {
+		t.Errorf("got %+v, want only the ip table deleted", *sent)
+	}
+}
+
+func TestSnapshotFallsBackToReplacedTable(t *testing.T) {
+	e, _ := familyEngine(t, config.FamilyIP, nftables.TableFamilyINet)
+	snap, err := e.Snapshot()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !snap.Present || snap.Family != config.FamilyINET {
+		t.Fatalf("want present inet snapshot, got %+v", snap)
+	}
+
+	// Reverting to the inet snapshot drops the tried ip table.
+	e, sent := familyEngine(t, config.FamilyIP, nftables.TableFamilyIPv4)
+	if err := e.Restore(&Snapshot{Table: "tomswall", Family: config.FamilyINET, Present: true}); err != nil {
+		t.Fatal(err)
+	}
+	want := []sentTable{{unix.NFT_MSG_DELTABLE, nftables.TableFamilyIPv4}, {unix.NFT_MSG_NEWTABLE, nftables.TableFamilyINet}}
+	if len(*sent) != 2 || (*sent)[0] != want[0] || (*sent)[1] != want[1] {
+		t.Errorf("got %+v, want %+v", *sent, want)
+	}
+}
+
+func TestRejectExprsFamily(t *testing.T) {
+	for af, want := range map[config.AddressFamily]expr.Reject{
+		config.FamilyINET: {Type: unix.NFT_REJECT_ICMPX_UNREACH, Code: unix.NFT_REJECT_ICMPX_PORT_UNREACH},
+		config.FamilyIP:   {Type: unix.NFT_REJECT_ICMP_UNREACH, Code: 3},
+		config.FamilyIP6:  {Type: unix.NFT_REJECT_ICMP_UNREACH, Code: 4},
+	} {
+		got := rejectExprs(unix.IPPROTO_UDP, af)[0].(*expr.Reject)
+		if *got != want {
+			t.Errorf("%s: got %+v, want %+v", af, *got, want)
+		}
+	}
+}

internal/nftables/snapshot.go

@@ -0,0 +1,116 @@
+package nftables
+
+import (
+	"encoding/binary"
+	"fmt"
+
+	"github.com/google/nftables/expr"
+	"github.com/mdlayher/netlink"
+	"golang.org/x/sys/unix"
+)
+
+// exprByName mirrors the expression types google/nftables can parse back from the kernel.
+var exprByName = map[string]func() expr.Any{
+	"ct":           func() expr.Any { return &expr.Ct{} },
+	"range":        func() expr.Any { return &expr.Range{} },
+	"meta":         func() expr.Any { return &expr.Meta{} },
+	"cmp":          func() expr.Any { return &expr.Cmp{} },
+	"counter":      func() expr.Any { return &expr.Counter{} },
+	"objref":       func() expr.Any { return &expr.Objref{} },
+	"payload":      func() expr.Any { return &expr.Payload{} },
+	"lookup":       func() expr.Any { return &expr.Lookup{} },
+	"immediate":    func() expr.Any { return &expr.Immediate{} },
+	"bitwise":      func() expr.Any { return &expr.Bitwise{} },
+	"redir":        func() expr.Any { return &expr.Redir{} },
+	"nat":          func() expr.Any { return &expr.NAT{} },
+	"limit":        func() expr.Any { return &expr.Limit{} },
+	"quota":        func() expr.Any { return &expr.Quota{} },
+	"dynset":       func() expr.Any { return &expr.Dynset{} },
+	"log":          func() expr.Any { return &expr.Log{} },
+	"exthdr":       func() expr.Any { return &expr.Exthdr{} },
+	"connlimit":    func() expr.Any { return &expr.Connlimit{} },
+	"queue":        func() expr.Any { return &expr.Queue{} },
+	"flow_offload": func() expr.Any { return &expr.FlowOffload{} },
+	"reject":       func() expr.Any { return &expr.Reject{} },
+	"masq":         func() expr.Any { return &expr.Masq{} },
+	"hash":         func() expr.Any { return &expr.Hash{} },
+	"notrack":      func() expr.Any { return &expr.Notrack{} },
+}
+
+func encodeState(state *FirewallState, fam byte) (map[string][]SnapshotRule, error) {
+	out := make(map[string][]SnapshotRule, len(state.Rules))
+	for chain, rules := range state.Rules {
+		for _, r := range rules {
+			sr := SnapshotRule{Tag: r.Tag}
+			for _, e := range r.Exprs {
+				b, err := expr.Marshal(fam, e)
+				if err != nil {
+					return nil, fmt.Errorf("encoding %s rule %q: %w", chain, r.Tag, err)
+				}
+				sr.Exprs = append(sr.Exprs, b)
+			}
+			out[chain] = append(out[chain], sr)
+		}
+	}
+	return out, nil
+}
+
+func decodeState(rules map[string][]SnapshotRule, fam byte) (*FirewallState, error) {
+	state := &FirewallState{Rules: make(map[string][]ManagedRule, len(rules))}
+	for chain, rs := range rules {
+		for _, sr := range rs {
+			r := ManagedRule{Chain: chain, Tag: sr.Tag}
+			for _, b := range sr.Exprs {
+				e, err := decodeExpr(b, fam)
+				if err != nil {
+					return nil, fmt.Errorf("decoding %s rule %q: %w", chain, sr.Tag, err)
+				}
+				r.Exprs = append(r.Exprs, e)
+			}
+			state.Rules[chain] = append(state.Rules[chain], r)
+		}
+	}
+	return state, nil
+}
+
+// decodeExpr reverses expr.Marshal, as google/nftables does when reading rules.
+func decodeExpr(b []byte, fam byte) (expr.Any, error) {
+	ad, err := netlink.NewAttributeDecoder(b)
+	if err != nil {
+		return nil, err
+	}
+	ad.ByteOrder = binary.BigEndian
+	var name string
+	var data []byte
+	for ad.Next() {
+		switch ad.Type() {
+		case unix.NFTA_EXPR_NAME:
+			name = ad.String()
+		case unix.NFTA_EXPR_DATA:
+			data = ad.Bytes()
+		}
+	}
+	if err := ad.Err(); err != nil {
+		return nil, err
+	}
+	newExpr, ok := exprByName[name]
+	if !ok {
+		return nil, fmt.Errorf("unsupported expression %q", name)
+	}
+	e := newExpr()
+	if name == "notrack" {
+		return e, nil
+	}
+	if err := expr.Unmarshal(fam, data, e); err != nil {
+		return nil, err
+	}
+	// A verdict is an immediate into the verdict register with no data.
+	if imm, ok := e.(*expr.Immediate); ok && imm.Register == unix.NFT_REG_VERDICT && len(imm.Data) == 0 {
+		v := &expr.Verdict{}
+		if err := expr.Unmarshal(fam, data, v); err != nil {
+			return nil, err
+		}
+		return v, nil
+	}
+	return e, nil
+}

