Fuzz NewManifestFromReader and cap the zstd decoder (closes #65)

FuzzNewManifestFromReader fails when the parser returns both or neither
of a manifest and an error, or allocates more than sixteen times its
input and the decompressed data it may read, plus room for the decoder's
window. make test runs the committed seed corpus; make fuzz fuzzes for
one minute, by hand only.

Parser bug: MaxDecompressedSize did not bound decompression. The zstd
decoder decoded a payload under 128 KiB in full, each frame up to its
own 64 GiB limit, before the LimitReader read any of it. It now decodes
synchronously, only what the LimitReader reads, with MaxDecompressedSize
as its limit. Regression seeds: a frame claiming 8 GiB, and two frames
each under the limit and together over it.

Model: opus-5-5

.dockerignore

@@ -1,4 +1,64 @@
+# .dockerignore does NOT use .gitignore semantics. Docker matches with
+# moby/patternmatcher: filepath.Match plus `**`, so `*` does not cross
+# `/` and an unprefixed pattern is anchored at the context root. Every
+# depth-independent pattern therefore needs `**/`, or `config/.env` and
+# `certs/server.key` still ship while this file reads as solved. Only
+# genuinely root-anchored entries go unprefixed. Never transplant these
+# into .gitignore, where `**/` is wrong.
+#
+# Matching is case-sensitive, so secrets use character ranges rather
+# than an ALL-CAPS twin, which would still miss `Server.Key`.
+#
+# Extend with this repo's own host-built artifacts, written anchored:
+# `/myapp`, never `**/myapp`, which also matches `cmd/myapp/` and
+# deletes the package directory from the context.
+
+# .git is sent without its config. Without a VERSION build argument the
+# stage that compiles runs `git describe --tags --always` on .git, which
+# does not need .git/config; that file can hold a credential, such as a
+# password in a remote URL or the token the CI checkout step stores there.
+.git/config
+
+# Agent scratch: one full checkout of the repo per in-flight agent.
+# Anchored because it occurs once where agents run at the repo root.
+# KNOWN GAP: a repo running agents in subdirectories still ships
+# `services/api/.claude/` and must add its own anchored entry.
+.claude
+
+# Environment files. `*.env` covers bare `.env` and the `prod.env`
+# convention. Re-include a committed template with a negation if the
+# build needs one: `!docs/example.env`.
+**/*.[eE][nN][vV]
+**/.[eE][nN][vV].*
+**/.[eE][nN][vV][rR][cC]
+
+# Private keys and the bundles carrying them. Public certificates
+# (*.crt, *.cer) are deliberately absent: they are legitimate inputs.
+**/*.[pP][eE][mM]
+**/*.[kK][eE][yY]
+**/*.[pP]12
+**/*.[pP][fF][xX]
+**/[iI][dD]_[rR][sS][aA]
+**/[iI][dD]_[dD][sS][aA]
+**/[iI][dD]_[eE][cC][dD][sS][aA]
+**/[iI][dD]_[eE][dD]25519
+
+# Dependencies: restored inside the image, never copied in.
+**/node_modules
+
+# OS metadata.
+**/.DS_Store
+**/Thumbs.db
+
+# Editor state: never a build input, and it churns COPY.
+**/*.swp
+**/*.swo
+**/*~
+**/*.bak
+**/.idea
+**/.vscode
+**/*.sublime-*
+
+# This repo's own host-built archives (Makefile).
 *.tmp
 *.dockerimage
-.git
-node_modules

Dockerfile

@@ -48,7 +48,26 @@ COPY . .
 RUN touch mfer/mf.pb.go
 
 RUN make test
-RUN cd cmd/mfer && go build -tags urfave_cli_no_docs -o /mfer .
+
+# A build context sent as a tar archive, as upaas sends it, keeps its files'
+# owners, and git refuses to read a checkout owned by another user.
+RUN git config --system --add safe.directory /src
+
+# The revision `mfer version` prints, stamped into main.Gitrev: the VERSION
+# build argument when one is given (script/docker passes one), otherwise
+# `git describe --tags --always` of the .git the build context carries: the
+# tag on a tagged commit, tag-N-gHASH on a commit after one, the short commit
+# when no tag is reachable. git ships in this base image. A context that
+# carries .git and still yields no version fails the build.
+ARG VERSION
+RUN version="${VERSION:-$(git describe --tags --always)}"; \
+    if [ -e .git ] && { [ -z "$version" ] || [ "$version" = dev ] || \
+        [ "$version" = unknown ]; }; then \
+        echo "no version could be derived although the build context carries .git" >&2; \
+        exit 1; \
+    fi; \
+    cd cmd/mfer && \
+    go build -tags urfave_cli_no_docs -ldflags "-X main.Gitrev=$version" -o /mfer .
 
 FROM scratch
 COPY --from=builder /mfer /mfer

Makefile

@@ -13,7 +13,7 @@ GOLDFLAGS += -X main.Version=$(VERSION)
 GOLDFLAGS += -X main.Gitrev=$(GITREV_BUILD)
 GOFLAGS := -ldflags "$(GOLDFLAGS)"
 
-.PHONY: bootstrap setup docker default run ci test check lint fmt fmt-check fmt-check-go fmt-check-md hooks fixme
+.PHONY: bootstrap setup docker default run ci test fuzz check lint fmt fmt-check fmt-check-go fmt-check-md hooks fixme
 
 default: fmt test
 
@@ -32,6 +32,9 @@ ci: test
 test:
 	@script/test
 
+fuzz:
+	@script/fuzz
+
 $(PROTOC_GEN_GO):
 	test -e $(PROTOC_GEN_GO) || go install -v google.golang.org/protobuf/cmd/protoc-gen-go@v1.28.1
 

README.md

@@ -1,12 +1,12 @@
 # mfer
 
-[mfer](https://git.eeqj.de/sneak/mfer) is a [WTFPL](https://wtfpl.net)-licensed
+[mfer](https://git.eeqj.de/sneak/mfer) is a reference implementation library and
-(public domain) [Go](https://golang.org) library and command-line tool by
+thin wrapper command-line utility written in [Go](https://golang.org) and first
-[@sneak](https://sneak.berlin) that specifies and generates `.mf` manifest files
+published in 2022 under the [WTFPL](https://wtfpl.net) (public domain) license.
-over a directory tree to encapsulate metadata about the files — such as
+It specifies and generates `.mf` manifest files over a directory tree of files
-cryptographic checksums and signatures over same — to aid in archiving,
+to encapsulate metadata about them (such as cryptographic checksums or
-downloading, streaming, and mirroring. It was first published in 2022. The
+signatures over same) to aid in archiving, downloading, and streaming, or
-manifest files' data is serialized with Google's
+mirroring. The manifest files' data is serialized with Google's
 [protobuf serialization format](https://developers.google.com/protocol-buffers).
 The structure of these files can be found
 [in the format specification](https://git.eeqj.de/sneak/mfer/src/branch/main/mfer/mf.proto)
@@ -21,40 +21,11 @@ This project was started by [@sneak](https://sneak.berlin) to scratch an itch in
 as a de-facto standard and be incorporated into other software. A compatible
 javascript library is planned.
 
-# Getting Started
-
-`mfer` builds from source with a Go 1.23+ toolchain. The generated protobuf code
-is committed, so no `protoc` toolchain is required:
-
-```sh
-git clone https://git.eeqj.de/sneak/mfer.git
-cd mfer
-go build -o bin/mfer ./cmd/mfer
-```
-
-Generate a manifest for a directory tree, verify it later, and fetch a published
-tree by URL:
-
-```sh
-# Write .index.mf describing every file under the current directory.
-bin/mfer gen .
-
-# Verify the files on disk against the manifest. Exits nonzero if any file
-# is missing or corrupted.
-bin/mfer check .index.mf
-
-# Download and cryptographically verify a tree published over HTTP: mfer
-# fetches <url>/index.mf, then downloads every file it lists.
-bin/mfer fetch https://example.com/tree/
-```
-
-Run `bin/mfer help` for the full command list, or `bin/mfer <command> --help`
-for a single command's options.
-
 # Build Status
 
-CI runs via `script/cibuild` (`docker build .`), which executes `make check`
+CI runs `script/cibuild`, which builds the Docker image with `--no-cache`, so
-(formatting, linting, tests). The `main` branch must always be green.
+the formatting, lint and test steps in the `Dockerfile` run on every build. The
+`main` branch must always be green.
 
 # Entrypoints
 
@@ -73,6 +44,9 @@ provide:
   such as `script/docker`
 - `script/test` — run the test suite (`go test`), regenerating the protobuf code
   first if it is stale
+- `script/fuzz` — fuzz the manifest parser for one minute; run by hand
+  (`make fuzz`), never by CI, while `script/test` runs its committed seed corpus
+  as ordinary tests
 - `script/lint` — run `golangci-lint` and verify `gofmt` cleanliness
 - `script/fmt` — format all code and docs (writes): `gofumpt`,
   `golangci-lint run --fix`, and `script/prettier --write`
@@ -86,8 +60,8 @@ provide:
   Docker lint stage, whose image has no node
 - `script/check` — run `script/test`, `script/lint`, and `script/fmt-check`
 - `script/docker` — build the Docker image tagged with the project name
-- `script/cibuild` — CI entrypoint: `docker build .` (the Dockerfile runs the
+- `script/cibuild` — CI entrypoint: builds the image with the same command as
-  checks)
+  `script/docker`, uncached, so the checks in the Dockerfile run every time
 - `script/precommit` — pre-commit checks: `go mod tidy` verification, then
   `script/check`
 - `script/install-precommit` — install the git pre-commit hook that runs
@@ -111,9 +85,7 @@ Any changes submitted to this project must also be
 See [`REPO_POLICIES.md`](REPO_POLICIES.md) for detailed coding standards,
 tooling requirements, and workflow conventions.
 
