microvms_app/control/artifact.rs
1// SPDX-License-Identifier: Apache-2.0
2//! The build artifact: a zip of the daemon binary, a Dockerfile, and — when the caller
3//! passes [`ProjectFiles`] — the project's dependency files.
4//!
5//! What `CreateMicrovmImage`'s `codeArtifact.uri` points at (`sandbox.py:523`
6//! `build_artifact`). Two entries always, both carrying a measured constraint, plus an
7//! ecosystem-named manifest/lockfile pair when the image bakes an environment layer (#74).
8//!
9//! # The execute bit has to be in the zip entry
10//!
11//! A build that copies a non-executable binary produces an image whose `CMD` fails, and
12//! the failure surfaces as a **run-hook timeout** — which says nothing about permissions,
13//! and sends the reader to look at the daemon's startup path instead of at the archive.
14//! So the `agentd` entry sets mode `0o755` explicitly rather than inheriting whatever the
15//! host file had.
16//!
17//! # The agent token is never in here (TRAP-5's other half)
18//!
19//! The artifact becomes a **shared image snapshot**: every MicroVM launched from the image
20//! sees the same bytes. A per-VM secret in there is a per-VM secret shared with every VM.
21//! So the token travels through `runHookPayload` at launch instead, and this function has
22//! no parameter that could carry one — see [`crate::control::microvm`]. The test at the
23//! bottom of this file scans the produced zip's raw bytes for a token value to prove the
24//! path stays closed, which is a byte scan rather than an API review because the leak
25//! would be a *value* appearing somewhere, not a parameter being declared.
26
27use std::io::Write as _;
28use std::time::Duration;
29
30use crate::error::{Error, ErrorKind};
31
32/// The Dockerfile entry's name, which the build looks for by convention.
33const DOCKERFILE_ENTRY: &str = "Dockerfile";
34
35/// The daemon entry's name, matching the `CMD ["/agentd"]` the Dockerfile sets.
36const AGENTD_ENTRY: &str = "agentd";
37
38/// The mode the daemon entry carries.
39///
40/// See the module docs: a non-executable binary here becomes a run-hook timeout later.
41const AGENTD_MODE: u32 = 0o755;
42
43/// An ecosystem whose dependency files may enter the build artifact (#74).
44///
45/// The variant fixes **both entry names**. That is the TRAP-5 half of the design: the
46/// artifact is a shared image snapshot, so what may enter it is an allowlist of
47/// dependency-defining files, not a caller-named list — there is no field anywhere in this
48/// module a caller could put `.env` or a credentials file's name into. The byte-scan test
49/// at the bottom of this file covers the entries this type admits.
50#[derive(Clone, Copy, Debug, Eq, PartialEq)]
51pub enum Ecosystem {
52 /// Python under uv: `pyproject.toml` + `uv.lock`.
53 Uv,
54 /// Node under npm: `package.json` + `package-lock.json`.
55 Npm,
56 /// Rust under cargo: `Cargo.toml` + `Cargo.lock`.
57 Cargo,
58}
59
60impl Ecosystem {
61 /// Every ecosystem, in detection order.
62 pub const ALL: [Self; 3] = [Self::Uv, Self::Npm, Self::Cargo];
63
64 /// The manifest's file name, which is also its zip entry name.
65 pub fn manifest_name(self) -> &'static str {
66 match self {
67 Self::Uv => "pyproject.toml",
68 Self::Npm => "package.json",
69 Self::Cargo => "Cargo.toml",
70 }
71 }
72
73 /// The lockfile's file name, which is also its zip entry name — and the file the
74 /// environment layer is keyed on (#74).
75 pub fn lockfile_name(self) -> &'static str {
76 match self {
77 Self::Uv => "uv.lock",
78 Self::Npm => "package-lock.json",
79 Self::Cargo => "Cargo.lock",
80 }
81 }
82
83 /// The whole `RUN` line that installs the environment layer: toolchain bootstrap on
84 /// the managed al2023 base, then the lockfile-faithful install.
85 ///
86 /// # The install commands are the lockfile-faithful spellings
87 ///
88 /// `uv sync --locked`, `npm ci`, and `cargo fetch --locked` each **refuse** a lockfile
89 /// that disagrees with its manifest rather than quietly re-resolving — which is what
90 /// keeps the baked layer the thing the content hash says it is. `uv sync` without
91 /// `--locked`, `npm install`, or a bare `cargo fetch` would install *something* and
92 /// the hash would then name an environment the build did not produce.
93 ///
94 /// # The bootstrap packages are measured, the full build is not yet
95 ///
96 /// The managed base is al2023-minimal and ships no toolchain, so each line installs
97 /// its own. Package names and the binaries they land were verified against the
98 /// Amazon Linux 2023 repos (2026-08-31): `python3.12-pip` puts `pip3.12` on PATH,
99 /// `nodejs22-npm` puts `npm` and `node` on PATH, `cargo` puts `cargo` on PATH.
100 /// `install_weak_deps=0` keeps the layer minimal. A full in-guest build of each line
101 /// is the live-conformance scenario #74's acceptance defers to — see the issue.
102 ///
103 /// Cargo is the one that needs a stub: `cargo fetch` refuses a package with no
104 /// targets, so the line creates an empty `src/main.rs` first. The working tree synced
105 /// at launch lands over it.
106 pub fn install_run_line(self) -> &'static str {
107 match self {
108 Self::Uv => {
109 "RUN dnf -y --setopt=install_weak_deps=0 install python3.12 python3.12-pip \
110 && dnf clean all && pip3.12 install --no-cache-dir uv && uv sync --locked"
111 }
112 Self::Npm => {
113 "RUN dnf -y --setopt=install_weak_deps=0 install nodejs22-npm \
114 && dnf clean all && npm ci"
115 }
116 Self::Cargo => {
117 "RUN dnf -y --setopt=install_weak_deps=0 install cargo \
118 && dnf clean all && mkdir -p src && touch src/main.rs \
119 && cargo fetch --locked"
120 }
121 }
122 }
123}
124
125/// A project's dependency files: the lockfile plus the manifest that owns it.
126///
127/// Bytes rather than paths, for the reason [`build_artifact`]'s `binary` is bytes: the
128/// caller owns the read, and this module keeps no filesystem behaviour to stub in a test.
129/// The entry names come from [`Ecosystem`], never from the caller — see the type's docs
130/// for why that is a TRAP-5 property rather than a convenience.
131#[derive(Clone, Debug, Eq, PartialEq)]
132pub struct ProjectFiles {
133 /// Which ecosystem's pair this is, fixing both entry names.
134 pub ecosystem: Ecosystem,
135 /// The manifest's bytes (`pyproject.toml`, `package.json`, `Cargo.toml`).
136 pub manifest: Vec<u8>,
137 /// The lockfile's bytes (`uv.lock`, `package-lock.json`, `Cargo.lock`).
138 pub lockfile: Vec<u8>,
139}
140
141/// Zips `binary` with `dockerfile` — and, when given, the project's dependency files —
142/// into the bytes `codeArtifact.uri` will point at.
143///
144/// `binary` is the daemon's bytes rather than a path, so the caller owns the read and this
145/// function has no filesystem behaviour to stub in a test. `project` adds exactly two more
146/// entries, named by its [`Ecosystem`], at the archive root beside the Dockerfile — which
147/// is the build context root, so a `COPY pyproject.toml uv.lock …` in the Dockerfile finds
148/// them.
149pub fn build_artifact(
150 binary: &[u8],
151 dockerfile: &str,
152 project: Option<&ProjectFiles>,
153) -> Result<Vec<u8>, Error> {
154 build_artifact_with_context(binary, dockerfile, project, None)
155}
156
157/// The mode every non-executable artifact entry carries.
158const FILE_MODE: u32 = 0o644;
159
160/// [`build_artifact`] plus a build context: the context's files at their relative paths
161/// beside the Dockerfile, which is the build context root, so the Dockerfile's `COPY` lines
162/// find them (IMAGE-7).
163///
164/// # One set of inputs, one object
165///
166/// Every entry carries the DOS epoch (1980-01-01) as its date and a fixed mode — `0o755` for
167/// the daemon and for context files with an execute bit, `0o644` for everything else — under
168/// Unix as the host system, and the entries are written in a fixed order: `Dockerfile`, `agentd`, the project pair, then
169/// the context sorted by name. So equal inputs produce byte-identical archives on any host
170/// at any time, and a content-addressed S3 key holds one object rather than whichever
171/// upload landed last. (Byte identity across `zip` crate versions is not promised; the image
172/// name hashes the inputs, not these bytes, for exactly that reason — see
173/// [`artifact_content_hash`].)
174///
175/// Refuses a `binary` that isn't an aarch64 ELF, empty included (#257): every artifact
176/// carries the daemon as its `CMD`, and a build finds the wrong one only as a run-hook
177/// timeout. Refuses a context entry that repeats a project file's name, since one archive
178/// cannot hold two entries of one name.
179pub fn build_artifact_with_context(
180 binary: &[u8],
181 dockerfile: &str,
182 project: Option<&ProjectFiles>,
183 context: Option<&super::context::BuildContext>,
184) -> Result<Vec<u8>, Error> {
185 use zip::write::{SimpleFileOptions, ZipWriter};
186
187 microvms_domain::provision::require_aarch64(binary)?;
188 if let (Some(project), Some(context)) = (project, context)
189 && let Some(entry) = context.entries().iter().find(|entry| {
190 entry.name == project.ecosystem.manifest_name()
191 || entry.name == project.ecosystem.lockfile_name()
192 })
193 {
194 return Err(Error::invalid_arg(format!(
195 "the build context has {:?} at its root, which the project files already carry; drop one of the two.",
196 entry.name
197 )));
198 }
199
200 let mut writer = ZipWriter::new(std::io::Cursor::new(Vec::new()));
201 // Unix as every entry's host system: left unset, zip records the host the build runs on,
202 // MS-DOS on Windows, where a reader that follows the record reads these modes as DOS
203 // attributes (zip 9 gives `0o664`, so the daemon loses its execute bit) and the bytes
204 // differ from a Linux or macOS build of the same inputs.
205 let options = |mode: u32| {
206 SimpleFileOptions::default()
207 .compression_method(zip::CompressionMethod::Deflated)
208 .last_modified_time(zip::DateTime::default())
209 .system(zip::System::Unix)
210 .unix_permissions(mode)
211 };
212
213 let mut write_entry = |name: &str, bytes: &[u8], mode: u32| {
214 writer
215 .start_file(name, options(mode))
216 .and_then(|()| writer.write_all(bytes).map_err(Into::into))
217 .map_err(|error| {
218 Error::new(
219 ErrorKind::Unexpected,
220 format!("could not add {name} to the build artifact: {error}"),
221 )
222 })
223 };
224
225 write_entry(DOCKERFILE_ENTRY, dockerfile.as_bytes(), FILE_MODE)?;
226 write_entry(AGENTD_ENTRY, binary, AGENTD_MODE)?;
227 if let Some(project) = project {
228 write_entry(
229 project.ecosystem.manifest_name(),
230 &project.manifest,
231 FILE_MODE,
232 )?;
233 write_entry(
234 project.ecosystem.lockfile_name(),
235 &project.lockfile,
236 FILE_MODE,
237 )?;
238 }
239 if let Some(context) = context {
240 for entry in context.entries() {
241 write_entry(&entry.name, &entry.bytes, entry.mode)?;
242 }
243 }
244
245 let bytes = writer
246 .finish()
247 .map_err(|error| {
248 Error::new(
249 ErrorKind::Unexpected,
250 format!("could not finish the build artifact: {error}"),
251 )
252 })?
253 .into_inner();
254 Ok(bytes)
255}
256
257/// The sha256 of the artifact's inputs, as lowercase hex.
258///
259/// # What is hashed, and why not the zip
260///
261/// The **inputs** — the daemon binary's bytes, the Dockerfile text, and the project
262/// files when the artifact carries them — rather than the bytes [`build_artifact`]
263/// produces. The zip is a container: its byte identity depends on the `zip` crate's
264/// version, its compression level, and its header defaults, so an upgraded dependency
265/// would silently change every image name and orphan every reuse. The inputs are what a
266/// build actually consumes, and two builds with equal inputs produce interchangeable
267/// images — which is the property content-addressed reuse rests on.
268///
269/// Each input is length-prefixed before hashing, so `(binary="ab", dockerfile="c")` and
270/// `(binary="a", dockerfile="bc")` are different hashes rather than one concatenation.
271/// A project file contributes its entry **name and bytes**, both length-prefixed, so the
272/// same bytes under a different ecosystem — `Cargo.lock` and `package-lock.json` holding
273/// identical text — are different identities too.
274///
275/// # `None` is the historical digest, deliberately
276///
277/// With no project files the digest stream is byte-identical to what this function
278/// produced before #74, so every projectless image name in every account survives the
279/// change — the alternative orphans exactly the reuse this hash exists to serve. The
280/// pinned vector in the tests is what holds that fixed. This is #74's key: two projects
281/// with identical lockfiles share an image, and a lockfile edit is a new name and a
282/// fresh build.
283pub fn artifact_content_hash(
284 binary: &[u8],
285 dockerfile: &str,
286 project: Option<&ProjectFiles>,
287) -> String {
288 artifact_content_hash_with_context(binary, dockerfile, project, None)
289}
290
291/// The tag that opens the build context's part of the digest stream.
292///
293/// A project file contributes a fixed file name, so no project entry can begin with this
294/// tag's length and bytes: the two sections cannot be confused for each other.
295const CONTEXT_TAG: &[u8] = b"microvms-build-context/1";
296
297/// [`artifact_content_hash`] extended with a build context (IMAGE-6).
298///
299/// With no context, or an empty one, the stream is byte-for-byte the one
300/// `artifact_content_hash` hashed before #221 — an empty context adds no entry, so it is the
301/// same artifact — and the pinned vector in the tests holds that fixed, so no `build --reuse`
302/// or `AgentVm` image name moves. A non-empty context appends [`CONTEXT_TAG`], the entry
303/// count, and each entry's name, mode, and bytes, the name and bytes length-prefixed. The
304/// entries are sorted by name ([`super::context::BuildContext`]), so the order a filesystem
305/// listed them in is not part of the identity; the mode is, because `COPY` keeps it.
306pub fn artifact_content_hash_with_context(
307 binary: &[u8],
308 dockerfile: &str,
309 project: Option<&ProjectFiles>,
310 context: Option<&super::context::BuildContext>,
311) -> String {
312 use sha2::{Digest as _, Sha256};
313
314 let mut hasher = Sha256::new();
315 hasher.update((binary.len() as u64).to_be_bytes());
316 hasher.update(binary);
317 hasher.update((dockerfile.len() as u64).to_be_bytes());
318 hasher.update(dockerfile.as_bytes());
319 if let Some(project) = project {
320 for (name, bytes) in [
321 (project.ecosystem.manifest_name(), &project.manifest),
322 (project.ecosystem.lockfile_name(), &project.lockfile),
323 ] {
324 hasher.update((name.len() as u64).to_be_bytes());
325 hasher.update(name.as_bytes());
326 hasher.update((bytes.len() as u64).to_be_bytes());
327 hasher.update(bytes);
328 }
329 }
330 if let Some(context) = context.filter(|context| !context.entries().is_empty()) {
331 hasher.update((CONTEXT_TAG.len() as u64).to_be_bytes());
332 hasher.update(CONTEXT_TAG);
333 hasher.update((context.entries().len() as u64).to_be_bytes());
334 for entry in context.entries() {
335 hasher.update((entry.name.len() as u64).to_be_bytes());
336 hasher.update(entry.name.as_bytes());
337 hasher.update(entry.mode.to_be_bytes());
338 hasher.update((entry.bytes.len() as u64).to_be_bytes());
339 hasher.update(&entry.bytes);
340 }
341 }
342 const_hex::encode(hasher.finalize())
343}
344
345/// A base image: the platform ARN, the Dockerfile `FROM` that pairs with it, and whether
346/// it declares a `WORKDIR`.
