fix: stop the job's docker socket from becoming a directory (#1215)

Fixes https://gitea.com/gitea/runner/issues/1213

Fix the DooD regression that mounts `/var/run/docker.sock` as a directory. Keep the Docker proxy available through job and post steps. Clean stale resources before opening it, then remove containers before their networks and volumes during teardown.

Use a unique filesystem probe and preserve socket ownership. Fall back to direct access when proxying is unsupported. Preserve exec output and clean up active streams and failed starts.

Add a real Docker job test for mounted socket access, post steps and resource cleanup.

---------

Co-authored-by: silverwind <[email protected]>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1215
Reviewed-by: silverwind <[email protected]>
Co-authored-by: Zettat123 <[email protected]>

.gitea/workflows/pull-pr-title.yml

@@ -18,8 +18,8 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 5
     steps:
-    - uses: actions/checkout@v7
-    - uses: actions/setup-node@v6
+    - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+    - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
       with:
         node-version: 24
     - run: make lint-pr-title

.gitea/workflows/release-nightly.yml

@@ -17,24 +17,35 @@ jobs:
   goreleaser:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0
-      - uses: actions/setup-go@v6
+      # Custom publishers (the R2 upload below) run as the very last
+      # step of goreleaser's publish pipeline, after the Gitea release
+      # has already been created. Fail here instead, before anything
+      # is built or published, if the R2 secrets are missing.
+      - name: check R2 configuration
+        run: sh scripts/upload-r2.sh --check-config
+        env:
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: "go.mod"
+          check-latest: true
       - name: goreleaser
-        uses: goreleaser/goreleaser-action@v7
+        uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7
         with:
           distribution: goreleaser-pro
           args: release --nightly
         env:
           GORELEASER_KEY: ${{ secrets.GORELEASER_KEY }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY_ID }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-          S3_REGION: ${{ secrets.AWS_REGION }}
-          S3_BUCKET: ${{ secrets.AWS_BUCKET }}
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
           GORELEASER_FORCE_TOKEN: "gitea"
           GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
@@ -58,18 +69,18 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@v4
+        uses: docker/setup-qemu-action@1f40c72289eff860ee54a304f1438e3cff362e0a # v4
 
       - name: Set up Docker BuildX
-        uses: docker/setup-buildx-action@v4
+        uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4
 
       - name: Login to DockerHub
-        uses: docker/login-action@v4
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
@@ -83,7 +94,7 @@ jobs:
           echo REPO_VERSION=$(git describe --tags --always | sed 's/-/+/' | sed 's/^v//') >> $GITHUB_OUTPUT
 
       - name: Build and push
-        uses: docker/build-push-action@v7
+        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
         with:
           context: .
           file: ./Dockerfile

.gitea/workflows/release-tag.yml

@@ -9,31 +9,42 @@ jobs:
   goreleaser:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
-      - uses: actions/setup-go@v6
+      # Custom publishers (the R2 upload below) run as the very last
+      # step of goreleaser's publish pipeline, after the Gitea release
+      # has already been created. Fail here instead, before anything
+      # is built or published, if the R2 secrets are missing.
+      - name: check R2 configuration
+        run: sh scripts/upload-r2.sh --check-config
+        env:
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: "go.mod"
+          check-latest: true
       - name: Import GPG key
         id: import_gpg
-        uses: crazy-max/ghaction-import-gpg@v7
+        uses: crazy-max/ghaction-import-gpg@2dc316deee8e90f13e1a351ab510b4d5bc0c82cd # v7
         with:
           gpg_private_key: ${{ secrets.GPG_PRIVATE_KEY }}
           passphrase: ${{ secrets.PASSPHRASE }}
           fingerprint: CC64B1DB67ABBEECAB24B6455FC346329753F4B0
       - name: goreleaser
-        uses: goreleaser/goreleaser-action@v7
+        uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7
         with:
           distribution: goreleaser-pro
           args: release
         env:
           GORELEASER_KEY: ${{ secrets.GORELEASER_KEY }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY_ID }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-          S3_REGION: ${{ secrets.AWS_REGION }}
-          S3_BUCKET: ${{ secrets.AWS_BUCKET }}
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
           GORELEASER_FORCE_TOKEN: "gitea"
           GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           GPG_FINGERPRINT: ${{ steps.import_gpg.outputs.fingerprint }}
@@ -61,25 +72,25 @@ jobs:
       DOCKER_LATEST: latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@v4
+        uses: docker/setup-qemu-action@1f40c72289eff860ee54a304f1438e3cff362e0a # v4
 
       - name: Set up Docker BuildX
-        uses: docker/setup-buildx-action@v4
+        uses: docker/setup-buildx-action@37fe631027851001ddb9b187196cc803df7f5f0e # v4
 
       - name: Login to DockerHub
-        uses: docker/login-action@v4
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
 
       - name: "Docker meta"
         id: docker_meta
-        uses: docker/metadata-action@v6
+        uses: docker/metadata-action@dc802804100637a589fabce1cb79ff13a1411302 # v6
         with:
           images: |
             ${{ env.DOCKER_ORG }}/runner
@@ -92,7 +103,7 @@ jobs:
             suffix=${{ matrix.variant.tag_suffix }},onlatest=true
 
       - name: Build and push
-        uses: docker/build-push-action@v7
+        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
         with:
           context: .
           file: ./Dockerfile

.gitea/workflows/test.yml

@@ -5,34 +5,45 @@ on:
       - main
   pull_request:
 
+env:
+  # The runner image ships a stale docker.io login; point docker at an empty config so
+  # image pulls go straight to anonymous instead of attempting (and failing) that auth
+  # first. The path must be a literal: the `runner` context is unavailable in workflow-level
+  # env, so `${{ runner.temp }}` would resolve to empty and config.Dir() would fall back
+  # to ~/.docker with the stale credentials.
+  DOCKER_CONFIG: /tmp/docker-noauth
+
 jobs:
   lint:
     name: check and test
     runs-on: ubuntu-latest
-    env:
-      # The runner image ships a stale docker.io login; point docker at an empty config so
-      # image pulls go straight to anonymous instead of attempting (and failing) that auth
-      # first. The path must be a literal: the `runner` context is unavailable in job-level
-      # env, so `${{ runner.temp }}` would resolve to empty and config.Dir() would fall back
-      # to ~/.docker with the stale credentials.
-      DOCKER_CONFIG: /tmp/docker-noauth
     steps:
-      - uses: actions/checkout@v7
-      - uses: actions/setup-go@v6
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: 'go.mod'
+          check-latest: true
       - name: prepare anonymous docker config
         run: mkdir -p "$DOCKER_CONFIG" && echo '{}' > "$DOCKER_CONFIG/config.json"
       # Pre-pull act/runner's two largest base images so a slow pull can't dominate `make test`;
       # the rest (alpine/ubuntu) pull on demand, absorbed by the make-test -timeout. The host
       # daemon retains them between runs, so this is usually a fast manifest re-check.
       - name: pre-pull test images
+        env:
+          TEST_JOB_IMAGE: node:24-bookworm-slim@sha256:ba849c60be29959425b8734d57b8b4b7d56f98edd9504c9af091d5281095a71e # renovate: datasource=docker
+          TEST_SERVICE_IMAGE: nginx:alpine@sha256:72ba65eb42c10344912a84ff42408db7d34f2feb642204570ab8fc5ffd29f1d3 # renovate: datasource=docker
         run: |
-          for img in node:24-bookworm-slim nginx:alpine; do
-            for try in 1 2 3; do docker pull "$img" && break || sleep 5; done
+          for image in "$TEST_JOB_IMAGE" "$TEST_SERVICE_IMAGE"; do
+            for attempt in 1 2 3; do
+              docker pull "$image" && break
+              [ "$attempt" = 3 ] || sleep 5
+            done
+            docker tag "$image" "${image%@*}"
           done
       - name: lint
         run: make lint
+      - name: checks
+        run: make checks
       - name: build
         run: make build
       - name: test
@@ -45,4 +56,23 @@ jobs:
       - name: coverage report
         run: |
           make coverage-report
-          cat .tmp/coverage.md >> "$GITHUB_STEP_SUMMARY"
+          cat .tmp/coverage.md >> "$GITHUB_STEP_SUMMARY"
+
+  e2e:
+    name: gitea compatibility (${{ matrix.gitea_image }})
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        gitea_image:
+          - gitea/gitea:latest
+          - gitea/gitea:main-nightly
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
+        with:
+          go-version-file: 'go.mod'
+          check-latest: true
+      - name: prepare anonymous docker config
+        run: mkdir -p "$DOCKER_CONFIG" && echo '{}' > "$DOCKER_CONFIG/config.json"
+      - name: e2e compatibility
+        run: make test-e2e E2E_GITEA_IMAGE=${{ matrix.gitea_image }}

.gitignore

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

.golangci.yml

@@ -11,6 +11,7 @@ linters:
     - dupl
     - errcheck
     - forbidigo
+    - forcetypeassert
     - gocheckcompilerdirectives
     - gocritic
     - goheader
@@ -36,12 +37,8 @@ linters:
       rules:
         main:
           deny:
-            - pkg: io/ioutil
-              desc: use os or io instead
             - pkg: golang.org/x/exp
               desc: it's experimental and unreliable
-            - pkg: github.com/pkg/errors
-              desc: use builtin errors package instead
     nolintlint:
       allow-unused: false
       require-explanation: true
@@ -49,8 +46,7 @@ linters:
     gocritic:
       enabled-checks:
         - equalFold
-      disabled-checks:
-        - ifElseChain
+      disabled-checks: []
     revive:
       severity: error
       rules:
@@ -74,10 +70,14 @@ linters:
         - name: unexported-return
         - name: var-declaration
         - name: var-naming
+          arguments:
+            - [] # AllowList - do not remove as args for the rule are positional and won't work without lists first
+            - [] # DenyList
+            - - skip-initialism-name-checks: true
     staticcheck:
       checks:
         - all
-        - -ST1005
+    testifylint: {}
     usetesting:
       os-temp-dir: true
     perfsprint:
@@ -95,13 +95,14 @@ linters:
     generated: lax
     presets:
       - comments
-      - common-false-positives
-      - legacy
       - std-error-handling
     rules:
       - linters:
           - forbidigo
         path: cmd
+      - linters:
+          - forcetypeassert
+        path: _test\.go
 issues:
   max-issues-per-linter: 0
   max-same-issues: 0
@@ -118,7 +119,8 @@ formatters:
         - blank
         - default
     gofumpt:
-      extra-rules: true
+      extra:
+        group-params: true
   exclusions:
     generated: lax
 run:

.goreleaser.yaml

@@ -83,18 +83,37 @@ builds:
       - cmd: sh .goreleaser.checksum.sh {{ .Path }}
       - cmd: sh .goreleaser.checksum.sh {{ .Path }}.xz
 
-blobs:
-  -
-    provider: s3
-    bucket: "{{ .Env.S3_BUCKET }}"
-    region: "{{ .Env.S3_REGION }}"
-    directory: "gitea-runner/{{.Version}}"
+# Uploads every release artifact to Cloudflare R2. The `blobs:` pipe
+# isn't usable here since it authenticates from the global AWS_* env
+# with no per-entry credentials; `publishers:` supports per-entry
+# `env:` instead, so it's used to invoke scripts/upload-r2.sh once per
+# artifact. Custom publishers inherit almost nothing from the
+# environment, hence the explicit R2_* forwarding below.
+#
+# This publisher fires 109 times for 73 distinct keys because
+# goreleaser's release pipe already registers `release.extra_files`
+# as UploadableFile artifacts, and `internal/exec`'s filterArtifacts
+# appends this block's own extra_files with no de-duplication. It
+# can't be globbed away, since gobwas/glob (via goreleaser/fileglob)
+# has no substring-exclusion matcher. It's harmless: PUT is
+# idempotent, and the `./**.xz` glob below is kept deliberately so
+# this publisher declares its own complete file set rather than
+# implicitly depending on the `release:` block's globs.
+publishers:
+  - name: cloudflare-r2
+    checksum: true
     extra_files:
       - glob: ./**.xz
       - glob: ./**.sha256
+    cmd: sh scripts/upload-r2.sh {{ abs .ArtifactPath }} gitea-runner/{{ .Version }}/{{ .ArtifactName }}
+    env:
+      - R2_ENDPOINT={{ index .Env "R2_ENDPOINT" }}
+      - R2_BUCKET={{ index .Env "R2_BUCKET" }}
+      - R2_ACCESS_KEY_ID={{ index .Env "R2_ACCESS_KEY_ID" }}
+      - R2_SECRET_ACCESS_KEY={{ index .Env "R2_SECRET_ACCESS_KEY" }}
 
 archives:
-  - format: binary
+  - formats: [binary]
     name_template: "{{ .Binary }}"
     allow_different_binary_count: true
 