-# Rationale
+# Problem Statement
-
-## The problem
 
 Given a plain URL, there is no standard way to safely and programmatically
 download everything "under" that URL path. `wget -r` can traverse directory
@@ -137,7 +109,7 @@ Real issues I face:
 - when I download a large file via HTTP, I have no way of knowing if the file
   content is what it's supposed to be
 
-## The solution
+# Proposed Solution
 
 A standard, a manifest file format, and a tool for generating same.
 
@@ -169,27 +141,6 @@ The manifest file would do several important things:
     - maybe a bittorrent chunklist for torrent client compatibility? perhaps a
       top-level infohash for the whole manifest?
 
-# Design
-
-The repository is split into a reusable library and a thin command-line wrapper
-around it.
-
-- `mfer/` is the reusable library and the heart of the project: it defines the
-  manifest format and implements building, scanning, checking, serialization,
-  and signing. The protobuf schema is `mfer/mf.proto`, and the generated code it
-  produces (`mfer/mf.pb.go`) is committed alongside it so the library builds
-  with `go get` and needs no `protoc` toolchain.
-- `internal/cli/` holds the command implementations — `gen`, `check`, `freshen`,
-  `export`, `list`, and `fetch` — that wire the library to the command-line
-  interface.
-- `internal/log/` provides the logging used across the commands.
-- `internal/bork/` provides error-handling support.
-- `cmd/mfer/` is the entrypoint: its `main` package assembles the pieces above
-  into the `mfer` binary.
-
-Everything under `internal/` is private to this repository; only the `mfer/`
-package is intended for import by other software.
-
 # Design Goals
 
 - Replace SHASUMS/SHASUMS.asc files
@@ -515,9 +466,9 @@ proto `go_package` option. Which is canonical?
 - Issues:
   [https://git.eeqj.de/sneak/mfer/issues](https://git.eeqj.de/sneak/mfer/issues)
 
-# Author
+# Authors
 
-- [@sneak](https://sneak.berlin)
+- [@sneak &lt;sneak@sneak.berlin&gt;](mailto:sneak@sneak.berlin)
 
 # License
 

TODO.md

@@ -24,11 +24,35 @@ only thing left of the `chore/align-repo-policies` branch is the list below.
 
 # Completed Steps
 
-- 2026-09-21: added the required README sections — named author, license, and
+- 2026-10-03: added `FuzzNewManifestFromReader` and its seed corpus, which
-  category in the Description first line; added Getting Started (verified
+  `make test` runs, plus `make fuzz` for a one-minute run by hand; the zstd
-  install/usage block), Rationale (folding in Problem Statement and Proposed
+  decoder now decodes only as many bytes as the parser reads, and has
-  Solution), and a Design section for the package layout; renamed Authors to
+  `MaxDecompressedSize` as its limit, so neither a frame claiming a large size
-  Author (#75)
+  nor several frames together can make the parser allocate past that limit (#65)
+- 2026-10-03: pinned the CLI error messages by driving the functions that emit
+  them in `internal/cli/errmsg_test.go`, and made the freshen mtime-presence
+  test distinguish an absent mtime from the epoch (#87)
+- 2026-10-03: `script/cibuild` builds the image with the same command as
+  `script/docker`, `--no-cache` included, so the checks in the Dockerfile run on
+  every build, also on an unchanged tree (#89)
+- 2026-10-03: `fetch` removes whatever sits at a file's temp name and then
+  creates the temp file only if that name is free, so a hard link left there
+  cannot make it write into a file outside the destination directory (#115)
+- 2026-10-03: `fetch` refuses any manifest path that runs through a symlink
+  already in the destination directory, checked before each of its writes
+  (directories, temp file, rename), so such a symlink cannot send a write
+  outside it (#86)
+- 2026-10-02: a plain `docker build .` of a clone now stamps the tag or short
+  commit into `mfer version` instead of nothing: `.dockerignore` sends `.git`
+  (not `.git/config`), and the build stage takes the `VERSION` build argument,
+  otherwise `git describe --tags --always`, failing if `.git` is present and no
+  version comes out. `script/docker` is the canonical copy, which passes
+  `VERSION`; `bin/gitrev.sh` uses `--tags` too (#112)
+- 2026-09-21: validate manifest entry paths on deserialize so untrusted `.mf`
+  files cannot make `Checker` stat or read outside `basePath` (#61)
+- 2026-09-21: rewrote `script/test` to the canonical pattern (30s timeout,
+  `-race -cover`, quiet-first with verbose-on-failure rerun) and fixed the
+  process-global logger data race it surfaced (#67)
 - 2026-09-21: added the canonical `.editorconfig`, made `.gitignore` cover
   secrets, OS, editor, and Go artifacts, and removed the dead Drone CI
   references from `.gitignore` and `bin/gitrev.sh` (#72)
@@ -70,7 +94,6 @@ only thing left of the `chore/align-repo-policies` branch is the list below.
     - Move FORMAT.md from repo root to docs/ and update the AGENTS.md reference
     - Pin Makefile-installed Go tools (`protoc-gen-go@v1.28.1`,
       `golangci-lint@v2.12.2`) by module hash, not mutable tag
-    - Set `make test` timeout to 30s (currently 10s)
     - Add explicit README "Rationale" heading (content exists under other
       names); name the author in the README Description first line
     - Reconcile root-level AGENTS.md with directory-hygiene policy (keep or
@@ -108,7 +131,7 @@ only thing left of the `chore/align-repo-policies` branch is the list below.
       rate-limit Checker progress output; add --deterministic flag or default;
       wire top-level --version properly
 - Testing:
-    - Fuzz NewManifestFromReader; end-to-end tests for freshen and fetch
+    - End-to-end tests for freshen and fetch
 - Documentation:
     - Promote docs/FORMAT.md as primary spec reference; audit error messages;
       document the signature scheme fully

bin/gitrev.sh

@@ -3,5 +3,5 @@
 if [[ ! -z "$GITREV" ]]; then
     echo $GITREV
 else
-    git describe --always --dirty=-dirty
+    git describe --tags --always --dirty=-dirty
 fi

internal/cli/entry_test.go

@@ -126,7 +126,7 @@ func TestHelpCommand(t *testing.T) {
 	stdout := testStdout(t, opts)
 	assert.Contains(t, stdout, cmdGenerate)
 	assert.Contains(t, stdout, cmdCheck)
-	assert.Contains(t, stdout, "fetch")
+	assert.Contains(t, stdout, cmdFetch)
 }
 
 func TestGenerateCommand(t *testing.T) {
@@ -679,11 +679,13 @@ func TestCheckDetectsManifestCorruption(t *testing.T) {
 	fs := afero.NewMemMapFs()
 	rng := rand.New(rand.NewSource(42)) //nolint:gosec // deterministic test data
 
-	// Create many small files with random names to generate a ~1MB manifest
+	// Create many small files with random names so the manifest has many
-	// Each manifest entry is roughly 50-60 bytes, so we need ~20000 files
+	// entries and random single-byte flips land at varied offsets. Each
+	// manifest entry is roughly 50-60 bytes. Kept modest so the suite stays
+	// within its wall-clock budget under -race.
 	require.NoError(t, fs.MkdirAll(testDir, 0o755))
 
-	numFiles := 20000
+	numFiles := 1500
 	for range numFiles {
 		// Generate random filename
 		filename := fmt.Sprintf("/testdir/%08x%08x%08x.dat",
@@ -699,11 +701,11 @@ func TestCheckDetectsManifestCorruption(t *testing.T) {
 	exitCode := runCLI(opts)
 	require.Equal(t, 0, exitCode, "generate should succeed")
 
-	// Read the valid manifest and verify it's approximately 1MB
+	// Read the valid manifest and verify it has real size.
 	validManifest, err := afero.ReadFile(fs, testManifest)
 	require.NoError(t, err)
-	require.GreaterOrEqual(t, len(validManifest), 1024*1024,
+	require.GreaterOrEqual(t, len(validManifest), 64*1024,
-		"manifest should be at least 1MB, got %d bytes", len(validManifest))
+		"manifest should be at least 64KB, got %d bytes", len(validManifest))
 	t.Logf("manifest size: %d bytes (%d files)", len(validManifest), numFiles)
 
 	// First corruption: truncate the manifest
@@ -726,8 +728,8 @@ func TestCheckDetectsManifestCorruption(t *testing.T) {
 	exitCode = runCLI(opts)
 	require.Equal(t, 0, exitCode, "check should pass with valid manifest")
 
-	// Now do 500 random corruption iterations
+	// Now do 100 random corruption iterations
-	for i := range 500 {
+	for i := range 100 {
 		// Corrupt: write a random byte at a random offset
 		corrupted := make([]byte, len(validManifest))
 		copy(corrupted, validManifest)

internal/cli/errmsg_test.go

@@ -2,167 +2,352 @@
 package cli
 
 import (
-	"fmt"
+	"bytes"
+	"context"
+	"flag"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"os/exec"
+	"path/filepath"
 	"testing"
 
+	"github.com/spf13/afero"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	urfcli "github.com/urfave/cli/v2"
+	"sneak.berlin/go/mfer/mfer"
 )
 
-// errMsgCase is one pinned user-visible error message.
+// These tests pin the exact rendered text of the CLI's user-visible error
-type errMsgCase struct {
+// messages. The messages are grepped for in CI pipelines and quoted in bug
-	name string
+// reports, so a reword is a deliberate change, never a refactoring side
-	err  error
+// effect.
-	want string
+//
-}
+// Every case drives the real function that emits the message and asserts on
+// what it returns. No production format string is restated here: a test that
+// only re-rendered a copied format string would keep passing after the real
+// message changed, which is exactly the regression these tests exist to
+// catch.
 