internal/nftables/snapshot_test.go

@@ -0,0 +1,259 @@
+package nftables
+
+import (
+	"bytes"
+	"encoding/binary"
+	"encoding/json"
+	"reflect"
+	"testing"
+
+	"github.com/google/nftables"
+	"github.com/google/nftables/expr"
+	"github.com/mdlayher/netlink"
+	"golang.org/x/sys/unix"
+
+	"git.unkin.net/unkin/tomswall/internal/config"
+)
+
+func TestSnapshotRulesRoundTrip(t *testing.T) {
+	exprs := []expr.Any{
+		&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
+		&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_TCP}},
+		&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
+		&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{0, 22}},
+		&expr.Ct{Register: 1, Key: expr.CtKeySTATE},
+		&expr.Notrack{},
+		&expr.Verdict{Kind: expr.VerdictAccept},
+	}
+	state := &FirewallState{Rules: map[string][]ManagedRule{
+		"input": {{Chain: "input", Tag: "ssh", Exprs: exprs}, {Chain: "input", Tag: "drop", Exprs: []expr.Any{&expr.Verdict{Kind: expr.VerdictDrop}}}},
+	}}
+
+	rules, err := encodeState(state, byte(nftables.TableFamilyINet))
+	if err != nil {
+		t.Fatal(err)
+	}
+	b, err := json.Marshal(&Snapshot{Table: "tomswall", Present: true, Rules: rules,
+		Policies: map[string]nftables.ChainPolicy{"input": nftables.ChainPolicyAccept}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	var snap Snapshot
+	if err := json.Unmarshal(b, &snap); err != nil {
+		t.Fatal(err)
+	}
+	if snap.Policies["input"] != nftables.ChainPolicyAccept {
+		t.Errorf("policy lost: %v", snap.Policies)
+	}
+	got, err := decodeState(snap.Rules, byte(nftables.TableFamilyINet))
+	if err != nil {
+		t.Fatal(err)
+	}
+	in := got.Rules["input"]
+	if len(in) != 2 || in[0].Tag != "ssh" || in[1].Tag != "drop" {
+		t.Fatalf("rules/order lost: %+v", in)
+	}
+	if !reflect.DeepEqual(in[0].Exprs, exprs) {
+		t.Errorf("exprs changed:\n got %#v\nwant %#v", in[0].Exprs, exprs)
+	}
+	if _, ok := in[1].Exprs[0].(*expr.Verdict); !ok {
+		t.Errorf("verdict decoded as %T", in[1].Exprs[0])
+	}
+}
+
+func TestEnsureChainsPolicyOverride(t *testing.T) {
+	e := testEngine(t, nil)
+	chains := e.ensureChains(e.ensureTable(), map[string]nftables.ChainPolicy{"input": nftables.ChainPolicyAccept})
+	if *chains["input"].Policy != nftables.ChainPolicyAccept {
+		t.Error("input policy not overridden")
+	}
+	if *chains["forward"].Policy != nftables.ChainPolicyDrop {
+		t.Error("forward policy should keep its default")
+	}
+}
+
+func TestSnapshotAndRestoreAbsentTable(t *testing.T) {
+	tablePresent := false
+	var sent []netlink.HeaderType
+	e := testEngine(t, func(req []netlink.Message) ([]netlink.Message, error) {
+		for _, m := range req {
+			sent = append(sent, m.Header.Type)
+			if m.Header.Type == nftType(unix.NFT_MSG_GETTABLE) && tablePresent {
+				data := []byte{byte(nftables.TableFamilyINet), 0, 0, 0}
+				attrs, _ := netlink.MarshalAttributes([]netlink.Attribute{{Type: unix.NFTA_TABLE_NAME, Data: []byte("tomswall\x00")}})
+				return []netlink.Message{{Header: netlink.Header{Type: nftType(unix.NFT_MSG_NEWTABLE), Sequence: m.Header.Sequence}, Data: append(data, attrs...)}}, nil