AGENTS.md

@@ -1,10 +1,19 @@
+- Never assume, verify before claiming
 - Use `make help` to find available development targets
-- Run `make fmt` to format `.go` files, and run `make lint-go` to lint them
-- Run `make tidy` after any `go.mod` changes
-- Run single go unit tests with `go test -run '^TestName$' ./modulepath/`
-- Add the current year into the copyright header of new `.go` files
-- Ensure no trailing whitespace in edited files
+- PR descriptions: minimal, only what and why, no task lists or file listings
+- Reference issues and PRs by full URL, not by number
+- Use Conventional Commits for commit messages and PR titles, plus the `enhance` type for user-facing enhancements
+- Add an `Assisted-by: AGENT_NAME:MODEL_VERSION` trailer to commit messages, never `Co-Authored-By` or `Signed-off-by`
+- Attribute agent authorship on one trailing line in issue and pull request comments, never as a PR description section
 - Never force-push, amend, or squash unless asked. Use new commits and normal push for pull request updates
-- Preserve existing code comments, do not remove or rewrite comments that are still relevant
-- Include authorship attribution in issue and pull request comments
-- Add `Co-Authored-By` lines to all commits, indicating name and model used
+- Comments: write almost none, short and preferably same-line, explaining why for a future reader. Never narrate code, the change or the prompt. Preserve existing ones that still apply
+- Add the current year into the copyright header of new `.go` files
+- Read `DEVELOPMENT.md` for internals and conventions
+- Ensure no trailing whitespace in edited files
+- Run `make fmt` after `.go` edits, `make tidy` after `go.mod` edits, and `make checks` for the non-lint source checks
+- Lint what changed with `make lint-go`, and `make lint-go-windows` for Windows and platform-split files
+- Fix the cause rather than disabling a linter or weakening a test. Where unavoidable, use the narrowest scope with a trailing comment giving the reason
+- Run single go tests with `go test -run '^TestName$' ./modulepath/`. `make test` self-skips the integration tests without docker or network, `make test-dind` runs the daemon-facing tests against the built dind image
+- Write the fewest, fastest tests covering the behavior, extending an existing one where possible. Prefer unit tests where logic is testable in isolation
+- Wait on a deterministic condition rather than `sleep`
+- Update the files under `docs/` when behavior documented there changes

DEVELOPMENT.md

@@ -0,0 +1,41 @@
+# Development
+
+## Job log line format
+
+Gitea stores one log row per line and its web UI decodes the payload, so getting the encoding
+wrong never fails a test here, it only shows up in the browser.
+
+**A row cannot contain a real newline.** `FormatLog` rewrites `\n` to a literal backslash-n and
+truncates at 64 KiB on a byte boundary.
+
+**The payload of a line starting with a recognised prefix is decoded**, with the escape set
+depending on the prefix:
+
+| prefix | decodes |
+| --- | --- |
+| `##[error]` `##[warning]` `##[notice]` `##[debug]` `##[group]` `##[endgroup]` `##[add-matcher]` | `%25` `%0D` `%0A` `%3B` `%5D` |
+| `::error::` `::warning::` `::notice::` `::debug::` (with or without ` key=value` properties), `::group::` `::endgroup::` `::add-matcher::` | `%25` `%0D` `%0A` |
+| `##[command]` `[command]`, or no recognised prefix | nothing |
+
+### Rules
+
+- **Emitting a command line?** Escape the payload with `runner.EscapeCommandData`. One escaper
+  covers both forms: it escapes `%` first, so a literal `%3B` becomes `%253B` that the extra
+  `##[…]` rules cannot match, and a raw `;` or `]` is never decoded. It is also what makes
+  multi-line work, `\n` becomes `%0A` and the UI turns it back into a line break.
+- **Forwarding a command from step output?** Leave the payload alone, it arrived escaped and is
+  decoded once. Decoding here double-decodes and destroys multi-line.
+- **No prefix?** Do not escape, and split multi-line values into one row each.
+- **Interpolating a secret?** Masking runs after escaping, so `AppendSecretMasker` registers the
+  encoded forms too.
+- Command *properties* also escape `%3A` and `%2C`, which the UI never decodes, so the reporter
+  decodes exactly those two when folding a location into an annotation.
+
+## End-to-end compatibility tests
+
+`make test-e2e` runs the runner against `E2E_GITEA_IMAGE`. It defaults to the nightly image.
+It requires Docker and is excluded from `make test`. CI runs stable and nightly variants in
+parallel.
+
+The suite shares one Gitea and regular runner. Cache and ephemeral scenarios use isolated
+repository runners. A run path is `<workflow>@<ref>` and a log row is `<timestamp>Z <payload>`.

Dockerfile

@@ -1,7 +1,7 @@
 ### BUILDER STAGE
 #
 #
-FROM golang:1.26-alpine3.23 AS builder
+FROM golang:1.27-alpine3.23 AS builder
 
 # Do not remove `git` here, it is required for getting runner version when executing `make build`
 RUN apk add --no-cache make git
@@ -17,7 +17,7 @@ RUN make clean && make build
 ### DIND VARIANT
 #
 #
-FROM docker:29.6.1-dind AS dind
+FROM docker:29.7.2-dind AS dind
 
 ARG VERSION=dev
 
@@ -37,7 +37,7 @@ ENTRYPOINT ["s6-svscan","/etc/s6"]
 ### DIND-ROOTLESS VARIANT
 #
 #
-FROM docker:29.6.1-dind-rootless AS dind-rootless
+FROM docker:29.7.2-dind-rootless AS dind-rootless
 
 ARG VERSION=dev
 

Makefile

@@ -4,9 +4,9 @@ DIST_DIRS := $(DIST)/binaries $(DIST)/release
 GO ?= go
 SHASUM ?= shasum -a 256
 HAS_GO = $(shell hash $(GO) > /dev/null 2>&1 && echo "GO" || echo "NOGO" )
-XGO_PACKAGE ?= src.techknowlogick.com/xgo@latest
-XGO_VERSION := go-1.26.x
-GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/[email protected]0
+XGO_PACKAGE ?= src.techknowlogick.com/xgo@v1.9.0 # renovate: datasource=go
+XGO_VERSION := go-1.27.x
+GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/[email protected]6 # renovate: datasource=go
 
 LINUX_ARCHS ?= linux/amd64,linux/arm64
 DARWIN_ARCHS ?= darwin-12/amd64,darwin-12/arm64
@@ -18,8 +18,10 @@ DOCKER_TAG ?= nightly
 DOCKER_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)
 DOCKER_ROOTLESS_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)-dind-rootless
 
-GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/[email protected]2.2
-GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.3.0
+GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/[email protected]3.2 # renovate: datasource=go
+GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.7.0 # renovate: datasource=go
+
+GOTEST_FLAGS ?= -race -timeout 20m -parallel 8
 
 STATIC ?=
 EXTLDFLAGS ?=
@@ -94,13 +96,8 @@ go-check:
 	fi
 
 .PHONY: fmt-check
-fmt-check: fmt
-	@diff=$$(git diff --color=always); \
-	if [ -n "$$diff" ]; then \
-		echo "Please run 'make fmt' and commit the result:"; \
-		printf "%s" "$${diff}"; \
-		exit 1; \
-	fi
+fmt-check: ## check the Go code formatting
+	$(GO) run $(GOLANGCI_LINT_PACKAGE) fmt --diff-colored
 
 .PHONY: deps-tools
 deps-tools: ## install tool dependencies
@@ -110,6 +107,9 @@ deps-tools: ## install tool dependencies
 	$(GO) install $(GOVULNCHECK_PACKAGE) & \
 	wait
 
+.PHONY: checks
+checks: tidy-check fmt-check security-check ## run the non-lint source checks
+
 .PHONY: lint
 lint: lint-go lint-go-windows ## lint everything
 
@@ -131,25 +131,25 @@ lint-pr-title: ## lint PR title against Conventional Commits (set PR_TITLE=...)
 	@node ./tools/lint-pr-title.ts
 
 .PHONY: security-check
-security-check: deps-tools
-	GOEXPERIMENT= $(GO) run $(GOVULNCHECK_PACKAGE) -show color ./... || true
+security-check:
+	$(GO) run $(GOVULNCHECK_PACKAGE) -show color ./... || true
 
 .PHONY: tidy
 tidy: ## run go mod tidy
+	$(eval GO_TOOLCHAIN := $(shell grep -Eo '^toolchain\s+go[0-9.]+' go.mod | cut -d' ' -f2))
 	$(GO) mod tidy
-
-.PHONY: tidy-check
-tidy-check: tidy
-	@diff=$$(git diff --color=always -- go.mod go.sum); \
-	if [ -n "$$diff" ]; then \
-		echo "Please run 'make tidy' and commit the result:"; \
-		printf "%s" "$${diff}"; \
-		exit 1; \
+	@# workaround https://github.com/golang/go/issues/75331: restore toolchain if tidy dropped it
+	@if [ -n "$(GO_TOOLCHAIN)" ] && ! grep -qE '^toolchain\s' go.mod; then \
+		$(GO) mod edit -toolchain=$(GO_TOOLCHAIN); \
 	fi
 
+.PHONY: tidy-check
+tidy-check: ## check that go.mod and go.sum are tidy
+	$(GO) mod tidy -diff
+
 .PHONY: test
-test: fmt-check security-check ## test everything (integration tests self-skip without docker/network)
-	@$(GO) test -race -timeout 20m -v -cover -coverprofile coverage.txt ./... && echo "\n==>\033[32m Ok\033[m\n" || exit 1
+test: ## test everything (integration tests self-skip without docker/network)
+	@$(GO) test $(GOTEST_FLAGS) -cover -coverprofile coverage.txt ./... && echo "\n==>\033[32m Ok\033[m\n" || exit 1
 
 .PHONY: coverage-report
 coverage-report: ## turn coverage.txt from `make test` into .tmp/coverage.md
@@ -161,6 +161,15 @@ coverage-report: ## turn coverage.txt from `make test` into .tmp/coverage.md
 test-dind: ## run the daemon-facing tests against the built dind image (TARGET=dind|dind-rootless)
 	@./scripts/test-dind.sh $(TARGET)
 
+E2E_JOB_IMAGE ?= node:24-bookworm@sha256:be23f54a88d34e8824c741b19b91064094f92c1c97b194144bfc8b50d67258e2 # renovate: datasource=docker
+SERVICE_IMAGE ?= nginx:1.31.4-alpine@sha256:db35bfc6b2951e7f8a72db5db120288c127ffaeeb4a6d4b95a26fead017d5913 # renovate: datasource=docker
+E2E_GITEA_IMAGE ?= gitea/gitea:main-nightly
+E2E_CONCURRENCY ?= 8
+
+.PHONY: test-e2e
+test-e2e: ## run Gitea compatibility tests against E2E_GITEA_IMAGE
+	@E2E_CONCURRENCY=$(E2E_CONCURRENCY) E2E_GITEA_IMAGE=$(E2E_GITEA_IMAGE) E2E_JOB_IMAGE=$(E2E_JOB_IMAGE) GO=$(GO) SERVICE_IMAGE=$(SERVICE_IMAGE) ./tools/test-e2e.sh
+
 .PHONY: install
 install: $(GOFILES) ## install the runner binary via `go install`
 	$(GO) install -v -tags '$(TAGS)' -ldflags '-s -w $(EXTLDFLAGS) $(LDFLAGS)'

README.md

@@ -129,27 +129,64 @@ Same idea as `dind`, but built on `docker:dind-rootless` so the bundled daemon a
 
 ### Configuration
 
-The runner is configured with a YAML file. Generate a starting point (this matches what ships in the tree):
+The runner reads a YAML file. Without one, every option keeps its default.
 
 ```bash
-./gitea-runner generate-config > config.yaml
+./gitea-runner config init             # write config.yaml, with no option set
+./gitea-runner config generate | less  # read what the options do
+./gitea-runner -c config.yaml daemon   # -c also works on register and cache-server
 ```
 
-Pass it with `-c` / `--config` on any command that loads configuration (`register`, `daemon`, `cache-server`):
+`config generate` prints [config.example.yaml](internal/pkg/config/config.example.yaml). Every value in it is commented out, so copy the lines you want to change into your own file and uncomment them.
+
+#### Editing a config file
+
+`config` edits a file in place, which is handy in provisioning scripts:
 
 ```bash
-./gitea-runner -c config.yaml register
-./gitea-runner -c config.yaml daemon
-./gitea-runner -c config.yaml cache-server
+./gitea-runner config set runner.capacity 4
+./gitea-runner config set runner.timeout 90m  # written as 1h30m0s
+./gitea-runner config set runner.envs.MY_VAR value
+./gitea-runner config add runner.labels 'ubuntu:docker://node:22'
+./gitea-runner config remove runner.labels 'ubuntu:docker://node:22'
+./gitea-runner config get runner.labels
 ```
 
-Every option is described in [config.example.yaml](internal/pkg/config/config.example.yaml) (the same content `generate-config` prints).
+A key is its dotted YAML path. An unknown key, a value of the wrong type, or `add`/`remove` on anything but a list is refused before the file is touched. `set` replaces a whole list when you give it several values.
 
-#### Without a config file
+An edit keeps the comments and the key order of the file. Indentation becomes two spaces, and a blank line between two values is dropped.
 
-If you omit `-c`, built-in defaults apply (same as an empty YAML document).
+`config get`, `set`, `add` and `remove` use `config.yaml` (or `config.yml`) from the working directory, then from the directory of the binary, and print their choice to stderr. `config init` writes `config.yaml` in the working directory, and refuses to overwrite an existing config without `--force`. Pass `-c` for another path.
 