+// Full 40-hex fingerprints used where a message embeds one.
 const (
 	msgFpA = "AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"
 	msgFpB = "BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB"
 )
 
-func checkErrMsgCases(t *testing.T, cases []errMsgCase) {
+// runLocked runs fn while holding runMu, so operations that write to the
+// process-global logger do not race the other CLI runs.
+func runLocked(fn func() error) error {
+	runMu.Lock()
+	defer runMu.Unlock()
+
+	return fn()
+}
+
+// unsignedChecker builds a Checker over a freshly scanned, unsigned manifest.
+func unsignedChecker(t *testing.T) *mfer.Checker {
 	t.Helper()
 
-	for _, tc := range cases {
+	fs := afero.NewMemMapFs()
-		t.Run(tc.name, func(t *testing.T) {
+	require.NoError(t, fs.MkdirAll("/d", 0o755))
-			t.Parallel()
+	require.NoError(t, afero.WriteFile(fs, "/d/f.txt", []byte("hi"), 0o644))
-			assert.Equal(t, tc.want, tc.err.Error())
+
-		})
+	s := mfer.NewScannerWithOptions(&mfer.ScannerOptions{Fs: fs})
-	}
+	require.NoError(t, s.EnumeratePath("/d", nil))
+
+	var buf bytes.Buffer
+
+	require.NoError(t, s.ToManifest(context.Background(), &buf, nil))
+	require.NoError(t, afero.WriteFile(fs, "/d/index.mf", buf.Bytes(), 0o644))
+
+	chk, err := mfer.NewChecker("/d/index.mf", "/d", fs)
+	require.NoError(t, err)
+	require.False(t, chk.IsSigned())
+
+	return chk
 }
 
-// TestErrorMessagesVerbatim pins the exact rendered text of the CLI's
+func TestNoManifestFoundMessage(t *testing.T) {
-// user-visible error messages.
+	t.Parallel()
+
+	_, err := findManifest(afero.NewMemMapFs(), "/tmp/x")
+	require.ErrorIs(t, err, errNoManifestFound)
+	assert.EqualError(t, err,
+		"no manifest found in /tmp/x (looked for index.mf and .index.mf)")
+}
+
+func TestVerifyRequiredSignerMessages(t *testing.T) {
+	t.Parallel()
+
+	t.Run("invalid fingerprint length", func(t *testing.T) {
+		t.Parallel()
+
+		err := verifyRequiredSigner(unsignedChecker(t), "12345678")
+		require.ErrorIs(t, err, errInvalidFingerprint)
+		assert.EqualError(t, err,
+			"invalid fingerprint: must be exactly 40 hex characters, got 8")
+	})
+
+	t.Run("manifest not signed", func(t *testing.T) {
+		t.Parallel()
+
+		err := verifyRequiredSigner(unsignedChecker(t), msgFpA)
+		require.ErrorIs(t, err, errManifestNotSigned)
+		assert.EqualError(t, err,
+			"manifest is not signed, but signature from "+msgFpA+" is required")
+	})
+}
+
+// TestSignerMismatchMessage drives verifyRequiredSigner against a real signed
+// manifest. The embedded fingerprint is whatever the generated key produced,
+// so it is read back from the checker and substituted into the expected
+// string; the required signer is a fixed value that cannot match it. Requires
+// gpg and is skipped where it is absent, as the other signing tests are.
 //
-// These strings are an interface: they are grepped for in CI pipelines
+//nolint:paralleltest // signedChecker calls t.Setenv, which bars t.Parallel
-// and quoted in bug reports. The messages are assembled by wrapping
+func TestSignerMismatchMessage(t *testing.T) {
-// static sentinels, and it is easy to change what a user sees while
+	chk := signedChecker(t)
-// only meaning to make an error matchable with errors.Is - which is
+
-// precisely what happened once already. Any change to a string below is
+	embeddedFP, err := chk.ExtractEmbeddedSigningKeyFP()
-// therefore a deliberate, separately stated change, never a side effect
+	require.NoError(t, err)
-// of a refactor.
+
-func TestErrorMessagesVerbatim(t *testing.T) {
+	err = verifyRequiredSigner(chk, msgFpB)
+	require.ErrorIs(t, err, errSignerMismatch)
+	assert.EqualError(t, err,
+		"embedded signing key fingerprint "+embeddedFP+
+			" does not match required "+msgFpB)
+}
+
+// signedChecker builds a Checker over a manifest signed by a throwaway GPG
+// key generated in a temporary GNUPGHOME.
+func signedChecker(t *testing.T) *mfer.Checker {
+	t.Helper()
+
+	_, err := exec.LookPath("gpg")
+	if err != nil {
+		t.Skip("gpg not installed, skipping signing test")
+	}
+
+	gpgHome := t.TempDir()
+	params := "%no-protection\n" +
+		"Key-Type: RSA\nKey-Length: 2048\n" +
+		"Name-Real: MFER Test Key\nName-Email: test@mfer.test\n" +
+		"Expire-Date: 0\n%commit\n"
+	paramsFile := filepath.Join(gpgHome, "key-params")
+	require.NoError(t, os.WriteFile(paramsFile, []byte(params), 0o600))
+
+	//nolint:gosec // paramsFile is a test-controlled path inside t.TempDir()
+	cmd := exec.CommandContext(context.Background(), "gpg",
+		"--batch", "--gen-key", paramsFile)
+
+	cmd.Env = append(os.Environ(), "GNUPGHOME="+gpgHome)
+
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		t.Skipf("failed to generate test GPG key: %v: %s", err, out)
+	}
+
+	t.Setenv("GNUPGHOME", gpgHome)
+
+	b := mfer.NewBuilder()
+	b.SetSigningOptions(&mfer.SigningOptions{KeyID: mfer.GPGKeyID("test@mfer.test")})
+
+	content := []byte("signed file")
+	_, err = b.AddFile("f.txt", mfer.FileSize(len(content)), mfer.ModTime{},
+		bytes.NewReader(content), nil)
+	require.NoError(t, err)
+
+	var buf bytes.Buffer
+
+	require.NoError(t, b.Build(&buf))
+
+	fs := afero.NewMemMapFs()
+	require.NoError(t, afero.WriteFile(fs, "/index.mf", buf.Bytes(), 0o644))
+
+	chk, err := mfer.NewChecker("/index.mf", "/", fs)
+	require.NoError(t, err)
+	require.True(t, chk.IsSigned())
+
+	return chk
+}
+
+func TestPathDoesNotExistMessage(t *testing.T) {
 	t.Parallel()
 