+			}
+		}
+		return nil, nil
+	})
+
+	snap, err := e.Snapshot()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if snap.Present || snap.Table != "tomswall" {
+		t.Fatalf("want absent tomswall snapshot, got %+v", snap)
+	}
+
+	// The try created the table; restoring the absent snapshot deletes it.
+	tablePresent = true
+	sent = nil
+	if err := e.Restore(snap); err != nil {
+		t.Fatal(err)
+	}
+	deleted := false
+	for _, ht := range sent {
+		deleted = deleted || ht == nftType(unix.NFT_MSG_DELTABLE)
+	}
+	if !deleted {
+		t.Errorf("table not deleted; sent %v", sent)
+	}
+}
+
+func TestRestorePresentTable(t *testing.T) {
+	want := []SnapshotRule{}
+	for _, r := range []ManagedRule{
+		{Tag: "ssh", Exprs: []expr.Any{&expr.Ct{Register: 1, Key: expr.CtKeySTATE}, &expr.Verdict{Kind: expr.VerdictAccept}}},
+		{Tag: "drop", Exprs: []expr.Any{&expr.Verdict{Kind: expr.VerdictDrop}}},
+	} {
+		enc, err := encodeState(&FirewallState{Rules: map[string][]ManagedRule{"input": {r}}}, byte(nftables.TableFamilyINet))
+		if err != nil {
+			t.Fatal(err)
+		}
+		want = append(want, enc["input"]...)
+	}
+	snap := &Snapshot{Table: "tomswall", Present: true,
+		Policies: map[string]nftables.ChainPolicy{"input": nftables.ChainPolicyAccept},
+		Rules:    map[string][]SnapshotRule{"input": want}}
+
+	// Live state: the tried ruleset left one managed rule (handle 7) in input.
+	attrs := func(a ...netlink.Attribute) []byte {
+		b, err := netlink.MarshalAttributes(a)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return append([]byte{byte(nftables.TableFamilyINet), 0, 0, 0}, b...)
+	}
+	handle := make([]byte, 8)
+	binary.BigEndian.PutUint64(handle, 7)
+	var batch []netlink.Message
+	e := testEngine(t, func(req []netlink.Message) ([]netlink.Message, error) {
+		if len(req) == 0 {
+			return nil, nil
+		}
+		reply := func(msg int, data []byte) ([]netlink.Message, error) {
+			return []netlink.Message{{Header: netlink.Header{Type: nftType(msg), Sequence: req[0].Header.Sequence}, Data: data}}, nil
+		}
+		switch req[0].Header.Type {
+		case nftType(unix.NFT_MSG_GETTABLE):
+			return reply(unix.NFT_MSG_NEWTABLE, attrs(netlink.Attribute{Type: unix.NFTA_TABLE_NAME, Data: []byte("tomswall\x00")}))
+		case nftType(unix.NFT_MSG_GETCHAIN):
+			return reply(unix.NFT_MSG_NEWCHAIN, attrs(
+				netlink.Attribute{Type: unix.NFTA_CHAIN_TABLE, Data: []byte("tomswall\x00")},
+				netlink.Attribute{Type: unix.NFTA_CHAIN_NAME, Data: []byte("input\x00")}))
+		case nftType(unix.NFT_MSG_GETRULE):
+			return reply(unix.NFT_MSG_NEWRULE, attrs(
+				netlink.Attribute{Type: unix.NFTA_RULE_TABLE, Data: []byte("tomswall\x00")},
+				netlink.Attribute{Type: unix.NFTA_RULE_CHAIN, Data: []byte("input\x00")},
+				netlink.Attribute{Type: unix.NFTA_RULE_HANDLE, Data: handle},
+				netlink.Attribute{Type: unix.NFTA_RULE_USERDATA, Data: []byte("tried")}))
+		}
+		batch = append(batch, req...)
+		return nil, nil
+	})
+
+	if err := e.Restore(snap); err != nil {
+		t.Fatal(err)
+	}
+
+	var deleted []uint64
+	var added []SnapshotRule
+	policy := map[string]uint32{}