-Earlier releases let a small set of environment variables (`GITEA_DEBUG`, `GITEA_TRACE`, `GITEA_RUNNER_CAPACITY`, `GITEA_RUNNER_FILE`, `GITEA_RUNNER_ENVIRON`, `GITEA_RUNNER_ENV_FILE`) override parts of the default config. Those overrides have been removed — use a YAML config file for all settings instead. For the Docker images, the entrypoint still understands a separate set of variables (such as `RUNNER_STATE_FILE`); see [scripts/run.sh](scripts/run.sh) and the container documentation below.
+#### Tool cache
+
+Setup actions like `setup-go` install tools into `RUNNER_TOOL_CACHE`, which is `/opt/hostedtoolcache` inside a job. `runner.tool_cache_mode` selects what backs it:
+
+| Mode | Tool cache | Trade-off |
+| --- | --- | --- |
+| `none` (default) | Per job, provided by the job image | A version the image lacks is downloaded in every job |
+| `shared` | One volume reused by every job | Two jobs writing the same tool version at once corrupt it, so use it only with `runner.capacity: 1` |
+
+With `none`, tools must come from the job image. Install them into `/opt/hostedtoolcache/<tool>/<version>/<arch>`, with an empty `<arch>.complete` file next to the directory:
+
+```dockerfile
+RUN GO=$(curl -fsSL 'https://go.dev/dl/?mode=json' | grep -oP '"version": "\Kgo1\.26\.[0-9]*' | head -1); \
+  DIR="/opt/hostedtoolcache/go/${GO#go}/x64" && \
+  mkdir -p "$(dirname "$DIR")" && \
+  curl -fsSL "https://dl.google.com/go/${GO}.linux-amd64.tar.gz" | tar -xz -C /tmp && \
+  mv /tmp/go "$DIR" && \
+  touch "${DIR}.complete"
+```
+
+A workflow requesting a minor version, `go-version: "1.26"`, resolves to the newest matching version in the cache, so a patch update in the image still hits it.
+
+Of the [runner images](https://gitea.com/gitea/runner-images), the `-full` flavour is the one that ships tools in this layout.
+
+`gitea-runner exec` reads no config file and takes `--tool-cache-mode` instead, defaulting to `none`.
+
+#### Environment variables
+
+Earlier releases let a few environment variables (`GITEA_DEBUG`, `GITEA_TRACE`, `GITEA_RUNNER_CAPACITY`, `GITEA_RUNNER_FILE`, `GITEA_RUNNER_ENVIRON`, `GITEA_RUNNER_ENV_FILE`) override parts of the config. They are gone, use the YAML file for all settings. The Docker images still read their own variables, such as `RUNNER_STATE_FILE`, see [scripts/run.sh](scripts/run.sh) and the container documentation below.
 
 ### Labels
 
@@ -191,7 +228,7 @@ a workflow with `runs-on: ubuntu-latest` is executed in the `runner-images:ubunt
 
 Names may themselves contain a colon (for example `pool:e57e18d4-10d4-406f-93bf-60f127221bdd`); only `host` and `docker` are treated as schemas.
 
-If a job's `runs-on` matches none of the runner's labels, the job still runs, in the default `docker.gitea.com/runner-images:ubuntu-latest` image. Images maintained for this purpose are listed at [gitea/runner-images](https://gitea.com/gitea/runner-images).
+If a job's `runs-on` matches none of the runner's labels, or sets no `runs-on` at all, it still runs: in `runner.default_image` where docker is available, on the host where it is not. Images maintained for this purpose are listed at [gitea/runner-images](https://gitea.com/gitea/runner-images).
 
 Labels are chosen at registration time (`--labels`, or the interactive prompt) and can be changed afterwards by editing `runner.labels` in the config file, or in the Gitea UI under the runner's settings.
 
@@ -209,10 +246,87 @@ Whenever the resulting labels differ from the ones in the registration file, the
 
 > **Note:** A runner that only exposes `host` labels still needs access to a Docker daemon (e.g. a mounted `/var/run/docker.sock`) whenever a job uses a `docker://` action or a service container. `host` labels only change where the job's own steps run; container-based steps and actions are still executed with Docker.
 
+#### Service containers
+
+A job's `services` are started before its steps run. When a service's image or its `options` declare a healthcheck, the runner waits for it to report healthy, so a workflow does not have to poll for its own services:
+
+```yaml
+services:
+  postgres:
+    image: postgres:17
+    options: >-
+      --health-cmd pg_isready
+      --health-interval 5s
+      --health-retries 10
+```
+
+A service that reports unhealthy fails the job right away, with its container log. One that never becomes healthy fails it after `container.service_ready_timeout` (default `5m`, negative disables the wait). A service that exits without declaring a healthcheck only gets its log and a warning.
+
+A job in a container reaches a service by its id on the job network, on the port the service listens on, for example `psql -h postgres -p 5432`. The started containers also fill the `job` context: `job.container.{id,network}` and `job.services.<id>.{id,network,ports}`, where `ports` maps a container port to the host port Docker published it on, for the services that publish one.
+
+Unlike GitHub, a job whose steps run on the host (a `host` label without `container:`) starts no service containers, so `job.services` and `job.container` stay empty. Give such a job a `container:` when it needs services.
+
+#### Docker from a job (`GITEA_DOCKER_WORKSPACE`)
+
+A container a job starts through the Docker socket cannot bind-mount the workspace by the job's own path, the daemon does not have it. `GITEA_DOCKER_WORKSPACE` holds the path the daemon sees. Use it as the prefix of workspace binds, with `.` as the fallback for local use, here in a `docker-compose.yaml`:
+
+```yaml
+volumes:
+  - ${GITEA_DOCKER_WORKSPACE:-.}/data:/app/data
+```
+
+Linux container jobs use a Docker proxy when socket sharing and permissions allow it. Other setups use the daemon socket directly. The proxy removes containers, networks and volumes after post steps and the completed hook. Named volumes created through it are job-scoped. To retain resources, mount the daemon socket explicitly in `container.options`. Host jobs use their existing Docker access, where `.` works.
+
+#### Proxy
+
+Set these variables in the runner's environment, with systemd `Environment=`, `docker run -e`, or Kubernetes `env:`:
+
+```sh
+http_proxy=http://proxy.example:3128
+https_proxy=http://proxy.example:3128
+no_proxy=gitea.internal,.example.local
+```
+
+The runner uses them for its own requests and gives them to every job, in lower and upper case.
+
+These hosts are added to `no_proxy` for jobs, so they are always reached directly:
+
+- the cache server
+- `localhost`, `127.0.0.1` and `::1`
+- the job's service containers
+- the Docker daemon, when it is reached over `tcp://`
+
+Gitea is not added. Add it to `no_proxy` yourself if it should be reached directly.
+
+To change a value for one job, set it in a step's `env:` or in the job's `container.env`. Setting it at workflow or job level has no effect. To change it for the whole runner, set it in `runner.envs`. A `no_proxy` set there is added to the list above instead of replacing it.
+
+Images are pulled by the Docker daemon, which needs its own proxy setting. In the `dind` images the daemon runs in the same container and reads the variables above. For any other daemon, see [the Docker documentation](https://docs.docker.com/engine/daemon/proxy/). The runner logs a warning at startup if it has a proxy and the daemon does not.
+
+Dockerfile actions are built with these variables as build arguments, so their `RUN` steps can reach the network.
+
+A password in a proxy URL is hidden in job logs. Any step can still read it, because the step is given the proxy URL in its environment.
+
 #### Caching (`actions/cache`)
 
 Each runner starts its own cache server automatically. Cache entries are local to that runner — runners do not share a cache by default.
 
+**Eviction**
+
+An entry nothing has read or written for `retention` is removed, and a repository past `repo_size_limit` loses its least recently accessed entries until it fits; `size_limit` caps the whole cache the same way. Age alone never retires an entry still in use, and whatever these allow, the cache keeps free space above `health_check.min_free_disk_space_mb` when health checks are enabled.
+
+These apply where the cache server runs, so on a shared server they belong in *its* config, not the runners'. See `retention`, `repo_size_limit`, `size_limit` and `sweep_interval` in [config.example.yaml](internal/pkg/config/config.example.yaml) for units and defaults.
+
+**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 on its way into the job, undone whenever the action is downloaded again. 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, so it is made whatever `v2` says: that setting only governs the API the runner advertises. Set `runner.patch_actions: false` to leave every bundle exactly as shipped, an escape hatch for an action the edit breaks. The artifact actions then refuse again and the cache client keeps to v1.
+
 **Shared cache across multiple runners**
 
 Run one dedicated `gitea-runner cache-server` that all runners point at.
@@ -224,6 +338,7 @@ Run one dedicated `gitea-runner cache-server` that all runners point at.
      dir: /data/actcache
      port: 8088
      external_secret: "replace-with-a-strong-random-secret"
+     # external_secret_file: /path/to/secret # secret can also be passed via a file
    ```
 
 2. Start the server:
@@ -238,8 +353,11 @@ Run one dedicated `gitea-runner cache-server` that all runners point at.
    cache:
      external_server: "http://<cache-server-host>:8088/"
      external_secret: "replace-with-a-strong-random-secret"  # must match the server
+     # external_secret_file: /path/to/secret # secret can also be passed via a file
    ```
 
+Jobs reach the cache server at `external_server`, so when a reverse proxy fronts the server, point `external_server` at the proxy. The cache server itself needs no extra configuration.
+
 Alternatively, mount the same NFS/CIFS share on every runner and point `cache.dir` at it — simpler, but with weaker isolation between repositories.
 
 **S3 / MinIO** — mount object storage as a FUSE filesystem (e.g. [s3fs](https://github.com/s3fs-fuse/s3fs-fuse) or [goofys](https://github.com/kahing/goofys)) and set `cache.dir` to the mount point.
@@ -252,11 +370,12 @@ 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).
 
-When `container.bind_workdir` is enabled, stale task workspace directories can be cleaned while the runner is idle:
-- 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)
+While the runner is idle it cleans up after earlier jobs:
+- when `container.bind_workdir` is enabled, stale task workspace directories older than `runner.workdir_cleanup_age` are removed (default: `24h`; set `0` to disable)
 - 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
+- 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`)
 
@@ -268,6 +387,26 @@ On Windows, use `.exe`, `.bat`, or `.cmd` paths; **PowerShell (`.ps1`) is not su
 
 See **[docs/post-task-script.md](docs/post-task-script.md)** for lifecycle details, environment variables, timeout interaction, and platform notes.
 
+#### Job hooks (`runner.hooks.job_started`, `runner.hooks.job_completed`)
+
+Optional scripts that run **inside the job environment** (the job container, or the host in host mode), before the job's first step and after its last one. They are the equivalent of GitHub's `ACTIONS_RUNNER_HOOK_JOB_STARTED` / `ACTIONS_RUNNER_HOOK_JOB_COMPLETED`, which are read when the settings are unset.
+
+Because they run where the steps run and see the job's environment, they are the place for per-job setup no workflow should have to carry: registry logins, mirror configuration, or masking runner-wide secrets with `::add-mask::`. Their output is part of the job log and is scanned for workflow commands, and they can export to the job through `$GITHUB_ENV` and `$GITHUB_PATH`.
+
+Both hooks are synchronous and block the job while they run. Either one exiting non-zero fails the job, and there is no per-hook timeout.
+
+See **[docs/job-hooks.md](docs/job-hooks.md)** for the execution order, environment, and platform notes.
+
+#### Local job logs (`log.job.dir`)
+
+Set `log.job.dir` to a path and the runner writes a copy of every task's log there as `<start time>-task-<id>.log`: the rows exactly as Gitea received them, with the same secrets masked and the job's result on the last line. Off by default, and what Gitea shows does not change.
+
+#### Secret masking
+
+A job's secrets and its `::add-mask::` values are hidden from what the runner writes and uploads: the job log, the local copy above, job summaries, and the names of the containers it creates. A job output carrying one is skipped with a warning rather than sent masked, as GitHub does, so a downstream `needs.<job>.outputs.<name>` reading it is empty.
+
+`log.job.retention` (default `168h`) is how long a log is kept, expired ones being deleted as new tasks start, and `log.job.max_size` (default `1GB`) caps one log. Keep `retention` above `runner.timeout` so a long job cannot outlive its own log, and prefer local disk, the file is written while the job runs. Only the runner's own user can read it.
+
 ### Example Deployments
 
 Check out the [examples](examples) directory for sample deployment types.

act/artifactcache/handler_v2.go

@@ -0,0 +1,300 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"cmp"
+	"encoding/json/jsontext"
+	"encoding/json/v2"
+	"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.
+//
+// Responses carry the proto field names, which is what Gitea's own results API emits and the only
+// spelling the Go clients parse. The JavaScript toolkit accepts either.
+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.touch(db, existing) // the client skips the upload, so this is the only sign the entry is still in use
+		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,
+		"signed_upload_url": h.signedURL(cred, 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 = int64(cmp.Or(req.SizeBytes, req.SizeBytesCamel))
+	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.
+		"entry_id": 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,
+		"signed_download_url": h.signedArtifactURL(cred, cache.ID, time.Now().Add(artifactURLTTL)),
+		"matched_key":         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
+	}
+
+	// The Azure SDK client dereferences this without checking, so its absence panics the caller.
+	w.Header().Set("x-ms-request-id", strconv.FormatInt(time.Now().UnixNano(), 10))
+	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      twirpInt64 `json:"size_bytes"`
+		SizeBytesCamel twirpInt64 `json:"sizeBytes"`
+	}
+
+	v2DownloadRequest struct {
+		Key              string   `json:"key"`
+		Version          string   `json:"version"`
+		RestoreKeys      []string `json:"restore_keys"`
+		RestoreKeysCamel []string `json:"restoreKeys"`
+	}
+)
+
+// twirpInt64 accepts its value as the JSON string the mapping prescribes or as a bare number.
+type twirpInt64 int64
+
+func (n *twirpInt64) UnmarshalJSONFrom(dec *jsontext.Decoder) error {
+	val, err := dec.ReadValue()
+	if err != nil {
+		return err
+	}
+	digits := []byte(val)
+	switch val.Kind() {
+	case 'n': // absent, keep the zero value
+		return nil
+	case '"':
+		if digits, err = jsontext.AppendUnquote(nil, val); err != nil {
+			return err
+		}
+	}
+	parsed, err := strconv.ParseInt(string(digits), 10, 64)
+	if err != nil {
+		return err
+	}
+	*n = twirpInt64(parsed)
+	return nil
+}
+
+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.UnmarshalRead(io.LimitReader(r.Body, 1<<20), &req)
+	return req, err
+}