-	checkErrMsgCases(t, []errMsgCase{
+	set := flag.NewFlagSet("gen", flag.ContinueOnError)
-		{
+	require.NoError(t, set.Parse([]string{"nope"}))
-			name: "check: no manifest found",
+
-			err: fmt.Errorf("%w in %s (looked for index.mf and .index.mf)",
+	mfa := &CLIApp{Fs: afero.NewMemMapFs()}
-				errNoManifestFound, "/tmp/x"),
+	ctx := urfcli.NewContext(nil, set, nil)
-			want: "no manifest found in /tmp/x " +
+
-				"(looked for index.mf and .index.mf)",
+	_, err := mfa.collectInputPaths(ctx.Args())
-		},
+	require.ErrorIs(t, err, errPathNotExist)
-		{
+	assert.EqualError(t, err, "path does not exist: nope")
-			name: "check: invalid fingerprint length",
+}
-			err:  fmt.Errorf("%w, got %d", errInvalidFingerprint, 8),
+
-			want: "invalid fingerprint: must be exactly 40 hex characters, got 8",
+func TestOutputFileExistsMessage(t *testing.T) {
-		},
+	t.Parallel()
-		{
+
-			name: "check: manifest not signed",
+	fs := afero.NewMemMapFs()
-			err: fmt.Errorf("%w, but signature from %s is required",
+	require.NoError(t, fs.MkdirAll("/d", 0o755))
-				errManifestNotSigned, msgFpA),
+	require.NoError(t, afero.WriteFile(fs, "/d/f.txt", []byte("hi"), 0o644))
-			want: "manifest is not signed, but signature from " + msgFpA +
+	require.NoError(t, afero.WriteFile(fs, "/out.mf", []byte("old"), 0o644))
-				" is required",
+
-		},
+	set := flag.NewFlagSet("gen", flag.ContinueOnError)
-		{
+	set.String("output", "", "")
-			name: "check: signer mismatch",
+	set.Bool("force", false, "")
-			err: fmt.Errorf("embedded signing key fingerprint %s %w %s",
+	require.NoError(t, set.Parse([]string{"/d"}))
-				msgFpA, errSignerMismatch, msgFpB),
+	require.NoError(t, set.Set("output", "/out.mf"))
-			want: "embedded signing key fingerprint " + msgFpA +
+
-				" does not match required " + msgFpB,
+	mfa := &CLIApp{Fs: fs}
-		},
+	ctx := urfcli.NewContext(nil, set, nil)
-		{
+
-			name: "gen: path does not exist",
+	// generateManifestOperation writes to the process-global logger during
-			err:  fmt.Errorf("%w: %s", errPathNotExist, "nope"),
+	// enumeration, so serialize with the other CLI runs.
-			want: "path does not exist: nope",
+	err := runLocked(func() error { return mfa.generateManifestOperation(ctx) })
-		},
+	require.ErrorIs(t, err, errOutputExists)
-		{
+	assert.EqualError(t, err,
-			name: "gen: output file exists",
+		"output file /out.mf already exists (use --force to overwrite)")
-			err:  fmt.Errorf("output file %s %w", "index.mf", errOutputExists),
+}
-			want: "output file index.mf already exists " +
+
-				"(use --force to overwrite)",
+// TestUnknownCommandMessage drives the root command's action. run only logs
-		},
+// the error that action returns, so the test lets run build the app with no
-		{
+// command given and then runs that same app on an unknown command to get the
-			name: "mfer: unknown command",
+// error itself.
-			err:  fmt.Errorf("%w %q", errUnknownCommand, "bogus"),
+func TestUnknownCommandMessage(t *testing.T) {
-			want: `unknown command "bogus"`,
+	t.Parallel()
-		},
+
+	mfa := &CLIApp{
+		appname: testApp,
+		Stdout:  &bytes.Buffer{},
+		Stderr:  &bytes.Buffer{},
+		Fs:      afero.NewMemMapFs(),
+	}
+
+	// run points the process-global logger at this app's output, so
+	// serialize with the other CLI runs.
+	err := runLocked(func() error {
+		mfa.run([]string{testApp})
+
+		return mfa.app.Run([]string{testApp, "bogus"})
+	})
+	require.ErrorIs(t, err, errUnknownCommand)
+	assert.EqualError(t, err, `unknown command "bogus"`)
+}
+
+func TestManifestLoaderHTTPStatusMessage(t *testing.T) {
+	t.Parallel()
+
+	server := httptest.NewServer(
+		http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+			w.WriteHeader(http.StatusNotFound)
+		}))
+	defer server.Close()
+
+	mfa := &CLIApp{Fs: afero.NewMemMapFs()}
+
+	_, err := mfa.openManifestReader(server.URL + "/foo.mf")
+	require.ErrorIs(t, err, errHTTPStatus)
+	assert.EqualError(t, err,
+		"failed to fetch "+server.URL+"/foo.mf: HTTP 404")
+}
+
+func TestFetchManifestHTTPStatusMessage(t *testing.T) {
+	t.Parallel()
+
+	server := httptest.NewServer(
+		http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+			w.WriteHeader(http.StatusNotFound)
+		}))
+	defer server.Close()
+
+	set := flag.NewFlagSet("fetch", flag.ContinueOnError)
+	require.NoError(t, set.Parse([]string{server.URL}))
+
+	mfa := &CLIApp{Fs: afero.NewMemMapFs()}
+	ctx := urfcli.NewContext(nil, set, nil)
+
+	// fetchManifestOperation logs to the process-global logger.
+	err := runLocked(func() error { return mfa.fetchManifestOperation(ctx) })
+	require.ErrorIs(t, err, errHTTPStatus)
+	assert.EqualError(t, err, "failed to fetch manifest: HTTP 404")
+}
+
+func TestFetchFileHTTPStatusMessage(t *testing.T) {
+	t.Parallel()
+
+	server := httptest.NewServer(
+		http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+			w.WriteHeader(http.StatusInternalServerError)
+		}))
+	defer server.Close()
+
+	err := downloadFile(context.Background(), server.URL+"/x", "x",
+		&mfer.MFFilePath{}, nil)
+	require.ErrorIs(t, err, errHTTPStatus)
+	assert.EqualError(t, err, "HTTP 500")
+}
+
+func TestURLRequiredMessage(t *testing.T) {
+	t.Parallel()
+
+	set := flag.NewFlagSet("fetch", flag.ContinueOnError)
+	require.NoError(t, set.Parse([]string{}))
+
+	mfa := &CLIApp{Fs: afero.NewMemMapFs()}
+	ctx := urfcli.NewContext(nil, set, nil)
+
+	// fetchManifestOperation logs to the process-global logger.
+	err := runLocked(func() error { return mfa.fetchManifestOperation(ctx) })
+	require.ErrorIs(t, err, errURLRequired)
+	assert.EqualError(t, err, "URL argument required")
+}
+
+func TestSanitizePathMessages(t *testing.T) {
+	t.Parallel()
+
+	t.Run("empty", func(t *testing.T) {
+		t.Parallel()
+
+		_, err := sanitizePath("")
+		require.ErrorIs(t, err, errEmptyPath)
+		assert.EqualError(t, err, "empty path")
+	})
+
+	t.Run("absolute", func(t *testing.T) {
+		t.Parallel()
+
+		_, err := sanitizePath("/etc/passwd")
+		require.ErrorIs(t, err, errAbsolutePath)
+		assert.EqualError(t, err, "absolute path not allowed: /etc/passwd")
+	})
+
+	t.Run("traversal", func(t *testing.T) {
+		t.Parallel()
+
+		_, err := sanitizePath("../x")
+		require.ErrorIs(t, err, errPathTraversal)
+		assert.EqualError(t, err, "path traversal not allowed: ../x")
 	})
 }
 
-// TestFetchErrorMessagesVerbatim pins the fetch and manifest-loader
+func TestSizeMismatchMessage(t *testing.T) {
-// messages; see TestErrorMessagesVerbatim for why.
-func TestFetchErrorMessagesVerbatim(t *testing.T) {
 	t.Parallel()
 
-	checkErrMsgCases(t, []errMsgCase{
+	// finishDownload returns the size-mismatch error before it touches the
-		{
+	// paths, digest, or entry, so those can be zero here.
-			name: "manifest_loader: http status",
+	err := finishDownload("", "", 9, 10, nil, nil, nil, nil)
-			err: fmt.Errorf("failed to fetch %s: %w %d",
+	require.ErrorIs(t, err, errSizeMismatch)
-				"https://example.com/index.mf", errHTTPStatus, 404),
+	assert.EqualError(t, err, "size mismatch: expected 10 bytes, got 9")
-			want: "failed to fetch https://example.com/index.mf: HTTP 404",
-		},
-		{
-			name: "fetch: manifest http status",
-			err: fmt.Errorf("failed to fetch manifest: %w %d",
-				errHTTPStatus, 404),
-			want: "failed to fetch manifest: HTTP 404",
-		},
-		{
-			name: "fetch: file http status",
-			err:  fmt.Errorf("%w %d", errHTTPStatus, 500),
-			want: "HTTP 500",
-		},
-		{
-			name: "fetch: empty path",
-			err:  errEmptyPath,
-			want: "empty path",
-		},
-		{
-			name: "fetch: absolute path",
-			err:  fmt.Errorf("%w: %s", errAbsolutePath, "/etc/passwd"),
-			want: "absolute path not allowed: /etc/passwd",
-		},
-		{
-			name: "fetch: path traversal",
-			err:  fmt.Errorf("%w: %s", errPathTraversal, "../x"),
-			want: "path traversal not allowed: ../x",
-		},
-		{
-			name: "fetch: size mismatch",
-			err: fmt.Errorf("%w: expected %d bytes, got %d",
-				errSizeMismatch, 10, 9),
-			want: "size mismatch: expected 10 bytes, got 9",
-		},
-		{
-			name: "fetch: url required",
-			err:  errURLRequired,
-			want: "URL argument required",
-		},
-		{
-			name: "fetch: hash mismatch",
-			err:  errHashMismatch,
-			want: "hash mismatch",
-		},
-	})
 }
 
-// TestSentinelsAreMatchable checks that the wrapped forms of the
+func TestHashMismatchMessage(t *testing.T) {
-// messages above remain matchable with errors.Is, which is the reason
-// the sentinels exist at all.
-func TestSentinelsAreMatchable(t *testing.T) {
 	t.Parallel()
 
-	wrapped := fmt.Errorf("embedded signing key fingerprint %s %w %s",
+	// A 32-byte digest that matches none of the (empty) manifest hashes.
-		"a", errSignerMismatch, "b")
+	err := verifyDownloadedHash(make([]byte, 32), &mfer.MFFilePath{})
-	require.ErrorIs(t, wrapped, errSignerMismatch)
+	require.ErrorIs(t, err, errHashMismatch)
-
+	require.NotErrorIs(t, err, errSizeMismatch)
-	wrapped = fmt.Errorf("output file %s %w", "index.mf", errOutputExists)
+	assert.EqualError(t, err, "hash mismatch")
-	require.ErrorIs(t, wrapped, errOutputExists)
-
-	wrapped = fmt.Errorf("failed to fetch manifest: %w %d", errHTTPStatus, 404)
-	require.ErrorIs(t, wrapped, errHTTPStatus)
-
-	assert.NotErrorIs(t, errHashMismatch, errSizeMismatch)
 }

internal/cli/fetch.go

@@ -36,6 +36,11 @@ const (
 	// traversal bit for group and other must stay set.
 	dirPerms os.FileMode = 0o755
 