+	for _, m := range batch {
+		ad, err := netlink.NewAttributeDecoder(m.Data[4:])
+		if err != nil {
+			t.Fatal(err)
+		}
+		ad.ByteOrder = binary.BigEndian
+		var name string
+		var r SnapshotRule
+		var h uint64
+		var pol *uint32
+		for ad.Next() {
+			switch {
+			case m.Header.Type == nftType(unix.NFT_MSG_NEWCHAIN) && ad.Type() == unix.NFTA_CHAIN_NAME:
+				name = ad.String()
+			case m.Header.Type == nftType(unix.NFT_MSG_NEWCHAIN) && ad.Type() == unix.NFTA_CHAIN_POLICY:
+				v := ad.Uint32()
+				pol = &v
+			case m.Header.Type == nftType(unix.NFT_MSG_DELRULE) && ad.Type() == unix.NFTA_RULE_HANDLE:
+				h = ad.Uint64()
+			case m.Header.Type == nftType(unix.NFT_MSG_NEWRULE) && ad.Type() == unix.NFTA_RULE_USERDATA:
+				r.Tag = string(ad.Bytes())
+			case m.Header.Type == nftType(unix.NFT_MSG_NEWRULE) && ad.Type() == unix.NFTA_RULE_EXPRESSIONS:
+				ad.Nested(func(nad *netlink.AttributeDecoder) error {
+					for nad.Next() {
+						r.Exprs = append(r.Exprs, bytes.Clone(nad.Bytes()))
+					}
+					return nil
+				})
+			}
+		}
+		switch m.Header.Type {
+		case nftType(unix.NFT_MSG_NEWCHAIN):
+			if pol != nil {
+				policy[name] = *pol
+			}
+		case nftType(unix.NFT_MSG_DELRULE):
+			deleted = append(deleted, h)
+		case nftType(unix.NFT_MSG_NEWRULE):
+			added = append(added, r)
+		}
+	}
+
+	if !reflect.DeepEqual(deleted, []uint64{7}) {
+		t.Errorf("deleted handles %v, want [7]", deleted)
+	}
+	if !reflect.DeepEqual(added, want) {
+		t.Errorf("restored rules differ from snapshot:\n got %+v\nwant %+v", added, want)
+	}
+	if policy["input"] != uint32(nftables.ChainPolicyAccept) || policy["forward"] != uint32(nftables.ChainPolicyDrop) {
+		t.Errorf("chain policies %v: input must be restored to accept, forward keep drop", policy)
+	}
+}
+
+func TestRestoreRejectsOtherTable(t *testing.T) {
+	e := testEngine(t, nil)
+	if err := e.Restore(&Snapshot{Table: "other"}); err == nil {
+		t.Error("expected table mismatch error")
+	}
+}
+
+func nftType(msg int) netlink.HeaderType {
+	return netlink.HeaderType(unix.NFNL_SUBSYS_NFTABLES<<8 | msg)
+}
+
+func testEngine(t *testing.T, dial func([]netlink.Message) ([]netlink.Message, error)) *Engine {
+	if dial == nil {
+		dial = func([]netlink.Message) ([]netlink.Message, error) { return nil, nil }
+	}
+	conn, err := nftables.New(nftables.WithTestDial(dial))
+	if err != nil {
+		t.Fatal(err)
+	}
+	return &Engine{cfg: &config.Config{Settings: config.Settings{TableName: "tomswall"}}, conn: conn}
+}
+
+func TestEnsureChainsRawPriority(t *testing.T) {
+	e := testEngine(t, nil)
+	chains := e.ensureChains(e.ensureTable(), nil)
+	for name, hook := range map[string]*nftables.ChainHook{"raw_prerouting": nftables.ChainHookPrerouting, "raw_output": nftables.ChainHookOutput} {
+		c, ok := chains[name]
+		if !ok {
+			t.Fatalf("%s chain not declared", name)
+		}
+		if *c.Priority != *nftables.ChainPriorityRaw || *c.Hooknum != *hook || c.Type != nftables.ChainTypeFilter || *c.Policy != nftables.ChainPolicyAccept {
+			t.Errorf("%s: got type %s hook %d prio %d", name, c.Type, *c.Hooknum, *c.Priority)
+		}
+	}
+}

