Compare commits

...

10 Commits

Author SHA1 Message Date
silverwind
34bfa19150 fix: resolve symlinked container paths before building tar entries (#1130)
Docker 29.7 extracts copied archives through `os.Root`, which refuses to follow a symlink to an absolute target, so copying into `/var/run/act` fails with `path escapes from parent` on the many images that link `/var/run` to `/run`. The daemon now resolves every path component for us before any tar entry name is built, and the destination is created with one directory entry per missing component, which no daemon version rejects.

Verified against real daemons (29.4.0, 29.5.0, 29.5.1, 29.5.3, 29.6.2, 29.7.0-rc.1) with `debian:bookworm` (absolute symlink) and `alpine:3` (relative symlink), and against `moby/go-archive` v0.2.0 through the pending fix branch.

1. Fixes https://gitea.com/gitea/runner/issues/1128
1. Upstream bug: https://github.com/moby/moby/issues/53258
1. Supersedes the no-op change in https://gitea.com/gitea/runner/pulls/1129, which cannot help since the daemon strips leading slashes itself

Reviewed-on: https://gitea.com/gitea/runner/pulls/1130
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
2026-07-31 16:09:27 +00:00
bircni
96d9f491db fix: strip leading slash from mkdir tarball in CopyTarStream (#1129)
`CopyTarStream` creates the destination directory by extracting a one-entry tarball with `DestinationPath: "/"`, but named that entry with the absolute `destPath`. Docker Engine 29.5+ tightened path validation on the copy API and rejects absolute entry names against a `/` destination with `statat var/run/act/actions/<sha>: path escapes from parent`, so the action directory never reached the job container and `actions/checkout` failed during "Set up job". Stripping the leading slash makes the entry relative, matching what the sibling `copyDir` already does and the upstream fix in nektos/act v0.2.89. Adds a regression test asserting the mkdir tarball entry is relative.

 Fixes #1128

Reviewed-on: https://gitea.com/gitea/runner/pulls/1129
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
2026-07-31 15:11:44 +00:00
bircni
14ec00b66e fix!: strip host-escape container options when privileged mode is disabled (#1058)
Workflow-controlled `jobs.<job>.container.options` were merged directly into the
Docker `HostConfig`. When the runner's privileged mode is disabled, only
`Privileged` was forced to `false` — host namespace flags, capability expansion,
security-profile overrides, and device/runtime access from the workflow YAML
survived into the final `HostConfig`. A workflow author could therefore enter
host PID/IPC namespaces and execute commands as root on the runner host:

```yaml
container:
  image: ubuntu:22.04
  options: >-
    --pid=host --ipc=host --cap-add=ALL
    --security-opt seccomp=unconfined --security-opt apparmor=unconfined
```

## Fix

`mergeContainerConfigs()` now strips the dangerous options-derived `HostConfig`
fields before merging when privileged mode is off: `PidMode`, `IpcMode`,
`UTSMode`, `CgroupnsMode`, `UsernsMode`, `CapAdd`, `SecurityOpt`, `Devices`,
`DeviceCgroupRules`, `DeviceRequests`, `VolumesFrom`, `Runtime`, `CgroupParent`,
and `Sysctls`. Each strip emits a warning, matching the existing
`--network ignored` handling. Options remain honored when privileged mode is
enabled, since the administrator has already opted into host access.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1058
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
2026-07-31 12:43:19 +00:00
bircni
68c6a5b4f1 feat: mask secrets that reach the log in an encoded form (#1108)
Only the verbatim value of a secret was masked, so a secret leaked through an action that serialized it stayed readable: `toJSON(secrets)` escapes it, an Authorization header carries it base64-encoded, a URL percent-encodes it. Each secret and `::add-mask::` value is now masked in those forms too, matching the value encoders of GitHub's runner. Encodings that leave the value unchanged are skipped, so a plain token still costs a single replacement pair. Includes regression tests.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1108
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
2026-07-31 12:35:10 +00:00
bircni
0cd0e52a24 fix!: guard against two runner processes sharing one runner file (#1099)
Starting two runner daemons with the same `.runner` file makes both present an
identical UUID+token, so Gitea treats them as one runner and they cancel each
other's jobs. This adds a non-blocking advisory lock on a sibling
`<runner-file>.lock`: the daemon (and `register`) acquire it at startup, and a
second process on the same host fails fast with a clear error instead of silently
interfering. The OS releases the lock when the process exits — including a hard
kill — so no stale lock is left behind. Legitimate multi-runner setups are
unaffected since each already uses its own `runner.file`.

Note: this covers the common single-host case; two hosts sharing a copied
`.runner` (e.g. over NFS) would still need server-side detection in Gitea.

---------

Co-authored-by: Zettat123 <zettat123@gmail.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1099
Reviewed-by: Zettat123 <39446+zettat123@noreply.gitea.com>
2026-07-31 12:15:04 +00:00
bircni
47d5b5ad03 feat!: add cache service v2, add toolkit patches (#1110)
Serves `github.actions.results.api.v1.CacheService` next to the v1 cache API, sharing its store, plus the subset of the Azure blob protocol the toolkit uploads with. On by default via `cache.v2`, and works with `external_server`.

Clients reach it through two edits in the action's own bundle: the GHES check is opened, and the cache service URL is taken from `ACTIONS_CACHE_URL`.

The same GHES check is what makes the stock `actions/upload-artifact` and `download-artifact` abort on Gitea. Opening it makes them work without the `gitea-upload-artifact` fork, from `upload-artifact@v4.4.0` on.

Verified against 118 real bundles, every major version of 16 actions: 92 patched, the rest deliberately left alone, and every patched bundle checked with `node --check`. Also end to end against pinned `actions/cache@v6.1.0` with an unreachable results URL, so only the patch can make the cache work.

---------

Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: bircni <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1110
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
2026-07-31 12:08:44 +00:00
Renovate Bot
2398d4a527 fix(deps): update module github.com/ulikunitz/xz to v0.5.15 [security] (#1127)
This PR contains the following updates:

| Package | Change | [Age](https://docs.renovatebot.com/merge-confidence/) | [Confidence](https://docs.renovatebot.com/merge-confidence/) |
|---|---|---|---|
| [github.com/ulikunitz/xz](https://github.com/ulikunitz/xz) | `v0.5.10` → `v0.5.15` | ![age](https://developer.mend.io/api/mc/badges/age/go/github.com%2fulikunitz%2fxz/v0.5.15?slim=true) | ![confidence](https://developer.mend.io/api/mc/badges/confidence/go/github.com%2fulikunitz%2fxz/v0.5.10/v0.5.15?slim=true) |

---

### github.com/ulikunitz/xz leaks memory when decoding a corrupted multiple LZMA archives
[CVE-2025-58058](https://nvd.nist.gov/vuln/detail/CVE-2025-58058) / [GHSA-jc7w-c686-c4v9](https://github.com/advisories/GHSA-jc7w-c686-c4v9) / [GO-2025-3922](https://pkg.go.dev/vuln/GO-2025-3922)

<details>
<summary>More information</summary>

#### Details
##### Summary

It is possible to put data in front of an LZMA-encoded byte stream without detecting the situation while reading the header. This can lead to increased memory consumption because the current implementation allocates the full decoding buffer directly after reading the header. The LZMA header doesn't include a magic number or has  a checksum to detect such an issue according to the [specification](https://github.com/jljusten/LZMA-SDK/blob/master/DOC/lzma-specification.txt).

Note that the code recognizes the issue later while reading the stream, but at this time the memory allocation has already been done.

##### Mitigations

The release v0.5.15 includes following mitigations:

- The ReaderConfig DictCap field is now interpreted as a limit for the dictionary size.
- The default is 2 Gigabytes - 1 byte (2^31-1 bytes).
- Users can check with the [Reader.Header] method what the actual values are in  their LZMA files and set a smaller limit using ReaderConfig.
- The dictionary size will not exceed the larger of the file size and the minimum dictionary size. This is another measure to prevent huge memory allocations for the dictionary.
- The code supports stream sizes only up to a pebibyte (1024^5).

Note that the original v0.5.14 version had a compiler error for 32 bit platforms, which has been fixed by v0.5.15.

##### Methods affected

Only software that uses [lzma.NewReader](https://pkg.go.dev/github.com/ulikunitz/xz/lzma#NewReader) or [lzma.ReaderConfig.NewReader](https://pkg.go.dev/github.com/ulikunitz/xz/lzma#ReaderConfig.NewReader) is affected. There is no issue for software using the xz functionality.

I thank  @&#8203;GregoryBuligin for his report, which is provided below.

##### Summary
When unpacking a large number of LZMA archives, even in a single goroutine, if the first byte of the archive file is 0 (a zero byte added to the beginning), an error __writeMatch: distance out of range__ occurs. Memory consumption spikes sharply, and the GC clearly cannot handle this situation.

##### Details
Judging by the error  __writeMatch: distance out of range__, the problems occur in the code around this function.
c8314b8f21/lzma/decoderdict.go (L81)

##### PoC
Run a function similar to this one in 1 or several goroutines on a multitude of LZMA archives that have a 0 (a zero byte) added to the beginning.
```
const ProjectLocalPath = "some/path"
const TmpDir = "tmp"

func UnpackLZMA(lzmaFile string) error {
	file, err := os.Open(lzmaFile)
	if err != nil {
		return err
	}
	defer file.Close()

	reader, err := lzma.NewReader(bufio.NewReader(file))
	if err != nil {
		return err
	}

	tmpFile, err := os.CreateTemp(TmpDir, TmpLZMAPrefix)
	if err != nil {
		return err
	}
	defer func() {
		tmpFile.Close()
		_ = os.Remove(tmpFile.Name())
	}()

	sha256Hasher := sha256.New()
	multiWriter := io.MultiWriter(tmpFile, sha256Hasher)

	if _, err = io.Copy(multiWriter, reader); err != nil {
		return err
	}

	unpackHash := hex.EncodeToString(sha256Hasher.Sum(nil))
	unpackDir := filepath.Join(
		ProjectLocalPath, unpackHash[:2],
	)
	_ = os.MkdirAll(unpackDir, DirPerm)

	unpackPath := filepath.Join(unpackDir, unpackHash)

	return os.Rename(tmpFile.Name(), unpackPath)
}
```

##### Impact
Servers with a small amount of RAM that download and unpack a large number of unverified LZMA archives

#### Severity
- CVSS Score: 5.3 / 10 (Medium)
- Vector String: `CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L`

#### References
- [https://github.com/ulikunitz/xz/security/advisories/GHSA-jc7w-c686-c4v9](https://github.com/ulikunitz/xz/security/advisories/GHSA-jc7w-c686-c4v9)
- [https://nvd.nist.gov/vuln/detail/CVE-2025-58058](https://nvd.nist.gov/vuln/detail/CVE-2025-58058)
- [88ddf1d0d9)
- [https://github.com/ulikunitz/xz](https://github.com/ulikunitz/xz)

This data is provided by [OSV](https://osv.dev/vulnerability/GHSA-jc7w-c686-c4v9) and the [GitHub Advisory Database](https://github.com/github/advisory-database) ([CC-BY 4.0](https://github.com/github/advisory-database/blob/main/LICENSE.md)).
</details>

---

### Memory leaks when decoding a corrupted multiple LZMA archives in github.com/ulikunitz/xz
[CVE-2025-58058](https://nvd.nist.gov/vuln/detail/CVE-2025-58058) / [GHSA-jc7w-c686-c4v9](https://github.com/advisories/GHSA-jc7w-c686-c4v9) / [GO-2025-3922](https://pkg.go.dev/vuln/GO-2025-3922)

<details>
<summary>More information</summary>

#### Details
Memory leaks when decoding a corrupted multiple LZMA archives in github.com/ulikunitz/xz

#### Severity
Unknown

#### References
- [https://github.com/ulikunitz/xz/security/advisories/GHSA-jc7w-c686-c4v9](https://github.com/ulikunitz/xz/security/advisories/GHSA-jc7w-c686-c4v9)
- [88ddf1d0d9)

This data is provided by [OSV](https://osv.dev/vulnerability/GO-2025-3922) and the [Go Vulnerability Database](https://github.com/golang/vulndb) ([CC-BY 4.0](https://github.com/golang/vulndb#license)).
</details>

---

### Release Notes

<details>
<summary>ulikunitz/xz (github.com/ulikunitz/xz)</summary>

### [`v0.5.15`](https://github.com/ulikunitz/xz/compare/v0.5.14...v0.5.15)

[Compare Source](https://github.com/ulikunitz/xz/compare/v0.5.14...v0.5.15)

### [`v0.5.14`](https://github.com/ulikunitz/xz/compare/v0.5.13...v0.5.14)

[Compare Source](https://github.com/ulikunitz/xz/compare/v0.5.13...v0.5.14)

### [`v0.5.13`](https://github.com/ulikunitz/xz/compare/v0.5.12...v0.5.13)

[Compare Source](https://github.com/ulikunitz/xz/compare/v0.5.12...v0.5.13)

### [`v0.5.12`](https://github.com/ulikunitz/xz/compare/v0.5.11...v0.5.12)

[Compare Source](https://github.com/ulikunitz/xz/compare/v0.5.11...v0.5.12)

### [`v0.5.11`](https://github.com/ulikunitz/xz/compare/v0.5.10...v0.5.11)

[Compare Source](https://github.com/ulikunitz/xz/compare/v0.5.10...v0.5.11)

</details>

---

### Configuration

📅 **Schedule**: (UTC)

- Branch creation
  - ""
- Automerge
  - At any time (no schedule defined)

🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied.

♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 **Ignore**: Close this PR and you won't be reminded about this update again.

---

 - [ ] <!-- rebase-check -->If you want to rebase/retry this PR, check this box

---

This PR has been generated by [Mend Renovate](https://github.com/renovatebot/renovate).
<!--renovate-debug:eyJjcmVhdGVkSW5WZXIiOiI0My4xOTEuMiIsInVwZGF0ZWRJblZlciI6IjQzLjE5MS4yIiwidGFyZ2V0QnJhbmNoIjoibWFpbiIsImxhYmVscyI6W119-->

Reviewed-on: https://gitea.com/gitea/runner/pulls/1127
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
Co-authored-by: Renovate Bot <renovate-bot@gitea.com>
2026-07-30 17:05:53 +00:00
silverwind
b7a3bf98bc fix: stop leaking per-job docker networks (#1124)
Leaked per-job networks each hold a subnet of the daemon's address pool and nothing reclaimed them, so a host eventually fails every job with `all predefined address pools have been fully subnetted`. This is what CI hit in https://gitea.com/gitea/runner/actions/runs/742177.

- teardown no longer loses a container, and its network with it: a failed `ContainerRemove` was reported as success, a container whose id was never learned was skipped, and the daemon's own `AutoRemove` teardown was raced
- the idle cleanup reclaims what teardown cannot: networks carry `com.gitea.runner.uuid`, so a runner only touches its own, and a cutoff keeps a job starting during the pass out of scope
- pull failures reported mid-stream were discarded, surfacing later as a confusing `No such image`; they now propagate, and fall back to a local copy instead of failing the job
- `NetworkCreate` retries pool exhaustion briefly, then says which knobs to turn
- digest-pinned images are not re-pulled, and removal kills first so it never waits out Podman's stop timeout (measured 10.1s → 0.09s per container)

---------

Co-authored-by: bircni <bircni@icloud.com>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1124
Reviewed-by: bircni <bircni@icloud.com>
2026-07-30 16:25:47 +00:00
silverwind
da4037899a chore: add renovate markers to Makefile tool packages (#1126)
The shared renovate config at https://gitea.com/gitea/renovate-config already updates `_PACKAGE` variables carrying a trailing `# renovate: datasource=` marker, so add them here.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1126
Co-authored-by: silverwind <me@silverwind.io>
2026-07-30 16:09:28 +00:00
silverwind
e4fe49dba4 chore: scope fmt-check to Go sources (#1125)
`fmt-check` runs `make fmt` and then diffs the whole working tree, so any unrelated uncommitted file makes it fail with `Please run 'make fmt' and commit the result` even when the formatter changed nothing.

Restrict the diff to the files the formatter can touch, matching what [gitea's Makefile](https://github.com/go-gitea/gitea/blob/main/Makefile) does.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1125
Reviewed-by: techknowlogick <9+techknowlogick@noreply.gitea.com>
Co-authored-by: silverwind <me@silverwind.io>
2026-07-30 13:06:28 +00:00
39 changed files with 2771 additions and 215 deletions

1
.gitignore vendored
View File

@@ -2,6 +2,7 @@
.env .env
!/act/runner/testdata/secrets/.env !/act/runner/testdata/secrets/.env
.runner .runner
.runner.lock
coverage.txt coverage.txt
.tmp/ .tmp/
/config.yaml /config.yaml

View File

@@ -4,9 +4,9 @@ DIST_DIRS := $(DIST)/binaries $(DIST)/release
GO ?= go GO ?= go
SHASUM ?= shasum -a 256 SHASUM ?= shasum -a 256
HAS_GO = $(shell hash $(GO) > /dev/null 2>&1 && echo "GO" || echo "NOGO" ) HAS_GO = $(shell hash $(GO) > /dev/null 2>&1 && echo "GO" || echo "NOGO" )
XGO_PACKAGE ?= src.techknowlogick.com/xgo@latest XGO_PACKAGE ?= src.techknowlogick.com/xgo@v1.9.0 # renovate: datasource=go
XGO_VERSION := go-1.26.x XGO_VERSION := go-1.26.x
GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/gxz@v0.5.10 GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/gxz@v0.5.15 # renovate: datasource=go
LINUX_ARCHS ?= linux/amd64,linux/arm64 LINUX_ARCHS ?= linux/amd64,linux/arm64
DARWIN_ARCHS ?= darwin-12/amd64,darwin-12/arm64 DARWIN_ARCHS ?= darwin-12/amd64,darwin-12/arm64
@@ -18,8 +18,8 @@ DOCKER_TAG ?= nightly
DOCKER_REF := $(DOCKER_IMAGE):$(DOCKER_TAG) DOCKER_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)
DOCKER_ROOTLESS_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)-dind-rootless DOCKER_ROOTLESS_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)-dind-rootless
GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/golangci-lint@v2.12.2 GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/golangci-lint@v2.12.2 # renovate: datasource=go
GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.3.0 GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.3.0 # renovate: datasource=go
GOTEST_FLAGS ?= -race -timeout 20m -parallel 8 GOTEST_FLAGS ?= -race -timeout 20m -parallel 8
@@ -97,7 +97,7 @@ go-check:
.PHONY: fmt-check .PHONY: fmt-check
fmt-check: fmt fmt-check: fmt
@diff=$$(git diff --color=always); \ @diff=$$(git diff --color=always -- '*.go'); \
if [ -n "$$diff" ]; then \ if [ -n "$$diff" ]; then \
echo "Please run 'make fmt' and commit the result:"; \ echo "Please run 'make fmt' and commit the result:"; \
printf "%s" "$${diff}"; \ printf "%s" "$${diff}"; \

View File

@@ -242,6 +242,17 @@ A password in a proxy URL is hidden in job logs. Any step can still read it, bec
Each runner starts its own cache server automatically. Cache entries are local to that runner — runners do not share a cache by default. Each runner starts its own cache server automatically. Cache entries are local to that runner — runners do not share a cache by default.
**Cache service v2**
`actions/cache@v4.2` and later can use the *cache service v2* API. The runner serves it from the same store as v1, on by default, and it works with `external_server`. Turn it off with:
```yaml
cache:
v2: false
```
Those actions refuse any host they do not take for GitHub. Rather than misreport the server URL, the runner edits that check out of the action's own bundle and keeps the untouched copy beside it; a bundle it does not recognise is left alone and keeps to v1. The same edit lets the stock `actions/upload-artifact` and `actions/download-artifact` work from `v4.4.0` on, without the `gitea-upload-artifact` fork.
**Shared cache across multiple runners** **Shared cache across multiple runners**
Run one dedicated `gitea-runner cache-server` that all runners point at. Run one dedicated `gitea-runner cache-server` that all runners point at.
@@ -283,11 +294,12 @@ Besides `GITEA_INSTANCE_URL` and `GITEA_RUNNER_REGISTRATION_TOKEN`, the image en
For a fuller container-oriented walkthrough, see [examples/docker](examples/docker/README.md). For a fuller container-oriented walkthrough, see [examples/docker](examples/docker/README.md).
When `container.bind_workdir` is enabled, stale task workspace directories can be cleaned while the runner is idle: While the runner is idle it cleans up after earlier jobs:
- directories older than `runner.workdir_cleanup_age` are removed (default: `24h`; set `0` to disable) - when `container.bind_workdir` is enabled, stale task workspace directories older than `runner.workdir_cleanup_age` are removed (default: `24h`; set `0` to disable)
- cleanup runs every `runner.idle_cleanup_interval` (default: `10m`; set `0` to disable)
- only purely numeric subdirectories under `container.workdir_parent` are treated as task workspaces and may be removed - only purely numeric subdirectories under `container.workdir_parent` are treated as task workspaces and may be removed
- cleanup assumes `container.workdir_parent` is not shared across multiple runners - cleanup assumes `container.workdir_parent` is not shared across multiple runners
- on runners that use docker, per-job networks left behind by jobs the runner did not live to tear down are removed, identified by the `com.gitea.runner.uuid` label carrying this runner's uuid
- cleanup runs every `runner.idle_cleanup_interval` (default: `10m`; set `0` to disable), and setting either knob to `0` disables all of the above
#### Post-task script (`runner.post_task_script`) #### Post-task script (`runner.post_task_script`)