+	// filePerms is the permission mode, before the umask, for downloaded
+	// files. It is the mode os.Create uses; like dirPerms, it keeps group
+	// and other read access.
+	filePerms os.FileMode = 0o666
+
 	// Bitrate unit thresholds in bits per second.
 	bpsPerGbps = 1e9
 	bpsPerMbps = 1e6
@@ -53,6 +58,9 @@ var (
 	// errPathTraversal indicates a manifest path escaping the target
 	// directory.
 	errPathTraversal = errors.New("path traversal not allowed")
+	// errSymlinkInPath indicates a manifest path running through a
+	// symlink that already exists in the target directory.
+	errSymlinkInPath = errors.New("symlink in path not allowed")
 	// errSizeMismatch indicates a downloaded file with an unexpected
 	// size.
 	errSizeMismatch = errors.New("size mismatch")
@@ -274,6 +282,35 @@ func sanitizePath(p string) (string, error) {
 	return cleaned, nil
 }
 
+// checkNoSymlinks returns an error if any part of the relative path p
+// already exists as a symlink. sanitizePath checks p only as text, so
+// without this a symlink inside the target directory could send a write
+// to p outside of it. Parts that do not exist yet are fine: fetch creates
+// them as plain directories and files. Call it immediately before each
+// write: a symlink created after it returns is not caught.
+func checkNoSymlinks(p string) error {
+	current := ""
+
+	for _, part := range strings.Split(p, string(filepath.Separator)) {
+		current = filepath.Join(current, part)
+
+		info, err := os.Lstat(current)
+		if errors.Is(err, os.ErrNotExist) {
+			return nil
+		}
+
+		if err != nil {
+			return fmt.Errorf("failed to check %s for a symlink: %w", current, err)
+		}
+
+		if info.Mode()&os.ModeSymlink != 0 {
+			return fmt.Errorf("%w: %s", errSymlinkInPath, current)
+		}
+	}
+
+	return nil
+}
+
 // resolveManifestURL takes a URL and returns the manifest URL.
 // If the URL already ends with .mf, it's returned as-is.
 // Otherwise, index.mf is appended.
@@ -419,7 +456,12 @@ func downloadFile(
 	// Create parent directories if needed
 	dir := filepath.Dir(localPath)
 	if dir != "" && dir != "." {
-		err := os.MkdirAll(dir, dirPerms)
+		err = checkNoSymlinks(dir)
+		if err != nil {
+			return err
+		}
+
+		err = os.MkdirAll(dir, dirPerms)
 		if err != nil {
 			return fmt.Errorf("failed to create directory %s: %w", dir, err)
 		}
@@ -447,15 +489,25 @@ func downloadFile(
 		totalBytes = expectedSize
 	}
 
-	// Create temp file.
+	err = checkNoSymlinks(tmpPath)
+	if err != nil {
+		return err
+	}
+
+	// Remove whatever is at tmpPath, such as a leftover from an
+	// interrupted run, rather than write into it: it may be a hard link
+	// to a file outside the target directory, and removing a hard link
+	// removes only this name. If the removal fails, O_EXCL below makes
+	// the create fail.
+	_ = os.Remove(tmpPath)
+
+	// Create the temp file only if nothing is at tmpPath (O_EXCL).
 	//
-	// G304: tmpPath is derived from localPath, which sanitizePath above
+	// G304: tmpPath is a relative path that sanitizePath keeps inside the
-	// constrains lexically to a relative path that does not escape the
+	// target directory as text, and checkNoSymlinks just found no symlink
-	// destination directory. That is a purely lexical guarantee: it does
+	// in it.
-	// not resolve symlinks, so a pre-existing symlink inside the
+	out, err := os.OpenFile( //nolint:gosec // G304: see comment above
-	// destination tree can still redirect this write outside of it
+		tmpPath, os.O_RDWR|os.O_CREATE|os.O_EXCL, filePerms)
-	// (tracked in issue #86).
-	out, err := os.Create(tmpPath) //nolint:gosec // G304: see comment above
 	if err != nil {
 		return fmt.Errorf("failed to create temp file: %w", err)
 	}
@@ -519,6 +571,11 @@ func finishDownload(
 		return err
 	}
 
+	err = checkNoSymlinks(localPath)
+	if err != nil {
+		return err
+	}
+
 	// Rename temp file to final path
 	err = os.Rename(tmpPath, localPath)
 	if err != nil {

internal/cli/fetch_test.go

@@ -440,3 +440,86 @@ func TestFetchProgress(t *testing.T) {
 	require.NoError(t, err)
 	assert.Equal(t, content, downloaded)
 }
+
+// TestFetchRefusesSymlinks runs fetch into a destination directory that
+// holds a symlink pointing outside it, in each of the three places fetch
+// writes: a parent directory, the temp file, and the file itself, which
+// the temp file is renamed onto; and once as a directory inside a plain
+// directory. The fetch must fail and nothing outside may change.
+//
+//nolint:paralleltest // changes the process-global working directory
+func TestFetchRefusesSymlinks(t *testing.T) {
+	tests := []struct {
+		name   string
+		entry  string // the manifest's only file
+		link   string // symlink placed in the destination directory
+		target string // what link points to, relative to the outside directory
+	}{
+		{"parent directory", "sub/deeper/file.txt", "sub", "."},
+		{"directory inside a plain directory", "docs/data/passwd", "docs/data", "."},
+		{"temp file", testFileTxt, ".file.txt.tmp", "new.txt"},
+		{"file", testFileTxt, testFileTxt, "new.txt"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			content := []byte("fetched")
+			sourceFs := afero.NewMemMapFs()
+			require.NoError(t, sourceFs.MkdirAll(filepath.Dir("/"+tt.entry), 0o755))
+			require.NoError(t, afero.WriteFile(sourceFs, "/"+tt.entry, content, 0o644))
+
+			server := httptest.NewServer(fetchTestHandler(
+				scanToManifest(t, sourceFs), map[string][]byte{tt.entry: content}))
+			defer server.Close()
+
+			outside := t.TempDir()
+
+			chdirTemp(t)
+			require.NoError(t, os.MkdirAll(filepath.Dir(tt.link), 0o750))
+			require.NoError(t, os.Symlink(filepath.Join(outside, tt.target), tt.link))
+
+			opts := testOpts([]string{testApp, cmdFetch, "-q", server.URL}, afero.NewOsFs())
+			assert.Equal(t, 1, runCLI(opts))
+			assert.Contains(t, testStderr(t, opts), "failed to download "+tt.entry+
+				": symlink in path not allowed: "+tt.link)
+
+			written, err := os.ReadDir(outside)
+			require.NoError(t, err)
+			assert.Empty(t, written, "fetch wrote outside the destination")
+		})
+	}
+}
+
+// TestFetchReplacesHardLinkAtTempName runs fetch into a destination
+// directory that holds, at the temp file's name, a hard link to a file
+// outside it. To fetch that is an ordinary leftover from an interrupted
+// earlier run: it must replace it and succeed, and the outside file must
+// not change.
+//
+//nolint:paralleltest // changes the process-global working directory
+func TestFetchReplacesHardLinkAtTempName(t *testing.T) {
+	content := []byte("fetched")
+	sourceFs := afero.NewMemMapFs()
+	require.NoError(t, afero.WriteFile(sourceFs, "/"+testFileTxt, content, 0o644))
+
+	server := httptest.NewServer(fetchTestHandler(
+		scanToManifest(t, sourceFs), map[string][]byte{testFileTxt: content}))
+	defer server.Close()
+
+	outsideFile := filepath.Join(t.TempDir(), "secret.txt")
+	require.NoError(t, os.WriteFile(outsideFile, []byte("outside"), 0o600))
+
+	chdirTemp(t)
+	require.NoError(t, os.Link(outsideFile, ".file.txt.tmp"))
+
+	opts := testOpts([]string{testApp, cmdFetch, "-q", server.URL}, afero.NewOsFs())
+	require.Equal(t, 0, runCLI(opts), testStderr(t, opts))
+
+	fetched, err := os.ReadFile(testFileTxt)
+	require.NoError(t, err)
+	assert.Equal(t, content, fetched)
+
+	outside, err := os.ReadFile(outsideFile) //nolint:gosec // test-controlled path
+	require.NoError(t, err)
+	assert.Equal(t, "outside", string(outside), "fetch wrote outside the destination")
+}

internal/cli/freshen_test.go

@@ -110,7 +110,10 @@ func TestFreshenRecordEntryMtimePresence(t *testing.T) {
 
 	const relPath = "file1.txt"
 