347///
348/// All three together because the first two **must** agree and used to be able to
349/// disagree: the Python client's `DEFAULT_BASE_IMAGE` named the managed base for
350/// `baseImageArn` while `default_dockerfile` hardcoded an unrelated registry literal in
351/// its `FROM`, so changing either left the other pointing somewhere else
352/// (`sandbox.py:410-444`). Pairing them means a caller selects one thing and both fields
353/// follow.
354#[derive(Clone, Debug, Eq, PartialEq)]
355pub struct BaseImage {
356 /// Goes into `baseImageArn` — the platform's managed base, not a registry ref.
357 pub name: String,
358 /// Goes into the Dockerfile `FROM` — the registry ref measured alongside `name`.
359 pub docker_ref: String,
360 /// What `docker inspect` reports for `WorkingDir`. Empty means it declares none.
361 ///
362 /// A field rather than a lookup because a caller with a purpose-built image is the
363 /// only one who can say what their image declares, and this client cannot read it
364 /// without pulling the manifest.
365 pub working_dir: String,
366}
367
368impl BaseImage {
369 /// The managed base every `docs/PLATFORM.md` measurement from 2026-08-06 onward used,
370 /// paired with the registry ref those same builds used as `FROM`.
371 ///
372 /// `working_dir` is empty, and that is measured rather than assumed:
373 /// `al2023-minimal`, `python:3.12-slim`, and `node:20-slim` all leave `WorkingDir`
374 /// empty (2026-08-05), which is what makes [`require_workdir`] necessary.
375 pub fn al2023() -> Self {
376 Self {
377 name: "al2023-1".to_string(),
378 docker_ref: "public.ecr.aws/amazonlinux/amazonlinux:2023-minimal".to_string(),
379 working_dir: String::new(),
380 }
381 }
382
383 /// The base image a task Dockerfile pairs with: the managed base's `name` (so
384 /// `baseImageArn` is unchanged) and the Dockerfile's own first `FROM` as `docker_ref`
385 /// (IMAGE-4).
386 ///
387 /// # The pairing the other way round
388 ///
389 /// [`require_matching_from`] refuses a Dockerfile whose first `FROM` is not the base's
390 /// `docker_ref`. That is the right guard where this client derives the Dockerfile from
391 /// the base; a task Dockerfile chooses its own `FROM`, so the pairing has to run from the
392 /// Dockerfile to the base, and every harness wrote the same inversion by hand (#220).
393 /// Here the ref is taken whole — `--platform` and `AS` decoration dropped by
394 /// [`dockerfile_from_ref`], a digest pin kept — so the guard's comparison is the ref
395 /// against itself and it passes by construction.
396 ///
397 /// `working_dir` is empty: this cannot read the `FROM` image's `WorkingDir` without
398 /// pulling its manifest, and empty is what every measured public base declares (see
399 /// [`BaseImage::al2023`]). A task that relies on one sets `WORKDIR`, which
400 /// [`require_workdir`] reads from the Dockerfile.
401 ///
402 /// Refuses a Dockerfile with no `FROM`: there is no ref to take.
403 pub fn from_dockerfile(dockerfile: &str) -> Result<Self, Error> {
404 match dockerfile_from_ref(dockerfile) {
405 Some(docker_ref) => Ok(Self {
406 docker_ref: docker_ref.to_string(),
407 ..Self::al2023()
408 }),
409 None => Err(Error::invalid_arg(
410 "the Dockerfile has no FROM, so there is no image reference to pair the \
411 managed base with. Derive the base from the Dockerfile the task builds \
412 from."
413 .to_string(),
414 )),
415 }
416 }
417
418 /// The `baseImageArn` for this base in `region`.
419 ///
420 /// # `microvm-image:<name>`, with a colon, and it is not just the managed base
421 ///
422 /// This was the only place in the repo building the colon form; the fakes and the
423 /// encoding test all used `microvm-image/<name>`, so the repo held two beliefs about the
424 /// shape of the identifier every image call takes. The model's `TaggableResource` pattern
425 /// admits only `(capacity-provider|network-connector|microvm-image):[a-zA-Z0-9-_]+`, and
426 /// a live read settled it: `ListMicrovmImages` in us-east-1 returns
427 /// `arn:aws:lambda:us-east-1:<account>:microvm-image:coding-agents-on-bedrock` for a
428 /// customer image, and `GetMicrovmImage` accepts exactly that (measured 2026-08-15).
429 ///
430 /// So the colon is right for **both** the managed base and a customer image; there is no
431 /// two-form rule, and this comment used to imply there was. The slash form fails as
432 /// `AccessDeniedException` rather than as a validation error, because IAM evaluates the
433 /// malformed ARN as a resource with no matching policy — a permissions message for a
434 /// resource that exists.
435 pub fn arn(&self, region: &crate::region::Region) -> String {
436 format!(
437 "arn:aws:lambda:{}:aws:microvm-image:{}",
438 region.as_str(),
439 self.name
440 )
441 }
442}
443
444impl Default for BaseImage {
445 fn default() -> Self {
446 Self::al2023()
447 }
448}
449
450/// Where the environment layer lives when the caller named no workdir of their own.
451///
452/// A constant rather than `/` because the install commands resolve everything against the
453/// current directory — `uv sync` writes `.venv` beside the manifest, `npm ci` writes
454/// `node_modules` — and the working tree synced at launch has to land in the same place
455/// for the layer to be *its* environment.
456pub const DEFAULT_PROJECT_WORKDIR: &str = "/project";
457
458/// A Dockerfile that makes the daemon the container `CMD`.
459///
460/// `ENTRYPOINT []` plus `CMD ["/agentd"]` is the deployment invariant the trust boundary
461/// rests on: it is what guarantees no task workload runs before the platform's run hook
462/// lands. It is also what makes an omitted `cwd` inherit the image `WORKDIR`, since the
463/// daemon's own cwd is the image's.
464///
465/// The `FROM` is derived from `base` rather than written here, so it cannot disagree with
466/// the `baseImageArn` the create call sends.
467///
468/// `project` bakes the environment layer (#74): the manifest+lockfile pair is copied into
469/// the working directory — `workdir` when given, [`DEFAULT_PROJECT_WORKDIR`] otherwise —
470/// and the ecosystem's [`Ecosystem::install_run_line`] installs from the lockfile. Without
471/// it the derived Dockerfile installs nothing, which was the measured state this closes:
472/// every launch then pays dependency installation inside the guest, the 31–48% of launch
473/// time `docs/STRATEGY.md` attributes to environment init.
474///
475/// The invariant is *unenforced*: a base image that starts its own background process
476/// before bootstrap breaks it, and enforcing that belongs to whoever builds the image
477/// (`docs/PROTOCOL.md`, "Trust boundary").
478pub fn default_dockerfile(
479 port: u16,
480 workdir: Option<&str>,
481 base: &BaseImage,
482 project: Option<Ecosystem>,
483) -> String {
484 format!(
485 "FROM {}\n{}",
486 base.docker_ref,
487 agentd_stanza(port, workdir, project)
488 )
489}
490
491/// The agentd stanza: every line [`default_dockerfile`] writes after its `FROM`, and every
492/// line [`wrap_dockerfile`] appends to a task Dockerfile (IMAGE-1).
493///
494/// # One source, and why it is this function rather than a constant
495///
496/// Two copies of these lines drift: a harness that carried the stanza as a string literal
497/// of its own (#220) had to track every change here by hand, and the day one side gained a
498/// line the other did not, the image a task built and the image the default build produced
499/// stopped agreeing about how the daemon starts. So both callers render this one function,
500/// and `wrapping_a_bare_from_is_the_default_dockerfile_minus_nothing` holds them equal:
501/// `wrap_dockerfile("FROM x\n")` is the default Dockerfile for a base whose ref is `x`.
502///
503/// The text is byte-for-byte what `default_dockerfile` emitted before the split, which is
504/// what keeps every content-addressed image name (`build --reuse`, `AgentVm::image_name`)
505/// derived from the default Dockerfile valid: the name hashes the Dockerfile text.
506fn agentd_stanza(port: u16, workdir: Option<&str>, project: Option<Ecosystem>) -> String {
507 let mut lines = vec![
508 "COPY agentd /agentd".to_string(),
509 "RUN chmod 0755 /agentd".to_string(),
510 ];
511 let workdir = match (workdir.filter(|dir| !dir.is_empty()), project) {
512 (Some(dir), _) => Some(dir),
513 // The layer needs a directory to live in even when the caller named none.
514 (None, Some(_)) => Some(DEFAULT_PROJECT_WORKDIR),
515 (None, None) => None,
516 };
517 if let Some(workdir) = workdir {
518 lines.push(format!("RUN mkdir -p {workdir}"));
519 lines.push(format!("WORKDIR {workdir}"));
520 }
521 if let Some(ecosystem) = project {
522 // Relative destination, resolving against the WORKDIR set above — the same
523 // directory the install command runs in.
524 lines.push(format!(
525 "COPY {} {} ./",
526 ecosystem.manifest_name(),
527 ecosystem.lockfile_name(),
528 ));
529 lines.push(ecosystem.install_run_line().to_string());
530 }
531 lines.extend([
532 format!("ENV AGENTD_PORT={port}"),
533 "ENV AGENTD_LOG=info".to_string(),
534 format!("EXPOSE {port}"),
535 "ENTRYPOINT []".to_string(),
536 r#"CMD ["/agentd"]"#.to_string(),
537 String::new(),
538 ]);
539 lines.join("\n")
540}
541
542/// What [`wrap_dockerfile`] needs besides the task text.
543#[derive(Clone, Debug, Eq, PartialEq)]
544pub struct WrapOptions {
545 /// The agent port the stanza's `ENV AGENTD_PORT` and `EXPOSE` name. It must be the port
546 /// the create call sends as `hooks.port` — [`ControlPlane::port`](super::ControlPlane::port)
547 /// — which [`require_matching_agentd_port`] checks at the create call.
548 pub port: u16,
549 /// A working directory for the stanza to create and set, exactly as
550 /// [`default_dockerfile`] writes one. One absolute path; `Some("")` reads as `None`, the
551 /// same way the default generator reads it.
552 pub workdir: Option<String>,
553 /// The caller relies on the image `WORKDIR` — an exec with no `cwd` runs there — so one
554 /// must be declared: by the task Dockerfile or by [`WrapOptions::workdir`]. Off by
555 /// default, because a task with no `WORKDIR` runs its commands in `/` under Docker too,
556 /// and a harness reproducing that has nothing to refuse.
557 pub inherit_workdir: bool,
558}
559
560impl Default for WrapOptions {
561 fn default() -> Self {
562 Self {
563 port: super::DEFAULT_AGENT_PORT,
564 workdir: None,
565 inherit_workdir: false,
566 }
567 }
568}
569
570/// A task Dockerfile with the agentd stanza appended: the one call a harness needs to go
571/// from the Dockerfile a task brings to one this client can build (#220).
572///
573/// The result is the task text verbatim, a newline if it lacked one, `USER root` when the
574/// task's last `USER` is anyone else, and then [`default_dockerfile`]'s stanza for
575/// `opts.port` and `opts.workdir` — the same function, so the two cannot drift (IMAGE-1).
576/// The stanza's `ENTRYPOINT []` and `CMD ["/agentd"]` are the result's last instructions,
577/// so whatever entrypoint, command, or port the task set, the daemon is the container's
578/// process and listens where the hooks are called (IMAGE-2). Pair it with
579/// [`BaseImage::from_dockerfile`] for the create call's `FROM` guard (IMAGE-4).
580///
581/// # `USER root`, only when the task changed user
582///
583/// The daemon demotes each exec to the user it names, which takes root; a task that ends on
584/// `USER app` would otherwise start the daemon as `app`. The line is written only when the
585/// last `USER` is not root, so a task that never changes user wraps to exactly the default
586/// stanza — writing it unconditionally would make the default Dockerfile and the wrap of a
587/// bare `FROM` two different texts (`crates/model/src/wrap.rs`, `AlwaysUserRoot`).
588///
589/// # What is refused, and why each is a refusal (IMAGE-3)
590///
591/// Each is a Dockerfile the build would accept and the guest would then fail on, one build
592/// cycle later and naming none of the cause:
593///
594/// * **No `FROM`.** Nothing to append to, and no base to derive.
595/// * **An unfinished last instruction.** A trailing line continuation — in the escape
596/// character the `# escape=` directive selects, across the blank and comment lines Docker
597/// skips inside one — joins the stanza's `COPY agentd /agentd` into the task's last
598/// instruction; an unterminated heredoc swallows the whole stanza as its body. Either way
599/// the image has no daemon and the build fails as a run-hook timeout.
600/// * **A keepalive at or over the client's stream idle timeout**, for the reason
601/// [`require_keepalive_under_idle_timeout`] gives.
602/// * **A port of 0, or a workdir that is not one absolute path.** A relative `WORKDIR`
603/// resolves against whatever the task left, whitespace splits `RUN mkdir -p` into several
604/// directories, and a line break writes an instruction the caller never saw.
605/// * **`inherit_workdir` with no `WORKDIR` anywhere**, for the reason [`require_workdir`]
606/// gives.
607///
608/// The scan reads instructions the way the Dockerfile parser does for these questions —
609/// escape directive, continuations, comments, heredoc bodies — and no further: it finds a
610/// Dockerfile that cannot take the stanza, it does not validate one.
611pub fn wrap_dockerfile(task: &str, opts: &WrapOptions) -> Result<String, Error> {
612 if dockerfile_from_ref(task).is_none() {
613 return Err(Error::invalid_arg(
614 "the task Dockerfile has no FROM, so there is no image to append the agentd \
615 stanza to and no base image to pair with it. Wrap the Dockerfile the task \
616 builds from, or pass `FROM <image>` for a task that names only an image."
617 .to_string(),
618 ));
619 }
620 let scan = Instructions::scan(task);
621 if let Some(terminator) = scan.open_heredoc {
622 return Err(Error::invalid_arg(format!(
623 "the task Dockerfile ends inside a heredoc that is never terminated (no line \
624 {terminator:?} closes it), so every line appended after it — the whole agentd \
625 stanza — would become the heredoc's body. The image would build with no daemon \
626 in it and fail as a run-hook timeout. Terminate the heredoc."