View File

@@ -158,12 +158,13 @@ func StartHandler(dir, outboundIP string, port uint16, internalSecret string, lo
router.POST(apiPath+"/clean", h.bearerAuth(h.clean)) router.POST(apiPath+"/clean", h.bearerAuth(h.clean))
// Artifact GET is signed via query-string HMAC because @actions/cache // Artifact GET is signed via query-string HMAC because @actions/cache
// does not attach Authorization when downloading archiveLocation. // does not attach Authorization when downloading archiveLocation.
router.GET(apiPath+"/artifacts/:id", h.signedURLAuth(h.get)) router.GET(apiPath+"/artifacts/:id", h.signedAuth("", h.get))
// Control-plane: a remote runner registers/revokes per-job tokens so the // Control-plane: a remote runner registers/revokes per-job tokens so the
// cache API can authenticate them. Always wired so the routes exist; the // cache API can authenticate them. Always wired so the routes exist; the
// handlers themselves 401 when internalSecret is unset. // handlers themselves 401 when internalSecret is unset.
router.POST(internalPath+"/register", h.internalAuth(h.internalRegister)) router.POST(internalPath+"/register", h.internalAuth(h.internalRegister))
router.POST(internalPath+"/revoke", h.internalAuth(h.internalRevoke)) router.POST(internalPath+"/revoke", h.internalAuth(h.internalRevoke))
h.registerV2Routes(router)
h.router = router h.router = router
@@ -339,7 +340,7 @@ func (h *Handler) find(w http.ResponseWriter, r *http.Request, _ httprouter.Para
} }
defer db.Close() defer db.Close()
cache, err := findCache(db, cred.Repo, keys, version) cache, err := h.lookupCache(db, cred.Repo, keys, version)
if err != nil { if err != nil {
h.responseJSON(w, r, 500, err) h.responseJSON(w, r, 500, err)
return return
@@ -348,15 +349,6 @@ func (h *Handler) find(w http.ResponseWriter, r *http.Request, _ httprouter.Para
h.responseJSON(w, r, 204) h.responseJSON(w, r, 204)
return return
} }
if ok, err := h.storage.Exist(cache.ID); err != nil {
h.responseJSON(w, r, 500, err)
return
} else if !ok {
_ = db.Delete(cache.ID, cache)
h.responseJSON(w, r, 204)
return
}
h.responseJSON(w, r, 200, map[string]any{ h.responseJSON(w, r, 200, map[string]any{
"result": "hit", "result": "hit",
"archiveLocation": h.signedArtifactURL(cache.ID, time.Now().Add(artifactURLTTL)), "archiveLocation": h.signedArtifactURL(cache.ID, time.Now().Add(artifactURLTTL)),
@@ -364,6 +356,25 @@ func (h *Handler) find(w http.ResponseWriter, r *http.Request, _ httprouter.Para
}) })
} }
// lookupCache returns the entry to restore for these keys, or (nil, nil) when there is none:
// either nothing matched, or the match had lost its blob to a prune, in which case the dangling
// entry is dropped on the way out.
func (h *Handler) lookupCache(db *bolthold.Store, repo string, keys []string, version string) (*Cache, error) {
cache, err := findCache(db, repo, keys, version)
if err != nil || cache == nil {
return nil, err
}
ok, err := h.storage.Exist(cache.ID)
if err != nil {
return nil, err
}
if !ok {
_ = db.Delete(cache.ID, cache)
return nil, nil //nolint:nilnil // absence is not an error here
}
return cache, nil
}
// POST /_apis/artifactcache/caches // POST /_apis/artifactcache/caches
func (h *Handler) reserve(w http.ResponseWriter, r *http.Request, _ httprouter.Params) { func (h *Handler) reserve(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
cred := credFromContext(r.Context()) cred := credFromContext(r.Context())
@@ -438,7 +449,7 @@ func (h *Handler) upload(w http.ResponseWriter, r *http.Request, params httprout
h.responseJSON(w, r, 500, err) h.responseJSON(w, r, 500, err)
return return
} }
h.useCache(id) _ = h.touchCache(uint64(id), false)
h.responseJSON(w, r, 200) h.responseJSON(w, r, 200)
} }
@@ -479,23 +490,7 @@ func (h *Handler) commit(w http.ResponseWriter, r *http.Request, params httprout
db.Close() db.Close()
size, err := h.storage.Commit(cache.ID, cache.Size) if err := h.commitCache(cache); err != nil {
if err != nil {
h.responseJSON(w, r, 500, err)
return
}
// write real size back to cache, it may be different from the current value when the request doesn't specify it.
cache.Size = size
db, err = h.openDB()
if err != nil {
h.responseJSON(w, r, 500, err)
return
}
defer db.Close()
cache.Complete = true
if err := db.Update(cache.ID, cache); err != nil {
h.responseJSON(w, r, 500, err) h.responseJSON(w, r, 500, err)
return return
} }
@@ -503,8 +498,28 @@ func (h *Handler) commit(w http.ResponseWriter, r *http.Request, params httprout
h.responseJSON(w, r, 200) h.responseJSON(w, r, 200)
} }
// commitCache assembles the uploaded parts and marks the entry complete. The caller must
// have closed its store first: Commit concatenates the whole archive and would otherwise
// hold bolt's exclusive file lock for the duration.
func (h *Handler) commitCache(cache *Cache) error {
written, err := h.storage.Commit(cache.ID, cache.Size)
if err != nil {
return err
}
// write real size back to cache, it may be different from the current value when the request doesn't specify it.
cache.Size = written
cache.Complete = true
db, err := h.openDB()
if err != nil {
return err
}
defer db.Close()
return db.Update(cache.ID, cache)
}
// GET /_apis/artifactcache/artifacts/:id // GET /_apis/artifactcache/artifacts/:id
// Authenticated via signed URL (see signedURLAuth), not bearer, because the // Authenticated via signed URL (see signedAuth), not bearer, because the
// @actions/cache toolkit downloads archiveLocation without Authorization. // @actions/cache toolkit downloads archiveLocation without Authorization.
// Repository scoping is already enforced at find() time; the signature binds // Repository scoping is already enforced at find() time; the signature binds
// the URL to the specific cache ID and an expiry. // the URL to the specific cache ID and an expiry.
@@ -514,7 +529,7 @@ func (h *Handler) get(w http.ResponseWriter, r *http.Request, params httprouter.
h.responseJSON(w, r, 400, err) h.responseJSON(w, r, 400, err)
return return
} }
h.useCache(id) _ = h.touchCache(uint64(id), false)
h.storage.Serve(w, r, uint64(id)) h.storage.Serve(w, r, uint64(id))
} }
@@ -548,7 +563,9 @@ func (h *Handler) bearerAuth(handler httprouter.Handle) httprouter.Handle {
} }
} }
func (h *Handler) signedURLAuth(handler httprouter.Handle) httprouter.Handle { // signedAuth authenticates a signed URL. purpose separates the flavours of URL the
// handler hands out, so one cannot be replayed as another; see computeSignature.
func (h *Handler) signedAuth(purpose string, handler httprouter.Handle) httprouter.Handle {
return func(w http.ResponseWriter, r *http.Request, params httprouter.Params) { return func(w http.ResponseWriter, r *http.Request, params httprouter.Params) {
h.logger.Debugf("%s %s", r.Method, r.URL.Path) h.logger.Debugf("%s %s", r.Method, r.URL.Path)
id, err := strconv.ParseInt(params.ByName("id"), 10, 64) id, err := strconv.ParseInt(params.ByName("id"), 10, 64)
@@ -571,7 +588,7 @@ func (h *Handler) signedURLAuth(handler httprouter.Handle) httprouter.Handle {
h.responseJSON(w, r, http.StatusUnauthorized, errors.New("signature expired")) h.responseJSON(w, r, http.StatusUnauthorized, errors.New("signature expired"))
return return
} }
expected := h.computeSignature(id, exp) expected := h.computeSignature(purpose, id, exp)
if !hmac.Equal([]byte(sig), []byte(expected)) { if !hmac.Equal([]byte(sig), []byte(expected)) {
h.responseJSON(w, r, http.StatusUnauthorized, errors.New("bad signature")) h.responseJSON(w, r, http.StatusUnauthorized, errors.New("bad signature"))
return return
@@ -655,19 +672,26 @@ func credFromContext(ctx context.Context) JobCredential {
return JobCredential{} return JobCredential{}
} }
func (h *Handler) computeSignature(cacheID, exp int64) string { // computeSignature signs a URL for one cache entry and expiry. purpose is mixed into the
// message so a URL handed out for writing an entry cannot be replayed to read one, and the
// other way round. Downloads use the empty purpose, the message v1 has always signed.
func (h *Handler) computeSignature(purpose string, cacheID, exp int64) string {
mac := hmac.New(sha256.New, h.secret) mac := hmac.New(sha256.New, h.secret)
fmt.Fprintf(mac, "%d:%d", cacheID, exp) fmt.Fprintf(mac, "%s%d:%d", purpose, cacheID, exp)
return hex.EncodeToString(mac.Sum(nil)) return hex.EncodeToString(mac.Sum(nil))
} }
func (h *Handler) signedArtifactURL(cacheID uint64, exp time.Time) string { // signedURL builds a URL under path that signedAuth accepts for the same purpose.
func (h *Handler) signedURL(path, purpose string, cacheID uint64, exp time.Time) string {
expUnix := exp.Unix() expUnix := exp.Unix()
sig := h.computeSignature(int64(cacheID), expUnix)
q := url.Values{} q := url.Values{}
q.Set("exp", strconv.FormatInt(expUnix, 10)) q.Set("exp", strconv.FormatInt(expUnix, 10))
q.Set("sig", sig) q.Set("sig", h.computeSignature(purpose, int64(cacheID), expUnix))
return fmt.Sprintf("%s%s/artifacts/%d?%s", h.ExternalURL(), apiPath, cacheID, q.Encode()) return fmt.Sprintf("%s%s/%d?%s", h.ExternalURL(), path, cacheID, q.Encode())
}
func (h *Handler) signedArtifactURL(cacheID uint64, exp time.Time) string {
return h.signedURL(apiPath+"/artifacts", "", cacheID, exp)
} }
// if not found, return (nil, nil) instead of an error. // if not found, return (nil, nil) instead of an error.
@@ -675,16 +699,12 @@ func findCache(db *bolthold.Store, repo string, keys []string, version string) (
cache := &Cache{} cache := &Cache{}
for _, prefix := range keys { for _, prefix := range keys {
// if a key in the list matches exactly, don't return partial matches // if a key in the list matches exactly, don't return partial matches
if err := db.FindOne(cache, exact, err := findExactCache(db, repo, prefix, version, true)
bolthold.Where("Repo").Eq(repo). if err != nil {
And("Key").Eq(prefix). return nil, err
And("Version").Eq(version). }
And("Complete").Eq(true). if exact != nil {
SortBy("CreatedAt").Reverse()); err == nil || !errors.Is(err, bolthold.ErrNotFound) { return exact, nil
if err != nil {
return nil, fmt.Errorf("find cache: %w", err)
}
return cache, nil
} }
prefixPattern := "^" + regexp.QuoteMeta(prefix) prefixPattern := "^" + regexp.QuoteMeta(prefix)
re, err := regexp.Compile(prefixPattern) re, err := regexp.Compile(prefixPattern)
@@ -707,6 +727,34 @@ func findCache(db *bolthold.Store, repo string, keys []string, version string) (
return nil, nil //nolint:nilnil // pre-existing issue from nektos/act return nil, nil //nolint:nilnil // pre-existing issue from nektos/act
} }
// findExactCache returns the entry for exactly this key and version, or (nil, nil) if there is
// none. Unlike findCache it never falls back to a prefix (restore-key) match, which is what both
// its callers need: a new key that is only a prefix of an existing key is not the same entry.
//
// A completed entry is the one to restore, sorted by when it was written. An incomplete one is a
// reservation being uploaded to, sorted by when it was last written to, because the upload route
// touches UsedAt on every part.
func findExactCache(db *bolthold.Store, repo, key, version string, complete bool) (*Cache, error) {
sortBy := "UsedAt"
if complete {
sortBy = "CreatedAt"
}
cache := &Cache{}
err := db.FindOne(cache,
bolthold.Where("Repo").Eq(repo).
And("Key").Eq(key).
And("Version").Eq(version).
And("Complete").Eq(complete).
SortBy(sortBy).Reverse())
if errors.Is(err, bolthold.ErrNotFound) {
return nil, nil //nolint:nilnil // absence is not an error here
}
if err != nil {
return nil, fmt.Errorf("find cache: %w", err)
}
return cache, nil
}
func insertCache(db *bolthold.Store, cache *Cache) error { func insertCache(db *bolthold.Store, cache *Cache) error {
if err := db.Insert(bolthold.NextSequence(), cache); err != nil { if err := db.Insert(bolthold.NextSequence(), cache); err != nil {
return fmt.Errorf("insert cache: %w", err) return fmt.Errorf("insert cache: %w", err)
@@ -718,18 +766,30 @@ func insertCache(db *bolthold.Store, cache *Cache) error {
return nil return nil
} }
func (h *Handler) useCache(id int64) { // touchCache stamps UsedAt so gcCache does not reap an entry mid-upload. With requireIncomplete
// it also refuses an entry that is already complete, which is what the v2 blob route needs: its
// upload URL outlives the finalize call, and overwriting a finished entry would leave the blob
// other jobs restore no longer matching its recorded size. An entry missing from the store is
// accepted, since the signature proves the id was handed out.
func (h *Handler) touchCache(id uint64, requireIncomplete bool) error {
db, err := h.openDB() db, err := h.openDB()
if err != nil { if err != nil {
return return err
} }
defer db.Close() defer db.Close()
cache := &Cache{} cache := &Cache{}
if err := db.Get(id, cache); err != nil { if err := db.Get(id, cache); err != nil {
return if errors.Is(err, bolthold.ErrNotFound) {
return nil
}
return err
}
if requireIncomplete && cache.Complete {
return fmt.Errorf("cache %d: already complete", id)
} }
cache.UsedAt = time.Now().Unix() cache.UsedAt = time.Now().Unix()
_ = db.Update(cache.ID, cache) return db.Update(cache.ID, cache)
} }
const ( const (

View File

@@ -1041,14 +1041,14 @@ func TestHandler_SecretPersistsAcrossRestarts(t *testing.T) {
first, err := StartHandler(dir, "127.0.0.1", 0, "", nil) first, err := StartHandler(dir, "127.0.0.1", 0, "", nil)
require.NoError(t, err) require.NoError(t, err)
exp := time.Now().Add(artifactURLTTL).Unix() exp := time.Now().Add(artifactURLTTL).Unix()
sig := first.computeSignature(42, exp) sig := first.computeSignature("", 42, exp)
require.NoError(t, first.Close()) require.NoError(t, first.Close())
second, err := StartHandler(dir, "127.0.0.1", 0, "", nil) second, err := StartHandler(dir, "127.0.0.1", 0, "", nil)
require.NoError(t, err) require.NoError(t, err)
defer second.Close() defer second.Close()
assert.Equal(t, sig, second.computeSignature(42, exp)) assert.Equal(t, sig, second.computeSignature("", 42, exp))
} }
// TestHandler_ArtifactSignatureDownload is a happy-path round trip that // TestHandler_ArtifactSignatureDownload is a happy-path round trip that

View File

@@ -0,0 +1,268 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package artifactcache
import (
"cmp"
"encoding/json"
"encoding/xml"
"errors"
"fmt"
"io"
"net/http"
"strconv"
"strings"
"time"
"github.com/julienschmidt/httprouter"
)
// The cache service v2 API. A client on this version talks twirp to
// `github.actions.results.api.v1.CacheService` instead of the /_apis/artifactcache
// endpoints, and uploads the archive to the returned URL with the Azure blob protocol.
// Both API versions are served from the same store, so a repository keeps its cache
// when a workflow moves between action versions.
const (
cacheServiceV2Path = "/twirp/github.actions.results.api.v1.CacheService"
// blobPath authenticates by signature, because the client uploads without an
// Authorization header. Downloads are handed the v1 artifact URL instead.
blobPath = apiPath + "/blobs"
// blobUploadPurpose keeps an upload URL from being replayed to read an entry.
blobUploadPurpose = "upload:"
blobUploadURLTTL = time.Hour
// twirpInternal is the only error code that is not the client's fault.
twirpInternal = "internal"
)
func (h *Handler) registerV2Routes(router *httprouter.Router) {
router.POST(cacheServiceV2Path+"/CreateCacheEntry", h.bearerAuth(h.v2CreateCacheEntry))
router.POST(cacheServiceV2Path+"/FinalizeCacheEntryUpload", h.bearerAuth(h.v2FinalizeCacheEntryUpload))
router.POST(cacheServiceV2Path+"/GetCacheEntryDownloadURL", h.bearerAuth(h.v2GetCacheEntryDownloadURL))
router.PUT(blobPath+"/:id", h.signedAuth(blobUploadPurpose, h.v2UploadBlob))
}
// An entry that already exists is reported as not ok, which is how the client learns to skip
// the upload.
func (h *Handler) v2CreateCacheEntry(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
cred := credFromContext(r.Context())
req, err := decodeTwirpRequest[v2CreateRequest](r)
if err != nil {
h.twirpError(w, r, "malformed_request", err)
return
}
if req.Key == "" || req.Version == "" {
h.twirpError(w, r, "invalid_argument", errors.New("key and version are required"))
return
}
db, err := h.openDB()
if err != nil {
h.twirpError(w, r, twirpInternal, err)
return
}
defer db.Close()
// An exact (key, version) match means the entry is already cached; the client then skips
// the upload. A prefix match must not count here, or a shorter key would be reported as
// existing and silently never saved.
if existing, err := findExactCache(db, cred.Repo, req.Key, req.Version, true); err != nil {
h.twirpError(w, r, twirpInternal, err)
return
} else if existing != nil {
h.twirpNotOK(w, r)
return
}
now := time.Now().Unix()
cache := &Cache{
Repo: cred.Repo,
Key: req.Key,
Version: req.Version,
Size: -1, // the size is only known at finalize time
CreatedAt: now,
UsedAt: now,
}
if err := insertCache(db, cache); err != nil {
h.twirpError(w, r, twirpInternal, err)
return
}
h.responseJSON(w, r, http.StatusOK, map[string]any{
"ok": true,
"signedUploadUrl": h.signedURL(blobPath, blobUploadPurpose, cache.ID, time.Now().Add(blobUploadURLTTL)),
})
}
func (h *Handler) v2FinalizeCacheEntryUpload(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
cred := credFromContext(r.Context())
req, err := decodeTwirpRequest[v2FinalizeRequest](r)
if err != nil {
h.twirpError(w, r, "malformed_request", err)
return
}
db, err := h.openDB()
if err != nil {
h.twirpError(w, r, twirpInternal, err)
return
}
defer db.Close()
cache, err := findExactCache(db, cred.Repo, req.Key, req.Version, false)
if err != nil {
h.twirpError(w, r, twirpInternal, err)
return
}
if cache == nil {
h.twirpNotOK(w, r)
return
}
db.Close() // commitCache needs the store closed
cache.Size, _ = cmp.Or(req.SizeBytes, req.SizeBytesCamel).Int64()
if err := h.commitCache(cache); err != nil {
h.logger.Errorf("finalize cache %d (%s): %v", cache.ID, cache.Key, err)
h.twirpNotOK(w, r)
return
}
h.responseJSON(w, r, http.StatusOK, map[string]any{
"ok": true,
// int64 fields travel as strings in the proto JSON mapping.
"entryId": strconv.FormatUint(cache.ID, 10),
})
}
func (h *Handler) v2GetCacheEntryDownloadURL(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
cred := credFromContext(r.Context())
req, err := decodeTwirpRequest[v2DownloadRequest](r)
if err != nil {
h.twirpError(w, r, "malformed_request", err)
return
}
db, err := h.openDB()
if err != nil {
h.twirpError(w, r, twirpInternal, err)
return
}
defer db.Close()
cache, err := h.lookupCache(db, cred.Repo, req.keys(), req.Version)
if err != nil {
h.twirpError(w, r, twirpInternal, err)
return
}
if cache == nil {
h.twirpNotOK(w, r)
return
}
h.responseJSON(w, r, http.StatusOK, map[string]any{
"ok": true,
"signedDownloadUrl": h.signedArtifactURL(cache.ID, time.Now().Add(artifactURLTTL)),
"matchedKey": cache.Key,
})
}
// The archive arrives over the subset of the Azure blob API the toolkit uses: a small
// cache is a single PUT, a large one is staged as blocks that a final block list puts
// in order.
func (h *Handler) v2UploadBlob(w http.ResponseWriter, r *http.Request, params httprouter.Params) {
id, err := strconv.ParseUint(params.ByName("id"), 10, 64)
if err != nil {
h.responseJSON(w, r, http.StatusBadRequest, err)
return
}
if err := h.touchCache(id, true); err != nil {
h.responseJSON(w, r, http.StatusBadRequest, err)
return
}
query := r.URL.Query()
switch strings.ToLower(query.Get("comp")) {
case "block":
blockID := query.Get("blockid")
if blockID == "" {
h.responseJSON(w, r, http.StatusBadRequest, errors.New("missing blockid"))
return
}
err = h.storage.WriteBlock(id, blockID, r.Body)
case "blocklist":
var list struct{ Latest []string }
if err := xml.NewDecoder(io.LimitReader(r.Body, 8<<20)).Decode(&list); err != nil {
h.responseJSON(w, r, http.StatusBadRequest, fmt.Errorf("malformed block list: %w", err))
return
}
err = h.storage.OrderBlocks(id, list.Latest)
default:
err = h.storage.Write(id, 0, r.Body)
}
if err != nil {
h.responseJSON(w, r, http.StatusInternalServerError, err)
return
}
w.WriteHeader(http.StatusCreated)
}
// twirpNotOK is the negative answer all three endpoints share: no such entry to restore, no
// reservation to finalize, or an entry that already exists and need not be uploaded again.
func (h *Handler) twirpNotOK(w http.ResponseWriter, r *http.Request) {
h.responseJSON(w, r, http.StatusOK, map[string]any{"ok": false})
}
// twirpError reports in the shape a twirp client expects, so the toolkit surfaces the message
// instead of a parse error.
func (h *Handler) twirpError(w http.ResponseWriter, r *http.Request, code string, err error) {
h.logger.Debugf("%s %s: %v", r.Method, r.URL.Path, err)
status := http.StatusBadRequest
if code == twirpInternal {
status = http.StatusInternalServerError
}
h.responseJSON(w, r, status, map[string]any{"code": code, "msg": err.Error()})
}
// The twirp request bodies. The toolkit's client serialises with useProtoFieldName, so the proto
// names are what arrive; the camelCase spellings of the same mapping are accepted too, as are
// int64s sent as a bare number rather than the string the mapping prescribes.
type (
v2CreateRequest struct {
Key string `json:"key"`
Version string `json:"version"`
}
v2FinalizeRequest struct {
Key string `json:"key"`
Version string `json:"version"`
SizeBytes json.Number `json:"size_bytes"`
SizeBytesCamel json.Number `json:"sizeBytes"`
}
v2DownloadRequest struct {
Key string `json:"key"`
Version string `json:"version"`
RestoreKeys []string `json:"restore_keys"`
RestoreKeysCamel []string `json:"restoreKeys"`
}
)
func (d v2DownloadRequest) keys() []string {
restoreKeys := d.RestoreKeys
if len(restoreKeys) == 0 {
restoreKeys = d.RestoreKeysCamel
}
return append([]string{d.Key}, restoreKeys...)
}
func decodeTwirpRequest[T any](r *http.Request) (T, error) {
var req T
err := json.NewDecoder(io.LimitReader(r.Body, 1<<20)).Decode(&req)
return req, err
}