-	mtime := time.Unix(1_700_000_000, 0)
+	// The scanned file's mtime is the Unix epoch. If recordEntry ever misreads
+	// an absent manifest mtime as the epoch, the "absent" case below would
+	// compare equal to this and be classified unchanged, so the test fails.
+	mtime := time.Unix(0, 0)
 	info := stubFileInfo{size: 8, mtime: mtime}
 
 	for _, tc := range []struct {

internal/cli/mfer.go

@@ -18,6 +18,7 @@ const (
 	cmdGenerate = "generate"
 	cmdCheck    = "check"
 	cmdExport   = "export"
+	cmdFetch    = "fetch"
 
 	flagProgress = "progress"
 
@@ -300,7 +301,7 @@ func (mfa *CLIApp) listCommand() *cli.Command {
 
 func (mfa *CLIApp) fetchCommand() *cli.Command {
 	return &cli.Command{
-		Name:  "fetch",
+		Name:  cmdFetch,
 		Usage: "fetch manifest and referenced files",
 		Action: func(c *cli.Context) error {
 			mfa.setVerbosity(c)
@@ -357,6 +358,6 @@ func (mfa *CLIApp) run(args []string) {
 	if err != nil {
 		mfa.exitCode = 1
 
-		log.WithError(err).Debugf("exiting")
+		log.Errorf("%s", err)
 	}
 }

internal/log/log.go

@@ -112,13 +112,16 @@ func DisableStyling() {
 }
 
 // Init initializes the logger with the CLI handler and default log level.
+//
+// It reconfigures the process-global apex/log logger under the write lock so
+// the global is never mutated while another goroutine holds the read lock to
+// read it in emit. Without this, parallel callers (e.g. the test suite) race
+// Init's SetLevel/SetHandler against concurrent log calls.
 func Init() {
-	mu.RLock()
+	mu.Lock()
+	defer mu.Unlock()
 
-	w := stderr
+	log.SetHandler(acli.New(stderr))
-
-	mu.RUnlock()
-	log.SetHandler(acli.New(w))
 	log.SetLevel(log.DebugLevel) // Let apex/log pass everything; we filter ourselves
 }
 
@@ -130,74 +133,66 @@ func isEnabled(l Level) bool {
 	return l >= currentLevel
 }
 
+// emit calls fn while holding the read lock if messages at level l are
+// enabled. Holding the read lock across the apex/log call keeps the global
+// logger from being read while Init reconfigures it under the write lock.
+func emit(l Level, fn func()) {
+	mu.RLock()
+	defer mu.RUnlock()
+
+	if l >= currentLevel {
+		fn()
+	}
+}
+
 // Fatalf logs a formatted message at fatal level.
 func Fatalf(format string, args ...any) {
-	if isEnabled(FatalLevel) {
+	emit(FatalLevel, func() { log.Fatalf(format, args...) })
-		log.Fatalf(format, args...)
-	}
 }
 
 // Fatal logs a message at fatal level.
 func Fatal(arg string) {
-	if isEnabled(FatalLevel) {
+	emit(FatalLevel, func() { log.Fatal(arg) })
-		log.Fatal(arg)
-	}
 }
 
 // Errorf logs a formatted message at error level.
 func Errorf(format string, args ...any) {
-	if isEnabled(ErrorLevel) {
+	emit(ErrorLevel, func() { log.Errorf(format, args...) })
-		log.Errorf(format, args...)
-	}
 }
 
 // Error logs a message at error level.
 func Error(arg string) {
-	if isEnabled(ErrorLevel) {
+	emit(ErrorLevel, func() { log.Error(arg) })
-		log.Error(arg)
-	}
 }
 
 // Warnf logs a formatted message at warn level.
 func Warnf(format string, args ...any) {
-	if isEnabled(WarnLevel) {
+	emit(WarnLevel, func() { log.Warnf(format, args...) })
-		log.Warnf(format, args...)
-	}
 }
 
 // Warn logs a message at warn level.
 func Warn(arg string) {
-	if isEnabled(WarnLevel) {
+	emit(WarnLevel, func() { log.Warn(arg) })
-		log.Warn(arg)
-	}
 }
 
 // Infof logs a formatted message at info level.
 func Infof(format string, args ...any) {
-	if isEnabled(InfoLevel) {
+	emit(InfoLevel, func() { log.Infof(format, args...) })
-		log.Infof(format, args...)
-	}
 }
 
 // Info logs a message at info level.
 func Info(arg string) {
-	if isEnabled(InfoLevel) {
+	emit(InfoLevel, func() { log.Info(arg) })
-		log.Info(arg)
-	}
 }
 
 // Verbosef logs a formatted message at verbose level.
 func Verbosef(format string, args ...any) {
-	if isEnabled(VerboseLevel) {
+	emit(VerboseLevel, func() { log.Infof(format, args...) })
-		log.Infof(format, args...)
-	}
 }
 
 // Verbose logs a message at verbose level.
 func Verbose(arg string) {
-	if isEnabled(VerboseLevel) {
+	emit(VerboseLevel, func() { log.Info(arg) })
-		log.Info(arg)
-	}
 }
 
 // Debugf logs a formatted message at debug level with caller location.
@@ -216,7 +211,10 @@ func Debug(arg string) {
 
 // DebugReal logs at debug level with caller info from the specified stack depth.
 func DebugReal(arg string, cs int) {
-	if !isEnabled(DebugLevel) {
+	mu.RLock()
+	defer mu.RUnlock()
+
+	if DebugLevel < currentLevel {
 		return
 	}
 
@@ -275,11 +273,6 @@ func GetLevel() Level {
 	return currentLevel
 }
 
-// WithError returns a log entry with the error attached.
-func WithError(e error) *log.Entry {
-	return log.Log.WithError(e)
-}
-
 // Progressf prints a progress message that overwrites the current line.
 // Use ProgressDone() when progress is complete to move to the next line.
 func Progressf(format string, args ...any) {

mfer/checker.go

@@ -312,6 +312,10 @@ func (c *Checker) FindExtraFiles(ctx context.Context, results chan<- Result) err
 }
 
 func (c *Checker) checkFile(entry *MFFilePath, checkedBytes *FileSize) Result {
+	// entry.GetPath() is safe to join here: a manifest's entry paths are
+	// validated against the path invariants when it is loaded (see
+	// deserializeInner) or built (see Builder.AddFile), so a traversal or
+	// absolute path can never reach this point.
 	absPath := filepath.Join(string(c.basePath), entry.GetPath())
 	relPath := RelFilePath(entry.GetPath())
 

mfer/deserialize.go

@@ -25,6 +25,7 @@ var (
 	errDecompressedTooLarge = errors.New("decompressed data exceeds maximum allowed size")
 	errUUIDMismatch         = errors.New("outer and inner UUID mismatch")
 	errInvalidFileFormat    = errors.New("invalid file format")
+	errInvalidManifestPath  = errors.New("manifest contains invalid path")
 )
 
 // validateUUID checks that the byte slice is a valid UUID (16 bytes, parseable).
@@ -110,7 +111,15 @@ func (m *manifest) verifyOuterIntegrity() error {
 func (m *manifest) decompressInner() ([]byte, error) {
 	bb := bytes.NewBuffer(m.pbOuter.GetInnerMessage())
 
-	zr, err := zstd.NewReader(bb)
+	// By default the decoder decodes a payload under 128 KiB in full,
+	// each frame up to the decoder's limit, before the LimitReader below
+	// reads any of it. Decoding synchronously and never in full makes it decode
+	// only what the LimitReader asks for. Its limit caps the window that
+	// a frame header can make it set aside.
+	zr, err := zstd.NewReader(bb,
+		zstd.WithDecoderConcurrency(1),
+		zstd.WithDecodeBuffersBelow(0),
+		zstd.WithDecoderMaxMemory(uint64(MaxDecompressedSize)))
 	if err != nil {
 		return nil, fmt.Errorf("deserialize: zstd reader: %w", err)
 	}
@@ -181,6 +190,19 @@ func (m *manifest) deserializeInner() error {
 		return errUUIDMismatch
 	}
 
+	// Enforce the manifest path invariants on every entry as it is loaded,
+	// so that no consumer of a manifest — Checker today, any restore or
+	// extract path tomorrow — acts on a traversal or absolute path from an
+	// untrusted .mf. Reject loudly on the first offender rather than
+	// dropping entries, which would let a hostile manifest hide files from a
+	// check.
+	for _, f := range m.pbInner.GetFiles() {
+		err = ValidatePath(f.GetPath())
+		if err != nil {
+			return fmt.Errorf("%w: %w", errInvalidManifestPath, err)
+		}
+	}
+
 	log.Infof("loaded manifest with %d files", len(m.pbInner.GetFiles()))
 