internal/shorewall/convert.go

@@ -99,6 +99,15 @@ func convertDir(dir string, ipv6 bool) (*config.Config, error) {
 	return cfg, nil
 }
 
+// disposition maps a shorewall *_DISPOSITION value; unset means CONTINUE and A_ (audit) variants map to their base action.
+func disposition(v string) config.PolicyAction {
+	v = strings.TrimPrefix(strings.ToLower(v), "a_")
+	if v == "" {
+		return config.PolicyContinue
+	}
+	return config.PolicyAction(v)
+}
+
 func subst(s string, params map[string]string) string {
 	if !strings.Contains(s, "$") {
 		return s
@@ -115,6 +124,8 @@ func convertConf(dir string, cfg *config.Config, params map[string]string, ipv6
 	if err != nil {
 		return err
 	}
+	cfg.Settings.InvalidDisposition = disposition(conf["INVALID_DISPOSITION"])
+	cfg.Settings.UntrackedDisposition = disposition(conf["UNTRACKED_DISPOSITION"])
 	if conf == nil {
 		return nil
 	}
@@ -136,7 +147,6 @@ func convertConf(dir string, cfg *config.Config, params map[string]string, ipv6
 	if v, ok := conf["IMPLICIT_CONTINUE"]; ok {
 		cfg.Settings.ImplicitContinue = v == "Yes"
 	}
-
 	return nil
 }
 

internal/shorewall/convert_test.go

@@ -725,3 +725,34 @@ func TestIsIPv6Dir(t *testing.T) {
 		}
 	})
 }
+
+func TestConvert_Dispositions(t *testing.T) {
+	cases := []struct {
+		conf               string
+		invalid, untracked config.PolicyAction
+	}{
+		{"", config.PolicyContinue, config.PolicyContinue},
+		{"IP_FORWARDING=Yes", config.PolicyContinue, config.PolicyContinue},
+		{"INVALID_DISPOSITION=CONTINUE\nUNTRACKED_DISPOSITION=ACCEPT", config.PolicyContinue, config.PolicyAccept},
+		{"INVALID_DISPOSITION=DROP\nUNTRACKED_DISPOSITION=A_DROP", config.PolicyDrop, config.PolicyDrop},
+		{"INVALID_DISPOSITION=A_REJECT", config.PolicyReject, config.PolicyContinue},
+	}
+	for _, tc := range cases {
+		dir := t.TempDir()
+		writeFile(t, dir, "shorewall.conf", tc.conf)
+		writeFile(t, dir, "zones", "fw firewall\nnet ipv4\n")
+		writeFile(t, dir, "interfaces", "net eth0 -\n")
+		writeFile(t, dir, "policy", "all all DROP\n")
+		cfg, err := Convert(dir)
+		if err != nil {
+			t.Fatalf("Convert(%q): %v", tc.conf, err)
+		}
+		if cfg.Settings.InvalidDisposition != tc.invalid || cfg.Settings.UntrackedDisposition != tc.untracked {
+			t.Errorf("%q: got invalid=%q untracked=%q, want %q/%q", tc.conf,
+				cfg.Settings.InvalidDisposition, cfg.Settings.UntrackedDisposition, tc.invalid, tc.untracked)
+		}
+		if err := cfg.Validate(); err != nil {
+			t.Errorf("%q: Validate: %v", tc.conf, err)
+		}
+	}
+}