View File

@@ -0,0 +1,236 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package artifactcache
import (
"bytes"
"encoding/base64"
"encoding/json"
"fmt"
"io"
"net/http"
"path/filepath"
"strconv"
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// v2Call posts a twirp request to the cache service and returns the decoded response.
// Field names are the proto ones, which is what the toolkit's client sends.
func v2Call(t *testing.T, handler *Handler, client *http.Client, method string, request any) map[string]any {
t.Helper()
body, err := json.Marshal(request)
require.NoError(t, err)
resp, err := client.Post(handler.ExternalURL()+cacheServiceV2Path+"/"+method, "application/json", bytes.NewReader(body))
require.NoError(t, err)
defer resp.Body.Close()
require.Equal(t, http.StatusOK, resp.StatusCode)
got := map[string]any{}
require.NoError(t, json.NewDecoder(resp.Body).Decode(&got))
return got
}
// putBlob uploads to a signed URL and returns the status, so a test can assert a refusal.
func putBlob(t *testing.T, url string, content []byte) int {
t.Helper()
req, err := http.NewRequestWithContext(t.Context(), http.MethodPut, url, bytes.NewReader(content))
require.NoError(t, err)
resp, err := http.DefaultClient.Do(req)
require.NoError(t, err)
defer resp.Body.Close()
return resp.StatusCode
}
func getURL(t *testing.T, url string) []byte {
t.Helper()
req, err := http.NewRequestWithContext(t.Context(), http.MethodGet, url, nil)
require.NoError(t, err)
resp, err := http.DefaultClient.Do(req)
require.NoError(t, err)
defer resp.Body.Close()
body, err := io.ReadAll(resp.Body)
require.NoError(t, err)
return body
}
func startTestHandler(t *testing.T) *Handler {
t.Helper()
handler, err := StartHandler(filepath.Join(t.TempDir(), "artifactcache"), "127.0.0.1", 0, "", nil)
require.NoError(t, err)
t.Cleanup(func() { _ = handler.Close() })
handler.RegisterJob(testToken, testRepo)
return handler
}
// saveV2 runs the reserve/upload/finalize sequence and returns the finalize response along
// with the upload URL it used.
func saveV2(t *testing.T, handler *Handler, key, version string, content []byte) (finalized map[string]any, uploadURL string) {
t.Helper()
created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": key, "version": version})
require.Equal(t, true, created["ok"])
uploadURL, _ = created["signedUploadUrl"].(string)
require.NotEmpty(t, uploadURL)
require.Equal(t, http.StatusCreated, putBlob(t, uploadURL, content))
return v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
"key": key, "version": version,
"size_bytes": strconv.Itoa(len(content)),
}), uploadURL
}
// The whole round trip an actions/cache v2 client makes, plus the guarantees on the signed
// URLs it is handed: unsigned requests are refused, an upload URL cannot be replayed to read
// or to replace a finalized entry.
func TestCacheServiceV2RoundTrip(t *testing.T) {
handler := startTestHandler(t)
content := []byte("the cached archive")
unsigned := fmt.Sprintf("%s%s/1", handler.ExternalURL(), blobPath)
assert.Equal(t, http.StatusUnauthorized, putBlob(t, unsigned, content))
finalized, uploadURL := saveV2(t, handler, "deps-v1", "abc123", content)
require.Equal(t, true, finalized["ok"])
assert.NotEmpty(t, finalized["entryId"])
// The upload URL outlives the finalize call, so replaying it must not poison the entry,
// and it is an upload URL only: nothing reads a blob back through it.
assert.Equal(t, http.StatusBadRequest, putBlob(t, uploadURL, []byte("poisoned")))
resp, err := http.Get(uploadURL) //nolint:noctx // the URL is the server under test
require.NoError(t, err)
defer resp.Body.Close()
assert.Equal(t, http.StatusMethodNotAllowed, resp.StatusCode)
got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{"key": "deps-v1", "version": "abc123"})
require.Equal(t, true, got["ok"])
assert.Equal(t, "deps-v1", got["matchedKey"])
downloadURL, _ := got["signedDownloadUrl"].(string)
require.NotEmpty(t, downloadURL)
assert.Equal(t, content, getURL(t, downloadURL))
}
// A large archive is staged as blocks and only put in order by the final block list, so
// blocks that arrive out of order must still be assembled the way the client asked.
func TestCacheServiceV2BlockUpload(t *testing.T) {
handler := startTestHandler(t)
created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "blocks", "version": "v1"})
uploadURL, _ := created["signedUploadUrl"].(string)
require.NotEmpty(t, uploadURL)
blocks := map[string][]byte{}
var order []string
for i, part := range []string{"hello ", "world", "!"} {
blockID := base64.StdEncoding.EncodeToString(fmt.Appendf(nil, "block-%d", i))
blocks[blockID] = []byte(part)
order = append(order, blockID)
}
// Upload in an order that is not the block list order.
for _, blockID := range []string{order[2], order[0], order[1]} {
require.Equal(t, http.StatusCreated, putBlob(t, uploadURL+"&comp=block&blockid="+blockID, blocks[blockID]))
}
var list bytes.Buffer
list.WriteString(`<?xml version="1.0" encoding="utf-8"?><BlockList>`)
for _, blockID := range order {
fmt.Fprintf(&list, "<Latest>%s</Latest>", blockID)
}
list.WriteString(`</BlockList>`)
require.Equal(t, http.StatusCreated, putBlob(t, uploadURL+"&comp=blocklist", list.Bytes()))
finalized := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
"key": "blocks", "version": "v1", "size_bytes": len("hello world!"),
})
require.Equal(t, true, finalized["ok"])
got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{"key": "blocks", "version": "v1"})
require.Equal(t, true, got["ok"])
assert.Equal(t, "hello world!", string(getURL(t, got["signedDownloadUrl"].(string))))
}
func TestCacheServiceV2Lookups(t *testing.T) {
handler := startTestHandler(t)
saved, _ := saveV2(t, handler, "deps-abc", "v1", []byte("x"))
require.Equal(t, true, saved["ok"])
t.Run("reports a miss for an unknown key", func(t *testing.T) {
got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{"key": "nothing", "version": "v1"})
assert.Equal(t, false, got["ok"])
})
// The toolkit serialises with the proto field names; the camelCase spellings of the same
// proto JSON mapping are accepted alongside them.
for _, field := range []string{"restore_keys", "restoreKeys"} {
t.Run("restore keys match by prefix, spelled "+field, func(t *testing.T) {
got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{
"key": "deps-zzz", field: []string{"deps-"}, "version": "v1",
})
require.Equal(t, true, got["ok"])
assert.Equal(t, "deps-abc", got["matchedKey"])
})
}
t.Run("an existing entry is not reserved twice", func(t *testing.T) {
again := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "deps-abc", "version": "v1"})
assert.Equal(t, false, again["ok"])
})
// A key that is only a prefix of an existing one is a different entry, so the
// reservation check must be exact and not a restore-key prefix match, or the shorter
// key would be reported as existing and silently never saved.
t.Run("a prefix of an existing key is still reserved", func(t *testing.T) {
reserved := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "deps", "version": "v1"})
require.Equal(t, true, reserved["ok"])
assert.NotEmpty(t, reserved["signedUploadUrl"])
})
t.Run("finalizing without a reservation is not ok", func(t *testing.T) {
got := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
"key": "never-reserved", "version": "v1", "size_bytes": 1,
})
assert.Equal(t, false, got["ok"])
})
// The size the client declares is what Commit validates the assembled archive against.
t.Run("finalizing with the wrong size is not ok", func(t *testing.T) {
created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "wrong-size", "version": "v1"})
require.Equal(t, http.StatusCreated, putBlob(t, created["signedUploadUrl"].(string), []byte("four")))
got := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
"key": "wrong-size", "version": "v1", "size_bytes": 99,
})
assert.Equal(t, false, got["ok"])
})
// Both API versions are served from one store, so an entry written through v2 is a hit for
// a v1 client asking for the same key and version.
t.Run("a v1 client sees an entry written through v2", func(t *testing.T) {
resp, err := testClient.Get(fmt.Sprintf("%s%s/cache?keys=deps-abc&version=v1", handler.ExternalURL(), apiPath))
require.NoError(t, err)
defer resp.Body.Close()
require.Equal(t, http.StatusOK, resp.StatusCode)
got := map[string]any{}
require.NoError(t, json.NewDecoder(resp.Body).Decode(&got))
assert.Equal(t, "deps-abc", got["cacheKey"])
assert.NotEmpty(t, got["archiveLocation"])
})
// The cache of one repository must stay invisible to another, as it does for the v1 API.
t.Run("another repository sees nothing", func(t *testing.T) {
handler.RegisterJob("other-runtime-token", "other/repo")
otherClient := &http.Client{Transport: &bearerTransport{token: "other-runtime-token"}}
got := v2Call(t, handler, otherClient, "GetCacheEntryDownloadURL", map[string]any{"key": "deps-abc", "version": "v1"})
assert.Equal(t, false, got["ok"])
})
}

View File

@@ -5,12 +5,15 @@
package artifactcache package artifactcache
import ( import (
"crypto/sha256"
"encoding/hex"
"fmt" "fmt"
"io" "io"
"net/http" "net/http"
"os" "os"
"path/filepath" "path/filepath"
"strconv" "strconv"
"strings"
) )
type Storage struct { type Storage struct {
@@ -37,7 +40,10 @@ func (s *Storage) Exist(id uint64) (bool, error) {
} }
func (s *Storage) Write(id uint64, offset int64, reader io.Reader) error { func (s *Storage) Write(id uint64, offset int64, reader io.Reader) error {
name := s.tempName(id, offset) return s.writeFile(s.tempName(id, offset), reader)
}
func (s *Storage) writeFile(name string, reader io.Reader) error {
if err := os.MkdirAll(filepath.Dir(name), 0o755); err != nil { if err := os.MkdirAll(filepath.Dir(name), 0o755); err != nil {
return err return err
} }
@@ -51,6 +57,26 @@ func (s *Storage) Write(id uint64, offset int64, reader io.Reader) error {
return err return err
} }
func (s *Storage) WriteBlock(id uint64, blockID string, reader io.Reader) error {
return s.writeFile(s.blockName(id, blockID), reader)
}
// OrderBlocks renames the staged blocks into the order the block list gives. A block the list
// does not name keeps its staged name, which is how Commit leaves it out, as Azure drops it. One
// rename pass is safe because a staged name always carries blockFilePrefix and a target name
// never does, so no rename can collide with a block not yet moved.
func (s *Storage) OrderBlocks(id uint64, blockIDs []string) error {
for i, blockID := range blockIDs {
if err := os.Rename(s.blockName(id, blockID), s.tempName(id, int64(i))); err != nil {
if os.IsNotExist(err) {
return fmt.Errorf("block %q of cache %d was never uploaded: %w", blockID, id, err)
}
return err
}
}
return nil
}
func (s *Storage) Commit(id uint64, size int64) (int64, error) { func (s *Storage) Commit(id uint64, size int64) (int64, error) {
defer func() { defer func() {
_ = os.RemoveAll(s.tempDir(id)) _ = os.RemoveAll(s.tempDir(id))
@@ -65,6 +91,31 @@ func (s *Storage) Commit(id uint64, size int64) (int64, error) {
if err := os.MkdirAll(filepath.Dir(name), 0o755); err != nil { if err := os.MkdirAll(filepath.Dir(name), 0o755); err != nil {
return 0, err return 0, err
} }
written, err := assemble(name, tempNames)
if err != nil {
return 0, err
}
// If size is less than 0, it means the size is unknown.
// We can't check the size of the file, just skip the check.
// It happens when the request comes from old versions of actions, like `actions/cache@v2`.
if size >= 0 && written != size {
_ = os.Remove(name)
return 0, fmt.Errorf("broken file: %v != %v", written, size)
}
return written, nil
}
// assemble concatenates the uploaded parts into name. A single part, which is what the v2 API
// produces below the client's block threshold, is already the whole archive and is moved.
func assemble(name string, tempNames []string) (int64, error) {
if len(tempNames) == 1 {
info, err := os.Stat(tempNames[0])
if err != nil {
return 0, err
}
return info.Size(), os.Rename(tempNames[0], name)
}
file, err := os.Create(name) file, err := os.Create(name)
if err != nil { if err != nil {
return 0, err return 0, err
@@ -84,16 +135,6 @@ func (s *Storage) Commit(id uint64, size int64) (int64, error) {
} }
written += n written += n
} }
// If size is less than 0, it means the size is unknown.
// We can't check the size of the file, just skip the check.
// It happens when the request comes from old versions of actions, like `actions/cache@v2`.
if size >= 0 && written != size {
_ = file.Close()
_ = os.Remove(name)
return 0, fmt.Errorf("broken file: %v != %v", written, size)
}
return written, nil return written, nil
} }
@@ -119,6 +160,17 @@ func (s *Storage) tempName(id uint64, offset int64) string {
return filepath.Join(s.tempDir(id), fmt.Sprintf("%016x", offset)) return filepath.Join(s.tempDir(id), fmt.Sprintf("%016x", offset))
} }
// blockFilePrefix marks a staged, not yet ordered block, so that tempNames can keep it out of
// Commit's name-ordered concatenation.
const blockFilePrefix = "block-"
func (s *Storage) blockName(id uint64, blockID string) string {
// The block id is client-chosen (base64), so it is hashed rather than trusted as a
// path element.
sum := sha256.Sum256([]byte(blockID))
return filepath.Join(s.tempDir(id), blockFilePrefix+hex.EncodeToString(sum[:]))
}
func (s *Storage) tempNames(id uint64) ([]string, error) { func (s *Storage) tempNames(id uint64) ([]string, error) {
dir := s.tempDir(id) dir := s.tempDir(id)
files, err := os.ReadDir(dir) files, err := os.ReadDir(dir)
@@ -127,7 +179,7 @@ func (s *Storage) tempNames(id uint64) ([]string, error) {
} }
var names []string var names []string
for _, v := range files { for _, v := range files {
if !v.IsDir() { if !v.IsDir() && !strings.HasPrefix(v.Name(), blockFilePrefix) {
names = append(names, filepath.Join(dir, v.Name())) names = append(names, filepath.Join(dir, v.Name()))
} }
} }

View File

@@ -89,6 +89,7 @@ type NewDockerBuildExecutorInput struct {
type NewDockerNetworkCreateExecutorInput struct { type NewDockerNetworkCreateExecutorInput struct {
EnableIPv4 *bool EnableIPv4 *bool
EnableIPv6 *bool EnableIPv6 *bool
RunnerUUID string
} }
// NewDockerPullExecutorInput the input for the NewDockerPullExecutor function // NewDockerPullExecutorInput the input for the NewDockerPullExecutor function

View File

@@ -57,7 +57,7 @@ func logDockerResponse(logger logrus.FieldLogger, dockerResponse io.ReadCloser,
if msg.ErrorDetail.Message != "" { if msg.ErrorDetail.Message != "" {
writeLog(logger, isError, "%s", msg.ErrorDetail.Message) writeLog(logger, isError, "%s", msg.ErrorDetail.Message)
return errors.New(msg.Error) return errors.New(msg.ErrorDetail.Message)
} }
if msg.Status != "" { if msg.Status != "" {

View File

@@ -8,12 +8,69 @@ package container
import ( import (
"context" "context"
"errors"
"fmt"
"strings"
"time"
"gitea.com/gitea/runner/act/common" "gitea.com/gitea/runner/act/common"
"github.com/moby/moby/client" "github.com/moby/moby/client"
) )
const (
networkCreateAttempts = 3
networkCreateRetryDelay = time.Second
// marks the networks a runner creates for its jobs, so it can tell its own leftovers from
// those of another runner sharing the daemon
runnerUUIDLabel = "com.gitea.runner.uuid"
)
// RemoveOrphanNetworks removes the networks this runner created for jobs whose teardown did
// not get to them: the runner died with the job, the teardown timed out, or the network still
// had an endpoint on it at the time. Each one holds a subnet of the daemon's address pool
// until it is removed. Networks created after createdBefore are left alone, so a job starting
// while this runs cannot lose the network it has created but not yet attached a container to.
func RemoveOrphanNetworks(ctx context.Context, runnerUUID string, createdBefore time.Time) error {
cli, err := GetDockerClient(ctx)
if err != nil {
return fmt.Errorf("failed to connect to the docker daemon: %w", err)
}
defer cli.Close()
return removeOrphanNetworks(ctx, cli, runnerUUID, createdBefore)
}
func removeOrphanNetworks(ctx context.Context, cli client.APIClient, runnerUUID string, createdBefore time.Time) error {
networks, err := cli.NetworkList(ctx, client.NetworkListOptions{
Filters: make(client.Filters).Add("label", runnerUUIDLabel+"="+runnerUUID),
})
if err != nil {
return err
}
var errs []error
for _, n := range networks.Items {
result, err := cli.NetworkInspect(ctx, n.ID, client.NetworkInspectOptions{})
if err != nil {
errs = append(errs, fmt.Errorf("failed to inspect network %s: %w", n.Name, err))
continue
}
// the emptiness check, not the label, is what keeps a live job of another process
// sharing this registration safe
if len(result.Network.Containers) != 0 || result.Network.Created.After(createdBefore) {
continue
}
if _, err := cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil {
errs = append(errs, fmt.Errorf("failed to remove network %s: %w", n.Name, err))
continue
}
common.Logger(ctx).Infof("removed docker network %s left behind by an earlier job", n.Name)
}
return errors.Join(errs...)
}
func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExecutorInput) common.Executor { func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExecutorInput) common.Executor {
return func(ctx context.Context) error { return func(ctx context.Context) error {
cli, err := GetDockerClient(ctx) cli, err := GetDockerClient(ctx)
@@ -36,18 +93,45 @@ func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExec
} }
} }
_, err = cli.NetworkCreate(ctx, name, client.NetworkCreateOptions{ for i := range networkCreateAttempts {
Driver: "bridge", if i > 0 {
Scope: "local", common.Logger(ctx).Infof("Waiting for a free docker address pool to create network %s", name)
EnableIPv4: opts.EnableIPv4, select {
EnableIPv6: opts.EnableIPv6, case <-ctx.Done():
}) return ctx.Err()
if err != nil { case <-time.After(time.Duration(i) * networkCreateRetryDelay):
return err }
}
if _, err = cli.NetworkCreate(ctx, name, client.NetworkCreateOptions{
Driver: "bridge",
Scope: "local",
EnableIPv4: opts.EnableIPv4,
EnableIPv6: opts.EnableIPv6,
Labels: runnerLabels(opts.RunnerUUID),
}); err == nil {
return nil
}
if !isAddressPoolExhausted(err) {
return err
}
} }
return fmt.Errorf("docker has no address pool left for this job's network, lower runner.capacity or widen default-address-pools in the docker daemon config: %w", err)
}
}
func runnerLabels(runnerUUID string) map[string]string {
if runnerUUID == "" {
return nil return nil
} }
return map[string]string{runnerUUIDLabel: runnerUUID}
}
// The daemon reports this as a plain invalid-parameter error, the same kind it uses for every
// malformed request, so the message is the only discriminator.
func isAddressPoolExhausted(err error) bool {
msg := err.Error()
return strings.Contains(msg, "all predefined address pools have been fully subnetted") ||
strings.Contains(msg, "could not find an available, non-overlapping IPv4 address pool among the defaults") // docker 24 and older
} }
func NewDockerNetworkRemoveExecutor(name string) common.Executor { func NewDockerNetworkRemoveExecutor(name string) common.Executor {
@@ -66,6 +150,7 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
} }
// For Gitea, reduce log noise // For Gitea, reduce log noise
// common.Logger(ctx).Debugf("%v", networks) // common.Logger(ctx).Debugf("%v", networks)
var errs []error
for _, n := range networks.Items { for _, n := range networks.Items {
if n.Name == name { if n.Name == name {
result, err := cli.NetworkInspect(ctx, n.ID, client.NetworkInspectOptions{}) result, err := cli.NetworkInspect(ctx, n.ID, client.NetworkInspectOptions{})
@@ -73,16 +158,17 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
return err return err
} }
if len(result.Network.Containers) == 0 { // it holds a subnet out of the daemon's pool until something reclaims it
if _, err = cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil { if len(result.Network.Containers) != 0 {
common.Logger(ctx).Debugf("%v", err) common.Logger(ctx).Warnf("Refusing to remove network %s because it still has active endpoints, the idle cleanup reclaims it once they are gone", name)
} continue
} else { }
common.Logger(ctx).Debugf("Refusing to remove network %v because it still has active endpoints", name) if _, err = cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil {
errs = append(errs, fmt.Errorf("failed to remove network %s: %w", name, err))
} }
} }
} }
return err return errors.Join(errs...)
} }
} }

View File

@@ -0,0 +1,50 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package container
import (
"context"
"errors"
"testing"
"time"
cerrdefs "github.com/containerd/errdefs"
"github.com/moby/moby/api/types/network"
mobyclient "github.com/moby/moby/client"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestIsAddressPoolExhausted(t *testing.T) {
assert.True(t, isAddressPoolExhausted(cerrdefs.ErrInvalidArgument.WithMessage("Error response from daemon: all predefined address pools have been fully subnetted")))
assert.True(t, isAddressPoolExhausted(errors.New("could not find an available, non-overlapping IPv4 address pool among the defaults to assign to the network")))
assert.False(t, isAddressPoolExhausted(cerrdefs.ErrInvalidArgument.WithMessage("invalid subnet 10.0.0.0/8: it overlaps with an existing network")))
}
// Of this runner's networks, only the ones nothing is attached to and old enough to predate
// any job now starting are the runner's to reclaim. An unexpected NetworkRemove fails the
// test on its own, since testify has no expectation to match it against.
func TestRemoveOrphanNetworks(t *testing.T) {
ctx := context.Background()
cutoff := time.Date(2026, time.April, 29, 20, 0, 0, 0, time.UTC)
client := &mockDockerClient{}
client.On("NetworkList", ctx, mobyclient.NetworkListOptions{
Filters: make(mobyclient.Filters).Add("label", runnerUUIDLabel+"=runner-1"),
}).Return(mobyclient.NetworkListResult{Items: []network.Summary{
{Network: network.Network{ID: "orphan"}},
{Network: network.Network{ID: "busy"}},
{Network: network.Network{ID: "starting"}},
}}, nil)
client.On("NetworkInspect", ctx, "orphan", mobyclient.NetworkInspectOptions{}).
Return(mobyclient.NetworkInspectResult{}, nil)
client.On("NetworkInspect", ctx, "busy", mobyclient.NetworkInspectOptions{}).
Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"c": {}}}}, nil)
client.On("NetworkInspect", ctx, "starting", mobyclient.NetworkInspectOptions{}).
Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Network: network.Network{Created: cutoff.Add(time.Second)}}}, nil)
client.On("NetworkRemove", ctx, "orphan", mobyclient.NetworkRemoveOptions{}).
Return(mobyclient.NetworkRemoveResult{}, nil)
require.NoError(t, removeOrphanNetworks(ctx, client, "runner-1", cutoff))
client.AssertExpectations(t)
}

View File

@@ -30,23 +30,19 @@ func NewDockerPullExecutor(input NewDockerPullExecutorInput) common.Executor {
return nil return nil
} }
pull := input.ForcePull // skip the pull when the image is already here: either none was forced, or a digest
if !pull { // pins the content so a forced pull could only fetch the same bytes again
if !input.ForcePull || isPinnedImage(input.Image) {
imageExists, err := ImageExistsLocally(ctx, input.Image, input.Platform) imageExists, err := ImageExistsLocally(ctx, input.Image, input.Platform)
logger.Debugf("Image exists? %v", imageExists) logger.Debugf("Image exists? %v", imageExists)
if err != nil { if err != nil {
return fmt.Errorf("unable to determine if image already exists for image '%s' (%s): %w", input.Image, input.Platform, err) return fmt.Errorf("unable to determine if image already exists for image '%s' (%s): %w", input.Image, input.Platform, err)
} }
if imageExists {
if !imageExists { return nil
pull = true
} }
} }
if !pull {
return nil
}
imageRef := cleanImage(ctx, input.Image) imageRef := cleanImage(ctx, input.Image)
logger.Debugf("pulling image '%v' (%s)", imageRef, input.Platform) logger.Debugf("pulling image '%v' (%s)", imageRef, input.Platform)
@@ -61,22 +57,32 @@ func NewDockerPullExecutor(input NewDockerPullExecutorInput) common.Executor {
return err return err
} }
reader, err := cli.ImagePull(ctx, imageRef, imagePullOptions) // the daemon reports a failure that happens after the first progress line in the
// stream rather than on the call itself, so both have to be checked
_ = logDockerResponse(logger, reader, err != nil) pullOnce := func(opts client.ImagePullOptions) error {
if err != nil { reader, err := cli.ImagePull(ctx, imageRef, opts)
if imagePullOptions.RegistryAuth != "" && strings.Contains(err.Error(), "unauthorized") { streamErr := logDockerResponse(logger, reader, err != nil)
logger.Errorf("pulling image '%v' (%s) failed with credentials %s retrying without them, please check for stale docker config files", imageRef, input.Platform, err.Error())
imagePullOptions.RegistryAuth = ""
reader, err = cli.ImagePull(ctx, imageRef, imagePullOptions)
_ = logDockerResponse(logger, reader, err != nil)
}
if err != nil { if err != nil {
return fmt.Errorf("failed to pull image '%s' (%s): %w", imageRef, input.Platform, err) return err
} }
return streamErr
} }
return nil
err = pullOnce(imagePullOptions)
if err != nil && imagePullOptions.RegistryAuth != "" && strings.Contains(err.Error(), "unauthorized") {
logger.Errorf("pulling image '%v' (%s) failed with credentials %s retrying without them, please check for stale docker config files", imageRef, input.Platform, err.Error())
imagePullOptions.RegistryAuth = ""
err = pullOnce(imagePullOptions)
}
if err == nil {
return nil
}
// a registry that is down should not fail a job whose image is already here
if exists, existsErr := ImageExistsLocally(ctx, input.Image, input.Platform); existsErr == nil && exists {
logger.Warnf("could not update image '%s' (%s), continuing with the local copy: %v", imageRef, input.Platform, err)
return nil
}
return fmt.Errorf("failed to pull image '%s' (%s): %w", imageRef, input.Platform, err)
} }
} }
@@ -122,6 +128,15 @@ func getImagePullOptions(ctx context.Context, input NewDockerPullExecutorInput)
return imagePullOptions, nil return imagePullOptions, nil
} }
func isPinnedImage(image string) bool {
ref, err := reference.ParseAnyReference(image)
if err != nil {
return false
}
_, pinned := ref.(reference.Canonical)
return pinned
}
func cleanImage(ctx context.Context, imageName string) string { func cleanImage(ctx context.Context, imageName string) string {
ref, err := reference.ParseAnyReference(imageName) ref, err := reference.ParseAnyReference(imageName)
if err != nil { if err != nil {