act/artifactcache/handler_v2_test.go

@@ -0,0 +1,243 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"bytes"
+	"encoding/base64"
+	"encoding/json/v2"
+	"fmt"
+	"io"
+	"net/http"
+	"strconv"
+	"strings"
+	"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.UnmarshalRead(resp.Body, &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()
+	if resp.StatusCode == http.StatusCreated {
+		// The Azure SDK client dereferences this header without checking, so a blob upload that
+		// omits it panics the caller rather than failing it.
+		require.NotEmpty(t, resp.Header.Get("x-ms-request-id"))
+	}
+	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
+}
+
+// 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["signed_upload_url"].(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 := newTestHandler(t, Policy{})
+	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["entry_id"])
+
+	// 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["matched_key"])
+	downloadURL, _ := got["signed_download_url"].(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 := newTestHandler(t, Policy{})
+
+	created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "blocks", "version": "v1"})
+	uploadURL, _ := created["signed_upload_url"].(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["signed_download_url"].(string))))
+}
+
+func TestCacheServiceV2Lookups(t *testing.T) {
+	handler := newTestHandler(t, Policy{})
+	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["matched_key"])
+		})
+	}
+
+	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["signed_upload_url"])
+	})
+
+	t.Run("a proxied job is handed the address its runner registered", func(t *testing.T) {
+		const proxy = "https://cache.example.invalid"
+		handler.RegisterJob("proxied", JobCredential{Repo: testRepo, PublicURL: proxy + "/"})
+		client := &http.Client{Transport: &bearerTransport{token: "proxied"}}
+
+		created := v2Call(t, handler, client, "CreateCacheEntry", map[string]any{"key": "proxied-key", "version": "v1"})
+		assert.True(t, strings.HasPrefix(created["signed_upload_url"].(string), proxy+blobPath+"/"))
+
+		got := v2Call(t, handler, client, "GetCacheEntryDownloadURL", map[string]any{"key": "deps-abc", "version": "v1"})
+		assert.True(t, strings.HasPrefix(got["signed_download_url"].(string), proxy+apiPath+"/artifacts/"))
+	})
+
+	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["signed_upload_url"].(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.UnmarshalRead(resp.Body, &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", JobCredential{Repo: "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"])
+	})
+}

act/artifactcache/model.go

@@ -10,23 +10,6 @@ type Request struct {
 	Size    int64  `json:"cacheSize"`
 }
 
-func (c *Request) ToCache() *Cache {
-	if c == nil {
-		return nil
-	}
-	ret := &Cache{
-		Key:     c.Key,
-		Version: c.Version,
-		Size:    c.Size,
-	}
-	if c.Size == 0 {
-		// So the request comes from old versions of actions, like `actions/cache@v2`.
-		// It doesn't send cache size. Set it to -1 to indicate that.
-		ret.Size = -1
-	}
-	return ret
-}
-
 type Cache struct {
 	ID        uint64 `json:"id" boltholdKey:"ID"`
 	Repo      string `json:"repo" boltholdIndex:"Repo"`

act/artifactcache/results.go

@@ -0,0 +1,59 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"crypto/tls"
+	"net/http"
+	"net/http/httputil"
+	"net/url"
+	"strings"
+)
+
+// The results service is one origin serving every github.actions.results.api.v1 service, and
+// Gitea implements only the artifact half of it. Forwarding that half from here makes this origin
+// the whole service, so ACTIONS_RESULTS_URL can point at it truthfully, which is what the clients
+// this runner cannot patch need, docker buildx among them.
+//
+// The instance to forward to travels with the job registration rather than with configuration, so
+// a cache server shared between runners serves each of their instances.
+const artifactServicePath = "/twirp/github.actions.results.api.v1.ArtifactService/"
+
+// forwardOrNotFound is the router's fallback: the artifact service of the instance the job
+// registered with, and the 404 the router would have written otherwise.
+func (h *Handler) forwardOrNotFound(w http.ResponseWriter, r *http.Request) {
+	cred, ok := h.lookupCredential(bearerToken(r))
+	if !ok || cred.Results == "" || !strings.HasPrefix(r.URL.Path, artifactServicePath) {
+		http.NotFound(w, r)
+		return
+	}
+	target, err := url.Parse(strings.TrimSuffix(cred.Results, "/"))
+	if err != nil {
+		h.logger.Errorf("artifact service forward to %q: %v", cred.Results, err)
+		w.WriteHeader(http.StatusBadGateway)
+		return
+	}
+	h.logger.Debugf("%s %s: forwarding to %s", r.Method, r.URL.Path, target)
+
+	proxy := &httputil.ReverseProxy{
+		Rewrite: func(r *httputil.ProxyRequest) {
+			r.SetURL(target)
+			// Gitea builds the URLs it hands back from this Host, and their scheme from the
+			// connection unless a forwarded header overrides it, so artifact bodies go to Gitea
+			// directly and never through here.
+			r.Out.Host = target.Host
+		},
+		ErrorHandler: func(w http.ResponseWriter, _ *http.Request, err error) {
+			h.logger.Warnf("artifact service forward to %s: %v", target, err)
+			w.WriteHeader(http.StatusBadGateway)
+		},
+	}
+	if cred.InsecureTLS {
+		proxy.Transport = insecureTransport
+	}
+	proxy.ServeHTTP(w, r)
+}
+
+// insecureTransport is shared, because a transport per request would pool no connections.
+var insecureTransport = &http.Transport{TLSClientConfig: &tls.Config{InsecureSkipVerify: true}} //nolint:gosec // the runner reaches its instance on the operator's say-so

act/artifactcache/results_test.go

@@ -0,0 +1,56 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// The artifact half is forwarded under the Host Gitea knows itself by, so the URLs it hands back
+// still point at Gitea, and nothing else is proxied.
+func TestFrontResultsService(t *testing.T) {
+	var gotHost, gotPath, gotProto string
+	gitea := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotHost, gotPath, gotProto = r.Host, r.URL.Path, r.Header.Get("X-Forwarded-Proto")
+		_, _ = io.WriteString(w, `{"ok":true}`)
+	}))
+	defer gitea.Close()
+
+	handler, err := StartHandler(Options{Dir: t.TempDir(), OutboundIP: "127.0.0.1"})
+	require.NoError(t, err)
+	defer handler.Close()
+	const token = "forward-token"
+
+	client := &http.Client{Transport: &bearerTransport{token: token}}
+	post := func(path string) int {
+		req, err := http.NewRequestWithContext(t.Context(), http.MethodPost, handler.ExternalURL()+path, nil)
+		require.NoError(t, err)
+		resp, err := client.Do(req)
+		require.NoError(t, err)
+		resp.Body.Close()
+		return resp.StatusCode
+	}
+
+	assert.Equal(t, http.StatusNotFound, post(artifactServicePath+"CreateArtifact"),
+		"an unregistered token is forwarded nowhere")
+
+	defer handler.RegisterJob(token, JobCredential{Repo: "owner/repo", Results: gitea.URL})()
+
+	assert.Equal(t, http.StatusOK, post(artifactServicePath+"CreateArtifact"))
+	assert.Equal(t, strings.TrimPrefix(gitea.URL, "http://"), gotHost, "Gitea must see the host it mints its URLs from")
+	assert.Empty(t, gotProto, "a forwarded scheme would make an https Gitea mint http URLs")
+	assert.Equal(t, artifactServicePath+"CreateArtifact", gotPath)
+
+	gotPath = ""
+	assert.Equal(t, http.StatusNotFound, post("/twirp/github.actions.results.api.v1.OtherService/Do"))
+	assert.Equal(t, http.StatusNotFound, post("/api/v1/repos/owner/repo"))
+	assert.Empty(t, gotPath, "only the artifact service is forwarded")
+}

act/artifactcache/storage.go

@@ -5,12 +5,15 @@
 package artifactcache
 
 import (
+	"crypto/sha256"
+	"encoding/hex"
 	"fmt"
 	"io"
 	"net/http"
 	"os"
 	"path/filepath"
 	"strconv"
+	"strings"
 )
 
 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 {
-	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 {
 		return err
 	}
@@ -51,6 +57,26 @@ func (s *Storage) Write(id uint64, offset int64, reader io.Reader) error {
 	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) {
 	defer func() {
 		_ = 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 {
 		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)
 	if err != nil {
 		return 0, err
@@ -84,16 +135,6 @@ func (s *Storage) Commit(id uint64, size int64) (int64, error) {
 		}
 		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
 }
 
@@ -102,9 +143,13 @@ func (s *Storage) Serve(w http.ResponseWriter, r *http.Request, id uint64) {
 	http.ServeFile(w, r, name)
 }
 
-func (s *Storage) Remove(id uint64) {
-	_ = os.Remove(s.filename(id))
-	_ = os.RemoveAll(s.tempDir(id))
+// Remove deletes an entry's blob and any staged parts. It reports failure so the caller can
+// keep the entry and retry, rather than dropping the only reference to bytes on disk.
+func (s *Storage) Remove(id uint64) error {
+	if err := os.Remove(s.filename(id)); err != nil && !os.IsNotExist(err) {
+		return err
+	}
+	return os.RemoveAll(s.tempDir(id))
 }
 
 func (s *Storage) filename(id uint64) string {
@@ -119,6 +164,17 @@ func (s *Storage) tempName(id uint64, offset int64) string {
 	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) {
 	dir := s.tempDir(id)
 	files, err := os.ReadDir(dir)
@@ -127,7 +183,7 @@ func (s *Storage) tempNames(id uint64) ([]string, error) {
 	}
 	var names []string
 	for _, v := range files {
-		if !v.IsDir() {
+		if !v.IsDir() && !strings.HasPrefix(v.Name(), blockFilePrefix) {
 			names = append(names, filepath.Join(dir, v.Name()))
 		}
 	}

act/artifacts/server.go

@@ -6,7 +6,7 @@ package artifacts
 
 import (
 	"context"
-	"encoding/json"
+	"encoding/json/v2"
 	"errors"
 	"fmt"
 	"io"
@@ -50,65 +50,29 @@ type ResponseMessage struct {
 	Message string `json:"message"`
 }
 
-type WritableFile interface {
-	io.WriteCloser
-}
-
-type WriteFS interface {
-	OpenWritable(name string) (WritableFile, error)
-	OpenAppendable(name string) (WritableFile, error)
-}
-
-type readWriteFSImpl struct{}
-
-func (fwfs readWriteFSImpl) Open(name string) (fs.File, error) {
-	return os.Open(name)
-}
-
-func (fwfs readWriteFSImpl) OpenWritable(name string) (WritableFile, error) {
-	if err := os.MkdirAll(filepath.Dir(name), os.ModePerm); err != nil {
-		return nil, err
-	}
-	return os.OpenFile(name, os.O_CREATE|os.O_RDWR|os.O_TRUNC, 0o644)
-}
-
-func (fwfs readWriteFSImpl) OpenAppendable(name string) (WritableFile, error) {
-	if err := os.MkdirAll(filepath.Dir(name), os.ModePerm); err != nil {
-		return nil, err
-	}
-	file, err := os.OpenFile(name, os.O_CREATE|os.O_RDWR, 0o644)
-	if err != nil {
-		return nil, err
-	}
-
-	_, err = file.Seek(0, io.SeekEnd)
-	if err != nil {
-		return nil, err
-	}
-	return file, nil
-}
-
 var gzipExtension = ".gz__"
 
 func safeResolve(baseDir, relPath string) string {
 	return filepath.Join(baseDir, filepath.Clean(filepath.Join(string(os.PathSeparator), relPath)))
 }
 
-func uploads(router *httprouter.Router, baseDir string, fsys WriteFS) {
+func writeJSON(w http.ResponseWriter, value any) {
+	data, err := json.Marshal(value)
+	if err != nil {
+		panic(err)
+	}
+	if _, err := w.Write(data); err != nil {
+		panic(err)
+	}
+}
+
+func uploads(router *httprouter.Router, baseDir string) {
 	router.POST("/_apis/pipelines/workflows/:runId/artifacts", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
 		runID := params.ByName("runId")
 
-		json, err := json.Marshal(FileContainerResourceURL{
+		writeJSON(w, FileContainerResourceURL{
 			FileContainerResourceURL: fmt.Sprintf("http://%s/upload/%s", req.Host, runID),
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.PUT("/upload/:runId", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
@@ -122,67 +86,47 @@ func uploads(router *httprouter.Router, baseDir string, fsys WriteFS) {
 		safeRunPath := safeResolve(baseDir, runID)
 		safePath := safeResolve(safeRunPath, itemPath)
 
-		file, err := func() (WritableFile, error) {
-			contentRange := req.Header.Get("Content-Range")
-			if contentRange != "" && !strings.HasPrefix(contentRange, "bytes 0-") {
-				return fsys.OpenAppendable(safePath)
-			}
-			return fsys.OpenWritable(safePath)
-		}()
+		if err := os.MkdirAll(filepath.Dir(safePath), os.ModePerm); err != nil {
+			panic(err)
+		}
+		flags := os.O_CREATE | os.O_WRONLY | os.O_TRUNC
+		appendUpload := req.Header.Get("Content-Range")
+		if appendUpload != "" && !strings.HasPrefix(appendUpload, "bytes 0-") {
+			flags = os.O_CREATE | os.O_WRONLY | os.O_APPEND
+		}
+		file, err := os.OpenFile(safePath, flags, 0o644)
 		if err != nil {
 			panic(err)
 		}
 		defer file.Close()
-
-		writer, ok := file.(io.Writer)
-		if !ok {
-			panic(errors.New("File is not writable"))
-		}
-
 		if req.Body == nil {
-			panic(errors.New("No body given"))
+			panic(errors.New("no body given"))
 		}
 