internal/tryapply/tryapply.go

@@ -0,0 +1,218 @@
+// Package tryapply keeps the state of a pending 'tomswall try' on disk so the
+// revert survives the try process, backed by a transient systemd timer.
+package tryapply
+
+import (
+	"crypto/rand"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"syscall"
+	"time"
+
+	"git.unkin.net/unkin/tomswall/internal/config"
+	"git.unkin.net/unkin/tomswall/internal/nftables"
+)
+
+// Unit is the transient systemd unit that reverts an unconfirmed try.
+const Unit = "tomswall-try-revert"
+
+var (
+	// Dir holds the lock and the pending snapshot.
+	Dir = "/var/lib/tomswall"
+	// run executes a systemd command; replaced in tests.
+	run = func(name string, args ...string) error {
+		if out, err := exec.Command(name, args...).CombinedOutput(); err != nil {
+			return fmt.Errorf("%s %s: %w: %s", name, strings.Join(args, " "), err, strings.TrimSpace(string(out)))
+		}
+		return nil
+	}
+	// restore rolls the live table back to a snapshot; replaced in tests.
+	restore = func(s *nftables.Snapshot) error {
+		engine, err := nftables.NewEngine(&config.Config{Settings: config.Settings{TableName: s.Table}})
+		if err != nil {
+			return err
+		}
+		return engine.Restore(s)
+	}
+)
+
+// ErrPending means a try awaits confirmation; nothing else may apply meanwhile.
+var ErrPending = errors.New("a 'tomswall try' is pending; run 'tomswall confirm' to keep it or 'tomswall revert' to restore the previous ruleset (also the recovery if an automatic revert failed)")
+
+type pending struct {
+	ID       string             `json:"id"`
+	PID      int                `json:"pid"`
+	Snapshot *nftables.Snapshot `json:"snapshot"`
+}
+
+func snapshotPath() string { return filepath.Join(Dir, "try-snapshot.json") }
+
+// Acquire takes the exclusive try lock, failing with ErrPending while a try is unconfirmed.
+func Acquire() (unlock func(), err error) {
+	unlock, err = lock()
+	if err != nil {
+		return nil, err
+	}
+	if _, err := os.Stat(snapshotPath()); err == nil {
+		unlock()
+		return nil, ErrPending
+	}
+	return unlock, nil
+}
+
+func lock() (func(), error) {
+	if err := os.MkdirAll(Dir, 0o755); err != nil {
+		return nil, err
+	}
+	f, err := os.OpenFile(filepath.Join(Dir, "try.lock"), os.O_CREATE|os.O_RDWR, 0o600)
+	if err != nil {
+		return nil, err
+	}
+	if err := syscall.Flock(int(f.Fd()), syscall.LOCK_EX); err != nil {
+		f.Close()
+		return nil, fmt.Errorf("locking %s: %w", f.Name(), err)
+	}
+	return func() { f.Close() }, nil
+}
+
+// Arm persists snap and schedules an out-of-process revert after delay,
+// returning the try ID that scopes later reverts to this try.
+// The caller must hold the lock from Acquire.
+func Arm(snap *nftables.Snapshot, pid int, delay time.Duration) (string, error) {
+	raw := make([]byte, 8)
+	if _, err := rand.Read(raw); err != nil {
+		return "", err
+	}
+	id := hex.EncodeToString(raw)
+	b, err := json.Marshal(pending{ID: id, PID: pid, Snapshot: snap})
+	if err != nil {
+		return "", err
+	}
+	f, err := os.CreateTemp(Dir, ".try-snapshot-*")
+	if err != nil {
+		return "", err
+	}
+	defer os.Remove(f.Name())
+	if _, err := f.Write(b); err != nil {
+		f.Close()
+		return "", err
+	}
+	if err := f.Sync(); err != nil {
+		f.Close()
+		return "", err
+	}
+	if err := f.Close(); err != nil {
+		return "", err
+	}
+	if err := os.Rename(f.Name(), snapshotPath()); err != nil {
+		return "", err
+	}
+
+	exe, err := os.Executable()
+	if err != nil {
+		return "", discardWith(err)
+	}
+	_ = disarm() // a leftover timer from an earlier try would block the unit name
+	if err := run("systemd-run", "--quiet", "--collect", "--unit", Unit,
+		fmt.Sprintf("--on-active=%ds", int(delay.Round(time.Second).Seconds())), exe, "revert", "--id", id); err != nil {
+		return "", discardWith(fmt.Errorf("arming revert timer: %w", err))
+	}
+	return id, nil
+}
+
+// Discard drops the pending snapshot and timer without restoring. The caller must hold the lock.
+func Discard() error {
+	_ = disarm()
+	if err := os.Remove(snapshotPath()); err != nil && !os.IsNotExist(err) {
+		return err
+	}
+	return nil
+}
+
+func discardWith(err error) error {
+	if derr := Discard(); derr != nil {
+		return fmt.Errorf("%w (discarding snapshot: %v)", err, derr)
+	}
+	return err
+}
+
+func disarm() error {
+	return run("systemctl", "stop", Unit+".timer")
+}
+
+// Confirm keeps the tried ruleset. ok is false when no try was pending, i.e.
+// it was already reverted; pid is the waiting try process, if any.
+func Confirm() (pid int, ok bool, err error) {
+	unlock, err := lock()
+	if err != nil {
+		return 0, false, err
+	}
+	defer unlock()
+	p, err := load()
+	if err != nil || p == nil {
+		return 0, false, err
+	}
+	return p.PID, true, Discard()
+}
+
+// Revert restores the pending snapshot. A non-empty id only reverts that try,
+// so a stale timer cannot undo a newer one. reverted is false when nothing
+// matching was pending (already confirmed or reverted). A failed restore keeps
+// the snapshot so 'tomswall revert' can retry.
+func Revert(id string) (reverted bool, err error) {
+	unlock, err := lock()
+	if err != nil {
+		return false, err
+	}
+	defer unlock()
+	p, err := load()
+	if err != nil || p == nil || (id != "" && p.ID != id) {
+		return false, err
+	}
+	return true, restorePending(p)
+}
+
+// Abort restores the pending snapshot after a failed apply, which may have
+// committed partially. A failed restore keeps the snapshot and timer so the
+// timer still reverts. The caller must hold the lock.
+func Abort() error {
+	p, err := load()
+	if err != nil {
+		return err
+	}
+	if p == nil {
+		return errors.New("no pending try to abort")
+	}
+	return restorePending(p)
+}
+
+func restorePending(p *pending) error {
+	if err := restore(p.Snapshot); err != nil {
+		return fmt.Errorf("restoring snapshot: %w", err)
+	}
+	return Discard()
+}
+
+func load() (*pending, error) {
+	b, err := os.ReadFile(snapshotPath())
+	if os.IsNotExist(err) {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	var p pending
+	if err := json.Unmarshal(b, &p); err != nil {
+		return nil, fmt.Errorf("parsing %s: %w", snapshotPath(), err)
+	}
+	if p.Snapshot == nil {
+		return nil, fmt.Errorf("%s has no snapshot", snapshotPath())
+	}
+	return &p, nil
+}