627 )));
628 }
629 if scan.continued {
630 return Err(Error::invalid_arg(format!(
631 "the task Dockerfile's last instruction ends with a line continuation ({:?}), so \
632 the first line appended after it — the agentd stanza's `COPY agentd /agentd` — \
633 would be joined into that instruction instead of running. The image would build \
634 with no daemon in it and fail as a run-hook timeout. Finish the instruction, or \
635 drop the trailing {:?}.",
636 scan.escape, scan.escape,
637 )));
638 }
639 require_keepalive_under_idle_timeout(crate::session::exec::DEFAULT_STREAM_IDLE_TIMEOUT, task)?;
640 super::require_valid_port("port", opts.port)?;
641 let workdir = opts.workdir.as_deref().filter(|dir| !dir.is_empty());
642 if let Some(dir) = workdir
643 && (!dir.starts_with('/') || dir.chars().any(|c| c.is_whitespace() || c.is_control()))
644 {
645 return Err(Error::invalid_arg(format!(
646 "the workdir {dir:?} is not one absolute path. The stanza writes it as \
647 `RUN mkdir -p {dir}` and `WORKDIR {dir}`: a relative path resolves against \
648 whatever WORKDIR the task left, whitespace splits the mkdir into several \
649 directories, and a line break writes an instruction nobody reviewed. Pass an \
650 absolute path such as /workspace."
651 )));
652 }
653 if opts.inherit_workdir && workdir.is_none() && !dockerfile_declares_workdir(task) {
654 return Err(Error::invalid_arg(
655 "inherit_workdir was requested but the task Dockerfile sets no WORKDIR and no \
656 workdir option was given, so there is nothing to inherit: every exec with no cwd \
657 would run in `/`, and every relative path in it would resolve there. Pass a \
658 workdir, or set WORKDIR in the task Dockerfile."
659 .to_string(),
660 ));
661 }
662
663 let stanza = agentd_stanza(opts.port, workdir, None);
664 let mut wrapped = String::with_capacity(task.len() + stanza.len() + 16);
665 wrapped.push_str(task);
666 if !task.ends_with('\n') {
667 wrapped.push('\n');
668 }
669 if scan
670 .last_user
671 .as_deref()
672 .is_some_and(|user| !is_root_user(user))
673 {
674 wrapped.push_str("USER root\n");
675 }
676 wrapped.push_str(&stanza);
677 Ok(wrapped)
678}
679
680/// Whether a `USER` value names root: `root` or `0`, with no group or a root group.
681///
682/// Anything else is not root, including a value built from a variable — the scan cannot
683/// expand `${TASK_USER}`, and restoring root for a value that turns out to be root costs one
684/// redundant line.
685fn is_root_user(value: &str) -> bool {
686 let is_root = |part: &str| part == "root" || part == "0";
687 match value.split_once(':') {
688 Some((user, group)) => is_root(user) && is_root(group),
689 None => is_root(value),
690 }
691}
692
693/// Whether any line of `dockerfile` is a `WORKDIR` with an argument.
694fn dockerfile_declares_workdir(dockerfile: &str) -> bool {
695 dockerfile.lines().any(|line| {
696 let mut words = line.split_whitespace();
697 words
698 .next()
699 .is_some_and(|first| first.eq_ignore_ascii_case("WORKDIR"))
700 && words.next().is_some()
701 })
702}
703
704/// What [`wrap_dockerfile`] needs to know about how a Dockerfile's instructions end.
705struct Instructions {
706 /// The escape character: `\` unless an `# escape=` directive chose a backtick.
707 escape: char,
708 /// The text ends inside a line continuation.
709 continued: bool,
710 /// The terminator of a heredoc the text never closes.
711 open_heredoc: Option<String>,
712 /// The value of the last `USER` instruction.
713 last_user: Option<String>,
714}
715
716impl Instructions {
717 /// Reads the parser directives, then every logical instruction: continuation lines
718 /// joined, blank and comment lines skipped (inside a continuation too, as the parser
719 /// does), heredoc bodies consumed up to their terminators.
720 fn scan(dockerfile: &str) -> Self {
721 let escape = escape_directive(dockerfile);
722 let mut heredocs: std::collections::VecDeque<(String, bool)> = Default::default();
723 let mut pending = String::new();
724 let mut continued = false;
725 let mut last_user = None;
726 for line in dockerfile.lines() {
727 if let Some((terminator, strip_tabs)) = heredocs.front() {
728 let candidate = if *strip_tabs {
729 line.trim_start_matches('\t')
730 } else {
731 line
732 };
733 if candidate.trim_end_matches('\r') == terminator {
734 heredocs.pop_front();
735 }
736 continue;
737 }
738 let trimmed = line.trim();
739 if trimmed.is_empty() || trimmed.starts_with('#') {
740 continue;
741 }
742 let body = line.trim_end();
743 if let Some(joined) = body.strip_suffix(escape) {
744 pending.push_str(joined);
745 pending.push(' ');
746 continued = true;
747 continue;
748 }
749 continued = false;
750 pending.push_str(body);
751 let instruction = std::mem::take(&mut pending);
752 let mut words = instruction.split_whitespace();
753 let Some(keyword) = words.next() else {
754 continue;
755 };
756 if keyword.eq_ignore_ascii_case("USER") {
757 last_user = words.next().map(str::to_string);
758 } else if ["RUN", "COPY", "ADD"]
759 .iter()
760 .any(|name| keyword.eq_ignore_ascii_case(name))
761 {
762 heredocs.extend(words.filter_map(heredoc_marker));
763 }
764 }
765 Self {
766 escape,
767 continued,
768 open_heredoc: heredocs.pop_front().map(|(terminator, _)| terminator),
769 last_user,
770 }
771 }
772}
773
774/// The escape character a Dockerfile's `# escape=` parser directive selects, or `\`.
775///
776/// Directives are only read at the very top: the first line that is not one — an
777/// instruction, a blank line, or an ordinary comment — ends them.
778fn escape_directive(dockerfile: &str) -> char {
779 for line in dockerfile.lines() {
780 let Some(directive) = line.trim().strip_prefix('#') else {
781 break;
782 };
783 let Some((key, value)) = directive.split_once('=') else {
784 break;
785 };
786 let key = key.trim().to_ascii_lowercase();
787 if key == "escape" {
788 return match value.trim() {
789 "`" => '`',
790 _ => '\\',
791 };
792 }
793 if !["syntax", "check"].contains(&key.as_str()) {
794 break;
795 }
796 }
797 '\\'
798}
799
800/// The terminator a heredoc marker word opens, and whether it strips leading tabs (`<<-`).
801///
802/// `<<EOF`, `<<-EOF`, `<<"EOF"`, `<<'EOF'`; not the here-string `<<<`, and not a bare `<<`.
803fn heredoc_marker(word: &str) -> Option<(String, bool)> {
804 let rest = word.strip_prefix("<<")?;
805 if rest.starts_with('<') {
806 return None;
807 }
808 let (rest, strip_tabs) = match rest.strip_prefix('-') {
809 Some(rest) => (rest, true),
810 None => (rest, false),
811 };
812 let name = rest
813 .strip_prefix('"')
814 .and_then(|name| name.strip_suffix('"'))
815 .or_else(|| {
816 rest.strip_prefix('\'')
817 .and_then(|name| name.strip_suffix('\''))
818 })
819 .unwrap_or(rest);
820 let valid = name
821 .chars()
822 .next()
823 .is_some_and(|first| first.is_ascii_alphabetic() || first == '_')
824 && name.chars().all(|c| c.is_ascii_alphanumeric() || c == '_');
825 valid.then(|| (name.to_string(), strip_tabs))
826}
827
828/// The image ref in a Dockerfile's first `FROM`, or `None` when it has none.
829///
830/// Deliberately loose on whitespace and case, and it ignores `--platform=` and `AS name`
831/// decoration, because the check exists to catch a base that disagrees rather than to
832/// validate Dockerfile syntax.
833///
834/// Split-and-scan rather than a regex: the pattern is "first token after FROM on a line
835/// whose first word is FROM", which `split_whitespace` states more directly than a
836/// pattern string would. (`regex-lite` is available if this ever grows.)
837pub fn dockerfile_from_ref(dockerfile: &str) -> Option<&str> {
838 for line in dockerfile.lines() {
839 let mut words = line.split_whitespace();
840 let Some(first) = words.next() else { continue };
841 if !first.eq_ignore_ascii_case("FROM") {
842 continue;
843 }
844 // Skip flag decoration such as `--platform=linux/arm64`.
845 return words.find(|word| !word.starts_with("--"));
846 }
847 None
848}
849
850/// Rejects working-directory inheritance when nothing declares one.
851///
852/// Measured 2026-08-05: `al2023-minimal`, `python:3.12-slim`, and `node:20-slim` all leave
853/// `WorkingDir` empty, so "inherit the image WORKDIR" inherits `/` and every relative path
854/// in the caller's commands resolves somewhere they did not mean.
855///
856/// Rejected rather than warned because the symptom appears in the **guest, one build cycle
857/// later**, as commands running in the wrong directory rather than as anything about
858/// `WORKDIR`.
859pub fn require_workdir(base: &BaseImage, dockerfile: Option<&str>) -> Result<(), Error> {
860 if !base.working_dir.is_empty() {
861 return Ok(());
862 }
863 let declares_workdir = dockerfile.is_some_and(|text| {
864 text.lines().any(|line| {
865 let mut words = line.split_whitespace();
866 words
867 .next()
868 .is_some_and(|first| first.eq_ignore_ascii_case("WORKDIR"))
869 && words.next().is_some()
870 })
871 });
872 if declares_workdir {
873 return Ok(());
874 }
875 Err(Error::invalid_arg(format!(
876 "inherit_workdir was requested but base image {:?} declares no WorkingDir and the \
877 Dockerfile sets none. Most public ARM64 base images leave it empty \
878 (docs/PLATFORM.md, 'Most public ARM64 base images have no WORKDIR'), so there is \
879 nothing to inherit and every relative path would resolve against `/`. Pass a workdir \
880 to default_dockerfile, or set WORKDIR in your own Dockerfile.",
881 base.name
882 )))
883}
884
885/// The port in a Dockerfile's last `ENV AGENTD_PORT=`, or `None` when it sets none.
886///
887/// Last rather than first: a later `ENV` of the same name wins at build time, so the last
888/// one is what the daemon reads. Accepts `ENV AGENTD_PORT=9000` and the legacy
889/// `ENV AGENTD_PORT 9000` spelling, since both set the variable and a guard that only
890/// understands one form passes the file it cannot parse.
891///
892/// Split-and-scan for the same reason [`dockerfile_from_ref`] is: `split_whitespace`
893/// states the pattern more directly than a regex string would here.
894pub fn dockerfile_agentd_port(dockerfile: &str) -> Option<u16> {
895 let mut found = None;
896 for line in dockerfile.lines() {
897 let mut words = line.split_whitespace();
898 let Some(first) = words.next() else { continue };
899 if !first.eq_ignore_ascii_case("ENV") {
900 continue;
901 }
902 let Some(assignment) = words.next() else {
903 continue;
904 };
905 let value = match assignment.split_once('=') {
906 Some(("AGENTD_PORT", value)) => value,
907 Some(_) => continue,
908 // `ENV AGENTD_PORT 9000`: the value is the next word.
909 None if assignment == "AGENTD_PORT" => match words.next() {
910 Some(value) => value,
911 None => continue,
912 },
913 None => continue,
914 };
915 // An unparseable value is not this guard's business: the daemon keeps its default
916 // for one (`crates/agentd/src/config.rs:118`), so there is no disagreement to report.
917 if let Ok(port) = value.parse() {
918 found = Some(port);
919 }
920 }
921 found
922}
923
924/// Rejects a Dockerfile whose `AGENTD_PORT` disagrees with the port the create call sends
925/// as `hooks.port`.
926///
927/// The two are set independently — `hooks.port` comes from `ControlPlane::port`, the
928/// variable from the caller's own Dockerfile — and the platform calls its build-time
929/// `ready`/`validate` hooks on the port in the create call. A guest listening elsewhere
930/// answers nothing, so the build fails.
931///
932/// Rejected rather than warned because the failure points away from its cause: the docker
933/// build succeeds, the log group holds a clean build *and* the daemon's own "agentd
934/// listening" line, and the image still lands in `CREATE_FAILED`. Neither
935/// `GetMicrovmImage` nor the build log names the port — only
936/// `GetMicrovmImageVersion`'s `hooks.port` does, compared by hand against the Dockerfile.
937///
938/// A Dockerfile that sets no `AGENTD_PORT` is checked against
939/// [`DEFAULT_AGENT_PORT`](crate::control::DEFAULT_AGENT_PORT) rather than passed, because
940/// silence is not neutral: `Config::from_env` keeps its own default for an unset variable
941/// (`crates/agentd/src/config.rs:118`, default `crates/agentd/src/config.rs:84`), so the guest listens on
942/// 9000 whether or not the Dockerfile mentions a port. Against a client that changed its
943/// port, the absent variable produces exactly the measured failure — and produces it for a
944/// caller who never typed a port anywhere, which is the harder version to diagnose.
945pub fn require_matching_agentd_port(port: u16, dockerfile: &str) -> Result<(), Error> {
946 // An unparseable value lands here too, and belongs here: the daemon warns and keeps the
947 // same default (`crates/agentd/src/config.rs:174-179`), so both spellings of "the Dockerfile
948 // named no usable port" have the same consequence in the guest.
949 let Some(found) = dockerfile_agentd_port(dockerfile) else {
950 if port == super::DEFAULT_AGENT_PORT {
951 return Ok(());
952 }
953 return Err(Error::invalid_arg(format!(
954 "this client sends hooks.port={port} but the Dockerfile sets no usable \
955 AGENTD_PORT, so the daemon will listen on its own default of {default}. These \
956 must agree: the platform calls the build-time ready/validate hooks on the port \
957 in the create call, and a daemon listening on {default} answers none of them — \
958 the docker build succeeds, the daemon logs that it is listening, and the image \
959 still fails with CREATE_FAILED naming no port. Add ENV AGENTD_PORT={port} to \
960 the Dockerfile, or build it with default_dockerfile, which derives the value \
961 from the same port.",
962 default = super::DEFAULT_AGENT_PORT,
963 )));
964 };
965 if found == port {
966 return Ok(());
967 }
968 Err(Error::invalid_arg(format!(
969 "the Dockerfile sets ENV AGENTD_PORT={found} but this client sends hooks.port={port}. \
970 These must agree: the platform calls the build-time ready/validate hooks on the port \
971 in the create call, and a daemon listening on {found} answers none of them — the \
972 docker build succeeds, the daemon logs that it is listening, and the image still \
973 fails with CREATE_FAILED naming no port. Set ENV AGENTD_PORT={port} in the \
974 Dockerfile, or build it with default_dockerfile, which derives the value from the \
975 same port."
976 )))
977}
978
979/// The seconds in a Dockerfile's last `ENV AGENTD_SSE_KEEPALIVE_SECS=`, or `None` when it
980/// sets none.
981///
982/// Same shape and same reasoning as [`dockerfile_agentd_port`]: last assignment wins, both
983/// `ENV` spellings, and an unparseable value reads as absent because the daemon warns and
984/// keeps its default for one (`crates/agentd/src/config.rs:174-179`).