-		_, err = io.Copy(writer, req.Body)
+		_, err = io.Copy(file, req.Body)
 		if err != nil {
 			panic(err)
 		}
 
-		json, err := json.Marshal(ResponseMessage{
+		writeJSON(w, ResponseMessage{
 			Message: "success",
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.PATCH("/_apis/pipelines/workflows/:runId/artifacts", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
-		json, err := json.Marshal(ResponseMessage{
+		writeJSON(w, ResponseMessage{
 			Message: "success",
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 }
 
-func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
+func downloads(router *httprouter.Router, baseDir string) {
 	router.GET("/_apis/pipelines/workflows/:runId/artifacts", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
 		runID := params.ByName("runId")
 
 		safePath := safeResolve(baseDir, runID)
 
-		entries, err := fs.ReadDir(fsys, safePath)
+		entries, err := os.ReadDir(safePath)
 		if err != nil {
 			panic(err)
 		}
@@ -195,18 +139,10 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 			})
 		}
 
-		json, err := json.Marshal(NamedFileContainerResourceURLResponse{
+		writeJSON(w, NamedFileContainerResourceURLResponse{
 			Count: len(list),
 			Value: list,
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.GET("/download/:container", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
@@ -215,7 +151,7 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 		safePath := safeResolve(baseDir, filepath.Join(container, itemPath))
 
 		var files []ContainerItem
-		err := fs.WalkDir(fsys, safePath, func(path string, entry fs.DirEntry, err error) error {
+		err := filepath.WalkDir(safePath, func(path string, entry fs.DirEntry, err error) error {
 			if !entry.IsDir() {
 				rel, err := filepath.Rel(safePath, path)
 				if err != nil {
@@ -241,17 +177,9 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 			panic(err)
 		}
 
-		json, err := json.Marshal(ContainerItemResponse{
+		writeJSON(w, ContainerItemResponse{
 			Value: files,
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.GET("/artifact/*path", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
@@ -259,15 +187,16 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 
 		safePath := safeResolve(baseDir, path)
 
-		file, err := fsys.Open(safePath)
+		file, err := os.Open(safePath)
 		if err != nil {
 			// try gzip file
-			file, err = fsys.Open(safePath + gzipExtension)
+			file, err = os.Open(safePath + gzipExtension)
 			if err != nil {
 				panic(err)
 			}
 			w.Header().Add("Content-Encoding", "gzip")
 		}
+		defer file.Close()
 
 		_, err = io.Copy(w, file)
 		if err != nil {
@@ -287,9 +216,8 @@ func Serve(ctx context.Context, artifactPath, addr, port string) context.CancelF
 	router := httprouter.New()
 
 	logger.Debugf("Artifacts base path '%s'", artifactPath)
-	fsys := readWriteFSImpl{}
-	uploads(router, artifactPath, fsys)
-	downloads(router, artifactPath, fsys)
+	uploads(router, artifactPath)
+	downloads(router, artifactPath)
 
 	server := &http.Server{
 		Addr:              fmt.Sprintf("%s:%s", addr, port),

act/artifacts/server_test.go

@@ -7,8 +7,7 @@ package artifacts
 import (
 	"bytes"
 	"compress/gzip"
-	"encoding/json"
-	"fmt"
+	"encoding/json/v2"
 	"io"
 	"maps"
 	"net/http"
@@ -18,238 +17,18 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
-	"testing/fstest"
 	"time"
 
 	"github.com/julienschmidt/httprouter"
-	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
-type writableMapFile struct {
-	fstest.MapFile
-}
-
-func (f *writableMapFile) Write(data []byte) (int, error) {
-	f.Data = data
-	return len(data), nil
-}
-
-func (f *writableMapFile) Close() error {
-	return nil
-}
-
-type writeMapFS struct {
-	fstest.MapFS
-}
-
-func (fsys writeMapFS) OpenWritable(name string) (WritableFile, error) {
-	file := &writableMapFile{
-		MapFile: fstest.MapFile{
-			Data: []byte("content2"),
-		},
-	}
-	fsys.MapFS[name] = &file.MapFile
-
-	return file, nil
-}
-
-func (fsys writeMapFS) OpenAppendable(name string) (WritableFile, error) {
-	file := &writableMapFile{
-		MapFile: fstest.MapFile{
-			Data: []byte("content2"),
-		},
-	}
-	fsys.MapFS[name] = &file.MapFile
-
-	return file, nil
-}
-
-func TestNewArtifactUploadPrepare(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPost, "http://localhost/_apis/pipelines/workflows/1/artifacts", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := FileContainerResourceURL{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("http://localhost/upload/1", response.FileContainerResourceURL)
-}
-
-func TestArtifactUploadBlob(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPut, "http://localhost/upload/1?itemPath=some/file", strings.NewReader("content"))
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := ResponseMessage{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("success", response.Message)
-	assert.Equal("content", string(memfs["artifact/server/path/1/some/file"].Data))
-}
-
-func TestFinalizeArtifactUpload(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPatch, "http://localhost/_apis/pipelines/workflows/1/artifacts", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := ResponseMessage{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("success", response.Message)
-}
-
-func TestListArtifacts(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/1/file.txt": {
-			Data: []byte(""),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/_apis/pipelines/workflows/1/artifacts", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	response := NamedFileContainerResourceURLResponse{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal(1, response.Count)
-	assert.Equal("file.txt", response.Value[0].Name)
-	assert.Equal("http://localhost/download/1", response.Value[0].FileContainerResourceURL)
-}
-
-func TestListArtifactContainer(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/1/some/file": {
-			Data: []byte(""),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/download/1?itemPath=some/file", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	response := ContainerItemResponse{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Len(response.Value, 1)
-	assert.Equal("some/file", response.Value[0].Path)
-	assert.Equal("file", response.Value[0].ItemType)
-	assert.Equal("http://localhost/artifact/1/some/file/.", response.Value[0].ContentLocation)
-}
-
-func TestDownloadArtifactFile(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/1/some/file": {
-			Data: []byte("content"),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/artifact/1/some/file", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	data := rr.Body.Bytes()
-
-	assert.Equal("content", string(data))
-}
-
-// TestArtifactFlow drives the real Serve() artifact server over a loopback socket, exercising
-// the same upload -> finalize -> list -> download protocol the upload-artifact/download-artifact
-// actions speak. Running it in-process (rather than from a job container) keeps it network-free
-// and reachable everywhere, including when the CI job is itself a container.
 func TestArtifactFlow(t *testing.T) {
 	artifactPath := t.TempDir()
 
-	// Serve the exact routes Serve() wires up, on a real loopback socket via httptest. httptest
-	// picks a free port and Close() tears the server down synchronously — avoiding both the
-	// port-rebind race and Serve()'s detached ListenAndServe goroutine, which logger.Fatal()s
-	// (process exit) on a bind error and can outlive the test's temp-dir cleanup.
 	router := httprouter.New()
-	fsys := readWriteFSImpl{}
-	uploads(router, artifactPath, fsys)
-	downloads(router, artifactPath, fsys)
+	uploads(router, artifactPath)
+	downloads(router, artifactPath)
 	server := httptest.NewServer(router)
 	defer server.Close()
 
@@ -257,8 +36,6 @@ func TestArtifactFlow(t *testing.T) {
 	client := server.Client()
 	client.Timeout = 5 * time.Second
 
-	// request performs one HTTP call and returns the status and body. The default transport adds
-	// Accept-Encoding: gzip and transparently decompresses, so gzipped downloads come back plain.
 	request := func(t *testing.T, method, rawURL string, body io.Reader, header http.Header) (int, []byte) {
 		t.Helper()
 		req, err := http.NewRequest(method, rawURL, body)
@@ -289,6 +66,8 @@ func TestArtifactFlow(t *testing.T) {
 
 		status, data = request(t, http.MethodPatch, baseURL+"/_apis/pipelines/workflows/"+runID+"/artifacts", nil, nil)
 		require.Equal(t, http.StatusOK, status, string(data))
+		require.NoError(t, json.Unmarshal(data, &msg))
+		require.Equal(t, "success", msg.Message)
 
 		status, data = request(t, http.MethodGet, baseURL+"/_apis/pipelines/workflows/"+runID+"/artifacts", nil, nil)
 		require.Equal(t, http.StatusOK, status, string(data))
@@ -314,6 +93,21 @@ func TestArtifactFlow(t *testing.T) {
 		require.Equal(t, content, string(stored))
 	})
 
+	t.Run("content-range", func(t *testing.T) {
+		const rawURL = "/upload/4?itemPath=chunks.txt"
+		status, data := request(t, http.MethodPut, baseURL+rawURL, strings.NewReader("first"),
+			http.Header{"Content-Range": []string{"bytes 0-4/11"}})
+		require.Equal(t, http.StatusOK, status, string(data))
+
+		status, data = request(t, http.MethodPut, baseURL+rawURL, strings.NewReader("-second"),
+			http.Header{"Content-Range": []string{"bytes 5-11/11"}})
+		require.Equal(t, http.StatusOK, status, string(data))
+
+		stored, err := os.ReadFile(filepath.Join(artifactPath, "4", "chunks.txt"))
+		require.NoError(t, err)
+		require.Equal(t, "first-second", string(stored))
+	})
+
 	t.Run("gzip-roundtrip", func(t *testing.T) {
 		const runID, item, content = "2", "logs/app.log", "compressed payload\n"
 
@@ -365,9 +159,7 @@ func TestArtifactFlow(t *testing.T) {
 	})
 }
 
-func TestMkdirFsImplSafeResolve(t *testing.T) {
-	assert := assert.New(t)
-
+func TestSafeResolve(t *testing.T) {
 	baseDir := "/foo/bar"
 
 	tests := map[string]struct {
@@ -385,97 +177,13 @@ func TestMkdirFsImplSafeResolve(t *testing.T) {
 
 	for name, tc := range tests {
 		t.Run(name, func(t *testing.T) {
-			assert.Equal(tc.want, safeResolve(baseDir, tc.input))
+			require.Equal(t, tc.want, safeResolve(baseDir, tc.input))
 		})
 	}
 }
 
-func TestReadWriteFSWritableAndAppendable(t *testing.T) {
-	fsys := readWriteFSImpl{}
-	name := filepath.Join(t.TempDir(), "nested", "artifact.txt")
-
-	w, err := fsys.OpenWritable(name)
-	require.NoError(t, err)
-	_, err = w.Write([]byte("first"))
-	require.NoError(t, err)
-	require.NoError(t, w.Close())
-
-	w, err = fsys.OpenAppendable(name)
-	require.NoError(t, err)
-	_, err = w.Write([]byte("-second"))
-	require.NoError(t, err)
-	require.NoError(t, w.Close())
-
-	got, err := os.ReadFile(name)
-	require.NoError(t, err)
-	require.Equal(t, "first-second", string(got))
-
-	w, err = fsys.OpenWritable(name)
-	require.NoError(t, err)
-	_, err = w.Write([]byte("replaced"))
-	require.NoError(t, err)
-	require.NoError(t, w.Close())
-
-	got, err = os.ReadFile(name)
-	require.NoError(t, err)
-	require.Equal(t, "replaced", string(got))
-}
-
 func TestServeEmptyArtifactPathReturnsCancelableNoop(t *testing.T) {
 	cancel := Serve(t.Context(), "", "127.0.0.1", "0")
 	require.NotNil(t, cancel)
 	cancel()
 }
-
-func TestDownloadArtifactFileUnsafePath(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/some/file": {
-			Data: []byte("content"),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/artifact/2/../../some/file", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	data := rr.Body.Bytes()
-
-	assert.Equal("content", string(data))
-}
-
-func TestArtifactUploadBlobUnsafePath(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPut, "http://localhost/upload/1?itemPath=../../some/file", strings.NewReader("content"))
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := ResponseMessage{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("success", response.Message)
-	assert.Equal("content", string(memfs["artifact/server/path/1/some/file"].Data))
-}

act/common/cartesian.go

@@ -1,60 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// Copyright 2020 The nektos/act Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import "slices"
-
-// CartesianProduct takes map of lists and returns list of unique tuples
-func CartesianProduct(mapOfLists map[string][]any) []map[string]any {
-	listNames := make([]string, 0)
-	lists := make([][]any, 0)
-	for k, v := range mapOfLists {
-		listNames = append(listNames, k)
-		lists = append(lists, v)
-	}
-
-	listCart := cartN(lists...)
-
-	rtn := make([]map[string]any, 0)
-	for _, list := range listCart {
-		vMap := make(map[string]any)
-		for i, v := range list {
-			vMap[listNames[i]] = v
-		}
-		rtn = append(rtn, vMap)
-	}
-	return rtn
-}
-
-func cartN(a ...[]any) [][]any {
-	c := 1
-	for _, a := range a {
-		c *= len(a)
-	}
-	if c == 0 || len(a) == 0 {
-		return nil
-	}
-	p := make([][]any, c)
-	b := make([]any, c*len(a))
-	n := make([]int, len(a))
-	s := 0
-	for i := range p {
-		e := s + len(a)
-		pi := b[s:e]
-		p[i] = pi
-		s = e
-		for j, n := range n {
-			pi[j] = a[j][n]
-		}
-		for j := range slices.Backward(n) {
-			n[j]++
-			if n[j] < len(a[j]) {
-				break
-			}
-			n[j] = 0
-		}
-	}
-	return p
-}

act/common/cartesian_test.go

@@ -1,43 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// Copyright 2020 The nektos/act Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-)
-
-func TestCartesianProduct(t *testing.T) {
-	assert := assert.New(t)
-	input := map[string][]any{
-		"foo": {1, 2, 3, 4},
-		"bar": {"a", "b", "c"},
-		"baz": {false, true},
-	}
-
-	output := CartesianProduct(input)
-	assert.Len(output, 24)
-
-	for _, v := range output {
-		assert.Len(v, 3)
-
-		assert.Contains(v, "foo")
-		assert.Contains(v, "bar")
-		assert.Contains(v, "baz")
-	}
-
-	input = map[string][]any{
-		"foo": {1, 2, 3, 4},
-		"bar": {},
-		"baz": {false, true},
-	}
-	output = CartesianProduct(input)
-	assert.Empty(output)
-
-	input = map[string][]any{}
-	output = CartesianProduct(input)
-	assert.Empty(output)
-}

act/common/executor.go

@@ -54,22 +54,6 @@ func NewPipelineExecutor(executors ...Executor) Executor {
 	return rtn
 }
 