internal/tryapply/tryapply_test.go

@@ -0,0 +1,249 @@
+package tryapply
+
+import (
+	"errors"
+	"os"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	"git.unkin.net/unkin/tomswall/internal/nftables"
+)
+
+func setup(t *testing.T) *[]string {
+	t.Helper()
+	Dir = t.TempDir()
+	var cmds []string
+	orig := run
+	run = func(name string, args ...string) error {
+		cmds = append(cmds, name+" "+strings.Join(args, " "))
+		return nil
+	}
+	t.Cleanup(func() { run = orig })
+	return &cmds
+}
+
+func arm(t *testing.T, snap *nftables.Snapshot) string {
+	t.Helper()
+	unlock, err := Acquire()
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer unlock()
+	id, err := Arm(snap, 4242, 90*time.Second)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return id
+}
+
+func TestArmPersistsSnapshotAndTimer(t *testing.T) {
+	cmds := setup(t)
+	snap := &nftables.Snapshot{Table: "tomswall", Present: true,
+		Rules: map[string][]nftables.SnapshotRule{"input": {{Tag: "ssh", Exprs: [][]byte{{1, 2, 3}}}}}}
+	id := arm(t, snap)
+
+	info, err := os.Stat(snapshotPath())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if info.Mode().Perm() != 0o600 {
+		t.Errorf("snapshot mode %v, want 0600", info.Mode().Perm())
+	}
+	p, err := load()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if p.ID != id || id == "" || p.PID != 4242 || !reflect.DeepEqual(p.Snapshot, snap) {
+		t.Errorf("round trip mismatch: %+v", p)
+	}
+	last := (*cmds)[len(*cmds)-1]
+	if !strings.HasPrefix(last, "systemd-run ") || !strings.Contains(last, "--unit "+Unit) ||
+		!strings.Contains(last, "--on-active=90s") || !strings.HasSuffix(last, " revert --id "+id) {
+		t.Errorf("unexpected arm command %q", last)
+	}
+}
+
+func TestAbsentTableSnapshotRoundTrip(t *testing.T) {
+	setup(t)
+	arm(t, &nftables.Snapshot{Table: "tomswall"})
+	p, err := load()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if p.Snapshot.Present || p.Snapshot.Table != "tomswall" {
+		t.Errorf("absent table not preserved: %+v", p.Snapshot)
+	}
+}
+
+func TestAcquireRefusesWhilePending(t *testing.T) {
+	setup(t)
+	arm(t, &nftables.Snapshot{Table: "tomswall"})
+	_, err := Acquire()
+	if !errors.Is(err, ErrPending) {
+		t.Fatalf("second try: got %v, want ErrPending", err)
+	}
+	if !strings.Contains(err.Error(), "tomswall revert") {
+		t.Errorf("ErrPending does not name the recovery: %v", err)
+	}
+}
+
+func TestArmFailureDiscardsSnapshot(t *testing.T) {
+	setup(t)
+	run = func(name string, args ...string) error {
+		if name == "systemd-run" {
+			return errors.New("no systemd")
+		}
+		return nil
+	}
+	unlock, err := Acquire()
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer unlock()
+	if _, err := Arm(&nftables.Snapshot{Table: "tomswall"}, 1, time.Minute); err == nil {
+		t.Fatal("expected arm error")
+	}
+	if _, err := os.Stat(snapshotPath()); !os.IsNotExist(err) {
+		t.Error("snapshot left behind without a revert timer")
+	}
+}
+
+func TestConfirmPendingDisarms(t *testing.T) {
+	cmds := setup(t)
+	arm(t, &nftables.Snapshot{Table: "tomswall"})
+	pid, ok, err := Confirm()
+	if err != nil || !ok || pid != 4242 {
+		t.Fatalf("Confirm = %d, %v, %v", pid, ok, err)
+	}
+	if _, err := os.Stat(snapshotPath()); !os.IsNotExist(err) {
+		t.Error("snapshot not removed")
+	}
+	if last := (*cmds)[len(*cmds)-1]; last != "systemctl stop "+Unit+".timer" {
+		t.Errorf("timer not stopped, last command %q", last)
+	}
+	unlock, err := Acquire()
+	if err != nil {
+		t.Fatalf("new try refused after confirm: %v", err)
+	}
+	unlock()
+}
+
+func TestConfirmAfterRevertFails(t *testing.T) {
+	setup(t)
+	_, ok, err := Confirm()
+	if err != nil || ok {
+		t.Fatalf("Confirm with nothing pending = %v, %v; want not ok", ok, err)
+	}
+	reverted, err := Revert("")
+	if err != nil || reverted {
+		t.Fatalf("Revert with nothing pending = %v, %v", reverted, err)
+	}
+}
+
+func stubRestore(t *testing.T, err error) *[]*nftables.Snapshot {
+	t.Helper()