985pub fn dockerfile_env_u64(dockerfile: &str, key: &str) -> Option<u64> {
986 let mut found = None;
987 for line in dockerfile.lines() {
988 let mut words = line.split_whitespace();
989 let Some(first) = words.next() else { continue };
990 if !first.eq_ignore_ascii_case("ENV") {
991 continue;
992 }
993 let Some(assignment) = words.next() else {
994 continue;
995 };
996 let value = match assignment.split_once('=') {
997 Some((name, value)) if name == key => value,
998 Some(_) => continue,
999 None if assignment == key => match words.next() {
1000 Some(value) => value,
1001 None => continue,
1002 },
1003 None => continue,
1004 };
1005 if let Ok(parsed) = value.parse() {
1006 found = Some(parsed);
1007 }
1008 }
1009 found
1010}
1011
1012/// Rejects a Dockerfile whose `AGENTD_SSE_KEEPALIVE_SECS` is not shorter than the client's
1013/// stream idle timeout.
1014///
1015/// The fourth pair of this shape, found by sweeping for the other three. The daemon's SSE
1016/// keepalive interval and the client's tolerance for silence are set in different
1017/// repositories of truth — the interval in the caller's Dockerfile
1018/// (`crates/agentd/src/config.rs:139`, default 15s at `:95`), the tolerance in
1019/// [`DEFAULT_STREAM_IDLE_TIMEOUT`](crate::session::exec::DEFAULT_STREAM_IDLE_TIMEOUT),
1020/// which is 60s *because* it is four times that 15. Raise the interval past the tolerance
1021/// and every attached stream treats a healthy connection as dead, reconnects
1022/// `max_reconnects` times, and fails.
1023///
1024/// The failure names both numbers and misattributes one of them: the client reports that the
1025/// stream "went silent for 60s, longer than the keepalive interval"
1026/// (`crates/microvms-edges/src/session/http.rs:223-227`), where 60 is its own timeout and the
1027/// keepalive interval is the number it does not know. So a reader is told the keepalive was
1028/// exceeded by a message that prints the wrong value for it, one build cycle after the
1029/// Dockerfile that caused it.
1030///
1031/// Equality is refused, not just excess: an interval exactly equal to the timeout races.
1032pub fn require_keepalive_under_idle_timeout(
1033 idle_timeout: Duration,
1034 dockerfile: &str,
1035) -> Result<(), Error> {
1036 let Some(secs) = dockerfile_env_u64(dockerfile, "AGENTD_SSE_KEEPALIVE_SECS") else {
1037 // Absent is safe here, unlike the port: the daemon's default of 15s is already
1038 // under every idle timeout this client will use, and a client that shortens its
1039 // timeout below 15s is not something a Dockerfile can be blamed for.
1040 return Ok(());
1041 };
1042 if secs < idle_timeout.as_secs() {
1043 return Ok(());
1044 }
1045 Err(Error::invalid_arg(format!(
1046 "the Dockerfile sets ENV AGENTD_SSE_KEEPALIVE_SECS={secs} but this client treats a \
1047 stream as dead after {timeout}s of silence. The keepalive must be shorter: the \
1048 daemon sends nothing between events except that keepalive, so an interval of \
1049 {secs}s makes a healthy stream look dead, and every attach reconnects until it \
1050 gives up. The error it raises then reports the client's own {timeout}s as though it \
1051 were the keepalive interval, so it names neither the Dockerfile nor {secs}. Leave \
1052 the variable unset for the daemon's 15s default, or keep it under {timeout}.",
1053 timeout = idle_timeout.as_secs(),
1054 )))
1055}
1056
1057/// The value of a Dockerfile's last `ENTRYPOINT`, or `None` when it sets none.
1058///
1059/// Last rather than first for the reason [`dockerfile_agentd_port`] gives: a later
1060/// instruction of the same name wins at build time, so the last one is what the container
1061/// runs. The value is the raw rest of the line — this scan finds a disagreement, it does
1062/// not parse Dockerfile syntax.
1063pub fn dockerfile_entrypoint(dockerfile: &str) -> Option<&str> {
1064 last_instruction_value(dockerfile, "ENTRYPOINT")
1065}
1066
1067/// The value of a Dockerfile's last `CMD`, or `None` when it sets none.
1068pub fn dockerfile_cmd(dockerfile: &str) -> Option<&str> {
1069 last_instruction_value(dockerfile, "CMD")
1070}
1071
1072/// The rest of the last line whose first word is `keyword`, trimmed, or `None`.
1073fn last_instruction_value<'a>(dockerfile: &'a str, keyword: &str) -> Option<&'a str> {
1074 let mut found = None;
1075 for line in dockerfile.lines() {
1076 let trimmed = line.trim_start();
1077 let Some(first) = trimmed.split_whitespace().next() else {
1078 continue;
1079 };
1080 if first.eq_ignore_ascii_case(keyword) {
1081 found = Some(trimmed[first.len()..].trim());
1082 }
1083 }
1084 found
1085}
1086
1087/// Whether an instruction value is the empty exec form — `[]`, with any spacing.
1088fn is_empty_exec_form(value: &str) -> bool {
1089 let mut meaningful = value.chars().filter(|c| !c.is_whitespace());
1090 meaningful.next() == Some('[') && meaningful.next() == Some(']') && meaningful.next().is_none()
1091}
1092
1093/// Rejects a Dockerfile whose image would never run the daemon: no `CMD`, or an
1094/// `ENTRYPOINT` that swallows it.
1095///
1096/// The artifact unconditionally carries the daemon as entry `agentd` ([`build_artifact`]),
1097/// and the deployment invariant is `ENTRYPOINT []` plus `CMD ["/agentd"]` — see
1098/// [`default_dockerfile`]. A caller Dockerfile with no `CMD` runs the base's own default
1099/// instead, and a non-empty `ENTRYPOINT` turns any `CMD` into that entrypoint's arguments
1100/// rather than the process. Either way the daemon never starts, the build-time
1101/// `ready`/`validate` hooks go unanswered, and the image lands in `CREATE_FAILED` — or the
1102/// launch dies as a run-hook timeout — with nothing naming `CMD`, `ENTRYPOINT`, or the
1103/// artifact entry.
1104///
1105/// Weak-form on purpose: it does **not** check that the `CMD` names a path anything was
1106/// copied to. That would be Dockerfile interpretation rather than agreement checking, and
1107/// the two mistakes people actually make are the two refused here. The unenforceable half —
1108/// a base image that starts its own background process before bootstrap — stays with
1109/// whoever builds the image (`docs/PROTOCOL.md`, "Trust boundary").
1110pub fn require_daemon_cmd(dockerfile: &str) -> Result<(), Error> {
1111 if let Some(entrypoint) = dockerfile_entrypoint(dockerfile)
1112 && !is_empty_exec_form(entrypoint)
1113 {
1114 return Err(Error::invalid_arg(format!(
1115 "the Dockerfile sets ENTRYPOINT {entrypoint}, which makes any CMD its arguments \
1116 rather than the container's process — so the daemon this client uploads never \
1117 starts. The build succeeds anyway: the ready/validate hooks just go unanswered \
1118 and the image fails as CREATE_FAILED, or the launch dies as a run-hook timeout, \
1119 and neither symptom names ENTRYPOINT. Set `ENTRYPOINT []` alongside \
1120 `CMD [\"/agentd\"]`, or build with default_dockerfile, which sets both."
1121 )));
1122 }
1123 match dockerfile_cmd(dockerfile) {
1124 Some(cmd) if !cmd.is_empty() && !is_empty_exec_form(cmd) => Ok(()),
1125 _ => Err(Error::invalid_arg(
1126 "the Dockerfile has no CMD, so the container runs the base image's default \
1127 process and the daemon this client uploads never starts. The build succeeds \
1128 anyway: the ready/validate hooks just go unanswered and the image fails as \
1129 CREATE_FAILED, or the launch dies as a run-hook timeout, and neither symptom \
1130 names CMD. Add `ENTRYPOINT []` and `CMD [\"/agentd\"]`, or build with \
1131 default_dockerfile, which sets both."
1132 .to_string(),
1133 )),
1134 }
1135}
1136
1137/// Rejects a caller Dockerfile that never mentions the lockfile the request carries.
1138///
1139/// The project files enter the artifact unconditionally once the request holds them, so a
1140/// custom Dockerfile that ignores them builds cleanly and produces an image with **no
1141/// environment layer** — and the symptom appears at launch, as every dependency
1142/// installing inside the guest, with nothing anywhere naming the `COPY` that was never
1143/// written. That is the silent-degradation shape: the caller asked for the layer, paid
1144/// the build, and got an image indistinguishable from one that never asked.
1145///
1146/// Weak-form on purpose, like [`require_daemon_cmd`]: a substring test for the lockfile's
1147/// name rather than parsing `COPY` syntax, because the check exists to catch a Dockerfile
1148/// that plainly ignores the files, not to validate how it consumes them.
1149pub fn require_project_install(project: &ProjectFiles, dockerfile: &str) -> Result<(), Error> {
1150 let lockfile = project.ecosystem.lockfile_name();
1151 if dockerfile.contains(lockfile) {
1152 return Ok(());
1153 }
1154 Err(Error::invalid_arg(format!(
1155 "the request carries project files ({lockfile} and {manifest}) but the Dockerfile \
1156 never mentions {lockfile}, so the image would bake no environment layer: the build \
1157 succeeds, and every launch still installs dependencies inside the guest with \
1158 nothing naming the COPY that was never written. Copy the pair and install from \
1159 the lockfile ({install}), or build with default_dockerfile, which derives both \
1160 lines — or drop the project files if the layer is not wanted.",
1161 manifest = project.ecosystem.manifest_name(),
1162 install = match project.ecosystem {
1163 Ecosystem::Uv => "uv sync --locked",
1164 Ecosystem::Npm => "npm ci",
1165 Ecosystem::Cargo => "cargo fetch --locked",
1166 },
1167 )))
1168}
1169
1170/// Rejects a Dockerfile whose `FROM` is not the selected base image.
1171///
1172/// The build runs the Dockerfile *on top of* the base named in `baseImageArn`, so the two
1173/// disagreeing produces an image built from something other than the platform base whose
1174/// behaviour every `docs/PLATFORM.md` measurement describes — and nothing in the result
1175/// says so.
1176pub fn require_matching_from(base: &BaseImage, dockerfile: &str) -> Result<(), Error> {
1177 let Some(found) = dockerfile_from_ref(dockerfile) else {
1178 // No FROM at all is not this check's business: the build will say so, and a
1179 // Dockerfile validator is not what this is.
1180 return Ok(());
1181 };
1182 if found == base.docker_ref {
1183 return Ok(());
1184 }
1185 // The digest-pinned spelling of the same ref: `<docker_ref>@sha256:<64 hex>`. A digest
1186 // names one specific manifest of the ref this check already accepts, so it is a stricter
1187 // statement of agreement rather than a disagreement — a supply-chain-conscious caller
1188 // pins exactly this way, and refusing it would force them to un-pin to pass. The tag
1189 // must still be present and identical: `alpine@sha256:...` against a docker_ref of
1190 // `alpine:3` is a different claim and stays refused.
1191 if let Some(digest) = found
1192 .strip_prefix(base.docker_ref.as_str())
1193 .and_then(|rest| rest.strip_prefix("@sha256:"))
1194 && digest.len() == 64
1195 && digest.bytes().all(|b| b.is_ascii_hexdigit())
1196 {
1197 return Ok(());
1198 }
1199 Err(Error::invalid_arg(format!(
1200 "the Dockerfile's FROM is {found:?} but base image {:?} pairs with {:?}. These must \
1201 agree: baseImageArn and the FROM select the same base, and a mismatch builds against \
1202 a base none of the measured platform behaviour applies to. Use default_dockerfile \
1203 with this base, or pass a BaseImage whose docker_ref matches.",
1204 base.name, base.docker_ref
1205 )))
1206}
1207
1208#[cfg(test)]
1209mod tests {
1210 use super::*;
1211 use crate::testing::aarch64_daemon;
1212
1213 const DEFAULT_AGENT_PORT_FOR_TESTS: u16 = crate::control::DEFAULT_AGENT_PORT;
1214
1215 /// The archive holds exactly the two entries the build expects, named as the build
1216 /// looks for them.
1217 #[test]
1218 fn the_artifact_holds_a_dockerfile_and_the_daemon() {
1219 let bytes =
1220 build_artifact(&aarch64_daemon(b"fake daemon"), "FROM scratch\n", None).expect("zips");
1221 let mut archive =
1222 zip::ZipArchive::new(std::io::Cursor::new(bytes)).expect("a readable zip");
1223
1224 let mut names: Vec<String> = (0..archive.len())
1225 .map(|index| {
1226 archive
1227 .by_index(index)
1228 .expect("entry")
1229 .name()
1230 .expect("a name")
1231 .into_owned()
1232 })
1233 .collect();
1234 names.sort();
1235 assert_eq!(names, ["Dockerfile", "agentd"]);
1236 }
1237
1238 /// **BIND-20 where the daemon enters an artifact (#257).** An x86_64 ELF, a script and no bytes
1239 /// at all are each refused as a precondition naming what the bytes are, so the bindings'
1240 /// two-step path (`build_artifact`, then the caller's own upload) never uploads one.
1241 ///
1242 /// **Falsification**: drop the `require_aarch64` call from `build_artifact_with_context`
1243 /// and all three zip.
1244 #[test]
1245 fn an_artifact_refuses_a_daemon_that_is_not_an_aarch64_elf() {
1246 for (binary, why) in [
1247 (crate::testing::elf_daemon(0x3E, b"x86"), "ELF machine 0x3e"),
1248 (b"#!/bin/sh\n".to_vec(), "not an ELF binary"),
1249 (Vec::new(), "not an ELF binary"),
1250 ] {
1251 let error = build_artifact(&binary, "FROM scratch\n", None).expect_err(why);
1252 assert_eq!(error.kind(), ErrorKind::Precondition, "{error}");
1253 assert!(error.to_string().contains(why), "{error}");
1254 }
1255 }
1256
1257 /// The daemon entry carries mode 0755, and the Dockerfile entry does not have to.
1258 ///
1259 /// Read back out of the archive rather than asserted on the constant, because the
1260 /// constant being right and the zip entry carrying it are two different facts — and it
1261 /// is the second one that turns into a run-hook timeout.
1262 ///
1263 /// The permission bits are masked out of the returned mode: `zip` reports the whole
1264 /// Unix mode, so the entry reads as `0o100755` — `S_IFREG | 0o755`. Comparing the
1265 /// unmasked value against `0o755` fails while the archive is perfectly correct, which
1266 /// is a test that would have sent someone to debug the writer.
1267 #[test]
1268 fn the_daemon_entry_carries_the_execute_bit() {
1269 let bytes =
1270 build_artifact(&aarch64_daemon(b"binary"), "FROM scratch\n", None).expect("zips");
1271 let mut archive =
1272 zip::ZipArchive::new(std::io::Cursor::new(bytes)).expect("a readable zip");
1273 let mode = archive
1274 .by_name("agentd")
1275 .expect("the daemon entry")
1276 .unix_mode()
1277 .expect("a mode was recorded");
1278 assert_eq!(
1279 mode & 0o777,
1280 AGENTD_MODE,
1281 "a non-executable binary surfaces as a run-hook timeout, not a permission error \
1282 (mode was {mode:o})"
1283 );
1284 assert!(
1285 mode & 0o111 != 0,
1286 "the execute bit is the whole point: {mode:o}"
1287 );
1288 }
1289
1290 /// The bytes round-trip. A zip that holds a truncated or re-encoded binary produces an
1291 /// image whose CMD fails for a reason no message names.