-// NewConditionalExecutor creates a new executor based on conditions
-func NewConditionalExecutor(conditional Conditional, trueExecutor, falseExecutor Executor) Executor {
-	return func(ctx context.Context) error {
-		if conditional(ctx) {
-			if trueExecutor != nil {
-				return trueExecutor(ctx)
-			}
-		} else {
-			if falseExecutor != nil {
-				return falseExecutor(ctx)
-			}
-		}
-		return nil
-	}
-}
-
 // NewErrorExecutor creates a new executor that always errors out
 func NewErrorExecutor(err error) Executor {
 	return func(ctx context.Context) error {
@@ -187,15 +171,8 @@ func (e Executor) Finally(finally Executor) Executor {
 		err := e(ctx)
 		err2 := finally(ctx)
 		if err2 != nil {
-			return fmt.Errorf("Error occurred running finally: %v (original error: %v)", err2, err)
+			return fmt.Errorf("error occurred running finally: %v (original error: %v)", err2, err)
 		}
 		return err
 	}
 }
-
-// Not return an inverted conditional
-func (c Conditional) Not() Conditional {
-	return func(ctx context.Context) bool {
-		return !c(ctx)
-	}
-}

act/common/executor_max_parallel_test.go

@@ -1,89 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import (
-	"context"
-	"sync"
-	"sync/atomic"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-)
-
-// Simple fast test that verifies max-parallel: 2 limits concurrency
-func TestMaxParallel2Quick(t *testing.T) {
-	ctx := context.Background()
-
-	var currentRunning atomic.Int32
-	var maxSimultaneous atomic.Int32
-
-	executors := make([]Executor, 4)
-	for i := range 4 {
-		executors[i] = func(ctx context.Context) error {
-			current := currentRunning.Add(1)
-
-			// Update max if needed
-			for {
-				maxValue := maxSimultaneous.Load()
-				if current <= maxValue || maxSimultaneous.CompareAndSwap(maxValue, current) {
-					break
-				}
-			}
-
-			time.Sleep(10 * time.Millisecond)
-			currentRunning.Add(-1)
-			return nil
-		}
-	}
-
-	err := NewParallelExecutor(2, executors...)(ctx)
-
-	assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.LessOrEqual(t, maxSimultaneous.Load(), int32(2),
-		"Should not exceed max-parallel: 2")
-}
-
-// Test that verifies max-parallel: 1 enforces sequential execution
-func TestMaxParallel1Sequential(t *testing.T) {
-	ctx := context.Background()
-
-	var currentRunning atomic.Int32
-	var maxSimultaneous atomic.Int32
-	var executionOrder []int
-	var orderMutex sync.Mutex
-
-	executors := make([]Executor, 5)
-	for i := range 5 {
-		taskID := i
-		executors[i] = func(ctx context.Context) error {
-			current := currentRunning.Add(1)
-
-			// Track execution order
-			orderMutex.Lock()
-			executionOrder = append(executionOrder, taskID)
-			orderMutex.Unlock()
-
-			// Update max if needed
-			for {
-				maxValue := maxSimultaneous.Load()
-				if current <= maxValue || maxSimultaneous.CompareAndSwap(maxValue, current) {
-					break
-				}
-			}
-
-			time.Sleep(20 * time.Millisecond)
-			currentRunning.Add(-1)
-			return nil
-		}
-	}
-
-	err := NewParallelExecutor(1, executors...)(ctx)
-
-	assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(t, int32(1), maxSimultaneous.Load(),
-		"max-parallel: 1 should only run 1 task at a time")
-	assert.Len(t, executionOrder, 5, "All 5 tasks should have executed")
-}

act/common/executor_parallel_advanced_test.go

@@ -1,221 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import (
-	"context"
-	"sync"
-	"sync/atomic"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-)
-
-// TestMaxParallelJobExecution tests actual job execution with max-parallel
-func TestMaxParallelJobExecution(t *testing.T) {
-	t.Run("MaxParallel=1 Sequential", func(t *testing.T) {
-		var currentRunning atomic.Int32
-		var maxConcurrent int32
-		var executionOrder []int
-		var mu sync.Mutex
-
-		executors := make([]Executor, 5)
-		for i := range 5 {
-			taskID := i
-			executors[i] = func(ctx context.Context) error {
-				current := currentRunning.Add(1)
-
-				// Track max concurrent
-				for {
-					maxValue := atomic.LoadInt32(&maxConcurrent)
-					if current <= maxValue || atomic.CompareAndSwapInt32(&maxConcurrent, maxValue, current) {
-						break
-					}
-				}
-
-				mu.Lock()
-				executionOrder = append(executionOrder, taskID)
-				mu.Unlock()
-
-				time.Sleep(10 * time.Millisecond)
-				currentRunning.Add(-1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(1, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.Equal(t, int32(1), maxConcurrent, "Should never exceed 1 concurrent execution")
-		assert.Len(t, executionOrder, 5, "All tasks should execute")
-	})
-
-	t.Run("MaxParallel=3 Limited", func(t *testing.T) {
-		var currentRunning atomic.Int32
-		var maxConcurrent int32
-
-		executors := make([]Executor, 10)
-		for i := range 10 {
-			executors[i] = func(ctx context.Context) error {
-				current := currentRunning.Add(1)
-
-				for {
-					maxValue := atomic.LoadInt32(&maxConcurrent)
-					if current <= maxValue || atomic.CompareAndSwapInt32(&maxConcurrent, maxValue, current) {
-						break
-					}
-				}
-
-				time.Sleep(20 * time.Millisecond)
-				currentRunning.Add(-1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(3, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.LessOrEqual(t, int(maxConcurrent), 3, "Should never exceed 3 concurrent executions")
-		assert.GreaterOrEqual(t, int(maxConcurrent), 1, "Should have at least 1 concurrent execution")
-	})
-
-	t.Run("MaxParallel=0 Uses1Worker", func(t *testing.T) {
-		var maxConcurrent int32
-		var currentRunning atomic.Int32
-
-		executors := make([]Executor, 5)
-		for i := range 5 {
-			executors[i] = func(ctx context.Context) error {
-				current := currentRunning.Add(1)
-
-				for {
-					maxValue := atomic.LoadInt32(&maxConcurrent)
-					if current <= maxValue || atomic.CompareAndSwapInt32(&maxConcurrent, maxValue, current) {
-						break
-					}
-				}
-
-				time.Sleep(10 * time.Millisecond)
-				currentRunning.Add(-1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		// When maxParallel is 0 or negative, it defaults to 1
-		err := NewParallelExecutor(0, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.Equal(t, int32(1), maxConcurrent, "Should use 1 worker when max-parallel is 0")
-	})
-}
-
-// TestMaxParallelWithErrors tests error handling with max-parallel
-func TestMaxParallelWithErrors(t *testing.T) {
-	t.Run("OneTaskFailsOthersContinue", func(t *testing.T) {
-		var successCount int32
-
-		executors := make([]Executor, 5)
-		for i := range 5 {
-			taskID := i
-			executors[i] = func(ctx context.Context) error {
-				if taskID == 2 {
-					return assert.AnError
-				}
-				atomic.AddInt32(&successCount, 1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(2, executors...)(ctx)
-
-		// Should return the error from task 2
-		assert.Error(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		// Other tasks should still execute
-		assert.Equal(t, int32(4), successCount, "4 tasks should succeed")
-	})
-
-	t.Run("ContextCancellation", func(t *testing.T) {
-		ctx, cancel := context.WithCancel(context.Background())
-
-		var startedCount int32
-		executors := make([]Executor, 10)
-		for i := range 10 {
-			executors[i] = func(ctx context.Context) error {
-				atomic.AddInt32(&startedCount, 1)
-				time.Sleep(100 * time.Millisecond)
-				return nil
-			}
-		}
-
-		// Cancel after a short delay
-		go func() {
-			time.Sleep(30 * time.Millisecond)
-			cancel()
-		}()
-
-		err := NewParallelExecutor(3, executors...)(ctx)
-		assert.Error(t, err)                     //nolint:testifylint // pre-existing issue from nektos/act
-		assert.ErrorIs(t, err, context.Canceled) //nolint:testifylint // pre-existing issue from nektos/act
-
-		// Not all tasks should start due to cancellation (but timing may vary)
-		// Just verify cancellation occurred
-		t.Logf("Started %d tasks before cancellation", startedCount)
-	})
-}
-
-// TestMaxParallelResourceSharing tests resource sharing scenarios
-func TestMaxParallelResourceSharing(t *testing.T) {
-	t.Run("SharedResourceWithMutex", func(t *testing.T) {
-		var sharedCounter int
-		var mu sync.Mutex
-
-		executors := make([]Executor, 100)
-		for i := range 100 {
-			executors[i] = func(ctx context.Context) error {
-				mu.Lock()
-				sharedCounter++
-				mu.Unlock()
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(10, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.Equal(t, 100, sharedCounter, "All tasks should increment counter")
-	})
-
-	t.Run("ChannelCommunication", func(t *testing.T) {
-		resultChan := make(chan int, 50)
-
-		executors := make([]Executor, 50)
-		for i := range 50 {
-			taskID := i
-			executors[i] = func(ctx context.Context) error {
-				resultChan <- taskID
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(5, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		close(resultChan)
-
-		results := make(map[int]bool)
-		for result := range resultChan {
-			results[result] = true
-		}
-
-		assert.Len(t, results, 50, "All task IDs should be received")
-	})
-}

act/common/executor_test.go

@@ -9,9 +9,9 @@ import (
 	"errors"
 	"reflect"
 	"strings"
+	"sync"
 	"sync/atomic"
 	"testing"
-	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -45,81 +45,45 @@ func TestNewWorkflow(t *testing.T) {
 	assert.Equal(2, runcount)
 }
 
-func TestNewConditionalExecutor(t *testing.T) {
-	assert := assert.New(t)
+// concurrencyProbe returns an executor recording the peak number of concurrent copies. Copies
+// block until wantActive are in flight so the peak is exact without sleeping, and later copies
+// find the gate already open so the last one still finishes with no partner left.
+func concurrencyProbe(wantActive int32) (exec Executor, count, maxActive *atomic.Int32) {
+	var counted, active, peak atomic.Int32
+	var once sync.Once
+	reached := make(chan struct{})
 
-	ctx := context.Background()
-
-	trueCount := 0
-	falseCount := 0
-
-	err := NewConditionalExecutor(func(ctx context.Context) bool {
-		return false
-	}, func(ctx context.Context) error {
-		trueCount++
-		return nil
-	}, func(ctx context.Context) error {
-		falseCount++
-		return nil
-	})(ctx)
-
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(0, trueCount)
-	assert.Equal(1, falseCount)
-
-	err = NewConditionalExecutor(func(ctx context.Context) bool {
-		return true
-	}, func(ctx context.Context) error {
-		trueCount++
-		return nil
-	}, func(ctx context.Context) error {
-		falseCount++
-		return nil
-	})(ctx)
-
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(1, trueCount)
-	assert.Equal(1, falseCount)
-}
-
-func TestNewParallelExecutor(t *testing.T) {
-	assert := assert.New(t)
-
-	ctx := context.Background()
-
-	var count, activeCount, maxCount atomic.Int32
-	emptyWorkflow := NewPipelineExecutor(func(ctx context.Context) error {
-		count.Add(1)
-
-		active := activeCount.Add(1)
+	return func(ctx context.Context) error {
+		counted.Add(1)
+		running := active.Add(1)
 		for {
-			m := maxCount.Load()
-			if active <= m || maxCount.CompareAndSwap(m, active) {
+			seen := peak.Load()
+			if running <= seen || peak.CompareAndSwap(seen, running) {
 				break
 			}
 		}
-		time.Sleep(2 * time.Second)
-		activeCount.Add(-1)
-
+		if running >= wantActive {
+			once.Do(func() { close(reached) })
+		}
+		<-reached
+		active.Add(-1)
 		return nil
-	})
+	}, &counted, &peak
+}
 
-	err := NewParallelExecutor(2, emptyWorkflow, emptyWorkflow, emptyWorkflow)(ctx)
+func TestNewParallelExecutor(t *testing.T) {
+	ctx := context.Background()
 
-	assert.Equal(int32(3), count.Load(), "should run all 3 executors")
-	assert.Equal(int32(2), maxCount.Load(), "should run at most 2 executors in parallel")
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
+	exec, count, maxActive := concurrencyProbe(2)
+	require.NoError(t, NewParallelExecutor(2, exec, exec, exec)(ctx))
+	assert.Equal(t, int32(3), count.Load(), "should run all 3 executors")
+	assert.Equal(t, int32(2), maxActive.Load(), "should run at most 2 executors in parallel")
 