 	return nil

mfer/deserialize_fuzz_test.go

@@ -0,0 +1,75 @@
+//nolint:testpackage // white-box tests exercise unexported internals
+package mfer
+
+import (
+	"bytes"
+	"runtime"
+	"testing"
+
+	"google.golang.org/protobuf/proto"
+)
+
+// FuzzNewManifestFromReader feeds arbitrary bytes to the manifest parser.
+// `make test` runs it on the seed corpus in
+// testdata/fuzz/FuzzNewManifestFromReader; `make fuzz` searches for new
+// inputs.
+//
+// For every input the parser must return a manifest or an error, not both
+// and not neither, and must not allocate more than a fixed multiple of its
+// input and of the decompressed data it may read, plus room for the
+// decoder's window. A panic or a hang fails the test on its own.
+func FuzzNewManifestFromReader(f *testing.F) {
+	// A signed manifest makes the parser write the key and signature to a
+	// temporary directory and run gpg on them. With gpg off the PATH and
+	// temporary files kept in the test's own directory, no process is
+	// started and nothing is written elsewhere; such input ends in an
+	// error instead.
+	f.Setenv("PATH", "")
+	f.Setenv("TMPDIR", f.TempDir())
+
+	f.Fuzz(func(t *testing.T, data []byte) {
+		var before, after runtime.MemStats
+
+		runtime.ReadMemStats(&before)
+
+		m, err := NewManifestFromReader(bytes.NewReader(data))
+
+		runtime.ReadMemStats(&after)
+
+		if (m == nil) == (err == nil) {
+			t.Fatalf("got manifest %p and error %v, want exactly one", m, err)
+		}
+
+		// The parser reads at most the declared size plus one byte of
+		// decompressed data, and never more than MaxDecompressedSize.
+		decompressed := uint64(MaxDecompressedSize)
+
+		outer := new(MFFileOuter)
+		if validateMagic(data) &&
+			proto.Unmarshal(data[len(MAGIC):], outer) == nil {
+			size := outer.GetSize()
+			if size > 0 && size < MaxDecompressedSize {
+				decompressed = uint64(size) + 1
+			}
+		}
+
+		// It also keeps a few copies of its input, and the decoder sets
+		// aside a window that a frame header sizes, up to a little over
+		// MaxDecompressedSize. Buffers grow by copying, so reaching those
+		// sizes allocates a few times them in total: sixteen times the
+		// input and the decompressed data, plus twice MaxDecompressedSize
+		// for the window, leaves room for that. The seed whose frame
+		// claims 8 GiB fails if the decoder sets that size aside; the seed
+		// whose two frames together exceed MaxDecompressedSize fails if
+		// the decoder decodes them in full instead of stopping at the
+		// declared size.
+		limit := 16*(uint64(len(data))+decompressed) +
+			2*uint64(MaxDecompressedSize)
+
+		allocated := after.TotalAlloc - before.TotalAlloc
+		if allocated > limit {
+			t.Fatalf("allocated %d bytes for %d bytes of input, limit %d",
+				allocated, len(data), limit)
+		}
+	})
+}

mfer/deserialize_path_test.go

@@ -0,0 +1,145 @@
+//nolint:testpackage // white-box tests exercise unexported internals
+package mfer
+
+import (
+	"bytes"
+	"crypto/sha256"
+	"fmt"
+	"testing"
+
+	"github.com/google/uuid"
+	"github.com/klauspost/compress/zstd"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"google.golang.org/protobuf/encoding/protowire"
+	"google.golang.org/protobuf/proto"
+)
+
+// craftInnerBytes builds the wire bytes of an inner MFFile holding a single
+// file entry whose path is exactly pathBytes. It writes the wire form by hand
+// so a hostile path — including one that is not valid UTF-8 — can be embedded
+// without proto.Marshal's own UTF-8 enforcement rejecting it first.
+func craftInnerBytes(id uuid.UUID, pathBytes string) []byte {
+	entry := protowire.AppendTag(nil, 1, protowire.BytesType) // MFFilePath.path
+	entry = protowire.AppendString(entry, pathBytes)
+
+	inner := protowire.AppendTag(nil, 100, protowire.VarintType) // MFFile.version
+	inner = protowire.AppendVarint(inner, uint64(MFFile_VERSION_ONE))
+	inner = protowire.AppendTag(inner, 101, protowire.BytesType) // MFFile.files
+	inner = protowire.AppendBytes(inner, entry)
+	inner = protowire.AppendTag(inner, 102, protowire.BytesType) // MFFile.uuid
+	inner = protowire.AppendBytes(inner, id[:])
+
+	return inner
+}
+
+// wrapInner wraps inner MFFile wire bytes in a complete, well-formed .mf
+// envelope (magic prefix, zstd-compressed payload, matching hash and UUID) so
+// that deserialization reaches path validation rather than failing earlier on
+// an integrity check.
+func wrapInner(t *testing.T, id uuid.UUID, innerData []byte) []byte {
+	t.Helper()
+
+	var cbuf bytes.Buffer
+
+	zw, err := zstd.NewWriter(&cbuf, zstd.WithEncoderLevel(zstd.SpeedBestCompression))
+	require.NoError(t, err)
+
+	_, err = zw.Write(innerData)
+	require.NoError(t, err)
+	require.NoError(t, zw.Close())
+
+	compressed := cbuf.Bytes()
+	sum := sha256.Sum256(compressed)
+
+	outer := &MFFileOuter{
+		InnerMessage:    compressed,
+		Size:            int64(len(innerData)),
+		Sha256:          sum[:],
+		Uuid:            id[:],
+		Version:         MFFileOuter_VERSION_ONE,
+		CompressionType: MFFileOuter_COMPRESSION_ZSTD,
+	}
+
+	ob, err := proto.Marshal(outer)
+	require.NoError(t, err)
+
+	return append([]byte(MAGIC), ob...)
+}
+
+func TestDeserializeRejectsInvalidEntryPaths(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		path string
+	}{
+		{"parent traversal", "../escape"},
+		{"interior traversal", "a/../../escape"},
+		{"absolute path", "/etc/passwd"},
+		{"backslash path", `a\b`},
+		{"double slash", "a//b"},
+		{"empty path", ""},
+		{"invalid utf-8", "abc\xff"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			id := uuid.New()
+			data := wrapInner(t, id, craftInnerBytes(id, tt.path))
+
+			_, err := NewManifestFromReader(bytes.NewReader(data))
+			require.Error(t, err)
+
+			if tt.path == "abc\xff" {
+				// A path that is not valid UTF-8 cannot survive the proto3
+				// string decoder, which rejects it before path validation
+				// runs; the manifest is still refused at load time.
+				return
+			}
+
+			require.ErrorIs(t, err, errInvalidManifestPath)
+
+			if tt.path != "" {
+				// ValidatePath quotes the path with %q; assert against the
+				// same rendering so escaped characters (e.g. a backslash)
+				// still match.
+				assert.Contains(t, err.Error(), fmt.Sprintf("%q", tt.path),
+					"error must name the offending path")
+			}
+		})
+	}
+}
+
+func TestDeserializeValidManifestRoundTrips(t *testing.T) {
+	t.Parallel()
+
+	hash := make([]byte, 34) // multihash: 2-byte prefix + 32-byte SHA-256
+
+	b := NewBuilder()
+	require.NoError(t, b.AddFileWithHash("dir/file.txt", 123, ModTime{}, hash))
+
+	var buf bytes.Buffer
+	require.NoError(t, b.Build(&buf))
+
+	m, err := NewManifestFromReader(bytes.NewReader(buf.Bytes()))
+	require.NoError(t, err)
+
+	files := m.Files()
+	require.Len(t, files, 1)
+	assert.Equal(t, "dir/file.txt", files[0].GetPath())
+	assert.Equal(t, int64(123), files[0].GetSize())
+}
+
+// TestValidatePathRejectsInvalidUTF8 pins the ValidatePath rule that a manifest
+// path must be valid UTF-8, independent of the proto decoder that also enforces
+// it on the wire.
+func TestValidatePathRejectsInvalidUTF8(t *testing.T) {
+	t.Parallel()
+
+	err := ValidatePath("abc\xff")
+	require.ErrorIs(t, err, errPathNotUTF8)
+	assert.Contains(t, err.Error(), "UTF-8")
+}

mfer/testdata/fuzz/FuzzNewManifestFromReader/empty

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("")

mfer/testdata/fuzz/FuzzNewManifestFromReader/magic-then-garbage

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFGy[i\x04\xe5O\x82A\x1d\xf4\xb0\xe2z7:U\xee\xa3\xf9\xd6m\xacZ\x9b\xce\x1d\xd9/{@\x1d\xa5y[i\x04\xe5O\x82A\x1d\xf4\xb0\xe2z7:U\xee\xa3\xf9\xd6m\xacZ\x9b\xce\x1d\xd9/{@\x1d\xa5")

mfer/testdata/fuzz/FuzzNewManifestFromReader/minimal

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06V\xc2\x06 \xa3\xf7\x97\xa7\xf3\x87:\x90)\\ӊj\xb9\xf7\xfaTJ\x1b\xe7:\xee\xbe\"V\xe0:\x8d(Z\xd6%\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\fc(\xb5/\xfd\x04\x00\xb1\x02\x00\xa0\x06\x01\xaa\x06=\n\x05a.txt\x10\x01\x1a$\n\"\x12 ʗ\x81\x12\xca\x1b\xbd\xca\xfa\xc21\xb3\x9a#\xdcM\xa7\x86\xef\xf8\x14|Nr\xb9\x80w\x85\xaf\xeeH\xbb\xf2\x12\v\b\x80\x92\xb8Ø\xfe\xff\xff\xff\x01\xb2\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3s\xeeG\x80")