1292 #[test]
1293 fn the_daemon_bytes_survive_the_round_trip() {
1294 use std::io::Read as _;
1295
1296 // Deliberately includes a NUL and a high byte, which is what a real ELF has and
1297 // what a text-mode write would mangle.
1298 let binary = aarch64_daemon(&(0u8..=255).cycle().take(4096).collect::<Vec<u8>>());
1299 let bytes = build_artifact(&binary, "FROM scratch\n", None).expect("zips");
1300 let mut archive =
1301 zip::ZipArchive::new(std::io::Cursor::new(bytes)).expect("a readable zip");
1302 let mut read = Vec::new();
1303 archive
1304 .by_name("agentd")
1305 .expect("entry")
1306 .read_to_end(&mut read)
1307 .expect("reads");
1308 assert_eq!(read, binary);
1309 }
1310
1311 /// **AC-2-3, the byte-scan guard.** The agent token must not appear anywhere in the
1312 /// artifact's raw bytes — including the project-file entries #74 added.
1313 ///
1314 /// A byte scan rather than an API review, because the leak this guards against is a
1315 /// *value* turning up somewhere — in the Dockerfile as an `ENV`, in a baked config
1316 /// file, in a stray argument — not a parameter being declared. Scanning the compressed
1317 /// bytes and the decompressed entries both, since deflate would hide a literal from a
1318 /// naive scan of the archive. The artifact under scan carries a full project-file
1319 /// pair, and the entry count is asserted so the scan cannot silently sweep fewer
1320 /// members than the artifact holds.
1321 ///
1322 /// **Falsification** — add the token to the Dockerfile (`ENV AGENT_TOKEN=…`, the
1323 /// plausible mistake) and the decompressed scan fails. Re-run for the #74 members
1324 /// (2026-08-31): append the token to the lockfile entry inside `build_artifact` and
1325 /// the per-entry scan fails naming `uv.lock`; it was restored.
1326 #[test]
1327 fn the_artifact_never_carries_the_agent_token() {
1328 use std::io::Read as _;
1329
1330 let token = "s3cr3t-agent-token-do-not-bake-me";
1331 let dockerfile = default_dockerfile(9000, Some("/opt/work"), &BaseImage::al2023(), None);
1332 assert!(
1333 !dockerfile.contains(token),
1334 "the default Dockerfile must not mention a token"
1335 );
1336
1337 let project = ProjectFiles {
1338 ecosystem: Ecosystem::Uv,
1339 manifest: b"[project]\nname = \"demo\"\n".to_vec(),
1340 lockfile: b"version = 1\n".to_vec(),
1341 };
1342 let bytes =
1343 build_artifact(&aarch64_daemon(b"binary"), &dockerfile, Some(&project)).expect("zips");
1344 assert!(
1345 !bytes
1346 .windows(token.len())
1347 .any(|window| window == token.as_bytes()),
1348 "the token appears in the artifact's raw bytes"
1349 );
1350
1351 let mut archive =
1352 zip::ZipArchive::new(std::io::Cursor::new(bytes)).expect("a readable zip");
1353 assert_eq!(
1354 archive.len(),
1355 4,
1356 "the scan below must sweep every member this artifact can hold"
1357 );
1358 for index in 0..archive.len() {
1359 let mut entry = archive.by_index(index).expect("entry");
1360 let name = entry.name().expect("a name").into_owned();
1361 let mut content = Vec::new();
1362 entry.read_to_end(&mut content).expect("reads");
1363 assert!(
1364 !content
1365 .windows(token.len())
1366 .any(|window| window == token.as_bytes()),
1367 "the token appears inside {name} — the image snapshot is shared by every VM \
1368 launched from it, so a token here is a token shared with every VM"
1369 );
1370 }
1371 }
1372
1373 /// **#74, the build-context half.** A project's manifest and lockfile enter the
1374 /// archive under their ecosystem's fixed names, at the root beside the Dockerfile —
1375 /// the build context root, where the generated `COPY` finds them — and their bytes
1376 /// survive the round trip.
1377 #[test]
1378 fn project_files_enter_the_artifact_under_their_ecosystem_names() {
1379 use std::io::Read as _;
1380
1381 for (ecosystem, manifest_name, lockfile_name) in [
1382 (Ecosystem::Uv, "pyproject.toml", "uv.lock"),
1383 (Ecosystem::Npm, "package.json", "package-lock.json"),
1384 (Ecosystem::Cargo, "Cargo.toml", "Cargo.lock"),
1385 ] {
1386 let project = ProjectFiles {
1387 ecosystem,
1388 manifest: b"manifest-bytes".to_vec(),
1389 lockfile: b"lockfile-bytes".to_vec(),
1390 };
1391 let bytes =
1392 build_artifact(&aarch64_daemon(b"binary"), "FROM scratch\n", Some(&project))
1393 .expect("zips");
1394 let mut archive =
1395 zip::ZipArchive::new(std::io::Cursor::new(bytes)).expect("a readable zip");
1396
1397 let mut names: Vec<String> = (0..archive.len())
1398 .map(|index| {
1399 archive
1400 .by_index(index)
1401 .expect("entry")
1402 .name()
1403 .expect("a name")
1404 .into_owned()
1405 })
1406 .collect();
1407 names.sort();
1408 let mut expected = vec![
1409 "Dockerfile".to_string(),
1410 "agentd".to_string(),
1411 manifest_name.to_string(),
1412 lockfile_name.to_string(),
1413 ];
1414 expected.sort();
1415 assert_eq!(names, expected, "{ecosystem:?}");
1416
1417 let mut lockfile = Vec::new();
1418 archive
1419 .by_name(lockfile_name)
1420 .expect("the lockfile entry")
1421 .read_to_end(&mut lockfile)
1422 .expect("reads");
1423 assert_eq!(lockfile, b"lockfile-bytes", "{ecosystem:?}");
1424 }
1425 }
1426
1427 /// **#74's TRAP-5 design guard, the compile surface.** [`ProjectFiles`] has no field
1428 /// that could carry an entry *name* — the names come from the [`Ecosystem`] variant —
1429 /// so a caller cannot put `.env` or a credentials file into the shared snapshot
1430 /// through this type. Asserted by destructuring, the same shape as TRAP-1's
1431 /// `no_request_type_carries_a_caller_supplied_client_token`: a `name`/`path` field
1432 /// added later fails to compile this test.
1433 #[test]
1434 fn project_files_carry_no_caller_named_entry() {
1435 let ProjectFiles {
1436 ecosystem: _,
1437 manifest: _,
1438 lockfile: _,
1439 } = ProjectFiles {
1440 ecosystem: Ecosystem::Npm,
1441 manifest: Vec::new(),
1442 lockfile: Vec::new(),
1443 };
1444 }
1445
1446 /// The content hash is a pure function of the two inputs: equal inputs agree, either
1447 /// input changing changes it, and the boundary between the two inputs is part of the
1448 /// identity.
1449 ///
1450 /// The boundary case is the one worth spelling out: without length prefixes,
1451 /// `("ab", "c")` and `("a", "bc")` hash the same concatenation — and a Dockerfile
1452 /// edit could then collide with a binary edit, serving a stale image for changed
1453 /// inputs, which is the exact hazard the hash exists to close.
1454 #[test]
1455 fn the_content_hash_follows_the_inputs_and_only_the_inputs() {
1456 let hash = artifact_content_hash(b"binary-bytes", "FROM scratch\n", None);
1457 assert_eq!(
1458 hash,
1459 artifact_content_hash(b"binary-bytes", "FROM scratch\n", None)
1460 );
1461 assert_eq!(hash.len(), 64, "sha256 as lowercase hex");
1462 assert!(
1463 hash.chars()
1464 .all(|c| c.is_ascii_hexdigit() && !c.is_ascii_uppercase())
1465 );
1466
1467 assert_ne!(
1468 hash,
1469 artifact_content_hash(b"other-bytes", "FROM scratch\n", None)
1470 );
1471 assert_ne!(
1472 hash,
1473 artifact_content_hash(b"binary-bytes", "FROM scratch\nRUN true\n", None)
1474 );
1475 assert_ne!(
1476 artifact_content_hash(b"ab", "c", None),
1477 artifact_content_hash(b"a", "bc", None),
1478 "the input boundary is part of the identity"
1479 );
1480
1481 // The projectless digest is pinned to the exact value this function produced
1482 // before #74. Every image name `build --reuse` ever derived embeds this digest's
1483 // first twelve characters, so changing the no-project stream silently orphans
1484 // every existing reuse — the module docs call that hazard out for the zip's own
1485 // bytes, and it applies to the algorithm the same way. Recomputed independently
1486 // (Python hashlib, 2026-08-31) rather than pasted from this function's output.
1487 assert_eq!(
1488 hash, "90a0ef925a4696b0c0f142241b2c714e63f40071d0d02951e26cb2a1d8bb24fc",
1489 "the no-project digest must stay what it was before #74"
1490 );
1491 }
1492
1493 /// **#74 step 3, the key.** The project files are part of the artifact identity:
1494 /// identical pairs agree, a lockfile edit is a new identity (and so a fresh image
1495 /// name and a fresh build), and the entry names count — the same bytes under a
1496 /// different ecosystem are a different identity, because the Dockerfile that consumes
1497 /// them installs a different environment.
1498 ///
1499 /// **Falsification** — run 2026-08-31 and again 2026-09-02: the lockfile's
1500 /// `hasher.update` lines were dropped from `artifact_content_hash` and the
1501 /// lockfile-edit assertion went red (two different lockfiles, one hash — the
1502 /// stale-reuse hazard verbatim); restored.
1503 #[test]
1504 fn the_content_hash_keys_on_the_lockfile() {
1505 let project = |ecosystem, lockfile: &[u8]| ProjectFiles {
1506 ecosystem,
1507 manifest: b"[project]".to_vec(),
1508 lockfile: lockfile.to_vec(),
1509 };
1510 let hash = artifact_content_hash(
1511 b"binary",
1512 "FROM scratch\n",
1513 Some(&project(Ecosystem::Uv, b"version = 1")),
1514 );
1515 assert_eq!(
1516 hash,
1517 artifact_content_hash(
1518 b"binary",
1519 "FROM scratch\n",
1520 Some(&project(Ecosystem::Uv, b"version = 1")),
1521 ),
1522 "identical dependency files share an image"
1523 );
1524 assert_ne!(
1525 hash,
1526 artifact_content_hash(
1527 b"binary",
1528 "FROM scratch\n",
1529 Some(&project(Ecosystem::Uv, b"version = 2")),
1530 ),
1531 "a lockfile edit is a new image name and a fresh build"
1532 );
1533 assert_ne!(
1534 hash,
1535 artifact_content_hash(b"binary", "FROM scratch\n", None),
1536 "carrying a layer is a different identity from carrying none"
1537 );
1538 assert_ne!(
1539 artifact_content_hash(
1540 b"binary",
1541 "FROM scratch\n",
1542 Some(&project(Ecosystem::Npm, b"version = 1")),
1543 ),
1544 hash,
1545 "the entry names count: the same bytes under another ecosystem install a \
1546 different environment"
1547 );
1548 }
1549
1550 /// The Dockerfile's `FROM` comes from the base image, so the two cannot disagree.
1551 #[test]
1552 fn the_default_dockerfile_derives_its_from_from_the_base_image() {
1553 let base = BaseImage::al2023();
1554 let dockerfile = default_dockerfile(9000, None, &base, None);
1555 assert_eq!(
1556 dockerfile_from_ref(&dockerfile),
1557 Some(base.docker_ref.as_str())
1558 );
1559 assert!(dockerfile.contains(r#"CMD ["/agentd"]"#));
1560 assert!(
1561 dockerfile.contains("ENTRYPOINT []"),
1562 "the trust boundary rests on this line"
1563 );
1564 assert!(dockerfile.contains("ENV AGENTD_PORT=9000"));
1565 }
1566
1567 /// A workdir is written as both a `mkdir` and a `WORKDIR`, and is absent when none was
1568 /// asked for. An empty string counts as none, since that is what an unset value looks
1569 /// like coming from a CLI flag.
1570 #[test]
1571 fn a_workdir_is_created_and_set_or_absent_entirely() {
1572 let base = BaseImage::al2023();
1573 let with = default_dockerfile(9000, Some("/opt/baked-workdir"), &base, None);
1574 assert!(with.contains("RUN mkdir -p /opt/baked-workdir"));
1575 assert!(with.contains("WORKDIR /opt/baked-workdir"));
1576
1577 for none in [None, Some("")] {
1578 let without = default_dockerfile(9000, none, &base, None);
1579 assert!(!without.contains("WORKDIR"), "{without}");
1580 }
1581 }
1582
1583 /// **#74 step 2, the install.** A project Dockerfile copies the ecosystem's pair into
1584 /// the working directory and installs from the lockfile with the lockfile-faithful
1585 /// spelling — `uv sync --locked`, `npm ci`, `cargo fetch --locked` — where the
1586 /// projectless Dockerfile installs nothing (the measured pre-#74 state).
1587 ///
1588 /// **Falsification** — run 2026-08-31: the `COPY` line was dropped from
1589 /// `default_dockerfile` and the copies-its-pair assertion went red for all three
1590 /// ecosystems; restored.
1591 #[test]
1592 fn a_project_dockerfile_installs_from_the_lockfile_it_can_see() {
1593 let base = BaseImage::al2023();
1594 for (ecosystem, install) in [
1595 (Ecosystem::Uv, "uv sync --locked"),
1596 (Ecosystem::Npm, "npm ci"),
1597 (Ecosystem::Cargo, "cargo fetch --locked"),
1598 ] {
1599 let dockerfile = default_dockerfile(9000, None, &base, Some(ecosystem));
1600 assert!(
1601 dockerfile.contains(&format!(
1602 "COPY {} {} ./",
1603 ecosystem.manifest_name(),
1604 ecosystem.lockfile_name(),
1605 )),
1606 "copies its pair: {dockerfile}"
1607 );
1608 assert!(dockerfile.contains(install), "{dockerfile}");
1609 // The layer needs a directory even when the caller named none, because the
1610 // install resolves everything against the current directory.
1611 assert!(
1612 dockerfile.contains(&format!("WORKDIR {DEFAULT_PROJECT_WORKDIR}")),
1613 "{dockerfile}"
1614 );
1615
1616 // A caller-named workdir is where the layer lives instead.
1617 let placed = default_dockerfile(9000, Some("/srv/app"), &base, Some(ecosystem));
1618 assert!(placed.contains("WORKDIR /srv/app"), "{placed}");
1619 assert!(!placed.contains(DEFAULT_PROJECT_WORKDIR), "{placed}");
1620 }
1621
1622 let without = default_dockerfile(9000, None, &base, None);
1623 for install in ["uv sync", "npm ci", "cargo fetch", "COPY pyproject"] {
1624 assert!(!without.contains(install), "{without}");
1625 }
1626 }
1627
1628 /// The project Dockerfile passes every guard the plain one does: the `FROM` matches
1629 /// its base, the port agrees, the daemon stays the `CMD`, and the pair it copies
1630 /// satisfies [`require_project_install`]. A derived Dockerfile that failed its own
1631 /// preflight would refuse every `--project` build at the door.