View File

@@ -6,11 +6,15 @@ package container
import ( import (
"context" "context"
"io"
"strings"
"testing" "testing"
"github.com/docker/cli/cli/config" "github.com/docker/cli/cli/config"
log "github.com/sirupsen/logrus" log "github.com/sirupsen/logrus"
"github.com/sirupsen/logrus/hooks/test"
assert "github.com/stretchr/testify/assert" assert "github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
) )
func init() { func init() {
@@ -65,3 +69,21 @@ func TestGetImagePullOptions(t *testing.T) {
assert.NoError(t, err, "Failed to create ImagePullOptions") //nolint:testifylint // pre-existing issue from nektos/act assert.NoError(t, err, "Failed to create ImagePullOptions") //nolint:testifylint // pre-existing issue from nektos/act
assert.Equal(t, "eyJ1c2VybmFtZSI6InVzZXJuYW1lIiwicGFzc3dvcmQiOiJwYXNzd29yZFxuIiwic2VydmVyYWRkcmVzcyI6Imh0dHBzOi8vaW5kZXguZG9ja2VyLmlvL3YxLyJ9", options.RegistryAuth, "RegistryAuth should be taken from local docker config") assert.Equal(t, "eyJ1c2VybmFtZSI6InVzZXJuYW1lIiwicGFzc3dvcmQiOiJwYXNzd29yZFxuIiwic2VydmVyYWRkcmVzcyI6Imh0dHBzOi8vaW5kZXguZG9ja2VyLmlvL3YxLyJ9", options.RegistryAuth, "RegistryAuth should be taken from local docker config")
} }
// A digest-pinned image is immutable, so its local copy is always current.
func TestIsPinnedImage(t *testing.T) {
assert.True(t, isPinnedImage("alpine@sha256:28bd5fe8b56d1bd048e5babf5b10710ebe0bae67db86916198a6eec434943f8b"))
assert.False(t, isPinnedImage("alpine:latest"))
}
// The pull path reports a failure the daemon sent mid-stream, so it must carry the reason
// whichever of the two shapes the daemon used.
func TestLogDockerResponseError(t *testing.T) {
logger, _ := test.NewNullLogger()
streamErr := func(line string) error {
return logDockerResponse(logger, io.NopCloser(strings.NewReader(line)), false)
}
require.EqualError(t, streamErr(`{"error":"toomanyrequests: rate limit exceeded"}`), "toomanyrequests: rate limit exceeded")
require.EqualError(t, streamErr(`{"errorDetail":{"message":"unexpected EOF"}}`), "unexpected EOF")
require.NoError(t, streamErr(`{"status":"Downloading"}`))
}

View File

@@ -14,6 +14,7 @@ import (
"fmt" "fmt"
"io" "io"
"os" "os"
"path"
"path/filepath" "path/filepath"
"regexp" "regexp"
"runtime" "runtime"
@@ -42,6 +43,7 @@ import (
"github.com/moby/moby/api/types/system" "github.com/moby/moby/api/types/system"
"github.com/moby/moby/client" "github.com/moby/moby/client"
specs "github.com/opencontainers/image-spec/specs-go/v1" specs "github.com/opencontainers/image-spec/specs-go/v1"
"github.com/sirupsen/logrus"
) )
// drainGracePeriod bounds how long we wait for an output-copy goroutine to // drainGracePeriod bounds how long we wait for an output-copy goroutine to
@@ -386,32 +388,71 @@ func (cr *containerReference) find() common.Executor {
} }
} }
// isContainerGone reports whether a failed remove still left the container gone (NotFound or Conflict).
func isContainerGone(err error) bool {
return cerrdefs.IsNotFound(err) || cerrdefs.IsConflict(err)
}
func (cr *containerReference) remove() common.Executor { func (cr *containerReference) remove() common.Executor {
return func(ctx context.Context) error { return func(ctx context.Context) error {
if cr.id == "" { idOrName := cr.id
if idOrName == "" && cr.input != nil {
idOrName = cr.input.Name
}
if idOrName == "" {
return nil return nil
} }
logger := common.Logger(ctx) logger := common.Logger(ctx)
_, err := cr.cli.ContainerRemove(ctx, cr.id, client.ContainerRemoveOptions{ // Kill first so removal never waits out a daemon's stop timeout: Docker kills outright
// on a forced remove, Podman sends SIGTERM and waits. Only worth it for a container
// this started, and removal can still deal with one it could not kill.
if cr.id != "" {
_, err := cr.cli.ContainerKill(ctx, cr.id, client.ContainerKillOptions{Signal: "SIGKILL"})
if err != nil && !cerrdefs.IsConflict(err) && !cerrdefs.IsNotFound(err) {
logger.Debugf("Container %s could not be killed: %v", cr.id, err)
}
}
_, err := cr.cli.ContainerRemove(ctx, idOrName, client.ContainerRemoveOptions{
RemoveVolumes: true, RemoveVolumes: true,
Force: true, Force: true,
}) })
if err != nil && !isContainerGone(err) { switch {
logger.Error(fmt.Errorf("failed to remove container: %w", err)) case cerrdefs.IsConflict(err):
// the daemon's own AutoRemove teardown is running, and it releases the volume
// references and the network endpoint only once it finishes
cr.waitForRemoval(ctx, idOrName)
case err != nil && !cerrdefs.IsNotFound(err):
logger.Error(fmt.Errorf("failed to remove container %s: %w", idOrName, err))
return nil // keep the id, the container is still there for a later Remove()
} }
logger.Debugf("Removed container: %v", cr.id) logger.Debugf("Removed container: %v", idOrName)
cr.id = "" cr.id = ""
return nil return nil
} }
} }
func (cr *containerReference) waitForRemoval(ctx context.Context, idOrName string) {
// per container, against the one minute the post-job executor allows for the whole
// cleanup, so a job with several services can spend most of that budget here
ctx, cancel := context.WithTimeout(ctx, 15*time.Second)
defer cancel()
waitResult := cr.cli.ContainerWait(ctx, idOrName, client.ContainerWaitOptions{
Condition: container.WaitConditionRemoved,
})
select {
case <-waitResult.Result:
case <-waitResult.Error:
case <-ctx.Done():
// the client delivers the result over an unbuffered channel, so leave a receiver
// behind or its goroutine parks on the send for the lifetime of the process
go func() {
select {
case <-waitResult.Result:
case <-waitResult.Error:
}
}()
common.Logger(ctx).Warnf("Timed out waiting for the daemon to remove container %s, its volumes and network may be left behind", idOrName)
}
}
func (cr *containerReference) mergeContainerConfigs(ctx context.Context, config *container.Config, hostConfig *container.HostConfig) (*container.Config, *container.HostConfig, error) { func (cr *containerReference) mergeContainerConfigs(ctx context.Context, config *container.Config, hostConfig *container.HostConfig) (*container.Config, *container.HostConfig, error) {
logger := common.Logger(ctx) logger := common.Logger(ctx)
input := cr.input input := cr.input
@@ -461,6 +502,16 @@ func (cr *containerReference) mergeContainerConfigs(ctx context.Context, config
return nil, nil, fmt.Errorf("Cannot process container options: '%s': '%w'", input.Options, err) return nil, nil, fmt.Errorf("Cannot process container options: '%s': '%w'", input.Options, err)
} }
// For Gitea
// When privileged mode is disabled, container.options is workflow-controlled
// untrusted input. Strip the HostConfig fields that would let a workflow break
// out of the container (host namespaces, capability expansion, security profile
// overrides, device and runtime access). Otherwise these survive into the final
// HostConfig even though --privileged is forced off.
if !hostConfig.Privileged {
sanitizeOptionsHostConfig(logger, containerConfig.HostConfig)
}
logger.Debugf("Custom container.Config from options ==> %+v", containerConfig.Config) logger.Debugf("Custom container.Config from options ==> %+v", containerConfig.Config)
err = mergo.Merge(config, containerConfig.Config, mergo.WithOverride, mergo.WithAppendSlice) err = mergo.Merge(config, containerConfig.Config, mergo.WithOverride, mergo.WithAppendSlice)
@@ -814,23 +865,59 @@ func (cr *containerReference) waitForCommand(ctx context.Context, isTerminal boo
} }
} }
// mkdirInContainer creates containerPath and returns it with the symlinked components
// replaced by the targets the daemon reports for them. Docker 29.7 rejects tar entries
// traversing a symlink to an absolute target, like the "/var/run" of most images, with
// "path escapes from parent", and not every daemon creates the implied parents of a
// directory entry, so one entry per missing component is extracted at the deepest
// existing ancestor.
// WORKAROUND: https://github.com/moby/moby/issues/53258
func (cr *containerReference) mkdirInContainer(ctx context.Context, containerPath string) (string, error) {
parts := strings.Split(strings.Trim(path.Clean(containerPath), "/"), "/")
existing := "/"
for i, part := range parts {
if part == "" {
return existing, nil
}
stat, err := cr.cli.ContainerStatPath(ctx, cr.id, client.ContainerStatPathOptions{Path: path.Join(existing, part)})
if err != nil {
// nothing below exists either, so create the remaining components
return path.Join(existing, path.Join(parts[i:]...)), cr.mkdirEntries(ctx, existing, parts[i:])
}
existing = path.Join(existing, part)
if target := stat.Stat.LinkTarget; target != "" {
if !path.IsAbs(target) {
target = path.Join(path.Dir(existing), target)
}
existing = target
}
}
return existing, nil
}
func (cr *containerReference) mkdirEntries(ctx context.Context, destPath string, missing []string) error {
buf := &bytes.Buffer{}
tw := tar.NewWriter(buf)
for i := range missing {
_ = tw.WriteHeader(&tar.Header{
Name: path.Join(missing[:i+1]...),
Mode: 0o777,
Typeflag: tar.TypeDir,
})
}
tw.Close()
_, err := cr.cli.CopyToContainer(ctx, cr.id, client.CopyToContainerOptions{
DestinationPath: destPath,
Content: buf,
})
return err
}
func (cr *containerReference) CopyTarStream(ctx context.Context, destPath string, tarStream io.Reader) error { func (cr *containerReference) CopyTarStream(ctx context.Context, destPath string, tarStream io.Reader) error {
if cr.id == "" { if cr.id == "" {
return cr.missingContainerError("copy to %s", destPath) return cr.missingContainerError("copy to %s", destPath)
} }
// Mkdir destPath, err := cr.mkdirInContainer(ctx, destPath)
buf := &bytes.Buffer{}
tw := tar.NewWriter(buf)
_ = tw.WriteHeader(&tar.Header{
Name: destPath,
Mode: 0o777,
Typeflag: tar.TypeDir,
})
tw.Close()
_, err := cr.cli.CopyToContainer(ctx, cr.id, client.CopyToContainerOptions{
DestinationPath: "/",
Content: buf,
})
if err != nil { if err != nil {
return fmt.Errorf("failed to mkdir to copy content to container: %w", err) return fmt.Errorf("failed to mkdir to copy content to container: %w", err)
} }
@@ -855,6 +942,10 @@ func (cr *containerReference) copyDir(dstPath, srcPath string, useGitIgnore bool
return cr.missingContainerError("copy directory to %s", dstPath) return cr.missingContainerError("copy directory to %s", dstPath)
} }
logger := common.Logger(ctx) logger := common.Logger(ctx)
dstPath, err := cr.mkdirInContainer(ctx, dstPath)
if err != nil {
return fmt.Errorf("failed to mkdir to copy directory to container: %w", err)
}
tarFile, err := os.CreateTemp("", "act") tarFile, err := os.CreateTemp("", "act")
if err != nil { if err != nil {
return err return err
@@ -1062,6 +1153,77 @@ func (cr *containerReference) wait() common.Executor {
} }
} }
// For Gitea
// sanitizeOptionsHostConfig clears the HostConfig fields parsed from a
// workflow-controlled container.options string that could be used to escape the
// container when privileged mode is disabled. It must only be called when the
// runner has privileged mode turned off; with privileged mode enabled the
// administrator has already opted into host access.
func sanitizeOptionsHostConfig(logger logrus.FieldLogger, hostConfig *container.HostConfig) {
warn := func(option string) {
logger.Warnf("container option %q is not allowed when privileged mode is disabled and will be ignored", option)
}
if hostConfig.PidMode != "" {
warn("--pid")
hostConfig.PidMode = ""
}
if hostConfig.IpcMode != "" {
warn("--ipc")
hostConfig.IpcMode = ""
}
if hostConfig.UTSMode != "" {
warn("--uts")
hostConfig.UTSMode = ""
}
if hostConfig.CgroupnsMode != "" {
warn("--cgroupns")
hostConfig.CgroupnsMode = ""
}
// UsernsMode is set from the runner-controlled input; never let options
// override it (e.g. --userns=host disables user namespace remapping).
if hostConfig.UsernsMode != "" {
warn("--userns")
hostConfig.UsernsMode = ""
}
if len(hostConfig.CapAdd) > 0 {
warn("--cap-add")
hostConfig.CapAdd = nil
}
if len(hostConfig.SecurityOpt) > 0 {
warn("--security-opt")
hostConfig.SecurityOpt = nil
}
if len(hostConfig.Devices) > 0 {
warn("--device")
hostConfig.Devices = nil
}
if len(hostConfig.DeviceCgroupRules) > 0 {
warn("--device-cgroup-rule")
hostConfig.DeviceCgroupRules = nil
}
if len(hostConfig.DeviceRequests) > 0 {
warn("--gpus")
hostConfig.DeviceRequests = nil
}
if len(hostConfig.VolumesFrom) > 0 {
warn("--volumes-from")
hostConfig.VolumesFrom = nil
}
if hostConfig.Runtime != "" {
warn("--runtime")
hostConfig.Runtime = ""
}
if hostConfig.CgroupParent != "" {
warn("--cgroup-parent")
hostConfig.CgroupParent = ""
}
if len(hostConfig.Sysctls) > 0 {
warn("--sysctl")
hostConfig.Sysctls = nil
}
}
// For Gitea // For Gitea
// sanitizeConfig remove the invalid configurations from `config` and `hostConfig` // sanitizeConfig remove the invalid configurations from `config` and `hostConfig`
func (cr *containerReference) sanitizeConfig(ctx context.Context, config *container.Config, hostConfig *container.HostConfig) (*container.Config, *container.HostConfig) { func (cr *containerReference) sanitizeConfig(ctx context.Context, config *container.Config, hostConfig *container.HostConfig) (*container.Config, *container.HostConfig) {