+	var got []*nftables.Snapshot
+	orig := restore
+	restore = func(s *nftables.Snapshot) error {
+		got = append(got, s)
+		return err
+	}
+	t.Cleanup(func() { restore = orig })
+	return &got
+}
+
+func TestRevertPendingRestoresAndDisarms(t *testing.T) {
+	cmds := setup(t)
+	restored := stubRestore(t, nil)
+	snap := &nftables.Snapshot{Table: "tomswall", Present: true,
+		Rules: map[string][]nftables.SnapshotRule{"input": {{Tag: "ssh", Exprs: [][]byte{{1}}}}}}
+	id := arm(t, snap)
+
+	reverted, err := Revert(id)
+	if err != nil || !reverted {
+		t.Fatalf("Revert = %v, %v", reverted, err)
+	}
+	if len(*restored) != 1 || !reflect.DeepEqual((*restored)[0], snap) {
+		t.Errorf("restored %+v, want the armed snapshot", *restored)
+	}
+	if _, err := os.Stat(snapshotPath()); !os.IsNotExist(err) {
+		t.Error("snapshot not removed")
+	}
+	if last := (*cmds)[len(*cmds)-1]; last != "systemctl stop "+Unit+".timer" {
+		t.Errorf("timer not stopped, last command %q", last)
+	}
+}
+
+func TestRevertFailureKeepsSnapshot(t *testing.T) {
+	setup(t)
+	boom := errors.New("netlink down")
+	stubRestore(t, boom)
+	id := arm(t, &nftables.Snapshot{Table: "tomswall"})
+
+	if _, err := Revert(id); !errors.Is(err, boom) {
+		t.Fatalf("Revert error = %v, want %v", err, boom)
+	}
+	if _, err := os.Stat(snapshotPath()); err != nil {
+		t.Fatalf("snapshot gone after failed revert: %v", err)
+	}
+	if _, err := Acquire(); !errors.Is(err, ErrPending) {
+		t.Errorf("failed revert must keep the try pending, got %v", err)
+	}
+}
+
+func TestRevertStaleIDIgnored(t *testing.T) {
+	setup(t)
+	restored := stubRestore(t, nil)
+	arm(t, &nftables.Snapshot{Table: "tomswall"})
+
+	reverted, err := Revert("stale-try")
+	if err != nil || reverted {
+		t.Fatalf("stale Revert = %v, %v; want no-op", reverted, err)
+	}
+	if len(*restored) != 0 {
+		t.Error("stale timer restored a newer try's snapshot")
+	}
+	if _, err := os.Stat(snapshotPath()); err != nil {
+		t.Errorf("newer try's snapshot removed: %v", err)
+	}
+}
+
+func TestAbortRestoresAndDisarms(t *testing.T) {
+	cmds := setup(t)
+	restored := stubRestore(t, nil)
+	snap := &nftables.Snapshot{Table: "tomswall", Present: true}
+	arm(t, snap)
+
+	if err := Abort(); err != nil {
+		t.Fatal(err)
+	}
+	if len(*restored) != 1 || !reflect.DeepEqual((*restored)[0], snap) {
+		t.Errorf("restored %+v, want the armed snapshot", *restored)
+	}
+	if _, err := os.Stat(snapshotPath()); !os.IsNotExist(err) {
+		t.Error("snapshot not removed")
+	}
+	if last := (*cmds)[len(*cmds)-1]; last != "systemctl stop "+Unit+".timer" {
+		t.Errorf("timer not stopped, last command %q", last)
+	}
+}
+
+func TestAbortFailureKeepsSnapshotAndTimer(t *testing.T) {
+	cmds := setup(t)
+	boom := errors.New("netlink down")
+	stubRestore(t, boom)
+	arm(t, &nftables.Snapshot{Table: "tomswall"})
+	armed := len(*cmds)
+
+	if err := Abort(); !errors.Is(err, boom) {
+		t.Fatalf("Abort error = %v, want %v", err, boom)
+	}
+	if _, err := os.Stat(snapshotPath()); err != nil {
+		t.Fatalf("snapshot gone after failed abort: %v", err)
+	}
+	if len(*cmds) != armed {
+		t.Errorf("revert timer touched after failed abort: %v", (*cmds)[armed:])
+	}
+}

tomswall.example.yaml

@@ -8,6 +8,10 @@ settings:
   log_level: info
   table_name: tomswall
   implicit_continue: false
+  # ct state invalid/untracked verdict: accept, drop, reject, continue (pass to rules)
+  # defaults: invalid drop, untracked continue (migrate defaults both to continue, as shorewall)
+  invalid_disposition: drop
+  untracked_disposition: continue
 
 # Named port groups — reusable port+protocol combos referenced in rules
 portgroups:
@@ -191,13 +195,12 @@ snat:
 # conntrack:
 #   - action: notrack
 #     source: net
-#     dest: fw
+#     dest: fw:203.0.113.1
 #     proto: udp
 #     dport: [53]
 #     comment: "Skip conntrack for DNS"
 #   - action: helper
 #     source: loc
-#     dest: net
 #     proto: tcp
 #     dport: [21]
 #     helper: ftp