1632 #[test]
1633 fn the_project_dockerfile_passes_its_own_guards() {
1634 let base = BaseImage::al2023();
1635 for ecosystem in Ecosystem::ALL {
1636 let dockerfile = default_dockerfile(9000, None, &base, Some(ecosystem));
1637 require_matching_from(&base, &dockerfile).expect("its FROM is its base");
1638 require_matching_agentd_port(9000, &dockerfile).expect("its port agrees");
1639 require_daemon_cmd(&dockerfile).expect("the daemon is still the CMD");
1640 let project = ProjectFiles {
1641 ecosystem,
1642 manifest: Vec::new(),
1643 lockfile: Vec::new(),
1644 };
1645 require_project_install(&project, &dockerfile)
1646 .expect("it copies the pair it was derived for");
1647 }
1648 }
1649
1650 /// **#74's silent-degradation guard.** A caller Dockerfile that never mentions the
1651 /// lockfile bakes no environment layer while building cleanly — the symptom is every
1652 /// launch still installing dependencies, with nothing naming the missing `COPY` — so
1653 /// it is refused up front, naming the lockfile, the install command, and both ways
1654 /// out.
1655 ///
1656 /// **Falsification** — run 2026-08-31: the `contains` check was inverted and both
1657 /// halves went red (the ignoring Dockerfile passed, the copying one was refused);
1658 /// restored.
1659 #[test]
1660 fn a_dockerfile_that_ignores_the_project_files_is_refused() {
1661 let project = ProjectFiles {
1662 ecosystem: Ecosystem::Npm,
1663 manifest: b"{}".to_vec(),
1664 lockfile: b"{}".to_vec(),
1665 };
1666 let error = require_project_install(
1667 &project,
1668 "FROM public.ecr.aws/amazonlinux/amazonlinux:2023-minimal\n\
1669 COPY agentd /agentd\n\
1670 ENTRYPOINT []\nCMD [\"/agentd\"]\n",
1671 )
1672 .expect_err("no mention of the lockfile bakes no layer");
1673 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1674 let message = error.to_string();
1675 assert!(message.contains("package-lock.json"), "{message}");
1676 assert!(message.contains("npm ci"), "{message}");
1677 assert!(message.contains("no environment layer"), "{message}");
1678
1679 require_project_install(
1680 &project,
1681 "FROM x\nCOPY package.json package-lock.json ./\nRUN npm ci\nCMD [\"/agentd\"]\n",
1682 )
1683 .expect("a Dockerfile that copies the pair agrees");
1684 }
1685
1686 /// `FROM` parsing tolerates the decoration a real Dockerfile carries: lowercase,
1687 /// leading whitespace, a `--platform` flag, and an `AS` alias.
1688 #[test]
1689 fn the_from_parser_ignores_decoration_rather_than_validating_syntax() {
1690 assert_eq!(dockerfile_from_ref("FROM alpine\n"), Some("alpine"));
1691 assert_eq!(dockerfile_from_ref(" from alpine:3\n"), Some("alpine:3"));
1692 assert_eq!(
1693 dockerfile_from_ref("FROM --platform=linux/arm64 alpine AS build\n"),
1694 Some("alpine")
1695 );
1696 assert_eq!(dockerfile_from_ref("RUN echo from nowhere\n"), None);
1697 assert_eq!(dockerfile_from_ref(""), None);
1698 }
1699
1700 /// A `FROM` that disagrees with the base image is refused, and the message names both
1701 /// refs — the one found and the one expected — because "these must agree" without both
1702 /// values leaves the caller to guess which to change.
1703 #[test]
1704 fn a_dockerfile_from_that_disagrees_with_the_base_is_refused() {
1705 let base = BaseImage::al2023();
1706 let error = require_matching_from(&base, "FROM ubuntu:24.04\nCOPY agentd /agentd\n")
1707 .expect_err("ubuntu is not the managed base");
1708 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1709 let message = error.to_string();
1710 assert!(message.contains("ubuntu:24.04"), "{message}");
1711 assert!(message.contains(&base.docker_ref), "{message}");
1712 assert!(
1713 message.contains("none of the measured platform behaviour applies"),
1714 "{message}"
1715 );
1716
1717 require_matching_from(&base, &default_dockerfile(9000, None, &base, None))
1718 .expect("the derived Dockerfile agrees with its own base");
1719 }
1720
1721 /// A digest-pinned `FROM` of the agreeing ref passes: it names one specific manifest of
1722 /// the base this check already accepts, which is the spelling a supply-chain-conscious
1723 /// caller uses. The boundary cases stay refused — a digest on a *different* ref is still
1724 /// a disagreement, and a malformed digest is not a digest.
1725 #[test]
1726 fn a_digest_pinned_from_of_the_agreeing_ref_passes() {
1727 let base = BaseImage::al2023();
1728 let digest = "c439fb4994ea7ca529233d6256446d3f8b7b4efb58956073e015303a170011de";
1729 let pinned = format!(
1730 "FROM {}@sha256:{digest}\nCOPY agentd /agentd\n",
1731 base.docker_ref
1732 );
1733 require_matching_from(&base, &pinned).expect("a digest pin of the same ref agrees");
1734
1735 require_matching_from(&base, &format!("FROM ubuntu:24.04@sha256:{digest}\n"))
1736 .expect_err("a digest does not launder a different ref");
1737 require_matching_from(&base, &format!("FROM {}@sha256:abc123\n", base.docker_ref))
1738 .expect_err("a 6-character digest is not a digest");
1739 require_matching_from(
1740 &base,
1741 &format!("FROM {}@sha256:{}\n", base.docker_ref, "g".repeat(64)),
1742 )
1743 .expect_err("64 non-hex characters are not a digest");
1744 }
1745
1746 /// A Dockerfile with no `FROM` is not this check's business: the build will say so, and
1747 /// refusing here would be a Dockerfile validator rather than an agreement check.
1748 #[test]
1749 fn a_dockerfile_with_no_from_is_left_to_the_build() {
1750 require_matching_from(&BaseImage::al2023(), "COPY agentd /agentd\n")
1751 .expect("no FROM is not a disagreement");
1752 }
1753
1754 /// The measured case, reproduced from the Dockerfile that spent it: a hand-written
1755 /// guest Dockerfile carrying `ENV AGENTD_PORT=8080` — the plausible port, and the wrong
1756 /// one — against a client sending `hooks.port=9000`. The build succeeds, the daemon logs
1757 /// `agentd listening`, and the image lands in `CREATE_FAILED` naming no port.
1758 #[test]
1759 fn a_dockerfile_port_that_disagrees_with_the_hook_port_is_refused() {
1760 let error = require_matching_agentd_port(
1761 crate::control::DEFAULT_AGENT_PORT,
1762 "FROM public.ecr.aws/amazonlinux/amazonlinux:2023-minimal\n\
1763 COPY agentd /agentd\n\
1764 ENV AGENTD_PORT=8080\n\
1765 EXPOSE 8080\n\
1766 CMD [\"/agentd\"]\n",
1767 )
1768 .expect_err("8080 is not the port the create call sends");
1769 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1770 let message = error.to_string();
1771 assert!(message.contains("8080"), "{message}");
1772 assert!(message.contains("9000"), "{message}");
1773 // The remedy is what a reader acts on, and the diagnostic's value is naming the
1774 // symptom that points away from the cause.
1775 assert!(message.contains("CREATE_FAILED"), "{message}");
1776 assert!(message.contains("ready/validate"), "{message}");
1777
1778 require_matching_agentd_port(
1779 crate::control::DEFAULT_AGENT_PORT,
1780 &default_dockerfile(
1781 crate::control::DEFAULT_AGENT_PORT,
1782 Some("/work"),
1783 &BaseImage::al2023(),
1784 None,
1785 ),
1786 )
1787 .expect("the derived Dockerfile agrees with the port it was derived from");
1788 }
1789
1790 /// Both `ENV` spellings set the variable, so a guard reading only `KEY=VALUE` would pass
1791 /// the legacy form it could not parse. The last assignment wins, as it does at build
1792 /// time.
1793 #[test]
1794 fn the_port_scan_reads_both_env_spellings_and_takes_the_last() {
1795 assert_eq!(dockerfile_agentd_port("ENV AGENTD_PORT=9000\n"), Some(9000));
1796 assert_eq!(dockerfile_agentd_port("ENV AGENTD_PORT 8080\n"), Some(8080));
1797 assert_eq!(dockerfile_agentd_port("env agentd_port=7000\n"), None);
1798 assert_eq!(
1799 dockerfile_agentd_port("ENV AGENTD_PORT=8080\nENV AGENTD_PORT=9000\n"),
1800 Some(9000),
1801 );
1802 assert_eq!(dockerfile_agentd_port("ENV AGENTD_LOG=info\n"), None);
1803 // Neither a missing variable nor an unparseable one is a disagreement: the daemon
1804 // keeps its own default for both (`crates/agentd/src/config.rs:118`).
1805 assert_eq!(
1806 dockerfile_agentd_port("FROM x\nCOPY agentd /agentd\n"),
1807 None
1808 );
1809 assert_eq!(dockerfile_agentd_port("ENV AGENTD_PORT=nine\n"), None);
1810 require_matching_agentd_port(
1811 crate::control::DEFAULT_AGENT_PORT,
1812 "FROM x\nCOPY agentd /agentd\n",
1813 )
1814 .expect("no variable agrees with the default, which is what the daemon will use");
1815 }
1816
1817 /// The other half of the pair, and the harder one to diagnose: a Dockerfile that names
1818 /// no port at all against a client that moved off the default. Silence is not neutral —
1819 /// `Config::from_env` keeps `9000` for an unset variable — so the guest listens on 9000
1820 /// while the hooks are dialled on the client's port, which is the measured failure with
1821 /// nothing in the Dockerfile to point at.
1822 #[test]
1823 fn a_dockerfile_naming_no_port_is_refused_when_the_client_moved_off_the_default() {
1824 let error = require_matching_agentd_port(
1825 8080,
1826 "FROM public.ecr.aws/amazonlinux/amazonlinux:2023-minimal\n\
1827 COPY agentd /agentd\n\
1828 CMD [\"/agentd\"]\n",
1829 )
1830 .expect_err("a silent Dockerfile leaves the daemon on 9000, not on 8080");
1831 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1832 let message = error.to_string();
1833 assert!(message.contains("8080"), "{message}");
1834 assert!(
1835 message.contains(&crate::control::DEFAULT_AGENT_PORT.to_string()),
1836 "the daemon's default is the value the reader has to learn: {message}"
1837 );
1838 assert!(message.contains("CREATE_FAILED"), "{message}");
1839 assert!(message.contains("ready/validate"), "{message}");
1840
1841 // An unparseable value has the same consequence in the guest as no value, so it is
1842 // refused for the same reason rather than passed as "not this guard's business".
1843 require_matching_agentd_port(8080, "FROM x\nENV AGENTD_PORT=nine\n")
1844 .expect_err("a value the daemon cannot parse leaves it on its own default");
1845
1846 // The default-port client is the common case and stays silent: the Dockerfile that
1847 // says nothing and the client that changed nothing already agree.
1848 require_matching_agentd_port(
1849 crate::control::DEFAULT_AGENT_PORT,
1850 "FROM x\nCOPY agentd /agentd\n",
1851 )
1852 .expect("silence agrees with the default");
1853 }
1854
1855 /// The fourth pair of the `FROM`/`WORKDIR`/`AGENTD_PORT` shape, found by sweeping for
1856 /// the others: a keepalive interval the client's silence tolerance is shorter than. The
1857 /// resulting error prints the client's own timeout as though it were the keepalive, so
1858 /// the number a reader would search for never appears.
1859 #[test]
1860 fn a_keepalive_at_or_over_the_client_idle_timeout_is_refused() {
1861 let timeout = crate::session::exec::DEFAULT_STREAM_IDLE_TIMEOUT;
1862 assert_eq!(timeout.as_secs(), 60, "the guard's arithmetic assumes this");
1863
1864 let error =
1865 require_keepalive_under_idle_timeout(timeout, "ENV AGENTD_SSE_KEEPALIVE_SECS=90\n")
1866 .expect_err("90s of scheduled silence exceeds a 60s tolerance");
1867 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1868 let message = error.to_string();
1869 assert!(message.contains("90"), "{message}");
1870 assert!(message.contains("60"), "{message}");
1871
1872 // Equality races rather than passing: the timeout fires on the same tick the
1873 // keepalive is due.
1874 require_keepalive_under_idle_timeout(timeout, "ENV AGENTD_SSE_KEEPALIVE_SECS=60\n")
1875 .expect_err("an interval equal to the tolerance is a race, not a margin");
1876
1877 require_keepalive_under_idle_timeout(timeout, "ENV AGENTD_SSE_KEEPALIVE_SECS 30\n")
1878 .expect("30s leaves a margin, in the legacy spelling");
1879 // The daemon's own default is 15s, so silence is safe for this pair.
1880 require_keepalive_under_idle_timeout(timeout, "FROM x\nCOPY agentd /agentd\n")
1881 .expect("an unset keepalive leaves the daemon at 15s");
1882 require_keepalive_under_idle_timeout(
1883 timeout,
1884 &default_dockerfile(
1885 crate::control::DEFAULT_AGENT_PORT,
1886 Some("/work"),
1887 &BaseImage::al2023(),
1888 None,
1889 ),
1890 )
1891 .expect("the derived Dockerfile sets no keepalive");
1892 }
1893
1894 /// **Issue #46, both refusable halves.** A Dockerfile with no `CMD` builds an image
1895 /// that runs the base's default process; a non-empty `ENTRYPOINT` turns the `CMD` into
1896 /// its arguments. Either way the daemon the artifact carries never starts, and the
1897 /// failure surfaces as `CREATE_FAILED` or a run-hook timeout naming neither
1898 /// instruction — so the message must name both the instruction and that symptom.
1899 #[test]
1900 fn a_dockerfile_that_never_runs_the_daemon_is_refused_naming_the_symptom() {
1901 // No CMD at all: the base's default applies, which is not /agentd.
1902 let error = require_daemon_cmd(
1903 "FROM public.ecr.aws/amazonlinux/amazonlinux:2023-minimal\n\
1904 COPY agentd /agentd\n\
1905 ENV AGENTD_PORT=9000\n",
1906 )
1907 .expect_err("no CMD means the daemon never starts");
1908 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1909 let message = error.to_string();
1910 assert!(message.contains("no CMD"), "{message}");
1911 assert!(message.contains("run-hook timeout"), "{message}");
1912 assert!(message.contains("CREATE_FAILED"), "{message}");
1913
1914 // A non-empty ENTRYPOINT swallows the CMD as its arguments.
1915 let error = require_daemon_cmd(
1916 "FROM public.ecr.aws/amazonlinux/amazonlinux:2023-minimal\n\
1917 COPY agentd /agentd\n\
1918 ENTRYPOINT [\"/bin/sh\", \"-c\"]\n\
1919 CMD [\"/agentd\"]\n",
1920 )
1921 .expect_err("a non-empty ENTRYPOINT makes CMD its arguments");
1922 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1923 let message = error.to_string();
1924 assert!(message.contains("ENTRYPOINT"), "{message}");
1925 assert!(message.contains("run-hook timeout"), "{message}");
1926
1927 // An empty CMD is the same absence in different spelling.