View File

@@ -5,6 +5,7 @@
package container package container
import ( import (
"archive/tar"
"bufio" "bufio"
"bytes" "bytes"
"context" "context"
@@ -92,6 +93,11 @@ func (m *mockDockerClient) ExecInspect(ctx context.Context, execID string, opts
return args.Get(0).(mobyclient.ExecInspectResult), args.Error(1) return args.Get(0).(mobyclient.ExecInspectResult), args.Error(1)
} }
func (m *mockDockerClient) ContainerStatPath(ctx context.Context, containerID string, opts mobyclient.ContainerStatPathOptions) (mobyclient.ContainerStatPathResult, error) {
args := m.Called(ctx, containerID, opts)
return args.Get(0).(mobyclient.ContainerStatPathResult), args.Error(1)
}
func (m *mockDockerClient) ContainerAttach(ctx context.Context, containerID string, opts mobyclient.ContainerAttachOptions) (mobyclient.ContainerAttachResult, error) { func (m *mockDockerClient) ContainerAttach(ctx context.Context, containerID string, opts mobyclient.ContainerAttachOptions) (mobyclient.ContainerAttachResult, error) {
args := m.Called(ctx, containerID, opts) args := m.Called(ctx, containerID, opts)
return args.Get(0).(mobyclient.ContainerAttachResult), args.Error(1) return args.Get(0).(mobyclient.ContainerAttachResult), args.Error(1)
@@ -122,6 +128,26 @@ func (m *mockDockerClient) ContainerRemove(ctx context.Context, id string, opts
return args.Get(0).(mobyclient.ContainerRemoveResult), args.Error(1) return args.Get(0).(mobyclient.ContainerRemoveResult), args.Error(1)
} }
func (m *mockDockerClient) ContainerKill(ctx context.Context, id string, opts mobyclient.ContainerKillOptions) (mobyclient.ContainerKillResult, error) {
args := m.Called(ctx, id, opts)
return args.Get(0).(mobyclient.ContainerKillResult), args.Error(1)
}
func (m *mockDockerClient) NetworkList(ctx context.Context, opts mobyclient.NetworkListOptions) (mobyclient.NetworkListResult, error) {
args := m.Called(ctx, opts)
return args.Get(0).(mobyclient.NetworkListResult), args.Error(1)
}
func (m *mockDockerClient) NetworkInspect(ctx context.Context, id string, opts mobyclient.NetworkInspectOptions) (mobyclient.NetworkInspectResult, error) {
args := m.Called(ctx, id, opts)
return args.Get(0).(mobyclient.NetworkInspectResult), args.Error(1)
}
func (m *mockDockerClient) NetworkRemove(ctx context.Context, id string, opts mobyclient.NetworkRemoveOptions) (mobyclient.NetworkRemoveResult, error) {
args := m.Called(ctx, id, opts)
return args.Get(0).(mobyclient.NetworkRemoveResult), args.Error(1)
}
type endlessReader struct { type endlessReader struct {
io.Reader io.Reader
} }
@@ -315,15 +341,37 @@ func TestDockerWaitFailure(t *testing.T) {
client.AssertExpectations(t) client.AssertExpectations(t)
} }
// stubStatPath answers path resolution: the given paths exist, mapped to their target
// when they are a symlink, everything else does not exist.
func stubStatPath(client *mockDockerClient, existing map[string]string) {
for containerPath, target := range existing {
client.On("ContainerStatPath", mock.Anything, "123", mobyclient.ContainerStatPathOptions{Path: containerPath}).
Return(mobyclient.ContainerStatPathResult{Stat: container.PathStat{LinkTarget: target}}, nil).Maybe()
}
client.On("ContainerStatPath", mock.Anything, "123", mock.Anything).
Return(mobyclient.ContainerStatPathResult{}, cerrdefs.ErrNotFound).Maybe()
}
// The mkdir tarball is extracted at the deepest existing ancestor, with entries relative
// to it that never traverse the "/var/run" symlink, see moby/moby#53258.
func TestDockerCopyTarStream(t *testing.T) { func TestDockerCopyTarStream(t *testing.T) {
ctx := context.Background() ctx := context.Background()
var mkdirNames []string
client := &mockDockerClient{} client := &mockDockerClient{}
stubStatPath(client, map[string]string{"/var": "", "/var/run": "/run", "/run": ""})
client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool { client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
return opts.DestinationPath == "/" && opts.Content != nil if opts.DestinationPath != "/run" || opts.Content == nil {
return false
}
tr := tar.NewReader(opts.Content)
for hdr, err := tr.Next(); err == nil; hdr, err = tr.Next() {
mkdirNames = append(mkdirNames, hdr.Name)
}
return true
})).Return(mobyclient.CopyToContainerResult{}, nil) })).Return(mobyclient.CopyToContainerResult{}, nil)
client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool { client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
return opts.DestinationPath == "/var/run/act" && opts.Content != nil return opts.DestinationPath == "/run/act" && opts.Content != nil
})).Return(mobyclient.CopyToContainerResult{}, nil) })).Return(mobyclient.CopyToContainerResult{}, nil)
cr := &containerReference{ cr := &containerReference{
id: "123", id: "123",
@@ -333,87 +381,85 @@ func TestDockerCopyTarStream(t *testing.T) {
}, },
} }
_ = cr.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{}) require.NoError(t, cr.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{}))
assert.Equal(t, []string{"act"}, mkdirNames)
client.AssertExpectations(t) client.AssertExpectations(t)
} }
func TestDockerCopyTarStreamErrorInCopyFiles(t *testing.T) { func TestDockerCopyTarStreamErrors(t *testing.T) {
ctx := context.Background()
merr := errors.New("Failure") merr := errors.New("Failure")
for _, testCase := range []struct {
name string
mkdirErr error
copyErr error
}{
{"mkdir", merr, nil},
{"copy content", nil, merr},
} {
t.Run(testCase.name, func(t *testing.T) {
ctx := context.Background()
client := &mockDockerClient{} client := &mockDockerClient{}
client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool { stubStatPath(client, map[string]string{"/var": "", "/var/run": ""})
return opts.DestinationPath == "/" && opts.Content != nil client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
})).Return(mobyclient.CopyToContainerResult{}, merr) return opts.DestinationPath == "/var/run" && opts.Content != nil
cr := &containerReference{ })).Return(mobyclient.CopyToContainerResult{}, testCase.mkdirErr)
id: "123", client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
cli: client, return opts.DestinationPath == "/var/run/act" && opts.Content != nil
input: &NewContainerInput{ })).Return(mobyclient.CopyToContainerResult{}, testCase.copyErr).Maybe()
Image: "image", cr := &containerReference{
}, id: "123",
cli: client,
input: &NewContainerInput{
Image: "image",
},
}
require.ErrorIs(t, cr.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{}), merr)
client.AssertExpectations(t)
})
} }
err := cr.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{})
assert.ErrorIs(t, err, merr) //nolint:testifylint // pre-existing issue from nektos/act
client.AssertExpectations(t)
}
func TestDockerCopyTarStreamErrorInMkdir(t *testing.T) {
ctx := context.Background()
merr := errors.New("Failure")
client := &mockDockerClient{}
client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
return opts.DestinationPath == "/" && opts.Content != nil
})).Return(mobyclient.CopyToContainerResult{}, nil)
client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
return opts.DestinationPath == "/var/run/act" && opts.Content != nil
})).Return(mobyclient.CopyToContainerResult{}, merr)
cr := &containerReference{
id: "123",
cli: client,
input: &NewContainerInput{
Image: "image",
},
}
err := cr.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{})
assert.ErrorIs(t, err, merr) //nolint:testifylint // pre-existing issue from nektos/act
client.AssertExpectations(t)
} }
// A remove that raced the daemon's AutoRemove teardown is not a failure and must not // A remove that raced the daemon's AutoRemove teardown is not a failure and must not
// be logged as one. // be logged as one.
func TestRemoveIgnoresAutoRemoveRace(t *testing.T) { func TestRemoveIgnoresAutoRemoveRace(t *testing.T) {
removeOpts := mobyclient.ContainerRemoveOptions{RemoveVolumes: true, Force: true} removeOpts := mobyclient.ContainerRemoveOptions{RemoveVolumes: true, Force: true}
killOpts := mobyclient.ContainerKillOptions{Signal: "SIGKILL"}
for _, tc := range []struct { for _, tc := range []struct {
name string name string
err error err error
wantLogs bool wantWait bool
wantFailure bool
}{ }{
{name: "removal in progress", err: cerrdefs.ErrConflict.WithMessage("removal of container abc is already in progress")}, {name: "removal in progress", err: cerrdefs.ErrConflict.WithMessage("removal of container abc is already in progress"), wantWait: true},
{name: "already removed", err: cerrdefs.ErrNotFound.WithMessage("No such container: abc")}, {name: "already removed", err: cerrdefs.ErrNotFound.WithMessage("No such container: abc")},
{name: "removed cleanly", err: nil}, {name: "removed cleanly", err: nil},
{name: "real failure", err: errors.New("driver failed to remove root filesystem"), wantLogs: true}, {name: "real failure", err: errors.New("driver failed to remove root filesystem"), wantFailure: true},
} { } {
t.Run(tc.name, func(t *testing.T) { t.Run(tc.name, func(t *testing.T) {
logger, hook := test.NewNullLogger() logger, hook := test.NewNullLogger()
ctx := common.WithLogger(context.Background(), logger) ctx := common.WithLogger(context.Background(), logger)
client := &mockDockerClient{} client := &mockDockerClient{}
client.On("ContainerKill", ctx, "abc", killOpts).Return(mobyclient.ContainerKillResult{}, nil)
client.On("ContainerRemove", ctx, "abc", removeOpts).Return(mobyclient.ContainerRemoveResult{}, tc.err) client.On("ContainerRemove", ctx, "abc", removeOpts).Return(mobyclient.ContainerRemoveResult{}, tc.err)
if tc.wantWait {
removed := make(chan container.WaitResponse, 1)
removed <- container.WaitResponse{}
client.On("ContainerWait", mock.Anything, "abc", mobyclient.ContainerWaitOptions{Condition: container.WaitConditionRemoved}).
Return(mobyclient.ContainerWaitResult{Result: removed})
}
cr := &containerReference{id: "abc", cli: client} cr := &containerReference{id: "abc", cli: client}
require.NoError(t, cr.remove()(ctx)) require.NoError(t, cr.remove()(ctx))
assert.Empty(t, cr.id) // a failure keeps the id, so a later Remove() can retry it
if tc.wantFailure {
if tc.wantLogs { assert.Equal(t, "abc", cr.id)
assert.Len(t, hook.AllEntries(), 1) assert.Len(t, hook.AllEntries(), 1)
} else { } else {
assert.Empty(t, cr.id)
assert.Empty(t, hook.AllEntries()) assert.Empty(t, hook.AllEntries())
} }
client.AssertExpectations(t) client.AssertExpectations(t)
@@ -421,6 +467,20 @@ func TestRemoveIgnoresAutoRemoveRace(t *testing.T) {
} }
} }
// A container whose id was never learned, because find() could not reach the daemon or
// create() lost its reply, must still be removed rather than leaking with its network. It
// was never started here, so it is not worth a kill of its own.
func TestRemoveWithoutIDUsesName(t *testing.T) {
ctx := context.Background()
client := &mockDockerClient{}
client.On("ContainerRemove", ctx, "job-1", mobyclient.ContainerRemoveOptions{RemoveVolumes: true, Force: true}).
Return(mobyclient.ContainerRemoveResult{}, nil)
cr := &containerReference{cli: client, input: &NewContainerInput{Name: "job-1"}}
require.NoError(t, cr.remove()(ctx))
client.AssertExpectations(t)
}
// find() must drop a stale cached id so later Copy/Exec don't hit the // find() must drop a stale cached id so later Copy/Exec don't hit the
// daemon with a torn-down container. // daemon with a torn-down container.
func TestFindRevalidatesStaleID(t *testing.T) { func TestFindRevalidatesStaleID(t *testing.T) {
@@ -542,9 +602,8 @@ func TestDockerCopyToSymlinkPath(t *testing.T) {
_ = rc.Close()(ctx) _ = rc.Close()(ctx)
}) })
// CopyTarStream first creates the destination directory by extracting a tar at "/", // CopyTarStream resolves the var/run symlink and creates act below its target, the
// which makes the daemon mkdir var, then var/run (the symlink), then act — the exact // exact step that fails on a broken daemon.
// step that fails on the broken daemon.
err := rc.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{}) err := rc.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{})
require.NoError(t, err) require.NoError(t, err)
} }
@@ -625,6 +684,110 @@ func TestCheckVolumes(t *testing.T) {
} }
} }
func TestSanitizeOptionsHostConfig(t *testing.T) {
logger, _ := test.NewNullLogger()
dangerous := func() *container.HostConfig {
return &container.HostConfig{
PidMode: "host",
IpcMode: "host",
UTSMode: "host",
CgroupnsMode: "host",
UsernsMode: "host",
CapAdd: []string{"ALL"},
SecurityOpt: []string{"seccomp=unconfined", "apparmor=unconfined"},
VolumesFrom: []string{"other"},
Runtime: "runc",
Resources: container.Resources{
CgroupParent: "/custom",
Devices: []container.DeviceMapping{{PathOnHost: "/dev/sda", PathInContainer: "/dev/sda", CgroupPermissions: "rwm"}},
DeviceCgroupRules: []string{"a *:* rwm"},
},
Sysctls: map[string]string{"net.ipv4.ip_forward": "1"},
}
}
hostConfig := dangerous()
sanitizeOptionsHostConfig(logger, hostConfig)
assert.Empty(t, string(hostConfig.PidMode))
assert.Empty(t, string(hostConfig.IpcMode))
assert.Empty(t, string(hostConfig.UTSMode))
assert.Empty(t, string(hostConfig.CgroupnsMode))
assert.Empty(t, string(hostConfig.UsernsMode))
assert.Empty(t, hostConfig.CapAdd)
assert.Empty(t, hostConfig.SecurityOpt)
assert.Empty(t, hostConfig.Devices)
assert.Empty(t, hostConfig.DeviceCgroupRules)
assert.Empty(t, hostConfig.VolumesFrom)
assert.Empty(t, hostConfig.Runtime)
assert.Empty(t, hostConfig.CgroupParent)
assert.Empty(t, hostConfig.Sysctls)
}
func TestMergeContainerConfigsStripsDangerousOptionsWhenUnprivileged(t *testing.T) {
// OS-independent options only: --device parsing requires a linux/windows
// server OS, which is not guaranteed for the test host.
const dangerousOptions = "--pid=host --ipc=host --uts=host --cgroupns=host " +
"--userns=host --cap-add=ALL --security-opt seccomp=unconfined " +
"--security-opt apparmor=unconfined --volumes-from other " +
"--runtime runc --cgroup-parent /custom --sysctl net.ipv4.ip_forward=1"
t.Run("unprivileged strips host-escape options", func(t *testing.T) {
logger, _ := test.NewNullLogger()
ctx := common.WithLogger(context.Background(), logger)
cr := &containerReference{
input: &NewContainerInput{
Options: dangerousOptions,
NetworkMode: "bridge",
UsernsMode: "private",
},
}
_, hostConfig, err := cr.mergeContainerConfigs(ctx, &container.Config{}, &container.HostConfig{
Privileged: false,
UsernsMode: container.UsernsMode("private"),
NetworkMode: container.NetworkMode("bridge"),
})
require.NoError(t, err)
assert.False(t, hostConfig.Privileged)
assert.Empty(t, string(hostConfig.PidMode))
assert.Empty(t, string(hostConfig.IpcMode))
assert.Empty(t, string(hostConfig.UTSMode))
assert.Empty(t, string(hostConfig.CgroupnsMode))
// UsernsMode must keep the runner-controlled value, not the one from options.
assert.Equal(t, "private", string(hostConfig.UsernsMode))
assert.Empty(t, hostConfig.CapAdd)
assert.Empty(t, hostConfig.SecurityOpt)
assert.Empty(t, hostConfig.VolumesFrom)
assert.Empty(t, hostConfig.Runtime)
assert.Empty(t, hostConfig.CgroupParent)
assert.Empty(t, hostConfig.Sysctls)
})
t.Run("privileged preserves options", func(t *testing.T) {
logger, _ := test.NewNullLogger()
ctx := common.WithLogger(context.Background(), logger)
cr := &containerReference{
input: &NewContainerInput{
Options: "--pid=host --cap-add=ALL --security-opt seccomp=unconfined",
NetworkMode: "bridge",
},
}
_, hostConfig, err := cr.mergeContainerConfigs(ctx, &container.Config{}, &container.HostConfig{
Privileged: true,
NetworkMode: container.NetworkMode("bridge"),
})
require.NoError(t, err)
assert.Equal(t, "host", string(hostConfig.PidMode))
assert.Equal(t, []string{"ALL"}, hostConfig.CapAdd)
assert.Equal(t, []string{"seccomp=unconfined"}, hostConfig.SecurityOpt)
})
}
func TestCheckVolumesRejectsEscapingHostPaths(t *testing.T) { func TestCheckVolumesRejectsEscapingHostPaths(t *testing.T) {
logger, _ := test.NewNullLogger() logger, _ := test.NewNullLogger()
ctx := common.WithLogger(context.Background(), logger) ctx := common.WithLogger(context.Background(), logger)

View File

@@ -9,6 +9,7 @@ package container
import ( import (
"context" "context"
"runtime" "runtime"
"time"
"gitea.com/gitea/runner/act/common" "gitea.com/gitea/runner/act/common"
@@ -72,3 +73,7 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
return nil return nil
} }
} }
func RemoveOrphanNetworks(ctx context.Context, runnerUUID string, createdBefore time.Time) error {
return nil
}

View File

@@ -595,8 +595,8 @@ func actionStagePaths(step actionStep) (actionDir, actionPath, actionName, conta
rc := step.getRunContext() rc := step.getRunContext()
stepModel := step.getStepModel() stepModel := step.getStepModel()
if _, ok := step.(*stepActionRemote); ok { if sar, ok := step.(*stepActionRemote); ok {
actionDir = fmt.Sprintf("%s/%s", rc.ActionCacheDir(), stepModel.UsesHash()) actionDir = sar.actionDir()
actionPath = newRemoteAction(stepModel.Uses).Path actionPath = newRemoteAction(stepModel.Uses).Path
} else { } else {
actionDir = filepath.Join(rc.Config.Workdir, stepModel.Uses) actionDir = filepath.Join(rc.Config.Workdir, stepModel.Uses)

View File

@@ -7,8 +7,11 @@ package runner
import ( import (
"bytes" "bytes"
"context" "context"
"encoding/base64"
"encoding/json"
"fmt" "fmt"
"io" "io"
"net/url"
"os" "os"
"slices" "slices"
"strings" "strings"
@@ -167,6 +170,76 @@ func withStepLogger(ctx context.Context, stepNumber int, stepID, stepName, stage
type entryProcessor func(entry *logrus.Entry) *logrus.Entry type entryProcessor func(entry *logrus.Entry) *logrus.Entry
// secretValueEncoders are the shapes a secret takes on its way into a log: a base64
// payload, a JSON string, or a URL component. An action that serializes a secret leaks
// it in one of these forms, which a mask of the verbatim value alone does not catch, so
// every form is masked as well. This mirrors the value encoders of GitHub's runner.
var secretValueEncoders = []func(string) string{
func(v string) string { return base64.StdEncoding.EncodeToString([]byte(v)) },
base64ShiftEncoder(1),
base64ShiftEncoder(2),
jsonStringEscape,
jsonStringEscapeNoHTML,
url.QueryEscape,
url.PathEscape,
}
// minShiftedBase64Len is the shortest shifted base64 fragment worth masking. A shorter
// one carries too few bytes of the secret to identify it and would mask unrelated output.
const minShiftedBase64Len = 8
// base64ShiftEncoder returns the part of a secret's base64 form that survives when the
// secret does not start on a 3-byte boundary of the payload it is embedded in. base64
// encodes three bytes at a time, so `Authorization: Basic base64("user:token")` contains
// the base64 of the token alone only when the prefix length happens to be a multiple of
// three; at the other two alignments the encoding of the whole value differs. Encoding
// the secret behind shift filler bytes reproduces those alignments, which is what the
// Base64StringEscapeShift1/2 encoders of GitHub's runner do.
//
// The leading group (filler mixed with the secret's first bytes) and the trailing group
// (padded here, but continuing into whatever follows the secret) are dropped, leaving the
// group-aligned middle that does appear verbatim in the log.
func base64ShiftEncoder(shift int) func(string) string {
return func(v string) string {
buf := make([]byte, shift+len(v))
copy(buf[shift:], v)
encoded := base64.StdEncoding.EncodeToString(buf)
// Keep only the aligned middle, and only when enough of it is left to be a
// distinctive pattern rather than a fragment that matches unrelated output.
if len(encoded) < 8+minShiftedBase64Len {
return ""
}
return encoded[4 : len(encoded)-4]
}
}
// jsonStringEscape returns v as it appears inside a JSON string, without the quotes,
// which is what `toJSON(secrets)` or any action logging a JSON body produces. Go's encoder
// escapes <, >, & (as act's own toJSON does); the non-HTML variant below covers the runtimes
// that do not. When v has none of those characters both forms are equal and deduplicated.
func jsonStringEscape(v string) string {
encoded, err := json.Marshal(v)
if err != nil {
return v
}
return string(encoded[1 : len(encoded)-1])
}
// jsonStringEscapeNoHTML is jsonStringEscape without HTML escaping, matching the JSON a
// JavaScript (JSON.stringify) or .NET action emits, so a secret containing < > or & is
// masked in that form too.
func jsonStringEscapeNoHTML(v string) string {
var buf bytes.Buffer
enc := json.NewEncoder(&buf)
enc.SetEscapeHTML(false)
if err := enc.Encode(v); err != nil {
return v
}
// Encode appends a newline; drop it along with the surrounding quotes.
encoded := strings.TrimRight(buf.String(), "\n")
return encoded[1 : len(encoded)-1]
}
func AppendSecretMasker(oldnew []string, v string) []string { func AppendSecretMasker(oldnew []string, v string) []string {
ret := oldnew ret := oldnew
@@ -182,6 +255,21 @@ func AppendSecretMasker(oldnew []string, v string) []string {
} }
} }
// The encoded forms are derived from the whole value: a multi-line secret is
// encoded as one string, not line by line.
trimmed := strings.TrimSpace(v)
if len(trimmed) <= 1 {
return ret
}
for _, encode := range secretValueEncoders {
encoded := encode(trimmed)
// An encoding that leaves the value unchanged is already masked above.
if encoded == trimmed || len(encoded) <= 1 || slices.Contains(ret, encoded) {
continue
}
ret = append(ret, encoded, "***")
}
return ret return ret
} }
@@ -194,6 +282,18 @@ func valueMasker(insecureSecrets bool, secrets map[string]string) entryProcessor
} }
oldnew = slices.Clip(oldnew) oldnew = slices.Clip(oldnew)
defReplacer := strings.NewReplacer(oldnew...) defReplacer := strings.NewReplacer(oldnew...)
// A ::add-mask:: only ever appends to the job's mask slice, so the replacer built for
// it stays valid until the slice grows. Cache it, keyed by the slice itself and its
// length, instead of encoding every secret and mask again for each log line.
var (
mu sync.Mutex
masksRef *[]string
pairs []string
masked int
replacer *strings.Replacer
)
return func(entry *logrus.Entry) *logrus.Entry { return func(entry *logrus.Entry) *logrus.Entry {
if insecureSecrets { if insecureSecrets {
return entry return entry
@@ -203,16 +303,27 @@ func valueMasker(insecureSecrets bool, secrets map[string]string) entryProcessor
if len(*masks) == 0 { if len(*masks) == 0 {
entry.Message = defReplacer.Replace(entry.Message) entry.Message = defReplacer.Replace(entry.Message)
} else { return entry
cmasker := oldnew
for _, v := range *masks {
cmasker = AppendSecretMasker(cmasker, v)
}
entry.Message = strings.NewReplacer(cmasker...).Replace(entry.Message)
} }
mu.Lock()
// A composite action logs through the same masker with its own mask slice, so a
// different slice starts the cache over.
if masksRef != masks {
masksRef, pairs, masked, replacer = masks, oldnew, 0, nil
}
if replacer == nil || masked != len(*masks) {
for _, v := range (*masks)[masked:] {
pairs = AppendSecretMasker(pairs, v)
}
masked = len(*masks)
replacer = strings.NewReplacer(pairs...)
}
cmasker := replacer
mu.Unlock()
entry.Message = cmasker.Replace(entry.Message)
return entry return entry
} }
} }

View File

@@ -4,7 +4,9 @@
package runner package runner
import ( import (
"encoding/base64"
"io" "io"
"net/url"
"strings" "strings"
"testing" "testing"
@@ -59,6 +61,136 @@ func TestValueMasker(t *testing.T) {
} }
} }
// A secret that reaches the log through an encoding — a base64 payload, a JSON body, a
// URL — must be masked as well: masking only the verbatim value leaks it.
func TestValueMaskerEncodedSecrets(t *testing.T) {
secret := `p@ss w"rd/1`
masker := valueMasker(false, map[string]string{"TOKEN": secret})
for _, tc := range []struct {
name string
line string
}{
{"verbatim", "the token is " + secret},
{"base64", "Authorization: Basic " + base64.StdEncoding.EncodeToString([]byte(secret))},
{"json", `{"token":"` + jsonStringEscape(secret) + `"}`},
{"query escaped", "https://example.com/?token=" + url.QueryEscape(secret)},
{"path escaped", "https://example.com/" + url.PathEscape(secret) + "/x"},
} {
t.Run(tc.name, func(t *testing.T) {
entry := masker(&logrus.Entry{Context: t.Context(), Message: tc.line})
assert.Contains(t, entry.Message, "***")
assert.NotContains(t, entry.Message, secret)
assert.NotContains(t, entry.Message, base64.StdEncoding.EncodeToString([]byte(secret)))
assert.NotContains(t, entry.Message, url.QueryEscape(secret))
})
}
}
// A secret containing " together with <, > or & serializes to JSON differently depending
// on the runtime: act's own toJSON (and Go) HTML-escape <>&, while a JavaScript
// (JSON.stringify) or .NET action leaves them literal. The secret must be masked in either
// form, so a JS-serialized JSON body does not leak it.
func TestValueMaskerJSONEscapesBothWays(t *testing.T) {
secret := `a"<b>&c`
masker := valueMasker(false, map[string]string{"TOKEN": secret})
for _, tc := range []struct {
name string
form string
}{
{"html escaped (act toJSON / Go)", jsonStringEscape(secret)},
{"literal (JS JSON.stringify / .NET)", jsonStringEscapeNoHTML(secret)},
} {
t.Run(tc.name, func(t *testing.T) {
entry := masker(&logrus.Entry{Context: t.Context(), Message: `{"t":"` + tc.form + `"}`})
assert.Contains(t, entry.Message, "***")
assert.NotContains(t, entry.Message, tc.form)
})
}
}
// ::add-mask:: values go through the same masker, so they get the same treatment.
func TestValueMaskerEncodedMasks(t *testing.T) {
masks := []string{"s3cr3t value"}
masker := valueMasker(false, nil)
entry := masker(&logrus.Entry{
Context: WithMasks(t.Context(), &masks),
Message: "encoded: " + base64.StdEncoding.EncodeToString([]byte("s3cr3t value")),
})
assert.Equal(t, "encoded: ***", entry.Message)
}
// A token in a Basic auth header is base64'd together with the user name, so the token's
// own base64 only appears when the prefix length is a multiple of three. The other two
// alignments must be masked as well, or `Authorization: Basic base64("user:token")` leaks
// the token to anyone who can decode the log.
func TestValueMaskerBase64Alignments(t *testing.T) {
secret := "s3cr3t-token-value"
masker := valueMasker(false, map[string]string{"TOKEN": secret})
// One prefix per alignment: len%3 of 0, 1 and 2.
for _, prefix := range []string{"x-access-token:", "user:", "ab:"} {
t.Run(prefix, func(t *testing.T) {
encoded := base64.StdEncoding.EncodeToString([]byte(prefix + secret))
entry := masker(&logrus.Entry{Context: t.Context(), Message: "Authorization: Basic " + encoded})
assert.Contains(t, entry.Message, "***")
// The aligned middle of the secret must be gone, so the payload can no longer be
// decoded back into the token.
assert.NotEqual(t, "Authorization: Basic "+encoded, entry.Message)
decodable := strings.TrimPrefix(entry.Message, "Authorization: Basic ")
decoded, err := base64.StdEncoding.DecodeString(decodable)
if err == nil {
assert.NotContains(t, string(decoded), secret)
}
})
}
}
// The masker caches its replacer, so it has to notice both a mask appended to the same
// slice and a composite action logging with a slice of its own.
func TestValueMaskerCachedReplacerSeesNewMasks(t *testing.T) {
masker := valueMasker(false, map[string]string{"TOKEN": "secret-token"})
mask := func(masks *[]string, message string) string {
return masker(&logrus.Entry{Context: WithMasks(t.Context(), masks), Message: message}).Message
}
job := []string{"first mask"}
assert.Equal(t, "a *** and ***", mask(&job, "a first mask and secret-token"))
// ::add-mask:: appends to the same slice
job = append(job, "second mask")
assert.Equal(t, "*** and ***", mask(&job, "first mask and second mask"))
// a composite action brings its own slice
composite := []string{"composite mask"}
assert.Equal(t, "*** but first mask", mask(&composite, "composite mask but first mask"))
// and the job's masks still apply once it is back
assert.Equal(t, "*** and *** but composite mask", mask(&job, "first mask and second mask but composite mask"))
}
func TestAppendSecretMaskerSkipsUselessEncodings(t *testing.T) {
// A token with no character an escape would touch only gains its base64 forms:
// JSON, query and path escaping all leave it unchanged.
pairs := AppendSecretMasker(nil, "plaintoken")
assert.Equal(t, []string{
"plaintoken", "***",
base64.StdEncoding.EncodeToString([]byte("plaintoken")), "***",
// The two shifted alignments, each without its leading and trailing group.
"YWludG9r", "***",
"bGFpbnRv", "***",
}, pairs)
// Too short to mask.
assert.Empty(t, AppendSecretMasker(nil, "x"))
}
func TestJobLogFormatterDecodesCommandData(t *testing.T) { func TestJobLogFormatterDecodesCommandData(t *testing.T) {
logger := logrus.New() logger := logrus.New()
logger.Out = io.Discard logger.Out = io.Discard