-	// Reset to test running the executor with 0 parallelism
-	count.Store(0)
-	activeCount.Store(0)
-	maxCount.Store(0)
-
-	errSingle := NewParallelExecutor(0, emptyWorkflow, emptyWorkflow, emptyWorkflow)(ctx)
-
-	assert.Equal(int32(3), count.Load(), "should run all 3 executors")
-	assert.Equal(int32(1), maxCount.Load(), "should run at most 1 executors in parallel")
-	assert.NoError(errSingle)
+	// parallelism below 1 falls back to a single worker
+	exec, count, maxActive = concurrencyProbe(1)
+	require.NoError(t, NewParallelExecutor(0, exec, exec, exec)(ctx))
+	assert.Equal(t, int32(3), count.Load(), "should run all 3 executors")
+	assert.Equal(t, int32(1), maxActive.Load(), "should run at most 1 executor in parallel")
 }
 
 func TestNewParallelExecutorEmpty(t *testing.T) {
@@ -173,6 +137,23 @@ func TestNewParallelExecutorCanceled(t *testing.T) {
 	assert.Error(errExpected, err) //nolint:testifylint // pre-existing issue from nektos/act
 }
 
+func TestNewParallelExecutorRunsRemainingAfterFailure(t *testing.T) {
+	var successCount atomic.Int32
+	executors := make([]Executor, 5)
+	for i := range executors {
+		executors[i] = func(ctx context.Context) error {
+			if i == 2 {
+				return errors.New("fake error")
+			}
+			successCount.Add(1)
+			return nil
+		}
+	}
+
+	require.Error(t, NewParallelExecutor(2, executors...)(context.Background()))
+	assert.Equal(t, int32(4), successCount.Load(), "a failing executor must not stop the others")
+}
+
 func TestExecutorConditionalsAndFinally(t *testing.T) {
 	ctx := context.Background()
 	var calls []string
@@ -205,10 +186,3 @@ func TestExecutorFinallyReturnsFinallyErrorWithOriginal(t *testing.T) {
 		t.Fatalf("finally error = %q, want both cleanup and original error", err)
 	}
 }
-
-func TestConditionalNot(t *testing.T) {
-	cond := Conditional(func(context.Context) bool { return false })
-	if !cond.Not()(context.Background()) {
-		t.Fatal("inverted conditional should be true")
-	}
-}

act/common/git/git.go

@@ -16,6 +16,7 @@ import (
 	"sync"
 
 	"gitea.com/gitea/runner/act/common"
+	"gitea.com/gitea/runner/internal/pkg/lock"
 
 	"github.com/go-git/go-git/v5"
 	"github.com/go-git/go-git/v5/config"
@@ -32,10 +33,9 @@ var (
 	githubHTTPRegex     = regexp.MustCompile(`^https?://.*github.com.*/(.+)/(.+?)(?:.git)?$`)
 	githubSSHRegex      = regexp.MustCompile(`github.com[:/](.+)/(.+?)(?:.git)?$`)
 
-	cloneLocks sync.Map // key: clone target directory; value: *sync.Mutex
+	cloneLocks lock.Keyed[string] // key: clone target directory
 
 	ErrShortRef = errors.New("short SHA references are not supported")
-	ErrNoRepo   = errors.New("unable to find git repo")
 )
 
 // AcquireCloneLock returns an unlock function after locking the per-directory mutex for dir.
@@ -43,10 +43,7 @@ var (
 // Callers reading files inside dir (e.g. tarring a checked-out action into a job container) must hold this lock too,
 // otherwise a concurrent NewGitCloneExecutor on the same dir can mutate the worktree mid-read.
 func AcquireCloneLock(dir string) func() {
-	v, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
-	mu := v.(*sync.Mutex)
-	mu.Lock()
-	return mu.Unlock
+	return cloneLocks.Lock(dir)
 }
 
 type Error struct {
@@ -189,19 +186,16 @@ func FindGitRef(ctx context.Context, file string) (string, error) {
 }
 
 // FindGithubRepo get the repo
-func FindGithubRepo(ctx context.Context, file, githubInstance, remoteName string) (string, error) {
+func FindGithubRepo(ctx context.Context, file, githubInstance string) (string, error) {
 	goGitMu.Lock()
 	defer goGitMu.Unlock()
-	if remoteName == "" {
-		remoteName = "origin"
-	}
 
-	url, err := findGitRemoteURL(ctx, file, remoteName)
+	url, err := findGitRemoteURL(ctx, file, "origin")
 	if err != nil {
 		return "", err
 	}
-	_, slug, err := findGitSlug(url, githubInstance)
-	return slug, err
+	_, slug := findGitSlug(url, githubInstance)
+	return slug, nil
 }
 
 func findGitRemoteURL(_ context.Context, file, remoteName string) (string, error) {
@@ -228,25 +222,25 @@ func findGitRemoteURL(_ context.Context, file, remoteName string) (string, error
 	return remote.Config().URLs[0], nil
 }
 
-func findGitSlug(url, githubInstance string) (string, string, error) { //nolint:unparam // pre-existing issue from nektos/act
+func findGitSlug(url, githubInstance string) (string, string) {
 	if matches := codeCommitHTTPRegex.FindStringSubmatch(url); matches != nil {
-		return "CodeCommit", matches[2], nil
+		return "CodeCommit", matches[2]
 	} else if matches := codeCommitSSHRegex.FindStringSubmatch(url); matches != nil {
-		return "CodeCommit", matches[2], nil
+		return "CodeCommit", matches[2]
 	} else if matches := githubHTTPRegex.FindStringSubmatch(url); matches != nil {
-		return "GitHub", fmt.Sprintf("%s/%s", matches[1], matches[2]), nil
+		return "GitHub", fmt.Sprintf("%s/%s", matches[1], matches[2])
 	} else if matches := githubSSHRegex.FindStringSubmatch(url); matches != nil {
-		return "GitHub", fmt.Sprintf("%s/%s", matches[1], matches[2]), nil
+		return "GitHub", fmt.Sprintf("%s/%s", matches[1], matches[2])
 	} else if githubInstance != "github.com" {
 		gheHTTPRegex := regexp.MustCompile(fmt.Sprintf(`^https?://%s/(.+)/(.+?)(?:.git)?$`, githubInstance))
 		gheSSHRegex := regexp.MustCompile(githubInstance + "[:/](.+)/(.+?)(?:.git)?$")
 		if matches := gheHTTPRegex.FindStringSubmatch(url); matches != nil {
-			return "GitHubEnterprise", fmt.Sprintf("%s/%s", matches[1], matches[2]), nil
+			return "GitHubEnterprise", fmt.Sprintf("%s/%s", matches[1], matches[2])
 		} else if matches := gheSSHRegex.FindStringSubmatch(url); matches != nil {
-			return "GitHubEnterprise", fmt.Sprintf("%s/%s", matches[1], matches[2]), nil
+			return "GitHubEnterprise", fmt.Sprintf("%s/%s", matches[1], matches[2])
 		}
 	}
-	return "", url, nil
+	return "", url
 }
 
 // NewGitCloneExecutorInput the input for the NewGitCloneExecutor
@@ -261,6 +255,10 @@ type NewGitCloneExecutorInput struct {
 	// 0 for full clone.
 	Depth int
 
+	// Quiet drops the informational clone line to debug level, for callers that log their own
+	// download summary (the setup section's action report).
+	Quiet bool
+
 	// For Gitea
 	InsecureSkipTLS bool
 }
@@ -276,11 +274,12 @@ func CloneIfRequired(ctx context.Context, refName plumbing.ReferenceName, input
 			return r, true, nil
 		}
 
-		if err != nil {
+		switch {
+		case err != nil:
 			logger.Debugf("Removing cached clone at %s because origin cannot be read: %v", input.Dir, err)
-		} else if len(remote.Config().URLs) == 0 {
+		case len(remote.Config().URLs) == 0:
 			logger.Debugf("Removing cached clone at %s because origin has no URL", input.Dir)
-		} else {
+		default:
 			logger.Debugf("Removing cached clone at %s because origin URL changed from %s to %s", input.Dir, remote.Config().URLs[0], input.URL)
 		}
 		if err := os.RemoveAll(input.Dir); err != nil {
@@ -343,11 +342,25 @@ func gitOptions(token string) (fetchOptions git.FetchOptions, pullOptions git.Pu
 	return fetchOptions, pullOptions
 }
 
+// staleRefreshErr reports why a failed refresh must abort: the resolve and
+// checkout that follow are local and succeed on a cancelled context, which
+// would hand back the cached revision as if it were fresh.
+func staleRefreshErr(ctx context.Context, err error) error {
+	if err == nil || errors.Is(err, git.NoErrAlreadyUpToDate) {
+		return nil
+	}
+	return ctx.Err()
+}
+
 // NewGitCloneExecutor creates an executor to clone git repos
 func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 	return func(ctx context.Context) error {
 		logger := common.Logger(ctx)
-		logger.Infof("git clone '%s' # ref=%s", input.URL, input.Ref)
+		if input.Quiet {
+			logger.Debugf("git clone '%s' # ref=%s", input.URL, input.Ref)
+		} else {
+			logger.Infof("git clone '%s' # ref=%s", input.URL, input.Ref)
+		}
 		logger.Debugf("  cloning %s to %s", input.URL, input.Dir)
 
 		defer AcquireCloneLock(input.Dir)()
@@ -378,8 +391,12 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 			}
 		}
 
-		if !isOfflineMode {
-			err = r.Fetch(&fetchOptions)
+		// A just-cloned ref is as current as a fetch would make it, and a commit hash never moves.
+		_, _, present := refRevision(r, input.Ref)
+		refresh := !isOfflineMode && (!present || (reused && !plumbing.IsHash(input.Ref)))
+
+		if refresh {
+			err = r.FetchContext(ctx, &fetchOptions)
 			if err != nil && !errors.Is(err, git.NoErrAlreadyUpToDate) {
 				return err
 			}
@@ -398,23 +415,7 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 			}
 		}
 
-		// At this point we need to know if it's a tag or a branch
-		// And the easiest way to do it is duck typing
-		//
-		// If err is nil, it's a tag so let's proceed with that hash like we would if
-		// it was a sha
-		refType := "tag"
-		rev = plumbing.Revision(path.Join("refs", "tags", input.Ref))
-		if _, err := r.Tag(input.Ref); errors.Is(err, git.ErrTagNotFound) {
-			rName := plumbing.ReferenceName(path.Join("refs", "remotes", "origin", input.Ref))
-			if _, err := r.Reference(rName, false); errors.Is(err, plumbing.ErrReferenceNotFound) {
-				refType = "sha"
-				rev = plumbing.Revision(input.Ref)
-			} else {
-				refType = "branch"
-				rev = plumbing.Revision(rName)
-			}
-		}
+		rev, refType, _ := refRevision(r, input.Ref)
 
 		if hash, err = r.ResolveRevision(rev); err != nil {
 			logger.Errorf("Unable to resolve %s: %v", input.Ref, err)
@@ -446,11 +447,14 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 		reusedMsg := ""
 
 		switch {
-		case !isOfflineMode && !shallow:
+		case refresh && !shallow:
 			// In shallow mode the depth-limited fetch above already advanced the ref.
-			if err = w.Pull(&pullOptions); err != nil && err != git.NoErrAlreadyUpToDate {
+			if err = w.PullContext(ctx, &pullOptions); err != nil && !errors.Is(err, git.NoErrAlreadyUpToDate) {
 				logger.Debugf("Unable to pull %s: %v", refName, err)
 			}
+			if err := staleRefreshErr(ctx, err); err != nil {
+				return err
+			}
 		case isOfflineMode && reused:
 			reusedMsg = " (reused in offline mode)"
 		}
@@ -485,27 +489,37 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 	}
 }
 
-// cloneAtDepth clones input.URL into input.Dir using opts.
-// With input.Depth > 0 it first tries a shallow, single-branch clone of input.Ref, falling back when error.
+// cloneAtDepth falls back to a full clone when the shallow attempt fails.
 func cloneAtDepth(ctx context.Context, input NewGitCloneExecutorInput, opts git.CloneOptions, logger log.FieldLogger) (*git.Repository, error) {
 	if input.Depth > 0 {
-		for _, refName := range []plumbing.ReferenceName{
-			plumbing.NewBranchReferenceName(input.Ref),
-			plumbing.NewTagReferenceName(input.Ref),
-		} {
-			shallowOpts := opts
-			shallowOpts.Depth = input.Depth
-			shallowOpts.SingleBranch = true
-			shallowOpts.ReferenceName = refName
-			shallowOpts.Tags = git.NoTags
-
-			r, err := git.PlainCloneContext(ctx, input.Dir, false, &shallowOpts)
+		if plumbing.IsHash(input.Ref) {
+			r, err := fetchPinnedSHA(ctx, input, opts)
 			if err == nil {
 				return r, nil
 			}
-			logger.Debugf("Shallow clone of %s as %s failed: %v", input.URL, refName, err)
-			if rmErr := os.RemoveAll(input.Dir); rmErr != nil {
-				return nil, fmt.Errorf("remove partial clone %s: %w", input.Dir, rmErr)
+			logger.Debugf("Shallow fetch of %s at %s failed: %v", input.URL, input.Ref, err)
+			if err := removePartialClone(input.Dir); err != nil {
+				return nil, err
+			}
+		} else {
+			for _, refName := range []plumbing.ReferenceName{
+				plumbing.NewBranchReferenceName(input.Ref),
+				plumbing.NewTagReferenceName(input.Ref),
+			} {
+				shallowOpts := opts
+				shallowOpts.Depth = input.Depth
+				shallowOpts.SingleBranch = true
+				shallowOpts.ReferenceName = refName
+				shallowOpts.Tags = git.NoTags
+
+				r, err := git.PlainCloneContext(ctx, input.Dir, false, &shallowOpts)
+				if err == nil {
+					return r, nil
+				}
+				logger.Debugf("Shallow clone of %s as %s failed: %v", input.URL, refName, err)
+				if err := removePartialClone(input.Dir); err != nil {
+					return nil, err
+				}
 			}
 		}
 		logger.Debugf("Falling back to a full clone of %s for ref %q", input.URL, input.Ref)
@@ -514,14 +528,65 @@ func cloneAtDepth(ctx context.Context, input NewGitCloneExecutorInput, opts git.
 	return git.PlainCloneContext(ctx, input.Dir, false, &opts)
 }
 