1928 require_daemon_cmd("FROM x\nCMD []\n").expect_err("CMD [] runs nothing");
1929 }
1930
1931 /// The shapes that must pass: the deployment invariant itself, in exec and spaced
1932 /// spellings, and the invariant's `ENTRYPOINT []` with any interior whitespace. The
1933 /// derived Dockerfile passes its own guard, which is what keeps the default path
1934 /// unaffected.
1935 #[test]
1936 fn the_deployment_invariant_passes_the_daemon_cmd_guard() {
1937 require_daemon_cmd(&default_dockerfile(
1938 9000,
1939 Some("/work"),
1940 &BaseImage::al2023(),
1941 None,
1942 ))
1943 .expect("the derived Dockerfile is the invariant");
1944 require_daemon_cmd("FROM x\nENTRYPOINT []\nCMD [\"/agentd\"]\n").expect("the invariant");
1945 require_daemon_cmd("FROM x\nENTRYPOINT [ ]\nCMD [\"/agentd\"]\n")
1946 .expect("interior whitespace is still the empty exec form");
1947 require_daemon_cmd("FROM x\ncmd [\"/agentd\"]\n").expect("case-insensitive, no ENTRYPOINT");
1948 // Last instruction wins, as at build time: a later empty ENTRYPOINT un-swallows.
1949 require_daemon_cmd("FROM x\nENTRYPOINT [\"/bin/sh\"]\nENTRYPOINT []\nCMD [\"/agentd\"]\n")
1950 .expect("the last ENTRYPOINT is the one the build uses");
1951 // And the reverse ordering is refused for the same reason.
1952 require_daemon_cmd("FROM x\nENTRYPOINT []\nENTRYPOINT [\"/bin/sh\"]\nCMD [\"/agentd\"]\n")
1953 .expect_err("the last ENTRYPOINT is non-empty");
1954 }
1955
1956 /// Workdir inheritance is refused when neither the base nor the Dockerfile declares
1957 /// one, and the message says where the symptom would have appeared — in the guest, a
1958 /// build cycle later.
1959 #[test]
1960 fn inheriting_a_workdir_nothing_declares_is_refused() {
1961 let base = BaseImage::al2023();
1962 assert!(base.working_dir.is_empty(), "the measured case");
1963
1964 let error = require_workdir(&base, None).expect_err("nothing declares a workdir");
1965 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1966 let message = error.to_string();
1967 assert!(message.contains("no WorkingDir"), "{message}");
1968 assert!(message.contains("nothing to inherit"), "{message}");
1969 assert!(message.contains("docs/PLATFORM.md"), "{message}");
1970 }
1971
1972 /// It is accepted when *either* side declares one — the Dockerfile's `WORKDIR` or the
1973 /// base image's own `working_dir`. Both directions, because a check that only read the
1974 /// Dockerfile would refuse a caller on a purpose-built image that declares one.
1975 #[test]
1976 fn either_the_dockerfile_or_the_base_may_supply_the_workdir() {
1977 let base = BaseImage::al2023();
1978 require_workdir(&base, Some("FROM x\nWORKDIR /srv\n"))
1979 .expect("the Dockerfile declares one");
1980
1981 let purpose_built = BaseImage {
1982 working_dir: "/app".to_string(),
1983 ..BaseImage::al2023()
1984 };
1985 require_workdir(&purpose_built, None).expect("the base image declares one");
1986 }
1987
1988 /// A bare `WORKDIR` with no argument does not count as declaring one. It is the shape a
1989 /// truncated edit leaves behind, and treating it as a declaration would pass the guard
1990 /// on a Dockerfile that sets nothing.
1991 #[test]
1992 fn a_workdir_with_no_argument_does_not_count_as_a_declaration() {
1993 let error = require_workdir(&BaseImage::al2023(), Some("FROM x\nWORKDIR\n"))
1994 .expect_err("WORKDIR with no path declares nothing");
1995 assert_eq!(error.kind(), ErrorKind::InvalidArg);
1996 }
1997
1998 /// **IMAGE-1, the acceptance test #220 names.** Wrapping a bare `FROM` produces exactly the
1999 /// default Dockerfile for that base, so the stanza has one source: the default generator's
2000 /// output minus its `FROM` line is what a wrap appends. Checked with and without a workdir,
2001 /// and on a non-default port, because each of those is a branch of the one stanza.
2002 #[test]
2003 fn wrapping_a_bare_from_is_the_default_dockerfile_minus_nothing() {
2004 for (port, workdir) in [
2005 (DEFAULT_AGENT_PORT_FOR_TESTS, None),
2006 (DEFAULT_AGENT_PORT_FOR_TESTS, Some("/srv/task")),
2007 (8080, None),
2008 ] {
2009 let base = BaseImage {
2010 docker_ref: "x".to_string(),
2011 ..BaseImage::al2023()
2012 };
2013 let default = default_dockerfile(port, workdir, &base, None);
2014 let opts = WrapOptions {
2015 port,
2016 workdir: workdir.map(str::to_string),
2017 ..WrapOptions::default()
2018 };
2019 assert_eq!(
2020 wrap_dockerfile("FROM x\n", &opts).expect("a bare FROM wraps"),
2021 default,
2022 "IMAGE-1: port {port} workdir {workdir:?}"
2023 );
2024 let (from, stanza) = default.split_once('\n').expect("lines");
2025 assert_eq!(from, "FROM x");
2026 assert!(stanza.ends_with("CMD [\"/agentd\"]\n"), "{stanza}");
2027 }
2028 }
2029
2030 /// **IMAGE-2.** Whatever the task set — its own entrypoint, command, port, user — the
2031 /// wrapped result ends on the bootstrap invariant with the client's port, and a task that
2032 /// ends on another user has `USER root` restored ahead of the stanza. A task that ends on
2033 /// root, or never changes user, gets no extra line, which is what keeps IMAGE-1's equality.
2034 #[test]
2035 fn a_wrapped_task_ends_on_the_invariant_with_root_restored_when_needed() {
2036 let task = "FROM python:3.12-slim\nWORKDIR /app\nENV AGENTD_PORT=8080\n\
2037 ENTRYPOINT [\"/bin/sh\", \"-c\"]\nCMD [\"python\", \"serve.py\"]\nUSER app\n";
2038 let wrapped = wrap_dockerfile(task, &WrapOptions::default()).expect("wraps");
2039 let added = wrapped
2040 .strip_prefix(task)
2041 .expect("the task text is kept verbatim");
2042 assert!(
2043 added.starts_with("USER root\nCOPY agentd /agentd\n"),
2044 "{added}"
2045 );
2046 assert_eq!(dockerfile_entrypoint(&wrapped), Some("[]"));
2047 assert_eq!(dockerfile_cmd(&wrapped), Some(r#"["/agentd"]"#));
2048 assert_eq!(dockerfile_agentd_port(&wrapped), Some(9000));
2049 require_daemon_cmd(&wrapped).expect("the daemon is the CMD");
2050 require_matching_agentd_port(9000, &wrapped).expect("the stanza's port wins");
2051
2052 for (last_user, restored) in [
2053 ("USER root", false),
2054 ("USER 0", false),
2055 ("USER root:root", false),
2056 ("USER 0:0", false),
2057 ("user app", true),
2058 ("USER 1000:1000", true),
2059 ("USER root:staff", true),
2060 ("USER ${TASK_USER}", true),
2061 ] {
2062 let task = format!("FROM x\n{last_user}\n");
2063 let added = wrap_dockerfile(&task, &WrapOptions::default())
2064 .expect("wraps")
2065 .strip_prefix(task.as_str())
2066 .expect("verbatim")
2067 .to_string();
2068 assert_eq!(
2069 added.starts_with("USER root\n"),
2070 restored,
2071 "IMAGE-2: {last_user:?}: {added}"
2072 );
2073 }
2074
2075 // Only the last USER counts, as at build time.
2076 let task = "FROM x\nUSER app\nUSER root\n";
2077 let wrapped = wrap_dockerfile(task, &WrapOptions::default()).expect("wraps");
2078 let added = wrapped.strip_prefix(task).expect("verbatim");
2079 assert!(added.starts_with("COPY agentd /agentd\n"), "{added}");
2080 }
2081
2082 /// **IMAGE-2, the newline.** A task file without a trailing newline gets one before the
2083 /// stanza, in either line ending, so the stanza's first line is its own instruction.
2084 #[test]
2085 fn a_task_without_a_trailing_newline_is_normalized() {
2086 let wrapped = wrap_dockerfile("FROM x", &WrapOptions::default()).expect("wraps");
2087 assert!(
2088 wrapped.starts_with("FROM x\nCOPY agentd /agentd\n"),
2089 "{wrapped}"
2090 );
2091 let wrapped =
2092 wrap_dockerfile("FROM x\r\nRUN true", &WrapOptions::default()).expect("wraps");
2093 assert!(
2094 wrapped.starts_with("FROM x\r\nRUN true\nCOPY agentd /agentd\n"),
2095 "{wrapped}"
2096 );
2097 let wrapped = wrap_dockerfile("FROM x\r\n", &WrapOptions::default()).expect("wraps");
2098 assert!(wrapped.starts_with("FROM x\r\nCOPY agentd"), "{wrapped}");
2099 }
2100
2101 /// **IMAGE-3.** Each refusal on its own, against an otherwise wrappable task, names its
2102 /// cause: no `FROM`, a trailing continuation (in the default and a custom escape
2103 /// character, across blank and comment lines, in either line ending), an open heredoc, a
2104 /// keepalive the client cannot tolerate, a bad workdir or port, and inheriting a workdir
2105 /// nothing declares.
2106 #[test]
2107 fn a_task_the_stanza_cannot_follow_is_refused_naming_the_cause() {
2108 let refused = |task: &str, opts: WrapOptions, cause: &str| {
2109 let error = wrap_dockerfile(task, &opts).expect_err(cause);
2110 assert_eq!(error.kind(), ErrorKind::InvalidArg, "{error}");
2111 let message = error.to_string();
2112 assert!(message.contains(cause), "IMAGE-3 {cause:?}: {message}");
2113 };
2114 let none = WrapOptions::default;
2115 refused("RUN echo hello\n", none(), "no FROM");
2116 refused("", none(), "no FROM");
2117 refused("FROM\n", none(), "no FROM");
2118 refused(
2119 "FROM x\nRUN apt-get update && \\\n",
2120 none(),
2121 "line continuation",
2122 );
2123 refused("FROM x\nRUN a \\ \n", none(), "line continuation");
2124 refused("FROM x\nRUN a \\\r\n", none(), "line continuation");
2125 refused(
2126 "FROM x\nRUN a \\\n\n# comment\n",
2127 none(),
2128 "line continuation",
2129 );
2130 refused("# escape=`\nFROM x\nRUN a `\n", none(), "line continuation");
2131 refused("FROM x\nRUN <<EOF\necho never closed\n", none(), "heredoc");
2132 refused("FROM x\nRUN <<-\"END\"\necho x\n", none(), "heredoc");
2133 refused("FROM x\nCOPY <<A <<B /dst/\na\nA\nb\n", none(), "heredoc");
2134 refused(
2135 "FROM x\nENV AGENTD_SSE_KEEPALIVE_SECS=90\n",
2136 none(),
2137 "AGENTD_SSE_KEEPALIVE_SECS=90",
2138 );
2139 for workdir in ["relative/dir", "/srv\nRUN evil", "/my dir", "/tab\there"] {
2140 refused(
2141 "FROM x\n",
2142 WrapOptions {
2143 workdir: Some(workdir.to_string()),
2144 ..none()
2145 },
2146 "absolute path",
2147 );
2148 }
2149 refused("FROM x\n", WrapOptions { port: 0, ..none() }, "port");
2150 refused(
2151 "FROM x\n",
2152 WrapOptions {
2153 inherit_workdir: true,
2154 ..none()
2155 },
2156 "nothing to inherit",
2157 );
2158 }
2159
2160 /// **IMAGE-3, the other side.** What looks like a refusal but is not: a backslash under a
2161 /// backtick escape, a continuation that was finished, a closed heredoc (including a `<<-`
2162 /// terminator indented with tabs), a here-string, an escaped keepalive under the timeout,
2163 /// and inheriting a workdir the task or the options declare.
2164 #[test]
2165 fn a_finished_task_is_wrapped_even_when_it_looks_unfinished() {
2166 let none = WrapOptions::default;
2167 for task in [
2168 "# escape=`\nFROM x\nRUN a \\\n",
2169 "FROM x\nRUN a \\\n && b\n",
2170 "FROM x\nRUN <<EOF\necho closed\nEOF\n",
2171 "FROM x\nRUN <<-EOF\n\techo closed\n\tEOF\n",
2172 "FROM x\nCOPY <<A <<B /dst/\na\nA\nb\nB\n",
2173 "FROM x\nRUN cat <<< word\n",
2174 "FROM x\n# a trailing comment \\\n",
2175 "FROM x\nENV AGENTD_SSE_KEEPALIVE_SECS=30\n",
2176 ] {
2177 wrap_dockerfile(task, &none())
2178 .unwrap_or_else(|error| panic!("{task:?} was refused: {error}"));
2179 }
2180 wrap_dockerfile(
2181 "FROM x\nWORKDIR /app\n",
2182 &WrapOptions {
2183 inherit_workdir: true,
2184 ..none()
2185 },
2186 )
2187 .expect("the task declares one");
2188 wrap_dockerfile(
2189 "FROM x\n",
2190 &WrapOptions {
2191 inherit_workdir: true,
2192 workdir: Some("/srv".to_string()),
2193 ..none()
2194 },
2195 )
2196 .expect("the options declare one");
2197 wrap_dockerfile(
2198 "FROM x\n",
2199 &WrapOptions {
2200 workdir: Some(String::new()),
2201 ..none()
2202 },
2203 )
2204 .expect("an empty workdir is none, as default_dockerfile reads it");
2205 }
2206
2207 /// **IMAGE-4.** A derived base takes its ref from the first `FROM`, keeps the managed
2208 /// base's name (so `baseImageArn` is unchanged) and an empty working dir, and pairs with
2209 /// the Dockerfile it came from under the `FROM` guard — digest pins included, since the
2210 /// ref is taken whole. A Dockerfile with no `FROM` has nothing to derive.
2211 #[test]
2212 fn a_base_derived_from_a_dockerfile_pairs_with_it_by_construction() {
2213 let digest = "c439fb4994ea7ca529233d6256446d3f8b7b4efb58956073e015303a170011de";
2214 for (task, docker_ref) in [
2215 ("FROM python:3.12-slim\n", "python:3.12-slim".to_string()),
2216 (
2217 "FROM --platform=linux/arm64 golang:1.23 AS build\nFROM python:3.12-slim\n",
2218 "golang:1.23".to_string(),
2219 ),
2220 (
2221 &*format!("FROM ubuntu:24.04@sha256:{digest}\n"),
2222 format!("ubuntu:24.04@sha256:{digest}"),
2223 ),
2224 ] {
2225 let base = BaseImage::from_dockerfile(task).expect("a FROM");
2226 assert_eq!(base.docker_ref, docker_ref);
2227 assert_eq!(
2228 base.name,
2229 BaseImage::al2023().name,
2230 "IMAGE-4 keeps the managed base"
2231 );
2232 assert_eq!(base.working_dir, "");
2233 let wrapped = wrap_dockerfile(task, &WrapOptions::default()).expect("wraps");
2234 require_matching_from(&base, &wrapped).expect("IMAGE-4: the pair agrees");
2235 assert_eq!(BaseImage::from_dockerfile(&wrapped).expect("a FROM"), base);
2236 }
2237 // The inversion #220 removes: under the managed base, a task on another base is
2238 // refused.