View File

@@ -69,6 +69,8 @@ func (sar *stepActionRemote) prepareActionExecutor() common.Executor {
github.Token = sar.RunContext.Config.ReplaceGheActionTokenWithGithubCom github.Token = sar.RunContext.Config.ReplaceGheActionTokenWithGithubCom
} }
} }
// Actions served from the action cache are read out of a git object store rather than a
// directory, so they never reach the bundle patch below and keep to the v1 cache API.
if sar.RunContext.Config.ActionCache != nil { if sar.RunContext.Config.ActionCache != nil {
cache := sar.RunContext.Config.ActionCache cache := sar.RunContext.Config.ActionCache
@@ -112,7 +114,7 @@ func (sar *stepActionRemote) prepareActionExecutor() common.Executor {
return err return err
} }
actionDir := fmt.Sprintf("%s/%s", sar.RunContext.ActionCacheDir(), sar.Step.UsesHash()) actionDir := sar.actionDir()
defaultActionURL := sar.RunContext.Config.DefaultActionURL() defaultActionURL := sar.RunContext.Config.DefaultActionURL()
// For Gitea // For Gitea
// A composite RunContext nils Config.Secrets, so getGitCloneToken would yield an // A composite RunContext nils Config.Secrets, so getGitCloneToken would yield an
@@ -171,6 +173,9 @@ func (sar *stepActionRemote) prepareActionExecutor() common.Executor {
sar.action = actionModel sar.action = actionModel
return err return err
}, },
// A stage of its own: it takes the same clone lock, and it has to land before
// runAction copies the action into the job container.
sar.patchActionToolkit,
)(ctx) )(ctx)
} }
} }
@@ -189,7 +194,7 @@ func (sar *stepActionRemote) pre() common.Executor {
return common.NewPipelineExecutor( return common.NewPipelineExecutor(
sar.prepareActionExecutor(), sar.prepareActionExecutor(),
runStepExecutor(sar, stepStagePre, runPreStep(sar)).If(hasPreStep(sar)).If(shouldRunPreStep(sar))) runStepExecutor(sar, stepStagePre, sar.revertToolkitOnFailure(runPreStep(sar))).If(hasPreStep(sar)).If(shouldRunPreStep(sar)))
} }
func (sar *stepActionRemote) main() common.Executor { func (sar *stepActionRemote) main() common.Executor {
@@ -211,15 +216,51 @@ func (sar *stepActionRemote) main() common.Executor {
return sar.RunContext.JobContainer.CopyDir(copyToPath, sar.RunContext.Config.Workdir+string(filepath.Separator)+".", sar.RunContext.Config.UseGitIgnore)(ctx) return sar.RunContext.JobContainer.CopyDir(copyToPath, sar.RunContext.Config.Workdir+string(filepath.Separator)+".", sar.RunContext.Config.UseGitIgnore)(ctx)
} }
actionDir := fmt.Sprintf("%s/%s", sar.RunContext.ActionCacheDir(), sar.Step.UsesHash()) actionDir := sar.actionDir()
return sar.runAction(sar, actionDir, sar.remoteAction)(ctx) return sar.revertToolkitOnFailure(sar.runAction(sar, actionDir, sar.remoteAction))(ctx)
}), }),
) )
} }
func (sar *stepActionRemote) post() common.Executor { func (sar *stepActionRemote) post() common.Executor {
return runStepExecutor(sar, stepStagePost, runPostStep(sar)).If(hasPostStep(sar)).If(shouldRunPostStep(sar)) return runStepExecutor(sar, stepStagePost, sar.revertToolkitOnFailure(runPostStep(sar))).If(hasPostStep(sar)).If(shouldRunPostStep(sar))
}
// toolkitBundles is the action directory and the entrypoints the toolkit may live in.
func (sar *stepActionRemote) toolkitBundles() (string, []string) {
if sar.remoteAction == nil {
return "", nil
}
dir := sar.actionDir()
return dir, actionScriptPaths(filepath.Join(dir, sar.remoteAction.Path), sar.action)
}
// patchActionToolkit edits the bundled toolkit so it works against Gitea, which lets the cache
// client use the v2 API this runner serves. A no-op unless the runner serves it.
func (sar *stepActionRemote) patchActionToolkit(ctx context.Context) error {
if sar.RunContext.GetEnv()[CacheServiceV2Env] != "" {
dir, scripts := sar.toolkitBundles()
patchToolkit(ctx, dir, scripts)
}
return nil
}
// revertToolkitOnFailure restores the untouched bundles when the action fails, so a later job
// runs it as shipped rather than repeating a failure the patch may have caused.
func (sar *stepActionRemote) revertToolkitOnFailure(exec common.Executor) common.Executor {
return func(ctx context.Context) error {
err := exec(ctx)
if err != nil {
dir, scripts := sar.toolkitBundles()
revertToolkit(ctx, dir, scripts)
}
return err
}
}
func (sar *stepActionRemote) actionDir() string {
return fmt.Sprintf("%s/%s", sar.RunContext.ActionCacheDir(), sar.Step.UsesHash())
} }
func (sar *stepActionRemote) getRunContext() *RunContext { func (sar *stepActionRemote) getRunContext() *RunContext {
@@ -270,7 +311,7 @@ func (sar *stepActionRemote) getActionModel() *model.Action {
func (sar *stepActionRemote) getCompositeRunContext(ctx context.Context) *RunContext { func (sar *stepActionRemote) getCompositeRunContext(ctx context.Context) *RunContext {
if sar.compositeRunContext == nil { if sar.compositeRunContext == nil {
actionDir := fmt.Sprintf("%s/%s", sar.RunContext.ActionCacheDir(), sar.Step.UsesHash()) actionDir := sar.actionDir()
actionLocation := path.Join(actionDir, sar.remoteAction.Path) actionLocation := path.Join(actionDir, sar.remoteAction.Path)
_, containerActionDir := getContainerActionPaths(sar.getStepModel(), actionLocation, sar.RunContext) _, containerActionDir := getContainerActionPaths(sar.getStepModel(), actionLocation, sar.RunContext)

262
act/runner/toolkit_patch.go Normal file
View File

@@ -0,0 +1,262 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package runner
import (
"bytes"
"context"
"os"
"path/filepath"
"regexp"
"strings"
"gitea.com/gitea/runner/act/common"
"gitea.com/gitea/runner/act/common/git"
"gitea.com/gitea/runner/act/model"
)
// Actions bundle the @actions toolkit into their own JavaScript, and two of its lines keep it
// from working against Gitea. Both are edited out of the bundle the runner downloaded.
//
// isGhes() takes any host that is not github.com, *.ghe.com or *.localhost for GitHub
// Enterprise. @actions/cache then forces the v1 API, and @actions/artifact refuses outright,
// which is why the stock upload-artifact aborts here. The edit empties the last of the three
// hostname tests, so `endsWith('.LOCALHOST')` becomes `endsWith(”)`, which every hostname
// satisfies: one string literal, no call sites to resolve, and the same answer the toolkit's own
// proposed ACTIONS_VENDOR switch would give. Gitea already makes this edit by hand in its fork
// of upload-artifact.
//
// getCacheServiceURL() then resolves the cache service from ACTIONS_RESULTS_URL alone, where v1
// reads ACTIONS_CACHE_URL first. Both reads there are given the same preference, which is what
// keeps the runner out of the artifact path: the results URL still points at Gitea.
//
// Either of these landing upstream makes this file deletable:
//
// https://github.com/actions/toolkit/pull/2123 — an ACTIONS_VENDOR switch, naming Gitea
// https://github.com/actions/toolkit/issues/2439 — treat ACTIONS_RESULTS_URL as the signal
const (
CacheServiceV2Env = "ACTIONS_CACHE_SERVICE_V2"
cacheURLEnv = "ACTIONS_CACHE_URL"
resultsURLEnv = "ACTIONS_RESULTS_URL"
// localhostHost is the suffix isGhes accepts; emptying the test is what opens the gate,
// because every hostname ends with the empty string.
localhostHost = ".LOCALHOST"
// artifactRefusal is the only thing the gate guards in @actions/artifact, which is what makes
// such a bundle safe to open. A bundle carrying neither toolkit uses isGhes for something this
// runner has not looked at, and is left alone.
artifactRefusal = "GHESNotSupportedError"
// sidecarSuffix names the directory of untouched copies, a sibling of the action directory
// because that directory is copied wholesale into job containers.
sidecarSuffix = ".toolkit-patch"
// skipMarker in the sidecar means a patched bundle already failed once here.
skipMarker = "skip"
maxBundleSize = 64 << 20
)
var (
// localhostTest matches the third hostname test of isGhes, in any quoting. The match is case
// sensitive on purpose, and that is load-bearing: isGhes uppercases the hostname before
// testing it, while undici, bundled into all of these actions, tests a lowercase ".localhost"
// in isURLPotentiallyTrustworthy. Opening that one would tell its HTTP client that every URL
// is trustworthy. Uppercase, the literal occurs nowhere but this test, across 118 bundles
// covering every major version of sixteen actions.
localhostTest = regexp.MustCompile(`endsWith\s*\(\s*` + quoted(regexp.QuoteMeta(localhostHost)) + `\s*\)`)
// serviceURLBranches matches both branches of getCacheServiceURL at once: the v1 branch reads
// the cache URL and falls back to the results URL, the v2 branch just below reads the results
// URL alone. That `||` pairing is the only place the two variables are read together, so
// matching them as one expression is what keeps the edit inside this function rather than
// anywhere they happen to sit near each other. The branches are 21 bytes apart minified and
// 63 not, across every bundle measured.
serviceURLBranches = regexp.MustCompile(`(` + envRead(cacheURLEnv) + `\s*\|\|\s*)(` +
envRead(resultsURLEnv) + `)((?s).{0,256}?)(` + envRead(resultsURLEnv) + `)`)
// cacheURLFirst gives both reads the preference the v1 branch already had.
cacheURLFirst = []byte(`${1}(process.env.` + cacheURLEnv + `||${2})${3}(process.env.` + cacheURLEnv + `||${4})`)
)
func envRead(name string) string {
return `process\s*\.\s*env\s*(?:\.\s*` + name + `\b|\[\s*` + quoted(name) + `\s*\])`
}
// quoted matches a string literal in any of the three quote characters. RE2 has no
// backreferences, so the pairs are spelled out.
func quoted(pattern string) string {
return "(?:'" + pattern + "'|\"" + pattern + "\"|`" + pattern + "`)"
}
// actionScriptPaths returns the entrypoints of a node action, the only kind with a bundle. Only
// remote actions get here: a local one lives in the user's checkout, which the runner does not
// rewrite.
func actionScriptPaths(dir string, action *model.Action) []string {
if action == nil || !action.Runs.Using.IsNode() {
return nil
}
var paths []string
for _, script := range []string{action.Runs.Pre, action.Runs.Main, action.Runs.Post} {
if script != "" {
paths = append(paths, filepath.Join(dir, script))
}
}
return paths
}
// patchToolkit edits the toolkit in an action's bundles, keeping each original beside them. Every
// failure is silent and leaves the bundle as it was, which costs the cache client the v2 API and
// an artifact action nothing at all.
func patchToolkit(ctx context.Context, actionDir string, scripts []string) {
if len(scripts) == 0 {
return
}
if _, err := os.Stat(filepath.Join(sidecarDir(actionDir), skipMarker)); err == nil {
return
}
defer git.AcquireCloneLock(actionDir)()
for _, script := range scripts {
if err := patchBundle(script, originalFor(actionDir, script)); err != nil {
common.Logger(ctx).Debugf("actions toolkit: %s left unpatched: %v", filepath.Base(script), err)
}
}
}
// revertToolkit puts the originals back and stops this action being patched again, so the next job
// runs it exactly as shipped. Called when a step failed with a patched bundle; it does not re-run
// the step, because a step's outputs and env-file writes are already recorded by then.
func revertToolkit(ctx context.Context, actionDir string, scripts []string) {
if len(scripts) == 0 {
return
}
if _, err := os.Stat(sidecarDir(actionDir)); err != nil {
return
}
defer git.AcquireCloneLock(actionDir)()
reverted := false
for _, script := range scripts {
original := originalFor(actionDir, script)
if !isPatchOf(original, script) {
continue
}
if err := os.Rename(original, script); err == nil {
reverted = true
}
}
if reverted {
_ = os.WriteFile(filepath.Join(sidecarDir(actionDir), skipMarker), nil, 0o600)
common.Logger(ctx).Warnf("actions toolkit: restored the original %s, it will not be patched again", filepath.Base(actionDir))
}
}
// sidecarDir holds an action's untouched bundles, and the marker that stops it being patched.
func sidecarDir(actionDir string) string {
return actionDir + sidecarSuffix
}
// originalFor is where a script's untouched copy lives, or "" for a script the action's own
// `runs` keys placed outside its directory, which is not this runner's to rewrite.
func originalFor(actionDir, script string) string {
rel, err := filepath.Rel(actionDir, script)
if err != nil || strings.HasPrefix(rel, "..") {
return ""
}
return filepath.Join(sidecarDir(actionDir), rel)
}
// patchBundle rewrites one entrypoint in place. The untouched copy kept beside it is what marks
// the bundle as already patched.
func patchBundle(script, original string) error {
if original == "" {
return nil
}
if _, err := os.Stat(original); err == nil {
if isPatchOf(original, script) {
return nil
}
// The action's ref moved and git checked the new bundle out over the patched one, so
// the pair no longer belongs together. Patch afresh rather than keep an original that
// would restore an older version of the action.
if err := os.Remove(original); err != nil {
return err
}
}
info, err := os.Stat(script)
if err != nil {
return err
}
if info.Size() > maxBundleSize {
return nil
}
data, err := os.ReadFile(script)
if err != nil {
return err
}
patched, ok := patchedBundle(data)
if !ok {
return nil
}
if err := os.MkdirAll(filepath.Dir(original), 0o755); err != nil {
return err
}
// The copy is taken before the bundle is replaced, so a write that fails part way can put the
// action back as it was. A crash needs no handling: the clone executor checks the action out
// and hard resets it on every prepare, so a half-written bundle never outlives the job.
if err := os.WriteFile(original, data, info.Mode().Perm()); err != nil {
return err
}
if err := os.WriteFile(script, patched, info.Mode().Perm()); err != nil {
_ = os.Rename(original, script)
return err
}
return nil
}
// isPatchOf reports whether script is exactly what patching original produced. It is what proves
// the two still belong together: an action whose ref moved is checked out over the patched bundle,
// leaving an original that would restore the version before the move.
func isPatchOf(original, script string) bool {
data, err := os.ReadFile(original)
if err != nil {
return false
}
current, err := os.ReadFile(script)
if err != nil {
return false
}
patched, ok := patchedBundle(data)
return ok && bytes.Equal(patched, current)
}
// patchedBundle opens the GHES gate, and where the cache toolkit is present, points the cache
// service at the cache server. A bundle this runner cannot account for comes back untouched.
func patchedBundle(data []byte) ([]byte, bool) {
if !localhostTest.Match(data) {
return data, false
}
switch {
case bytes.Contains(data, []byte(CacheServiceV2Env)):
// The cache toolkit: both edits or neither, because choosing v2 without redirecting the
// URL would send the client to a results URL that serves no cache service.
if !serviceURLBranches.Match(data) {
return data, false
}
case bytes.Contains(data, []byte(artifactRefusal)):
// The artifact toolkit, where the gate is a plain refusal and there is no URL to move:
// artifacts already go to Gitea, which implements that service.
default:
return data, false
}
opened := localhostTest.ReplaceAllFunc(data, func(test []byte) []byte {
// Drop the hostname from the test rather than rewriting the call, so the bundle's own
// quoting survives and the result stays valid even inside a string literal.
return bytes.Replace(test, []byte(localhostHost), nil, 1)
})
return serviceURLBranches.ReplaceAll(opened, cacheURLFirst), true
}

View File

@@ -0,0 +1,196 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package runner
import (
"context"
"io"
"net/http"
"os"
"os/exec"
"path/filepath"
"strings"
"testing"
"time"
"gitea.com/gitea/runner/act/artifactcache"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// actionsCacheRef pins the actions/cache release this is verified against. Bump it
// deliberately: a new release is exactly what can stop the patch matching.
const actionsCacheRef = "v6.1.0"
// bundleFromGitHub downloads one entrypoint, keeping it in the user cache dir so repeated runs
// cost nothing. The bundles are megabytes, too large to vendor.
func bundleFromGitHub(t *testing.T, repo, ref, path string) string {
t.Helper()
cacheDir, err := os.UserCacheDir()
require.NoError(t, err)
dir := filepath.Join(cacheDir, "gitea-runner-test", strings.ReplaceAll(repo, "/", "-")+"-"+ref)
bundle := filepath.Join(dir, strings.ReplaceAll(path, "/", "-"))
if _, err := os.Stat(bundle); err == nil {
return bundle
}
require.NoError(t, os.MkdirAll(dir, 0o755))
url := "https://raw.githubusercontent.com/" + repo + "/" + ref + "/" + path
ctx, cancel := context.WithTimeout(t.Context(), time.Minute)
defer cancel()
req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
require.NoError(t, err)
resp, err := http.DefaultClient.Do(req)
if err != nil {
t.Skipf("cannot reach %s: %v", url, err)
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
t.Skipf("GET %s: %s", url, resp.Status)
}
file, err := os.Create(bundle)
require.NoError(t, err)
_, err = io.Copy(file, resp.Body)
require.NoError(t, file.Close())
require.NoError(t, err)
return bundle
}
// runCacheAction runs one entrypoint the way a job would: a real Gitea server URL, and a results
// URL that points at Gitea rather than at the runner. Nothing about the environment is rewritten,
// so only the patch can make the client choose v2 and find the cache server.
func runCacheAction(t *testing.T, script, workspace, runnerTemp, cacheURL, token, key string) string {
t.Helper()
state := filepath.Join(runnerTemp, "state")
output := filepath.Join(runnerTemp, "output")
for _, name := range []string{state, output} {
require.NoError(t, os.WriteFile(name, nil, 0o600))
}
cmd := exec.CommandContext(t.Context(), "node", script)
cmd.Dir = workspace
cmd.Env = append(os.Environ(),
"INPUT_PATH=to-cache",
"INPUT_KEY="+key,
"ACTIONS_RUNTIME_TOKEN="+token,
"ACTIONS_CACHE_URL="+cacheURL+"/",
// Unreachable on purpose: the artifact service lives here, the cache service must not.
"ACTIONS_RESULTS_URL=https://gitea.example",
"ACTIONS_CACHE_SERVICE_V2=true",
"GITHUB_SERVER_URL=https://gitea.example.com",
"GITHUB_REF=refs/heads/main",
"GITHUB_EVENT_NAME=push",
"GITHUB_WORKSPACE="+workspace,
"RUNNER_TEMP="+runnerTemp,
"GITHUB_STATE="+state,
"GITHUB_OUTPUT="+output,
)
out, err := cmd.CombinedOutput()
t.Logf("%s:\n%s", filepath.Base(filepath.Dir(script)), out)
require.NoError(t, err, "%s failed", script)
return string(out)
}
// tempDirPath is TempDir with symlinks resolved, because macOS hands out /var paths that resolve
// to /private/var and the client derives archive paths relative to the workspace.
func tempDirPath(t *testing.T) string {
t.Helper()
dir, err := filepath.EvalSymlinks(t.TempDir())
require.NoError(t, err)
return dir
}
// The whole chain against the pinned release, whose bundles ship unminified: patch them, run the
// real client with an ordinary Gitea server URL and a results URL that goes nowhere, and have it
// save and restore through this runner's cache server. The unreachable results URL is the point,
// it is what proves the cache reaches the runner without the runner fronting Gitea. If a release
// stops matching the patch the client falls back to v1 and this fails on the version line, which
// is the signal to look at the new bundle.
func TestCacheServiceV2EndToEnd(t *testing.T) {
requireHostTools(t, "node")
// A stand-in action directory, patched exactly as a downloaded one would be.
actionDir := tempDirPath(t)
scripts := map[string]string{}
for _, stage := range []string{"restore", "save"} {
body, err := os.ReadFile(bundleFromGitHub(t, "actions/cache", actionsCacheRef, "dist/"+stage+"/index.js"))
require.NoError(t, err)
scripts[stage] = filepath.Join(actionDir, stage+".js")
require.NoError(t, os.WriteFile(scripts[stage], body, 0o600))
}
patchToolkit(t.Context(), actionDir, []string{scripts["restore"], scripts["save"]})
handler, err := artifactcache.StartHandler(filepath.Join(t.TempDir(), "cache"), "127.0.0.1", 0, "", nil)
require.NoError(t, err)
t.Cleanup(func() { _ = handler.Close() })
const token, repo = "e2e-runtime-token", "testuser/testrepo"
handler.RegisterJob(token, repo)
workspace, runnerTemp := tempDirPath(t), tempDirPath(t)
require.NoError(t, os.MkdirAll(filepath.Join(workspace, "to-cache"), 0o755))
content := []byte("cached through the patched gate")
require.NoError(t, os.WriteFile(filepath.Join(workspace, "to-cache", "data.txt"), content, 0o600))
const key = "patched-gate-key"
missed := runCacheAction(t, scripts["restore"], workspace, runnerTemp, handler.ExternalURL(), token, key)
require.Contains(t, missed, "Cache service version: v2", "the patch did not take, the client stayed on v1")
require.Contains(t, missed, "Cache not found for input keys: "+key)
saved := runCacheAction(t, scripts["save"], workspace, runnerTemp, handler.ExternalURL(), token, key)
require.Contains(t, saved, "Cache saved with key: "+key)
restored := tempDirPath(t)
hit := runCacheAction(t, scripts["restore"], restored, runnerTemp, handler.ExternalURL(), token, key)
require.Contains(t, hit, "Cache restored from key: "+key)
got, err := os.ReadFile(filepath.Join(restored, "to-cache", "data.txt"))
require.NoError(t, err)
assert.Equal(t, content, got)
}
// The gate and the URL getter are separate functions, and a bundler may put either first: the gap
// between them runs from 159 to 1179 bytes across these actions, which is why neither edit is
// anchored on that distance. One entrypoint from each of the families that bundle the cache
// toolkit, patched but not run, is what keeps a future release from quietly matching only one of
// the two shapes and leaving every cache on v1.
func TestToolkitPatchAcrossActions(t *testing.T) {
for _, tc := range []struct{ repo, ref, path string }{
{"actions/setup-go", "v7.0.0", "dist/setup/index.js"},
{"actions/setup-node", "v6.0.0", "dist/cache-save/index.js"},
{"actions/setup-python", "v6.0.0", "dist/setup/index.js"},
{"ruby/setup-ruby", "v1.271.0", "dist/index.js"},
{"pnpm/action-setup", "v6.0.9", "dist/index.js"},
// The artifact toolkit, where the gate is a refusal and there is nothing to redirect.
// v4.4.0 is the first release whose gate carries the localhost test this matches; the
// releases before it refuse in a shape the runner leaves alone.
{"actions/upload-artifact", "v4.4.0", "dist/upload/index.js"},
{"actions/upload-artifact", "v7.0.1", "dist/upload/index.js"},
{"actions/download-artifact", "v6.0.0", "dist/index.js"},
{"oven-sh/setup-bun", "v2.2.0", "dist/setup/index.js"},
} {
t.Run(tc.repo+"@"+tc.ref, func(t *testing.T) {
t.Parallel()
data, err := os.ReadFile(bundleFromGitHub(t, tc.repo, tc.ref, tc.path))
require.NoError(t, err)
out, patched := patchedBundle(data)
assert.True(t, patched, "the version gate was not patched")
assert.NotContains(t, string(out), ".LOCALHOST", "a copy of the gate was missed")
if !strings.Contains(string(data), CacheServiceV2Env) {
return // the artifact toolkit: a refusal to open, and no URL to move
}
// Only the reads inside getCacheServiceURL are rewritten. The others, such as the
// feature-availability check, must be left as they are.
assert.NotZero(t, strings.Count(string(out), "(process.env."+cacheURLEnv+"||process.env"),
"the cache service URL was not redirected")
assert.Equal(t, strings.Count(string(data), resultsURLEnv), strings.Count(string(out), resultsURLEnv),
"a read of the results URL was lost, it must stay as the fallback")
})
}
}

