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 plus MaxDecompressedSize. 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 kept its own 64 GiB limit, set aside whatever size a frame
header claimed, and decoded payloads under 128 KiB in full before the
LimitReader read any of it, so a 92-byte manifest claiming 8 GiB made
the parser allocate 8 GiB. The decoder now has MaxDecompressedSize as
its limit; that manifest is a regression seed.

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
-(public domain) [Go](https://golang.org) library and command-line tool by
-[@sneak](https://sneak.berlin) that specifies and generates `.mf` manifest files
-over a directory tree to encapsulate metadata about the files — such as
-cryptographic checksums and signatures over same — to aid in archiving,
-downloading, streaming, and mirroring. It was first published in 2022. The
-manifest files' data is serialized with Google's
+[mfer](https://git.eeqj.de/sneak/mfer) is a reference implementation library and
+thin wrapper command-line utility written in [Go](https://golang.org) and first
+published in 2022 under the [WTFPL](https://wtfpl.net) (public domain) license.
+It specifies and generates `.mf` manifest files over a directory tree of files
+to encapsulate metadata about them (such as cryptographic checksums or
+signatures over same) to aid in archiving, downloading, and streaming, or
+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`
-(formatting, linting, tests). The `main` branch must always be green.
+CI runs `script/cibuild`, which builds the Docker image with `--no-cache`, so
+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
-  checks)
+- `script/cibuild` — CI entrypoint: builds the image with the same command as
+  `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
-
-## The problem
+# Problem Statement
 
 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 +24,31 @@ 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
-  category in the Description first line; added Getting Started (verified
-  install/usage block), Rationale (folding in Problem Statement and Proposed
-  Solution), and a Design section for the package layout; renamed Authors to
-  Author (#75)
+- 2026-10-03: added `FuzzNewManifestFromReader` and its seed corpus, which
+  `make test` runs, plus `make fuzz` for a one-minute run by hand; the zstd
+  decoder now has `MaxDecompressedSize` as its limit, so a frame claiming a
+  large size can no longer make the parser allocate that size (#65)
+- 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 +90,6 @@ 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 +127,7 @@ 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
-	// Each manifest entry is roughly 50-60 bytes, so we need ~20000 files
+	// Create many small files with random names so the manifest has many
+	// 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,
-		"manifest should be at least 1MB, got %d bytes", len(validManifest))
+	require.GreaterOrEqual(t, len(validManifest), 64*1024,
+		"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
-	for i := range 500 {
+	// Now do 100 random corruption iterations
+	for i := range 100 {
 		// Corrupt: write a random byte at a random offset
 		corrupted := make([]byte, len(validManifest))
 		copy(corrupted, validManifest)

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
-	// constrains lexically to a relative path that does not escape the
-	// destination directory. That is a purely lexical guarantee: it does
-	// not resolve symlinks, so a pre-existing symlink inside the
-	// destination tree can still redirect this write outside of it
-	// (tracked in issue #86).
-	out, err := os.Create(tmpPath) //nolint:gosec // G304: see comment above
+	// G304: tmpPath is a relative path that sanitizePath keeps inside the
+	// target directory as text, and checkNoSymlinks just found no symlink
+	// in it.
+	out, err := os.OpenFile( //nolint:gosec // G304: see comment above
+		tmpPath, os.O_RDWR|os.O_CREATE|os.O_EXCL, filePerms)
 	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/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
-
-	mu.RUnlock()
-	log.SetHandler(acli.New(w))
+	log.SetHandler(acli.New(stderr))
 	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) {
-		log.Fatalf(format, args...)
-	}
+	emit(FatalLevel, func() { log.Fatalf(format, args...) })
 }
 
 // Fatal logs a message at fatal level.
 func Fatal(arg string) {
-	if isEnabled(FatalLevel) {
-		log.Fatal(arg)
-	}
+	emit(FatalLevel, func() { log.Fatal(arg) })
 }
 
 // Errorf logs a formatted message at error level.
 func Errorf(format string, args ...any) {
-	if isEnabled(ErrorLevel) {
-		log.Errorf(format, args...)
-	}
+	emit(ErrorLevel, func() { log.Errorf(format, args...) })
 }
 
 // Error logs a message at error level.
 func Error(arg string) {
-	if isEnabled(ErrorLevel) {
-		log.Error(arg)
-	}
+	emit(ErrorLevel, func() { log.Error(arg) })
 }
 
 // Warnf logs a formatted message at warn level.
 func Warnf(format string, args ...any) {
-	if isEnabled(WarnLevel) {
-		log.Warnf(format, args...)
-	}
+	emit(WarnLevel, func() { log.Warnf(format, args...) })
 }
 
 // Warn logs a message at warn level.
 func Warn(arg string) {
-	if isEnabled(WarnLevel) {
-		log.Warn(arg)
-	}
+	emit(WarnLevel, func() { log.Warn(arg) })
 }
 
 // Infof logs a formatted message at info level.
 func Infof(format string, args ...any) {
-	if isEnabled(InfoLevel) {
-		log.Infof(format, args...)
-	}
+	emit(InfoLevel, func() { log.Infof(format, args...) })
 }
 
 // Info logs a message at info level.
 func Info(arg string) {
-	if isEnabled(InfoLevel) {
-		log.Info(arg)
-	}
+	emit(InfoLevel, func() { log.Info(arg) })
 }
 
 // Verbosef logs a formatted message at verbose level.
 func Verbosef(format string, args ...any) {
-	if isEnabled(VerboseLevel) {
-		log.Infof(format, args...)
-	}
+	emit(VerboseLevel, func() { log.Infof(format, args...) })
 }
 
 // Verbose logs a message at verbose level.
 func Verbose(arg string) {
-	if isEnabled(VerboseLevel) {
-		log.Info(arg)
-	}
+	emit(VerboseLevel, func() { 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 +111,12 @@ func (m *manifest) verifyOuterIntegrity() error {
 func (m *manifest) decompressInner() ([]byte, error) {
 	bb := bytes.NewBuffer(m.pbOuter.GetInnerMessage())
 
-	zr, err := zstd.NewReader(bb)
+	// The decoder's own limit is 64 GiB. Up to that it sets aside as many
+	// bytes as a frame header claims, and it decodes a payload under
+	// 128 KiB in full before the LimitReader below reads any of it, so it
+	// needs MaxDecompressedSize as its limit too.
+	zr, err := zstd.NewReader(bb,
+		zstd.WithDecoderMaxMemory(uint64(MaxDecompressedSize)))
 	if err != nil {
 		return nil, fmt.Errorf("deserialize: zstd reader: %w", err)
 	}
@@ -181,6 +187,19 @@ 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 +1,54 @@
+//nolint:testpackage // white-box tests exercise unexported internals
+package mfer
+
+import (
+	"bytes"
+	"runtime"
+	"testing"
+)
+
+// 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 plus MaxDecompressedSize. 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 keeps a copy of its input and at most
+		// MaxDecompressedSize bytes of decompressed data. Buffers grow by
+		// copying, so reaching those sizes allocates a few times them in
+		// total. Sixteen times leaves room for that, and none for a
+		// decoder that sets aside whatever size its input claims.
+		limit := 16 * (uint64(len(data)) + 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
-# (via make check), so a successful build implies all checks pass.
-# Generic: needs no adaptation. The Gitea workflow runs this on push.
+# script/cibuild: run the CI build; the Gitea workflow runs this on push.
+# It builds the image with the same command as script/docker. --no-cache
+# 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 "$@"