2239 require_matching_from(
2240 &BaseImage::al2023(),
2241 &wrap_dockerfile("FROM python:3.12-slim\n", &WrapOptions::default()).expect("wraps"),
2242 )
2243 .expect_err("the managed base does not pair with python:3.12-slim");
2244
2245 let error = BaseImage::from_dockerfile("RUN echo hello\n").expect_err("no FROM");
2246 assert_eq!(error.kind(), ErrorKind::InvalidArg);
2247 assert!(error.to_string().contains("no FROM"), "{error}");
2248 }
2249
2250 /// **The table `crates/model/src/wrap.rs` specifies**, rendered as real Dockerfile text: every
2251 /// task and option combination the model enumerates, and the refusal (or the `USER root`
2252 /// verdict) the model's `specified` gives it. The model crate cannot be a dependency here
2253 /// (this crate is published and the model isn't), so the table is restated as `expected`
2254 /// below, and `crates/model-conformance/tests/tables.rs` drives this function over the model's rows.
2255 #[test]
2256 fn wrap_agrees_with_the_model_table() {
2257 #[derive(Debug)]
2258 enum Expect {
2259 Refused(&'static str),
2260 Wrapped { user_root: bool },
2261 }
2262 let digest = "c439fb4994ea7ca529233d6256446d3f8b7b4efb58956073e015303a170011de";
2263 let managed = BaseImage::al2023().docker_ref;
2264 let froms = [
2265 None,
2266 Some(managed.clone()),
2267 Some(format!("{managed}@sha256:{digest}")),
2268 Some("python:3.12-slim".to_string()),
2269 ];
2270 let mut checked = 0;
2271 for from in &froms {
2272 for tail in 0..3 {
2273 for user in 0..3 {
2274 for declares_workdir in [false, true] {
2275 for keepalive_too_long in [false, true] {
2276 for workdir in 0..3 {
2277 for inherit_workdir in [false, true] {
2278 let mut text = String::new();
2279 match from {
2280 Some(from) => text.push_str(&format!("FROM {from}\n")),
2281 None => text.push_str("RUN echo no base\n"),
2282 }
2283 if declares_workdir {
2284 text.push_str("WORKDIR /app\n");
2285 }
2286 if keepalive_too_long {
2287 text.push_str("ENV AGENTD_SSE_KEEPALIVE_SECS=60\n");
2288 }
2289 match user {
2290 1 => text.push_str("USER root\n"),
2291 2 => text.push_str("USER app\n"),
2292 _ => {}
2293 }
2294 match tail {
2295 1 => text.push_str("RUN make \\\n"),
2296 2 => text.push_str("RUN <<EOF\necho open\n"),
2297 _ => {}
2298 }
2299 let opts = WrapOptions {
2300 workdir: match workdir {
2301 1 => Some("/srv".to_string()),
2302 2 => Some("srv".to_string()),
2303 _ => None,
2304 },
2305 inherit_workdir,
2306 ..WrapOptions::default()
2307 };
2308 let expected = if from.is_none() {
2309 Expect::Refused("no FROM")
2310 } else if tail != 0 {
2311 Expect::Refused(if tail == 1 {
2312 "line continuation"
2313 } else {
2314 "heredoc"
2315 })
2316 } else if keepalive_too_long {
2317 Expect::Refused("AGENTD_SSE_KEEPALIVE_SECS")
2318 } else if workdir == 2 {
2319 Expect::Refused("absolute path")
2320 } else if inherit_workdir && !declares_workdir && workdir == 0 {
2321 Expect::Refused("nothing to inherit")
2322 } else {
2323 Expect::Wrapped {
2324 user_root: user == 2,
2325 }
2326 };
2327 let got = wrap_dockerfile(&text, &opts);
2328 match (&expected, &got) {
2329 (Expect::Refused(cause), Err(error)) => assert!(
2330 error.to_string().contains(cause),
2331 "{expected:?} for {text:?} {opts:?}: {error}"
2332 ),
2333 (Expect::Wrapped { user_root }, Ok(out)) => assert_eq!(
2334 out.strip_prefix(text.as_str())
2335 .expect("verbatim")
2336 .starts_with("USER root\n"),
2337 *user_root,
2338 "{text:?} {opts:?}"
2339 ),
2340 _ => panic!(
2341 "{expected:?} for {text:?} {opts:?}, got {got:?}"
2342 ),
2343 }
2344 checked += 1;
2345 }
2346 }
2347 }
2348 }
2349 }
2350 }
2351 }
2352 assert_eq!(
2353 checked,
2354 4 * 3 * 3 * 2 * 2 * 3 * 2,
2355 "every combination the model has"
2356 );
2357 }
2358
2359 fn context_of(entries: &[(&str, u32, &[u8])]) -> crate::control::context::BuildContext {
2360 crate::control::context::BuildContext::from_entries(
2361 entries
2362 .iter()
2363 .map(
2364 |(name, mode, bytes)| crate::control::context::ContextEntry {
2365 name: (*name).to_string(),
2366 mode: *mode,
2367 bytes: bytes.to_vec(),
2368 },
2369 )
2370 .collect(),
2371 )
2372 .expect("valid entries")
2373 }
2374
2375 /// **IMAGE-6, the pin.** A request with no build context — `None`, or an empty one — hashes
2376 /// to exactly what `artifact_content_hash` produced before #221, so no `build --reuse` or
2377 /// `AgentVm` image name moves. The vector is the one pinned for #74, recomputed
2378 /// independently with Python's hashlib then.
2379 #[test]
2380 fn a_request_without_a_context_keeps_its_digest() {
2381 let pinned = "90a0ef925a4696b0c0f142241b2c714e63f40071d0d02951e26cb2a1d8bb24fc";
2382 assert_eq!(
2383 artifact_content_hash_with_context(b"binary-bytes", "FROM scratch\n", None, None),
2384 pinned,
2385 "IMAGE-6: the no-context digest is the pre-#221 digest"
2386 );
2387 let empty = crate::control::context::BuildContext::default();
2388 assert_eq!(
2389 artifact_content_hash_with_context(
2390 b"binary-bytes",
2391 "FROM scratch\n",
2392 None,
2393 Some(&empty)
2394 ),
2395 pinned,
2396 "an empty context carries no extra entries, so it is the same artifact"
2397 );
2398 assert_eq!(
2399 artifact_content_hash(b"binary-bytes", "FROM scratch\n", None),
2400 pinned
2401 );
2402 }
2403
2404 /// **IMAGE-6, the extension.** Each entry's name, mode, and bytes are part of the
2405 /// identity; the order entries were read in is not, because a context is sorted by name.
2406 #[test]
2407 fn the_context_hash_follows_every_entry_and_not_their_order() {
2408 let hash = |context: &crate::control::context::BuildContext| {
2409 artifact_content_hash_with_context(b"bin", "FROM x\n", None, Some(context))
2410 };
2411 let base = hash(&context_of(&[
2412 ("a.txt", 0o644, b"a"),
2413 ("b.txt", 0o644, b"b"),
2414 ]));
2415 assert_ne!(
2416 base,
2417 artifact_content_hash_with_context(b"bin", "FROM x\n", None, None),
2418 "IMAGE-6: a context is part of the identity"
2419 );
2420 assert_eq!(
2421 base,
2422 hash(&context_of(&[
2423 ("b.txt", 0o644, b"b"),
2424 ("a.txt", 0o644, b"a")
2425 ])),
2426 "the read order is not"
2427 );
2428 for (changed, what) in [
2429 (
2430 context_of(&[("a.txt", 0o644, b"A"), ("b.txt", 0o644, b"b")]),
2431 "bytes",
2432 ),
2433 (
2434 context_of(&[("a.txt", 0o755, b"a"), ("b.txt", 0o644, b"b")]),
2435 "mode",
2436 ),
2437 (
2438 context_of(&[("c.txt", 0o644, b"a"), ("b.txt", 0o644, b"b")]),
2439 "name",
2440 ),
2441 (context_of(&[("a.txt", 0o644, b"ab")]), "boundary"),
2442 ] {
2443 assert_ne!(base, hash(&changed), "IMAGE-6: the entry {what} counts");
2444 }
2445 }
2446
2447 /// **IMAGE-7, the zip.** Equal inputs make byte-identical artifacts, so equal inputs are
2448 /// one S3 object: every entry carries the DOS epoch as its date and a fixed mode, the
2449 /// context entries sit beside the Dockerfile at the root, and the Dockerfile entry is the
2450 /// one passed in. The same build a second apart, which needs a real sleep this crate
2451 /// can't make, is `crates/microvms-core/tests/artifact_across_seconds.rs`.
2452 #[test]
2453 fn the_artifact_is_deterministic_with_fixed_dates_and_modes() {
2454 use std::io::Read as _;
2455
2456 let context = context_of(&[("app/run.sh", 0o755, b"#!/bin/sh\n"), ("data", 0o644, b"d")]);
2457 let build = || {
2458 build_artifact_with_context(
2459 &aarch64_daemon(b"daemon"),
2460 "FROM x\n",
2461 None,
2462 Some(&context),
2463 )
2464 .expect("zips")
2465 };
2466 let first = build();
2467 assert_eq!(first, build(), "IMAGE-7: equal inputs, identical bytes");
2468
2469 let mut archive = zip::ZipArchive::new(std::io::Cursor::new(first)).expect("a zip");
2470 let mut names = Vec::new();
2471 for index in 0..archive.len() {
2472 let entry = archive.by_index(index).expect("entry");
2473 let name = entry.name().expect("a name").into_owned();
2474 let modified = entry.last_modified().expect("a date");
2475 assert_eq!(
2476 (modified.year(), modified.month(), modified.day()),
2477 (1980, 1, 1),
2478 "IMAGE-7: {name} carries a fixed date"
2479 );
2480 let mode = entry.unix_mode().expect("a mode") & 0o777;
2481 let expected = match name.as_str() {
2482 "agentd" | "app/run.sh" => 0o755,
2483 _ => 0o644,
2484 };
2485 assert_eq!(mode, expected, "IMAGE-7: {name} mode {mode:o}");
2486 names.push(name);
2487 }
2488 assert_eq!(names, ["Dockerfile", "agentd", "app/run.sh", "data"]);
2489 let mut dockerfile = String::new();
2490 archive
2491 .by_name("Dockerfile")
2492 .expect("entry")
2493 .read_to_string(&mut dockerfile)
2494 .expect("reads");
2495 assert_eq!(dockerfile, "FROM x\n");
2496 }
2497
2498 /// **IMAGE-7, the host system.** Every entry's central-directory header records Unix (3)
2499 /// as the system that made it, whatever host builds the artifact. zip records the running
2500 /// host otherwise, MS-DOS (0) on Windows, and that one byte made a Windows build differ
2501 /// from a Linux build of the same inputs and let a reader that follows it take the
2502 /// entries' modes for DOS attributes.
2503 ///
2504 /// **Falsification**: set `.system(zip::System::Dos)`, the Windows default, in
2505 /// `build_artifact_with_context`'s options, and every host byte reads 0.
2506 #[test]
2507 fn every_artifact_entry_records_unix_as_its_host_system() {
2508 let context = context_of(&[("app/run.sh", 0o755, b"#!/bin/sh\n")]);
2509 let bytes = build_artifact_with_context(
2510 &aarch64_daemon(b"daemon"),
2511 "FROM x\n",
2512 None,
2513 Some(&context),
2514 )
2515 .expect("zips");
2516 assert_eq!(
2517 central_directory_hosts(&bytes),
2518 [3, 3, 3],
2519 "IMAGE-7: every entry's host system is Unix"
2520 );
2521 }
2522
2523 /// The host-system byte of each central-directory header, in order (APPNOTE 4.3.12 and
2524 /// 4.4.2: the high byte of "version made by"). The artifact carries no archive comment, so
2525 /// its end record is the last 22 bytes: the entry count at +10, the directory's offset at
2526 /// +16.
2527 fn central_directory_hosts(bytes: &[u8]) -> Vec<u8> {
2528 let u16_at = |at: usize| usize::from(u16::from_le_bytes([bytes[at], bytes[at + 1]]));
2529 let end = bytes.len() - 22;
2530 assert_eq!(&bytes[end..end + 4], b"PK\x05\x06", "the end record");
2531 let offset = bytes[end + 16..end + 20].try_into().expect("four bytes");
2532 let mut at = usize::try_from(u32::from_le_bytes(offset)).expect("fits");
2533 (0..u16_at(end + 10))
2534 .map(|_| {
2535 assert_eq!(
2536 &bytes[at..at + 4],
2537 b"PK\x01\x02",
2538 "a central-directory header"
2539 );
2540 let host = bytes[at + 5];
2541 at += 46 + u16_at(at + 28) + u16_at(at + 30) + u16_at(at + 32);
2542 host
2543 })
2544 .collect()
2545 }
2546
2547 /// **IMAGE-7, the byte scan extended.** The agent token has no path into a context either:
2548 /// the artifact built from a context is scanned the way AC-2-3 scans the plain one.
2549 #[test]
2550 fn a_context_artifact_never_carries_the_agent_token() {
2551 let token = "s3cr3t-agent-token-do-not-bake-me";
2552 let context = context_of(&[("app/main.py", 0o644, b"print('hi')\n")]);
2553 let bytes = build_artifact_with_context(
2554 &aarch64_daemon(b"binary"),
2555 &wrap_dockerfile("FROM x\n", &WrapOptions::default()).expect("wraps"),
2556 None,
2557 Some(&context),
2558 )
2559 .expect("zips");
2560 assert!(
2561 !bytes
2562 .windows(token.len())
2563 .any(|window| window == token.as_bytes())
2564 );
2565 }
2566
2567 /// The base image ARN, for the region it is requested in.
2568 ///
2569 /// `microvm-image:` with a **colon**, which is the spelling for a managed base *and* for
2570 /// a customer image alike — measured 2026-08-15 against real ARNs in us-east-1, and the
2571 /// only form the model's `TaggableResource` pattern admits. The docs above this function
2572 /// carry the measurement.
2573 ///
2574 /// **Falsification** — change the separator to `/` and both assertions go red. That was
2575 /// the state twelve fakes and the transport encoding test were in, which is why this
2576 /// test now asserts the separator explicitly rather than only the region.
2577 #[test]
2578 fn the_base_image_arn_names_the_request_region_with_a_colon_separator() {
2579 assert_eq!(
2580 BaseImage::al2023().arn(&crate::region::Region::UsEast1),
2581 "arn:aws:lambda:us-east-1:aws:microvm-image:al2023-1"
2582 );
2583 assert_eq!(
2584 BaseImage::al2023().arn(&crate::region::Region::ApNortheast1),
2585 "arn:aws:lambda:ap-northeast-1:aws:microvm-image:al2023-1"
2586 );
2587 assert!(
2588 !BaseImage::al2023()
2589 .arn(&crate::region::Region::UsEast1)
2590 .contains("microvm-image/"),
2591 "the slash form answers AccessDeniedException, not a validation error"
2592 );
2593 }
2594}