View File

@@ -0,0 +1,328 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package runner
import (
"context"
"errors"
"os"
"os/exec"
"path/filepath"
"strings"
"testing"
"gitea.com/gitea/runner/act/model"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// The three shapes real bundlers emit, reduced to the bytes that matter: the version gate, and
// the URL getter that follows it. tsc keeps the names, webpack prefixes them, esbuild mangles
// them, writes ternaries in place of the switch, and records the real name in the export
// assignment. Each carries both reads of the results URL, as the real getter does.
const (
urlTSC = `function getCacheServiceURL() {` + "\n" + ` switch (getCacheServiceVersion()) {` + "\n" + ` case 'v1':` + "\n" + ` return (process.env['ACTIONS_CACHE_URL'] || process.env['ACTIONS_RESULTS_URL'] || '');` + "\n" + ` case 'v2':` + "\n" + ` return process.env['ACTIONS_RESULTS_URL'] || '';` + "\n" + ` }` + "\n" + `}`
urlEsbuild = `function YK(){let e=XK();return e==="v1"?process.env.ACTIONS_CACHE_URL||process.env.ACTIONS_RESULTS_URL||"":e==="v2"?process.env.ACTIONS_RESULTS_URL||"":""}`
isGhesTSC = `function isGhes(){const h=new URL(process.env['GITHUB_SERVER_URL']||'https://github.com').hostname.toUpperCase();return h!=='GITHUB.COM'&&!h.endsWith('.GHE.COM')&&!h.endsWith('.LOCALHOST')}`
gateTSC = isGhesTSC + "\n" + `function getCacheServiceVersion() {` + "\n" + ` if (isGhes())` + "\n" + ` return 'v1';` + "\n" + ` return process.env['ACTIONS_CACHE_SERVICE_V2'] ? 'v2' : 'v1';` + "\n" + `}` + "\n" + urlTSC
gateWebpack = `function config_isGhes(){const h=new URL(process.env['GITHUB_SERVER_URL']||'https://github.com').hostname.toUpperCase();return h!=='GITHUB.COM'&&!h.endsWith('.GHE.COM')&&!h.endsWith('.LOCALHOST')}` + "\n" + `function config_getCacheServiceVersion() {` + "\n" + ` if (config_isGhes())` + "\n" + ` return 'v1';` + "\n" + ` return process.env['ACTIONS_CACHE_SERVICE_V2'] ? 'v2' : 'v1';` + "\n" + `}` + "\n" + urlTSC
gateEsbuild = `vu.isGhes=$K;vu.getCacheServiceVersion=XK;function $K(){let e=new URL(process.env.GITHUB_SERVER_URL||"https://github.com").hostname.toUpperCase(),r=e==="GITHUB.COM",n=e.endsWith(".GHE.COM"),i=e.endsWith(".LOCALHOST");return!r&&!n&&!i}function XK(){return $K()?"v1":process.env.ACTIONS_CACHE_SERVICE_V2?"v2":"v1"}` + urlEsbuild
)
func TestPatchedBundle(t *testing.T) {
for _, tc := range []struct {
name, body string
wantPatched bool
}{
{"tsc keeps the names", gateTSC, true},
{"webpack prefixes them", gateWebpack, true},
{"esbuild mangles and minifies them", gateEsbuild, true},
// A bundler picks its own quoting; gateTSC is single-quoted already.
{"double-quoted", requoted(`"`), true},
{"backtick-quoted", requoted("`"), true},
// sccache-action sets the variable itself; there is no gate to open.
{"mentions the variable without the gate", `core.exportVariable("ACTIONS_CACHE_SERVICE_V2","on")`, false},
// Both edits or neither: a gate patched without the URL would send the client to a
// results URL that serves no cache service.
{"gate without a recognisable url getter", strings.TrimSuffix(gateTSC, "\n"+urlTSC), false},
// And the other way round: an action that reads both variables but has no gate to open.
{"url getter without a gate", urlTSC, false},
} {
t.Run(tc.name, func(t *testing.T) {
out, patched := patchedBundle([]byte(tc.body))
assert.Equal(t, tc.wantPatched, patched)
if !tc.wantPatched {
assert.Equal(t, tc.body, string(out), "an unpatched bundle must come back byte for byte")
return
}
assert.True(t, gateOpened(string(out)))
// The other two hostname tests are left alone, so a host that really is GitHub or
// GHES is still recognised as such.
assert.NotContains(t, string(out), ".LOCALHOST", "the localhost test is the one that opens")
assert.Contains(t, string(out), ".GHE.COM")
// Every read of the results URL now prefers the cache URL, and none was lost: the
// results URL stays the fallback, so a runner not serving the cache still works.
assert.Equal(t, strings.Count(tc.body, "ACTIONS_RESULTS_URL"), strings.Count(string(out), "ACTIONS_RESULTS_URL"))
assert.Equal(t, strings.Count(tc.body, "ACTIONS_RESULTS_URL"),
strings.Count(string(out), "(process.env.ACTIONS_CACHE_URL||process.env"))
})
}
}
// undici, bundled into every one of these actions, decides whether to trust a URL with a
// lowercase test that reads almost the same. Opening it would tell the HTTP client that every URL
// is trustworthy, so the uppercase the toolkit produces is what separates them.
func TestPatchedBundleLeavesTrustworthyURLCheckAlone(t *testing.T) {
const undici = `if(n.hostname==="localhost"||n.hostname.includes("localhost.")||n.hostname.endsWith(".localhost")){return true}`
out, patched := patchedBundle([]byte(undici + gateTSC))
require.True(t, patched)
assert.Contains(t, string(out), undici, "the trustworthy-URL check must survive byte for byte")
assert.True(t, gateOpened(string(out)))
}
// The artifact toolkit puts the same gate in front of a plain refusal, with no URL to move, so
// opening it is what lets the stock upload-artifact work against Gitea instead of aborting.
func TestPatchedBundleOpensTheArtifactRefusal(t *testing.T) {
const artifact = isGhesTSC + "\n" + `uploadArtifact(){if(isGhes()){throw new GHESNotSupportedError()}}`
out, patched := patchedBundle([]byte(artifact))
assert.True(t, patched)
assert.True(t, gateOpened(string(out)))
assert.Contains(t, string(out), "GHESNotSupportedError", "the refusal itself is left in place, it just stops firing")
// A bundle using the gate for something this runner has not accounted for is not touched.
unknown := strings.Replace(artifact, "GHESNotSupportedError", "SomeOtherError", 1)
out, patched = patchedBundle([]byte(unknown))
assert.False(t, patched)
assert.Equal(t, unknown, string(out))
}
// requoted respells gateTSC's string literals with another quote character.
func requoted(quote string) string {
gate := strings.ReplaceAll(gateTSC, `'.LOCALHOST'`, quote+".LOCALHOST"+quote)
gate = strings.ReplaceAll(gate, `['ACTIONS_RESULTS_URL']`, "["+quote+"ACTIONS_RESULTS_URL"+quote+"]")
return strings.ReplaceAll(gate, `['ACTIONS_CACHE_URL']`, "["+quote+"ACTIONS_CACHE_URL"+quote+"]")
}
// gateOpened reports whether the hostname test was emptied, in whatever quoting the bundle used.
func gateOpened(body string) bool {
return strings.Contains(body, "endsWith(") && !strings.Contains(body, ".LOCALHOST")
}
// The patched bundle must still be JavaScript, and must resolve the way the runner needs: v2 for
// an ordinary Gitea host, the cache server for the service URL, and the results URL when there is
// no cache server. Unpatched, the same bundle must still choose v1, or the patch proves nothing.
func TestPatchedBundleBehavesInNode(t *testing.T) {
requireHostTools(t, "node")
eval := func(t *testing.T, bundle, prelude, cacheURL string) string {
t.Helper()
script := prelude + bundle + "\nprocess.stdout.write(getCacheServiceVersion()+' '+getCacheServiceURL())"
cmd := exec.CommandContext(t.Context(), "node", "-e", script)
cmd.Env = append(os.Environ(),
"ACTIONS_CACHE_SERVICE_V2=true",
"ACTIONS_CACHE_URL="+cacheURL,
"ACTIONS_RESULTS_URL=https://gitea.example",
"GITHUB_SERVER_URL=https://gitea.example",
)
out, err := cmd.CombinedOutput()
require.NoError(t, err, "%s", out)
return string(out)
}
for _, tc := range []struct{ name, bundle, prelude string }{
{"tsc", gateTSC, ""},
{"webpack", gateWebpack, "const getCacheServiceVersion=()=>config_getCacheServiceVersion();"},
{"esbuild", gateEsbuild, "var vu={};const getCacheServiceVersion=()=>XK(),getCacheServiceURL=()=>YK();"},
} {
t.Run(tc.name, func(t *testing.T) {
// Unpatched, a Gitea host is taken for GHES: v1, whose branch already reads the
// cache URL. The patch has to move the version without moving that.
assert.Equal(t, "v1 http://cache:8088/", eval(t, tc.bundle, tc.prelude, "http://cache:8088/"))
patched, ok := patchedBundle([]byte(tc.bundle))
require.True(t, ok)
assert.Equal(t, "v2 http://cache:8088/", eval(t, string(patched), tc.prelude, "http://cache:8088/"))
assert.Equal(t, "v2 https://gitea.example", eval(t, string(patched), tc.prelude, ""),
"with no cache server the results URL is still the fallback")
})
}
}
// A bundler that embeds module sources as strings, such as webpack with devtool: eval, carries
// the gate inside a double-quoted literal. Rewriting the call rather than emptying its argument
// would end that string early and leave the bundle unparseable.
func TestPatchedBundleSurvivesInsideAStringLiteral(t *testing.T) {
requireHostTools(t, "node")
escaped := strings.ReplaceAll(gateTSC, `"`, `\"`)
embedded := `eval("` + strings.ReplaceAll(escaped, "\n", `\n`) + `");`
out, patched := patchedBundle([]byte(embedded))
require.True(t, patched)
file := filepath.Join(t.TempDir(), "bundle.js")
require.NoError(t, os.WriteFile(file, out, 0o600))
checked, err := exec.CommandContext(t.Context(), "node", "--check", file).CombinedOutput()
require.NoError(t, err, "%s", checked)
}
func TestPatchBundleKeepsTheOriginal(t *testing.T) {
dir, script := bundleFile(t, gateTSC)
original := originalFor(dir, script)
require.NoError(t, patchBundle(script, original))
patched, err := os.ReadFile(script)
require.NoError(t, err)
assert.True(t, gateOpened(string(patched)))
kept, err := os.ReadFile(original)
require.NoError(t, err)
assert.Equal(t, gateTSC, string(kept), "the untouched bundle is kept outside the action tree")
assert.NotContains(t, original, dir+string(filepath.Separator), "originals must not ship into job containers")
// Patching again must not stack, and must not overwrite the kept original.
require.NoError(t, patchBundle(script, original))
again, err := os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, string(patched), string(again))
kept, err = os.ReadFile(original)
require.NoError(t, err)
assert.Equal(t, gateTSC, string(kept))
}
// A bundle with nothing to patch is left exactly as it was, with no original kept beside it.
func TestPatchBundleLeavesOtherActionsAlone(t *testing.T) {
dir, script := bundleFile(t, `console.log("checkout")`)
original := originalFor(dir, script)
require.NoError(t, patchBundle(script, original))
body, err := os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, `console.log("checkout")`, string(body))
_, err = os.Stat(original)
assert.True(t, os.IsNotExist(err), "no original is kept for a bundle that was not patched")
}
// bundleFile writes one entrypoint into a fresh action directory.
func bundleFile(t *testing.T, body string) (dir, script string) {
t.Helper()
dir = t.TempDir()
script = filepath.Join(dir, "index.js")
require.NoError(t, os.WriteFile(script, []byte(body), 0o600))
return dir, script
}
func TestActionScriptPaths(t *testing.T) {
node := &model.Action{Runs: model.ActionRuns{Using: "node20", Main: "dist/restore/index.js", Post: "dist/save/index.js"}}
assert.Equal(t, []string{"/a/dist/restore/index.js", "/a/dist/save/index.js"}, actionScriptPaths("/a", node))
// Only a node action has a bundle to patch.
assert.Nil(t, actionScriptPaths("/a", &model.Action{Runs: model.ActionRuns{Using: "docker", Image: "alpine"}}))
assert.Nil(t, actionScriptPaths("/a", nil))
}
// A step that fails with a patched bundle gets the untouched bundle back, and the action is not
// patched again, so later jobs run it exactly as its author shipped it.
func TestRevertToolkit(t *testing.T) {
dir, script := bundleFile(t, gateTSC)
scripts := []string{script}
patchToolkit(t.Context(), dir, scripts)
body, err := os.ReadFile(script)
require.NoError(t, err)
require.True(t, gateOpened(string(body)), "precondition: the bundle is patched")
revertToolkit(t.Context(), dir, scripts)
body, err = os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, gateTSC, string(body), "the original bundle is back")
// The skip marker survives, so the action stays unpatched from now on.
patchToolkit(t.Context(), dir, scripts)
body, err = os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, gateTSC, string(body), "a reverted action stays unpatched")
}
// An action whose ref moves is checked out over the patched bundle. The kept original then
// belongs to the version before the move, and must not be restored over the new one.
func TestPatchBundleAfterTheActionMoved(t *testing.T) {
dir, script := bundleFile(t, gateTSC)
original := originalFor(dir, script)
scripts := []string{script}
patchToolkit(t.Context(), dir, scripts)
require.NoError(t, os.WriteFile(script, []byte(gateWebpack), 0o600)) // the new version lands
// Reverting must not roll the action back to the version the original came from.
revertToolkit(t.Context(), dir, scripts)
body, err := os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, gateWebpack, string(body))
// Nothing was reverted, so the action is not marked off either: the new version is patched
// in its own right, and keeps its own original.
require.NoFileExists(t, filepath.Join(sidecarDir(dir), skipMarker))
require.NoError(t, patchBundle(script, original))
body, err = os.ReadFile(script)
require.NoError(t, err)
assert.True(t, gateOpened(string(body)))
kept, err := os.ReadFile(original)
require.NoError(t, err)
assert.Equal(t, gateWebpack, string(kept))
}
// The wiring: a step patches its own bundles only when the runner serves the v2 API, and a step
// that fails gets them back. The action's path inside its repository is part of where they live.
func TestStepActionRemoteToolkitPatch(t *testing.T) {
newStep := func(t *testing.T, env map[string]string) (*stepActionRemote, string) {
t.Helper()
sar := &stepActionRemote{
Step: &model.Step{Uses: "owner/repo/sub@v1"},
remoteAction: &remoteAction{Org: "owner", Repo: "repo", Path: "sub", Ref: "v1"},
action: &model.Action{Runs: model.ActionRuns{Using: "node20", Main: "index.js"}},
RunContext: &RunContext{
Env: env,
Config: &Config{ActionCacheDir: t.TempDir()},
},
}
script := filepath.Join(sar.actionDir(), "sub", "index.js")
require.NoError(t, os.MkdirAll(filepath.Dir(script), 0o755))
require.NoError(t, os.WriteFile(script, []byte(gateTSC), 0o600))
return sar, script
}
t.Run("left alone when the runner does not serve the v2 API", func(t *testing.T) {
sar, script := newStep(t, map[string]string{})
require.NoError(t, sar.patchActionToolkit(t.Context()))
body, err := os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, gateTSC, string(body))
})
t.Run("patched, and put back when the step fails", func(t *testing.T) {
sar, script := newStep(t, map[string]string{CacheServiceV2Env: "true"})
require.NoError(t, sar.patchActionToolkit(t.Context()))
body, err := os.ReadFile(script)
require.NoError(t, err)
require.True(t, gateOpened(string(body)))
failed := errors.New("the step failed")
require.ErrorIs(t, sar.revertToolkitOnFailure(func(context.Context) error { return failed })(t.Context()), failed)
body, err = os.ReadFile(script)
require.NoError(t, err)
assert.Equal(t, gateTSC, string(body))
})
}

View File

@@ -22,6 +22,7 @@ import (
"gitea.com/gitea/runner/internal/pkg/config" "gitea.com/gitea/runner/internal/pkg/config"
"gitea.com/gitea/runner/internal/pkg/envcheck" "gitea.com/gitea/runner/internal/pkg/envcheck"
"gitea.com/gitea/runner/internal/pkg/labels" "gitea.com/gitea/runner/internal/pkg/labels"
"gitea.com/gitea/runner/internal/pkg/lock"
"gitea.com/gitea/runner/internal/pkg/metrics" "gitea.com/gitea/runner/internal/pkg/metrics"
"gitea.com/gitea/runner/internal/pkg/ver" "gitea.com/gitea/runner/internal/pkg/ver"
@@ -49,6 +50,25 @@ func runDaemon(ctx context.Context, daemArgs *daemonArgs, configFile *string) fu
return fmt.Errorf("failed to load registration file: %w", err) return fmt.Errorf("failed to load registration file: %w", err)
} }
// Guard against a second runner process sharing this runner file: two
// processes with the same identity are indistinguishable to Gitea and
// end up cancelling each other's jobs.
releaseLock, err := lock.TryLock(cfg.Runner.File)
if errors.Is(err, lock.ErrLocked) {
log.Errorf("another gitea-runner process is already using %q; each runner process needs its own runner file (runner.file)", cfg.Runner.File)
return err
} else if err != nil {
// Best-effort guard: if the lock file can't be created (e.g. a
// read-only runner-file mount), warn and start anyway rather than
// refusing to run.
log.Warnf("could not lock runner file %q, continuing without the single-process guard: %v", cfg.Runner.File, err)
} else {
// Held until shutdown finishes: the draining runner still owns this
// identity on the server, so releasing early would let a restart
// reintroduce the duplicate-identity job cancellations.
defer func() { _ = releaseLock() }()
}
lbls := resolveLabels(daemArgs.Labels, cfg.Runner.Labels, reg.Labels) lbls := resolveLabels(daemArgs.Labels, cfg.Runner.Labels, reg.Labels)
ls := labels.Labels{} ls := labels.Labels{}
@@ -141,6 +161,11 @@ func runDaemon(ctx context.Context, daemArgs *daemonArgs, configFile *string) fu
) )
runner := run.NewRunner(cfg, reg, cli) runner := run.NewRunner(cfg, reg, cli)
defer func() {
if err := runner.Close(); err != nil {
log.Warnf("runner %s: cache server shutdown: %v", reg.Name, err)
}
}()
// declare the labels of the runner before fetching tasks // declare the labels of the runner before fetching tasks
resp, err := runner.Declare(ctx, ls.Names()) resp, err := runner.Declare(ctx, ls.Names())

View File

@@ -132,6 +132,8 @@ func (i *executeArgs) LoadEnvs() map[string]string {
_ = readEnvs(i.Envfile(), envs) _ = readEnvs(i.Envfile(), envs)
envs["ACTIONS_CACHE_URL"] = i.cacheHandler.ExternalURL() + "/" envs["ACTIONS_CACHE_URL"] = i.cacheHandler.ExternalURL() + "/"
// The same server answers the cache service v2 API, so let the actions reach it.
envs[runner.CacheServiceV2Env] = "true"
return envs return envs
} }

View File

@@ -18,6 +18,7 @@ import (
"gitea.com/gitea/runner/internal/pkg/client" "gitea.com/gitea/runner/internal/pkg/client"
"gitea.com/gitea/runner/internal/pkg/config" "gitea.com/gitea/runner/internal/pkg/config"
"gitea.com/gitea/runner/internal/pkg/labels" "gitea.com/gitea/runner/internal/pkg/labels"
"gitea.com/gitea/runner/internal/pkg/lock"
"gitea.com/gitea/runner/internal/pkg/ver" "gitea.com/gitea/runner/internal/pkg/ver"
"connectrpc.com/connect" "connectrpc.com/connect"
@@ -346,6 +347,19 @@ func registerNoInteractive(ctx context.Context, configFile string, regArgs *regi
} }
func doRegister(ctx context.Context, cfg *config.Config, inputs *registerInputs) error { func doRegister(ctx context.Context, cfg *config.Config, inputs *registerInputs) error {
// Refuse to rewrite the runner file while another process is using it.
releaseLock, err := lock.TryLock(cfg.Runner.File)
if errors.Is(err, lock.ErrLocked) {
return fmt.Errorf("another process is already using %q; stop it before re-registering", cfg.Runner.File)
} else if err != nil {
// Best-effort guard: if the lock file can't be created, warn and
// register anyway; writing the runner file will surface any real
// permission problem with a clearer error.
log.Warnf("could not lock runner file %q, continuing without the single-process guard: %v", cfg.Runner.File, err)
} else {
defer func() { _ = releaseLock() }()
}
// initial http client // initial http client
cli := client.New( cli := client.New(
inputs.InstanceAddr, inputs.InstanceAddr,

View File

@@ -54,6 +54,7 @@ func RunnerCapabilities() []string {
// Runner runs the pipeline. // Runner runs the pipeline.
type Runner struct { type Runner struct {
name string name string
uuid string
cfg *config.Config cfg *config.Config
@@ -108,6 +109,12 @@ func NewRunner(cfg *config.Config, reg *config.Registration, cli client.Client)
} }
} }
if envs["ACTIONS_CACHE_URL"] != "" && (cfg.Cache.V2 == nil || *cfg.Cache.V2) {
// act patches the GHES check out of an action's bundle when it sees this, so the client
// uses the cache service v2 API this server also answers; see act/runner/toolkit_patch.go.
envs[runner.CacheServiceV2Env] = "true"
}
// set artifact gitea api // set artifact gitea api
artifactGiteaAPI := strings.TrimSuffix(cli.Address(), "/") + "/api/actions_pipeline/" artifactGiteaAPI := strings.TrimSuffix(cli.Address(), "/") + "/api/actions_pipeline/"
envs["ACTIONS_RUNTIME_URL"] = artifactGiteaAPI envs["ACTIONS_RUNTIME_URL"] = artifactGiteaAPI
@@ -119,6 +126,7 @@ func NewRunner(cfg *config.Config, reg *config.Registration, cli client.Client)
runner := &Runner{ runner := &Runner{
name: reg.Name, name: reg.Name,
uuid: reg.UUID,
cfg: cfg, cfg: cfg,
client: cli, client: cli,
labels: ls, labels: ls,
@@ -130,6 +138,14 @@ func NewRunner(cfg *config.Config, reg *config.Registration, cli client.Client)
return runner return runner
} }
// Close shuts down the cache server this runner exposes to job containers.
func (r *Runner) Close() error {
return r.cacheHandler.Close()
}
// removeOrphanNetworks is a variable so tests can substitute one that needs no Docker daemon.
var removeOrphanNetworks = container.RemoveOrphanNetworks
// OnIdle performs lightweight maintenance during polling idle windows. // OnIdle performs lightweight maintenance during polling idle windows.
// It runs synchronously on the poller goroutine; shouldRunIdleCleanup // It runs synchronously on the poller goroutine; shouldRunIdleCleanup
// throttles invocations to runner.idle_cleanup_interval so the impact on // throttles invocations to runner.idle_cleanup_interval so the impact on
@@ -151,6 +167,21 @@ func (r *Runner) OnIdle(ctx context.Context) {
if hostRoot := filepath.FromSlash(r.cfg.Host.WorkdirParent); hostRoot != "" { if hostRoot := filepath.FromSlash(r.cfg.Host.WorkdirParent); hostRoot != "" {
r.cleanupStaleDirs(ctx, hostRoot, isHostScratchDir) r.cleanupStaleDirs(ctx, hostRoot, isHostScratchDir)
} }
r.cleanupOrphanNetworks(ctx)
}
// cleanupOrphanNetworks reclaims the per-job networks of jobs this runner did not live to
// tear down. A labelled network with no containers on it is finished with, and as for the
// directories above, a task beginning during the pass is safe because the cutoff keeps a
// network it has created but not yet attached a container to out of scope.
func (r *Runner) cleanupOrphanNetworks(ctx context.Context) {
if r.uuid == "" || !r.labels.RequireDocker() && !r.cfg.Container.RequireDocker {
return
}
cutoff := r.now().Add(-r.cfg.Runner.WorkdirCleanupAge)
if err := removeOrphanNetworks(ctx, r.uuid, cutoff); err != nil {
log.Warnf("failed to clean up networks left behind by earlier jobs: %v", err)
}
} }
func (r *Runner) shouldRunIdleCleanup() bool { func (r *Runner) shouldRunIdleCleanup() bool {
@@ -472,6 +503,9 @@ func (r *Runner) run(ctx context.Context, task *runnerv1.Task, reporter *report.
ContainerNetworkCreateOptions: container.NewDockerNetworkCreateExecutorInput{ ContainerNetworkCreateOptions: container.NewDockerNetworkCreateExecutorInput{
EnableIPv4: r.cfg.Container.NetworkCreateOptions.EnableIPv4, EnableIPv4: r.cfg.Container.NetworkCreateOptions.EnableIPv4,
EnableIPv6: r.cfg.Container.NetworkCreateOptions.EnableIPv6, EnableIPv6: r.cfg.Container.NetworkCreateOptions.EnableIPv6,
// so a network this job leaks, if the runner dies before its teardown, can be
// told apart from one belonging to another runner on the same daemon
RunnerUUID: r.uuid,
}, },
ContainerOptions: r.cfg.Container.Options, ContainerOptions: r.cfg.Container.Options,
ContainerDaemonSocket: r.cfg.Container.DockerHost, ContainerDaemonSocket: r.cfg.Container.DockerHost,