mfer/testdata/fuzz/FuzzNewManifestFromReader/signed

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06V\xc2\x06 \xa3\xf7\x97\xa7\xf3\x87:\x90)\\ӊj\xb9\xf7\xfaTJ\x1b\xe7:\xee\xbe\"V\xe0:\x8d(Z\xd6%\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\fc(\xb5/\xfd\x04\x00\xb1\x02\x00\xa0\x06\x01\xaa\x06=\n\x05a.txt\x10\x01\x1a$\n\"\x12 ʗ\x81\x12\xca\x1b\xbd\xca\xfa\xc21\xb3\x9a#\xdcM\xa7\x86\xef\xf8\x14|Nr\xb9\x80w\x85\xaf\xeeH\xbb\xf2\x12\v\b\x80\x92\xb8Ø\xfe\xff\xff\xff\x01\xb2\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3s\xeeG\x80\xca\f\xe8\x03-----BEGIN PGP SIGNATURE-----\n\niQEzBAABCgAdFiEET1Yr+4Y/3GtRtO6IhypRF2zvI64FAmrBIXAACgkQhypRF2zv\nI67BQAf/QrpX2MjY15YGMGkjR5oIhnx/YV96aGYZyZThzb+l/R/N75iVFVkhX21d\nZhQqdCsORrodTPAXic2g2UGVXP9PhNMh7n6Wm3LsvQYjrRQGrQnqtCkut+3tUt8K\n7pt4OAnnwRSieaVImA1COmzxIrQQKNOs6UkgmAstGuPV0XZoeDiSG8TUYJ/vieCn\np5hC0FFXtzfw4NtkxSmkewE0xBxIwFCA/RfSHCGH3m5K+tRz41vMEgGbL1iEp6+V\nuBaoEc4hqCgEt+Af2pA8VHfqeu2vKiwggOpYpaILXZKVqH9+tWHL1EBv9t0vTsYE\n9D57euuR9+kOdngYNPieP1yn5dOSHg==\n=kvgL\n-----END PGP SIGNATURE-----\n\xd2\f(4F562BFB863FDC6B51B4EE88872A51176CEF23AE\xda\f\xb5\a-----BEGIN PGP PUBLIC KEY BLOCK-----\n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\nKoTjp4nxgQGy5WmeBzs4Hdm/\n=TCB8\n-----END PGP PUBLIC KEY BLOCK-----\n")

mfer/testdata/fuzz/FuzzNewManifestFromReader/size-above-payload

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06!\xc2\x06 \x91\x90*\xa5>\fݐ \x87\xbeaL\xc1\x05?\x0eR\xc18\xa4eՕ\xa95\xb9KʺoZ\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\f-(\xb5/\xfd\x04\x00\x01\x01\x00\xa0\x06\x01\xaa\x06\a\n\x05a.txt\xb2\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3a[k'")

mfer/testdata/fuzz/FuzzNewManifestFromReader/size-below-payload

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06\x1f\xc2\x06 \x91\x90*\xa5>\fݐ \x87\xbeaL\xc1\x05?\x0eR\xc18\xa4eՕ\xa95\xb9KʺoZ\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\f-(\xb5/\xfd\x04\x00\x01\x01\x00\xa0\x06\x01\xaa\x06\a\n\x05a.txt\xb2\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3a[k'")

mfer/testdata/fuzz/FuzzNewManifestFromReader/truncated-at-172

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06V\xc2\x06 \xa3\xf7\x97\xa7\xf3\x87:\x90)\\ӊj\xb9\xf7\xfaTJ\x1b\xe7:\xee\xbe\"V\xe0:\x8d(Z\xd6%\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\fc(\xb5/\xfd\x04\x00\xb1\x02\x00\xa0\x06\x01\xaa\x06=\n\x05a.txt\x10\x01\x1a$\n\"\x12 ʗ\x81\x12\xca\x1b\xbd\xca\xfa\xc21\xb3\x9a#\xdcM\xa7\x86\xef\xf8\x14|Nr\xb9\x80w\x85\xaf\xeeH\xbb\xf2\x12\v\b\x80\x92\xb8Ø\xfe\xff\xff\xff\x01\xb2\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3s\xeeG")

mfer/testdata/fuzz/FuzzNewManifestFromReader/truncated-at-20

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06V\xc2\x06 ")

mfer/testdata/fuzz/FuzzNewManifestFromReader/truncated-at-4

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAV")

mfer/testdata/fuzz/FuzzNewManifestFromReader/truncated-at-8

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG")

mfer/testdata/fuzz/FuzzNewManifestFromReader/truncated-at-86

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06V\xc2\x06 \xa3\xf7\x97\xa7\xf3\x87:\x90)\\ӊj\xb9\xf7\xfaTJ\x1b\xe7:\xee\xbe\"V\xe0:\x8d(Z\xd6%\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\fc(\xb5/\xfd\x04\x00\xb1\x02\x00\xa0\x06\x01")

mfer/testdata/fuzz/FuzzNewManifestFromReader/wrong-magic

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFX\xa8\x06\x01\xb0\x06\x01\xb8\x06V\xc2\x06 \xa3\xf7\x97\xa7\xf3\x87:\x90)\\ӊj\xb9\xf7\xfaTJ\x1b\xe7:\xee\xbe\"V\xe0:\x8d(Z\xd6%\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\fc(\xb5/\xfd\x04\x00\xb1\x02\x00\xa0\x06\x01\xaa\x06=\n\x05a.txt\x10\x01\x1a$\n\"\x12 ʗ\x81\x12\xca\x1b\xbd\xca\xfa\xc21\xb3\x9a#\xdcM\xa7\x86\xef\xf8\x14|Nr\xb9\x80w\x85\xaf\xeeH\xbb\xf2\x12\v\b\x80\x92\xb8Ø\xfe\xff\xff\xff\x01\xb2\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3s\xeeG\x80")

mfer/testdata/fuzz/FuzzNewManifestFromReader/zstd-frame-claims-8gib

@@ -0,0 +1,2 @@
+go test fuzz v1
+[]byte("ZNAVSRFG\xa8\x06\x01\xb0\x06\x01\xb8\x06\x01\xc2\x06 \xd6aQ\xa4\x85\xc0+\xbb\xca\x11\x11\a<\x019\x97\xb3\xbb3\xd0 \xd5U\xfa!\xaeAf<N@\x9d\xca\x06\x10\x93\x85\vpu\x85\xe4\x04\xe4\x95\x1a=\xdc\x1f\x05\xa3\xba\f\x12(\xb5/\xfd\xc0\x00\x00\x00\x00\x00\x02\x00\x00\x00\v\x00\x00\x00")

script/cibuild

@@ -1,14 +1,24 @@
 #!/bin/sh
-# script/cibuild: run the CI build. The Dockerfile runs script/check
+# script/cibuild: run the CI build; the Gitea workflow runs this on push.
-# (via make check), so a successful build implies all checks pass.
+# It builds the image with the same command as script/docker. --no-cache
-# Generic: needs no adaptation. The Gitea workflow runs this on push.
+# because the checks the final stage depends on are RUN steps, and a
+# cached one is a check that did not run.
 set -eu
 
-ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
+ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build .
+    # Own line: a failing command substitution inside an argument does
+    # not trip `set -e`, so the inline form degrades silently to an
+    # empty constant. The VERSION build argument takes precedence over
+    # the version a build stage derives from the .git in the context.
+    version="$(git describe --tags --always --dirty 2>/dev/null || true)"
+    [ -n "$version" ] || version="unknown"
+    docker build --no-cache \
+        --build-arg VERSION="$version" \
+        -t "$("$SCRIPT_DIR/projectname")" .
 }
 
 main "$@"

script/docker

@@ -1,7 +1,8 @@
 #!/bin/sh
 # script/docker: build the Docker image tagged with the project name.
 # Identical in all repos; the tag comes from script/projectname.
-# Generic: needs no adaptation.
+# --no-cache because the gate phases the final stage depends on are RUN
+# steps, and a cached one is a check that did not run.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -9,7 +10,15 @@ ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build -t "$("$SCRIPT_DIR/projectname")" .
+    # Own line: a failing command substitution inside an argument does
+    # not trip `set -e`, so the inline form degrades silently to an
+    # empty constant. The VERSION build argument takes precedence over
+    # the version a build stage derives from the .git in the context.
+    version="$(git describe --tags --always --dirty 2>/dev/null || true)"
+    [ -n "$version" ] || version="unknown"
+    docker build --no-cache \
+        --build-arg VERSION="$version" \
+        -t "$("$SCRIPT_DIR/projectname")" .
 }
 
 main "$@"

script/fuzz

@@ -0,0 +1,17 @@
+#!/bin/sh
+# script/fuzz: fuzz the manifest parser for one minute. Run by hand only:
+# script/test already runs the committed seed corpus as ordinary tests,
+# and CI never fuzzes. An input that fails is written to
+# mfer/testdata/fuzz/FuzzNewManifestFromReader/; once the parser is fixed,
+# commit it there as a regression seed.
+set -eu
+
+ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
+
+main() {
+    cd "$ROOT"
+    go test -run '^$' -fuzz '^FuzzNewManifestFromReader$' \
+        -fuzztime 1m -parallel 2 ./mfer
+}
+
+main "$@"

script/test

@@ -17,7 +17,12 @@ ensure_pb() {
 main() {
     cd "$ROOT"
     ensure_pb
-    go test -v --timeout 10s ./...
+    go test -timeout 30s -race -cover ./... ||
+        {
+            echo "--- Rerunning with -v for details ---"
+            go test -timeout 30s -race -v ./...
+            exit 1
+        }
 }
 
 main "$@"