View File

@@ -297,3 +297,37 @@ func TestRunnerOnIdleSkipsWhenAlreadyCancelled(t *testing.T) {
assert.DirExists(t, stale) assert.DirExists(t, stale)
} }
// The idle sweep reclaims the docker networks of jobs this runner did not live to tear down,
// and stays out of the way of runners that share the daemon but not the registration.
func TestRunnerOnIdleRemovesOrphanNetworks(t *testing.T) {
now := time.Date(2026, time.April, 29, 20, 0, 0, 0, time.UTC)
cfg := &config.Config{
Container: config.Container{RequireDocker: true},
Runner: config.Runner{
WorkdirCleanupAge: 24 * time.Hour,
IdleCleanupInterval: time.Minute,
},
}
var swept []string
var sweptCutoff time.Time
origRemoveOrphanNetworks := removeOrphanNetworks
removeOrphanNetworks = func(_ context.Context, runnerUUID string, createdBefore time.Time) error {
swept = append(swept, runnerUUID)
sweptCutoff = createdBefore
return nil
}
t.Cleanup(func() { removeOrphanNetworks = origRemoveOrphanNetworks })
r := &Runner{uuid: "runner-1", cfg: cfg, now: func() time.Time { return now }}
r.OnIdle(context.Background())
assert.Equal(t, []string{"runner-1"}, swept)
// a network of a job starting during the pass is younger than this and so out of scope
assert.Equal(t, now.Add(-24*time.Hour), sweptCutoff)
// a host-only runner has no daemon to sweep
hostOnly := &Runner{uuid: "runner-2", cfg: &config.Config{Runner: cfg.Runner}, now: func() time.Time { return now }}
hostOnly.OnIdle(context.Background())
assert.Equal(t, []string{"runner-1"}, swept)
}

View File

@@ -7,12 +7,14 @@ import (
"context" "context"
"testing" "testing"
"gitea.com/gitea/runner/act/runner"
clientmocks "gitea.com/gitea/runner/internal/pkg/client/mocks" clientmocks "gitea.com/gitea/runner/internal/pkg/client/mocks"
"gitea.com/gitea/runner/internal/pkg/config" "gitea.com/gitea/runner/internal/pkg/config"
"gitea.com/gitea/runner/internal/pkg/ver" "gitea.com/gitea/runner/internal/pkg/ver"
"connectrpc.com/connect" "connectrpc.com/connect"
runnerv1 "gitea.dev/actions-proto-go/runner/v1" runnerv1 "gitea.dev/actions-proto-go/runner/v1"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/mock" "github.com/stretchr/testify/mock"
"github.com/stretchr/testify/require" "github.com/stretchr/testify/require"
"google.golang.org/protobuf/types/known/structpb" "google.golang.org/protobuf/types/known/structpb"
@@ -91,6 +93,7 @@ func TestNewRunnerInitializesLabelsAndEnvironment(t *testing.T) {
require.Equal(t, "true", r.envs["GITEA_ACTIONS"]) require.Equal(t, "true", r.envs["GITEA_ACTIONS"])
require.NotEmpty(t, r.envs["GITEA_ACTIONS_RUNNER_VERSION"]) require.NotEmpty(t, r.envs["GITEA_ACTIONS_RUNNER_VERSION"])
require.Nil(t, r.cacheHandler) require.Nil(t, r.cacheHandler)
require.Empty(t, r.envs[runner.CacheServiceV2Env], "no cache server, nothing to serve v2 from")
} }
// Proxy variables are assembled per task, because a job's service containers have to be // Proxy variables are assembled per task, because a job's service containers have to be
@@ -120,3 +123,22 @@ func taskWithDefaultActionsURL(url string) *runnerv1.Task {
}, },
} }
} }
// The cache service v2 API is announced to jobs unless it is turned off. Announcing it is what
// makes act patch the GHES check out of an action's bundle, so the client can reach it.
func TestNewRunnerCacheServiceV2(t *testing.T) {
announced := func(v2 *bool) string {
cfg := &config.Config{}
cfg.Cache.V2, cfg.Cache.Dir, cfg.Cache.Host = v2, t.TempDir(), "127.0.0.1"
cli := clientmocks.NewClient(t)
cli.On("Address").Return("https://gitea.example/").Maybe()
r := NewRunner(cfg, &config.Registration{Name: "runner"}, cli)
t.Cleanup(func() { _ = r.Close() })
return r.envs[runner.CacheServiceV2Env]
}
off := false
assert.Equal(t, "true", announced(nil))
assert.Empty(t, announced(&off))
}

View File

@@ -15,6 +15,9 @@ runner:
# Where to store the registration result. # Where to store the registration result.
file: .runner file: .runner
# Execute how many tasks concurrently at the same time. # Execute how many tasks concurrently at the same time.
# With `container.network` empty, each concurrent docker job takes a subnet from the
# daemon's address pool, so a high capacity can exhaust it. See `default-address-pools`
# in the docker daemon config.
capacity: 1 capacity: 1
# Extra environment variables to run jobs. # Extra environment variables to run jobs.
envs: envs:
@@ -43,9 +46,12 @@ runner:
# While idle, remove stale bind-workdir task directories and orphaned host-mode # While idle, remove stale bind-workdir task directories and orphaned host-mode
# scratch directories (left behind when a host cleanup delete stalls) older than # scratch directories (left behind when a host cleanup delete stalls) older than
# this duration. Setting either workdir_cleanup_age or idle_cleanup_interval to 0 # this duration. Setting either workdir_cleanup_age or idle_cleanup_interval to 0
# (or any non-positive value) disables stale-directory cleanup entirely. # (or any non-positive value) disables stale-directory cleanup entirely, along with
# the docker network cleanup below.
workdir_cleanup_age: 24h workdir_cleanup_age: 24h
# Cadence for the idle stale-directory cleanup pass. # Cadence for the idle cleanup pass. Besides the directories above, on runners that use
# docker it removes the per-job networks of jobs this runner did not live to tear down,
# which would otherwise hold a subnet of the daemon address pool until the host is rebuilt.
idle_cleanup_interval: 10m idle_cleanup_interval: 10m
# The base interval for periodic log flush to the Gitea instance. # The base interval for periodic log flush to the Gitea instance.
# Logs may be sent earlier if the buffer reaches log_report_batch_size # Logs may be sent earlier if the buffer reaches log_report_batch_size
@@ -149,6 +155,11 @@ cache:
# A moved tag (e.g. a re-tagged "v6") or an updated branch stays at the cached commit # A moved tag (e.g. a re-tagged "v6") or an updated branch stays at the cached commit
# until its cache entry expires or is manually removed. # until its cache entry expires or is manually removed.
offline_mode: false offline_mode: false
# Serve the actions cache service v2 API, used by actions/cache@v4.2 and later. Those actions
# refuse any host they do not take for GitHub, so reaching it means editing that check out of
# the action's own bundle, keeping the untouched copy beside it. The same edit lets the stock
# upload-artifact and download-artifact work here. A bundle that does not match is left alone.
v2: true
container: container:
# Specifies the network to which the container will connect. # Specifies the network to which the container will connect.
@@ -161,7 +172,9 @@ container:
# network_create_options only apply when `network` is left empty and the runner # network_create_options only apply when `network` is left empty and the runner
# auto-creates a per-job network that does not already exist. They have no effect # auto-creates a per-job network that does not already exist. They have no effect
# when a custom `network` name is set, because that network is used as-is and never # when a custom `network` name is set, because that network is used as-is and never
# created by the runner. Omit the entire block to use Docker's defaults. # created by the runner. Omit the entire block to use Docker's defaults. An auto-created
# network is labelled com.gitea.runner.uuid=<this runner's uuid>, which is how the idle
# cleanup tells its own leftovers apart from those of other runners on the same daemon.
network_create_options: network_create_options:
enable_ipv4: true # Omit to use Docker's default (IPv4 enabled). Set false to disable IPv4. enable_ipv4: true # Omit to use Docker's default (IPv4 enabled). Set false to disable IPv4.
enable_ipv6: false # Omit to use Docker's default (IPv6 disabled). Enabling it requires dockerd started with --ipv6. enable_ipv6: false # Omit to use Docker's default (IPv6 disabled). Enabling it requires dockerd started with --ipv6.
@@ -197,6 +210,9 @@ container:
docker_host: "" docker_host: ""
# Pull docker image(s) even if already present. # Pull docker image(s) even if already present.
# Defaults to false when the key is omitted. # Defaults to false when the key is omitted.
# Two exceptions: an image pinned by digest (image@sha256:...) cannot change, so it is never
# re-pulled, and a pull that fails while a copy is already on the host does not fail the job,
# which runs on that copy with a warning in its log.
force_pull: false force_pull: false
# Rebuild docker image(s) even if already present # Rebuild docker image(s) even if already present
force_rebuild: false force_rebuild: false

View File

@@ -42,7 +42,7 @@ type Runner struct {
FetchInterval time.Duration `yaml:"fetch_interval"` // FetchInterval specifies the interval duration for fetching resources. FetchInterval time.Duration `yaml:"fetch_interval"` // FetchInterval specifies the interval duration for fetching resources.
FetchIntervalMax time.Duration `yaml:"fetch_interval_max"` // FetchIntervalMax specifies the maximum backoff interval when idle. FetchIntervalMax time.Duration `yaml:"fetch_interval_max"` // FetchIntervalMax specifies the maximum backoff interval when idle.
WorkdirCleanupAge time.Duration `yaml:"workdir_cleanup_age"` // WorkdirCleanupAge removes stale bind-workdir task directories and orphaned host-mode scratch dirs older than this duration during idle cleanup. WorkdirCleanupAge time.Duration `yaml:"workdir_cleanup_age"` // WorkdirCleanupAge removes stale bind-workdir task directories and orphaned host-mode scratch dirs older than this duration during idle cleanup.
IdleCleanupInterval time.Duration `yaml:"idle_cleanup_interval"` // IdleCleanupInterval runs stale-directory cleanup periodically while the runner is idle. Set to 0 to disable cleanup cadence. IdleCleanupInterval time.Duration `yaml:"idle_cleanup_interval"` // IdleCleanupInterval runs the idle cleanup (stale directories and orphaned docker networks) periodically while the runner is idle. Set to 0 to disable cleanup cadence.
LogReportInterval time.Duration `yaml:"log_report_interval"` // LogReportInterval specifies the base interval for periodic log flush. LogReportInterval time.Duration `yaml:"log_report_interval"` // LogReportInterval specifies the base interval for periodic log flush.
LogReportMaxLatency time.Duration `yaml:"log_report_max_latency"` // LogReportMaxLatency specifies the max time a log row can wait before being sent. LogReportMaxLatency time.Duration `yaml:"log_report_max_latency"` // LogReportMaxLatency specifies the max time a log row can wait before being sent.
LogReportBatchSize int `yaml:"log_report_batch_size"` // LogReportBatchSize triggers immediate log flush when buffer reaches this size. LogReportBatchSize int `yaml:"log_report_batch_size"` // LogReportBatchSize triggers immediate log flush when buffer reaches this size.
@@ -74,6 +74,7 @@ type Cache struct {
ExternalSecret string `yaml:"external_secret"` // ExternalSecret is a shared secret between this runner and an external gitea-runner cache-server, enabling per-job ACTIONS_RUNTIME_TOKEN authentication and repo scoping over the network. Required whenever ExternalServer is set; ExternalSecretFile is the alternative way to provide it. ExternalSecret string `yaml:"external_secret"` // ExternalSecret is a shared secret between this runner and an external gitea-runner cache-server, enabling per-job ACTIONS_RUNTIME_TOKEN authentication and repo scoping over the network. Required whenever ExternalServer is set; ExternalSecretFile is the alternative way to provide it.
ExternalSecretFile string `yaml:"external_secret_file"` // ExternalSecretFile is the path to a file holding the ExternalSecret value, so the secret can be mounted instead of stored in the config file. LoadDefault reads it into ExternalSecret; setting both is an error. ExternalSecretFile string `yaml:"external_secret_file"` // ExternalSecretFile is the path to a file holding the ExternalSecret value, so the secret can be mounted instead of stored in the config file. LoadDefault reads it into ExternalSecret; setting both is an error.
OfflineMode bool `yaml:"offline_mode"` // OfflineMode reuses a cached action without fetching from the remote; a moved tag or branch stays at the cached commit until the cache entry is removed. OfflineMode bool `yaml:"offline_mode"` // OfflineMode reuses a cached action without fetching from the remote; a moved tag or branch stays at the cached commit until the cache entry is removed.
V2 *bool `yaml:"v2"` // V2 serves the actions cache service v2 API to jobs, used by actions/cache@v4.2 and later, and edits the action bundles that would otherwise refuse it. Unset means enabled.
} }
// Container represents the configuration for the container. // Container represents the configuration for the container.
@@ -86,7 +87,7 @@ type Container struct {
WorkdirParent string `yaml:"workdir_parent"` // WorkdirParent specifies the parent directory for the container's working directory. WorkdirParent string `yaml:"workdir_parent"` // WorkdirParent specifies the parent directory for the container's working directory.
ValidVolumes []string `yaml:"valid_volumes"` // ValidVolumes specifies the volumes (including bind mounts) can be mounted to containers. ValidVolumes []string `yaml:"valid_volumes"` // ValidVolumes specifies the volumes (including bind mounts) can be mounted to containers.
DockerHost string `yaml:"docker_host"` // DockerHost specifies the Docker host. It overrides the value specified in environment variable DOCKER_HOST. DockerHost string `yaml:"docker_host"` // DockerHost specifies the Docker host. It overrides the value specified in environment variable DOCKER_HOST.
ForcePull bool `yaml:"force_pull"` // Pull docker image(s) even if already present ForcePull bool `yaml:"force_pull"` // Pull docker image(s) even if already present, except digest-pinned ones. A pull that fails while a local copy exists is a warning, not a job failure.
ForceRebuild bool `yaml:"force_rebuild"` // Rebuild docker image(s) even if already present ForceRebuild bool `yaml:"force_rebuild"` // Rebuild docker image(s) even if already present
RequireDocker bool `yaml:"require_docker"` // Always require a reachable docker daemon, even if not required by runner RequireDocker bool `yaml:"require_docker"` // Always require a reachable docker daemon, even if not required by runner
DockerTimeout time.Duration `yaml:"docker_timeout"` // Timeout to wait for the docker daemon to be reachable, if docker is required by require_docker or runner DockerTimeout time.Duration `yaml:"docker_timeout"` // Timeout to wait for the docker daemon to be reachable, if docker is required by require_docker or runner

View File

@@ -347,3 +347,13 @@ cache:
require.Error(t, err) require.Error(t, err)
assert.Contains(t, err.Error(), "contains no secret") assert.Contains(t, err.Error(), "contains no secret")
} }
// The shipped example must parse, and every key in it must be one the config knows.
func TestLoadDefault_ExampleConfigParses(t *testing.T) {
hook := test.NewGlobal()
defer hook.Reset()
_, err := LoadDefault("config.example.yaml")
require.NoError(t, err)
assert.Empty(t, hook.AllEntries())
}

35
internal/pkg/lock/lock.go Normal file
View File

@@ -0,0 +1,35 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
// Package lock provides a cross-platform, non-blocking advisory file lock used
// to ensure a single runner process owns a given runner file.
package lock
import (
"errors"
"fmt"
)
// ErrLocked is returned by TryLock when another process already holds the lock.
var ErrLocked = errors.New("runner file is already locked by another process")
// TryLock takes a non-blocking exclusive advisory lock tied to runnerFile. The
// lock is placed on a sibling "<runnerFile>.lock" so it never interferes with
// in-place rewrites of the runner file itself.
//
// It returns a release function that drops the lock. The operating system also
// releases the lock automatically when the process exits, including on a hard
// kill, so a crashed runner never leaves a stale lock behind.
//
// If another process already holds the lock, it returns ErrLocked.
func TryLock(runnerFile string) (func() error, error) {
path := runnerFile + ".lock"
release, err := tryLock(path)
if errors.Is(err, ErrLocked) {
return nil, ErrLocked
}
if err != nil {
return nil, fmt.Errorf("lock %q: %w", path, err)
}
return release, nil
}

View File

@@ -0,0 +1,11 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
//go:build plan9
package lock
// tryLock is a best-effort no-op on plan9, which lacks flock/LockFileEx.
func tryLock(_ string) (func() error, error) {
return func() error { return nil }, nil
}

View File

@@ -0,0 +1,54 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
//go:build !plan9
package lock
import (
"errors"
"path/filepath"
"testing"
)
func TestTryLock(t *testing.T) {
runnerFile := filepath.Join(t.TempDir(), ".runner")
release, err := TryLock(runnerFile)
if err != nil {
t.Fatalf("first TryLock failed: %v", err)
}
// A second lock on the same file must be refused while the first is held.
if _, err := TryLock(runnerFile); !errors.Is(err, ErrLocked) {
t.Fatalf("second TryLock: want ErrLocked, got %v", err)
}
if err := release(); err != nil {
t.Fatalf("release failed: %v", err)
}
// After release the lock is available again.
release2, err := TryLock(runnerFile)
if err != nil {
t.Fatalf("TryLock after release failed: %v", err)
}
if err := release2(); err != nil {
t.Fatalf("second release failed: %v", err)
}
}
// TestTryLockUncreatable ensures a lock file that cannot be created reports a
// non-ErrLocked error, so callers can tell "already locked by another process"
// apart from "couldn't lock" and degrade gracefully (e.g. a read-only mount).
func TestTryLockUncreatable(t *testing.T) {
runnerFile := filepath.Join(t.TempDir(), "missing-dir", ".runner")
_, err := TryLock(runnerFile)
if err == nil {
t.Fatal("TryLock on an uncreatable lock file: want error, got nil")
}
if errors.Is(err, ErrLocked) {
t.Fatal("TryLock on an uncreatable lock file: want non-ErrLocked error, got ErrLocked")
}
}

View File

@@ -0,0 +1,35 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
//go:build !windows && !plan9
package lock
import (
"errors"
"os"
"golang.org/x/sys/unix"
)
// tryLock opens (creating if needed) the lock file and takes a non-blocking
// exclusive flock on it. The returned release closes the file, which drops the
// lock; the kernel also drops it when the process exits.
func tryLock(path string) (func() error, error) {
f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0o600)
if err != nil {
return nil, err
}
if err := unix.Flock(int(f.Fd()), unix.LOCK_EX|unix.LOCK_NB); err != nil {
_ = f.Close()
if errors.Is(err, unix.EWOULDBLOCK) || errors.Is(err, unix.EAGAIN) {
return nil, ErrLocked
}
return nil, err
}
return func() error {
// Best-effort unlock; closing the fd releases the lock regardless.
_ = unix.Flock(int(f.Fd()), unix.LOCK_UN)
return f.Close()
}, nil
}

View File

@@ -0,0 +1,36 @@
// Copyright 2026 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
//go:build windows
package lock
import (
"errors"
"os"
"golang.org/x/sys/windows"
)
// tryLock opens (creating if needed) the lock file and takes a non-blocking
// exclusive lock on it via LockFileEx. The returned release closes the file,
// which drops the lock; Windows also drops it when the process exits.
func tryLock(path string) (func() error, error) {
f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0o600)
if err != nil {
return nil, err
}
handle := windows.Handle(f.Fd())
overlapped := new(windows.Overlapped)
if err := windows.LockFileEx(handle, windows.LOCKFILE_EXCLUSIVE_LOCK|windows.LOCKFILE_FAIL_IMMEDIATELY, 0, 1, 0, overlapped); err != nil {
_ = f.Close()
if errors.Is(err, windows.ERROR_LOCK_VIOLATION) {
return nil, ErrLocked
}
return nil, err
}
return func() error {
_ = windows.UnlockFileEx(handle, 0, 1, 0, overlapped)
return f.Close()
}, nil
}

View File

@@ -5,15 +5,18 @@ package report
import ( import (
"context" "context"
"encoding/base64"
"errors" "errors"
"fmt" "fmt"
"maps" "maps"
"net/url"
"slices" "slices"
"strings" "strings"
"sync/atomic" "sync/atomic"
"testing" "testing"
"time" "time"
"gitea.com/gitea/runner/act/runner"
"gitea.com/gitea/runner/internal/pkg/client/mocks" "gitea.com/gitea/runner/internal/pkg/client/mocks"
"gitea.com/gitea/runner/internal/pkg/config" "gitea.com/gitea/runner/internal/pkg/config"
@@ -1096,3 +1099,24 @@ func TestReporter_ParseResult(t *testing.T) {
}) })
} }
} }
// A secret leaked in an encoded form — the shape it takes once an action puts it in a
// JSON body, a URL or a base64 payload — must be masked in the reported log as well.
func TestReporter_masksEncodedSecrets(t *testing.T) {
secret := `p@ss w"rd/1`
r := &Reporter{logReplacer: strings.NewReplacer()}
r.oldnew = runner.AppendSecretMasker(r.oldnew, secret)
r.logReplacer = strings.NewReplacer(r.oldnew...)
for _, line := range []string{
"token: " + secret,
"basic " + base64.StdEncoding.EncodeToString([]byte(secret)),
"https://example.com/?token=" + url.QueryEscape(secret),
} {
row := r.parseLogRow(&log.Entry{Message: line})
require.NotNil(t, row)
assert.Contains(t, row.Content, "***")
assert.NotContains(t, row.Content, secret)
assert.NotContains(t, row.Content, base64.StdEncoding.EncodeToString([]byte(secret)))
}
}