+func removePartialClone(dir string) error {
+	if err := os.RemoveAll(dir); err != nil {
+		return fmt.Errorf("remove partial clone %s: %w", dir, err)
+	}
+	return nil
+}
+
+func fetchPinnedSHA(ctx context.Context, input NewGitCloneExecutorInput, opts git.CloneOptions) (*git.Repository, error) {
+	r, err := git.PlainInit(input.Dir, false)
+	if err != nil {
+		return nil, err
+	}
+	if _, err := r.CreateRemote(&config.RemoteConfig{Name: "origin", URLs: []string{input.URL}}); err != nil {
+		return nil, err
+	}
+	if err := r.FetchContext(ctx, &git.FetchOptions{
+		RefSpecs:        []config.RefSpec{pinnedRefSpec(input.Ref)},
+		Depth:           input.Depth,
+		Tags:            git.NoTags,
+		Auth:            opts.Auth,
+		Progress:        opts.Progress,
+		InsecureSkipTLS: opts.InsecureSkipTLS,
+	}); err != nil {
+		return nil, err
+	}
+	return r, nil
+}
+
+// pinnedRefSpec keeps a hash-shaped name out of refs/heads and refs/tags.
+func pinnedRefSpec(sha string) config.RefSpec {
+	return config.RefSpec(fmt.Sprintf("+%s:refs/pinned/%s", sha, sha))
+}
+
+// refRevision picks the revision to check out and reports whether it resolves locally.
+func refRevision(r *git.Repository, ref string) (plumbing.Revision, string, bool) {
+	if plumbing.IsHash(ref) {
+		// git ignores a ref named as 40 hex digits, so a full hash always denotes the commit itself.
+		_, err := r.CommitObject(plumbing.NewHash(ref))
+		return plumbing.Revision(ref), "sha", err == nil
+	}
+	if _, err := r.Tag(ref); err == nil {
+		return plumbing.Revision(path.Join("refs", "tags", ref)), "tag", true
+	}
+	remoteRef := plumbing.ReferenceName(path.Join("refs", "remotes", "origin", ref))
+	if _, err := r.Reference(remoteRef, false); err == nil {
+		return plumbing.Revision(remoteRef), "branch", true
+	}
+	rev := plumbing.Revision(ref)
+	_, err := r.ResolveRevision(rev)
+	return rev, "sha", err == nil
+}
+
 // isShallow reports whether the local repository was cloned with a limited depth.
 func isShallow(r *git.Repository) bool {
 	shallows, err := r.Storer.Shallow()
 	return err == nil && len(shallows) > 0
 }
 
-// shallowFetchRefSpec returns the single refspec that updates only input.Ref, keeping a shallow clone from re-downloading every branch's history.
-// ok is false when the ref is not present locally as a tag or remote-tracking branch, in which case the broad default refspec is used.
+// shallowFetchRefSpec limits a shallow update to the requested ref, falling back to the broad refspec when it is not present locally.
 func shallowFetchRefSpec(r *git.Repository, ref string) (config.RefSpec, bool) {
 	tagRef := plumbing.NewTagReferenceName(ref)
 	if _, err := r.Reference(tagRef, false); err == nil {
@@ -532,5 +597,9 @@ func shallowFetchRefSpec(r *git.Repository, ref string) (config.RefSpec, bool) {
 		branchRef := plumbing.NewBranchReferenceName(ref)
 		return config.RefSpec(fmt.Sprintf("+%s:%s", branchRef, remoteRef)), true
 	}
+	if plumbing.IsHash(ref) {
+		// The broad refspec carries only advertised tips, never a pinned commit.
+		return pinnedRefSpec(ref), true
+	}
 	return "", false
 }

act/common/git/git_test.go

@@ -6,17 +6,30 @@ package git
 
 import (
 	"context"
+	"errors"
 	"fmt"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"os/exec"
 	"path/filepath"
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"syscall"
 	"testing"
 	"time"
 
+	"gitea.com/gitea/runner/act/common"
+
+	gogit "github.com/go-git/go-git/v5"
+	gogitconfig "github.com/go-git/go-git/v5/config"
+	"github.com/go-git/go-git/v5/plumbing/transport"
+	gogitclient "github.com/go-git/go-git/v5/plumbing/transport/client"
+	gogitfile "github.com/go-git/go-git/v5/plumbing/transport/file"
+	log "github.com/sirupsen/logrus"
+	logrustest "github.com/sirupsen/logrus/hooks/test"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -42,9 +55,7 @@ func TestFindGitSlug(t *testing.T) {
 	}
 
 	for _, tt := range slugTests {
-		provider, slug, err := findGitSlug(tt.url, "github.com")
-
-		assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
+		provider, slug := findGitSlug(tt.url, "github.com")
 		assert.Equal(tt.provider, provider)
 		assert.Equal(tt.slug, slug)
 	}
@@ -78,45 +89,20 @@ func cleanGitHooks(dir string) error {
 	return nil
 }
 
-func TestFindGitRemoteURL(t *testing.T) {
-	assert := assert.New(t)
-
-	basedir := t.TempDir()
-	err := gitCmd("init", basedir)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	err = cleanGitHooks(basedir)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-
-	remoteURL := "https://git-codecommit.us-east-1.amazonaws.com/v1/repos/my-repo-name"
-	err = gitCmd("-C", basedir, "remote", "add", "origin", remoteURL)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-
-	u, err := findGitRemoteURL(context.Background(), basedir, "origin")
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(remoteURL, u)
-
-	remoteURL = "[email protected]/AwesomeOwner/MyAwesomeRepo.git"
-	err = gitCmd("-C", basedir, "remote", "add", "upstream", remoteURL)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	u, err = findGitRemoteURL(context.Background(), basedir, "upstream")
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(remoteURL, u)
-}
-
-func TestFindGithubRepoUsesOriginAndCustomRemote(t *testing.T) {
+func TestFindGithubRepoUsesOrigin(t *testing.T) {
 	basedir := t.TempDir()
+	const remoteURL = "https://github.com/owner/repo.git"
 	require.NoError(t, gitCmd("init", basedir))
 	require.NoError(t, cleanGitHooks(basedir))
-	require.NoError(t, gitCmd("-C", basedir, "remote", "add", "origin", "https://github.com/owner/repo.git"))
-	require.NoError(t, gitCmd("-C", basedir, "remote", "add", "ghe", "[email protected]:team/project.git"))
+	require.NoError(t, gitCmd("-C", basedir, "remote", "add", "origin", remoteURL))
 
-	slug, err := FindGithubRepo(context.Background(), basedir, "github.com", "")
+	url, err := findGitRemoteURL(context.Background(), basedir, "origin")
+	require.NoError(t, err)
+	require.Equal(t, remoteURL, url)
+
+	slug, err := FindGithubRepo(context.Background(), basedir, "github.com")
 	require.NoError(t, err)
 	require.Equal(t, "owner/repo", slug)
-
-	slug, err = FindGithubRepo(context.Background(), basedir, "git.example.com", "ghe")
-	require.NoError(t, err)
-	require.Equal(t, "team/project", slug)
 }
 
 func TestGitFindRef(t *testing.T) {
@@ -404,6 +390,44 @@ func TestGitCloneExecutorOfflineMode(t *testing.T) {
 	})
 }
 
+func TestGitCloneExecutorQuietDemotesCloneLine(t *testing.T) {
+	remoteDir := t.TempDir()
+	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
+	workDir := t.TempDir()
+	require.NoError(t, gitCmd("clone", remoteDir, workDir))
+	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "initial"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+
+	// Quiet callers report the download themselves, so the clone line must not reach the job log.
+	for name, quiet := range map[string]bool{"quiet": true, "not quiet": false} {
+		t.Run(name, func(t *testing.T) {
+			logger, hook := logrustest.NewNullLogger()
+			logger.SetLevel(log.InfoLevel)
+			ctx := common.WithLogger(context.Background(), logger.WithField("job", "j1"))
+
+			require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+				URL:   remoteDir,
+				Ref:   "main",
+				Dir:   t.TempDir(),
+				Quiet: quiet,
+			})(ctx))
+
+			var cloneLines int
+			for _, entry := range hook.AllEntries() {
+				if strings.HasPrefix(entry.Message, "git clone ") {
+					cloneLines++
+				}
+			}
+			if quiet {
+				assert.Zero(t, cloneLines)
+			} else {
+				assert.Equal(t, 1, cloneLines)
+			}
+		})
+	}
+}
+
 func TestGitCloneExecutorShallow(t *testing.T) {
 	// Build a local "remote" with several commits on main plus a tag, so a full clone would pull noticeably more history than a shallow one.
 	remoteDir := t.TempDir()
@@ -415,7 +439,7 @@ func TestGitCloneExecutorShallow(t *testing.T) {
 		require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", m))
 	}
 	require.NoError(t, gitCmd("-C", workDir, "tag", "v1"))
-	sha := gitRevParse(t, workDir, "HEAD~1") // c2, a SHA that go-git cannot shallow-clone
+	unadvertisedSHA := gitRevParse(t, workDir, "HEAD~1")
 	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
 	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "v1"))
 
@@ -441,14 +465,43 @@ func TestGitCloneExecutorShallow(t *testing.T) {
 		assert.Equal(t, "c3", gitHeadSubject(t, dir))
 	})
 
-	t.Run("SHA falls back to a full clone", func(t *testing.T) {
+	unadvertisedRemote := func(t *testing.T, allowUnadvertised bool) string {
+		remote := filepath.Join(t.TempDir(), "remote.git")
+		require.NoError(t, gitCmd("clone", "--bare", remoteDir, remote))
+		require.NoError(t, gitCmd("-C", remote, "config", "uploadpack.allowAnySHA1InWant", strconv.FormatBool(allowUnadvertised)))
+		return remote
+	}
+
+	t.Run("commit hash falls back to a full clone when the remote refuses unadvertised objects", func(t *testing.T) {
 		dir := t.TempDir()
 		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
-			URL: remoteDir, Ref: sha, Dir: dir, Depth: 1,
+			URL: unadvertisedRemote(t, false), Ref: unadvertisedSHA, Dir: dir, Depth: 1,
 		})(t.Context()))
-		// go-git cannot shallow-clone a raw SHA, so it falls back to a full clone; the absence of a shallow marker proves the fallback happened.
-		assert.NoFileExists(t, shallowMarker(dir), "a SHA ref must not produce a shallow clone")
-		assert.Equal(t, sha, gitRevParse(t, dir, "HEAD"))
+		assert.NoFileExists(t, shallowMarker(dir))
+		assert.Equal(t, unadvertisedSHA, gitRevParse(t, dir, "HEAD"))
+	})
+
+	t.Run("commit hash is fetched shallowly, moved to another hash, then reused without reaching the remote", func(t *testing.T) {
+		remote := unadvertisedRemote(t, true)
+		dir := t.TempDir()
+		cloneAt := func(ref string) error {
+			return NewGitCloneExecutor(NewGitCloneExecutorInput{
+				URL: remote, Ref: ref, Dir: dir, Depth: 1,
+			})(t.Context())
+		}
+
+		require.NoError(t, cloneAt(unadvertisedSHA))
+		assert.FileExists(t, shallowMarker(dir))
+		assert.Equal(t, 1, gitRevCount(t, dir))
+		assert.Equal(t, unadvertisedSHA, gitRevParse(t, dir, "HEAD"))
+
+		olderSHA := gitRevParse(t, workDir, "HEAD~2")
+		require.NoError(t, cloneAt(olderSHA))
+		assert.Equal(t, olderSHA, gitRevParse(t, dir, "HEAD"))
+
+		require.NoError(t, os.RemoveAll(remote))
+		require.NoError(t, cloneAt(olderSHA))
+		assert.Equal(t, olderSHA, gitRevParse(t, dir, "HEAD"))
 	})
 
 	t.Run("moving branch updates while staying shallow", func(t *testing.T) {
@@ -471,6 +524,77 @@ func TestGitCloneExecutorShallow(t *testing.T) {
 	})
 }
 
+func TestGitCloneExecutorColdCloneSkipsRefresh(t *testing.T) {
+	remoteDir := t.TempDir()
+	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
+	workDir := t.TempDir()
+	require.NoError(t, gitCmd("clone", remoteDir, workDir))
+	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "c1"))
+	require.NoError(t, gitCmd("-C", workDir, "tag", "v1"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "v1"))
+
+	for name, tt := range map[string]struct {
+		Ref   string
+		Depth int
+	}{
+		"shallow branch": {"main", 1},
+		"full clone tag": {"v1", 0},
+	} {
+		t.Run(name, func(t *testing.T) {
+			counter := installCountingTransport(t)
+			require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+				URL: remoteDir, Ref: tt.Ref, Dir: t.TempDir(), Depth: tt.Depth,
+			})(t.Context()))
+			assert.Equal(t, int64(1), counter.sessions.Load())
+		})
+	}
+}
+
+func TestGitCloneExecutorPinnedHashIgnoresShadowingRef(t *testing.T) {
+	remoteDir := t.TempDir()
+	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
+	workDir := t.TempDir()
+	require.NoError(t, gitCmd("clone", remoteDir, workDir))
+	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "pinned"))
+	pinned := gitRevParse(t, workDir, "HEAD")
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "decoy"))
+	require.NoError(t, gitCmd("-C", workDir, "tag", pinned))
+	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "origin", pinned))
+
+	for name, depth := range map[string]int{"shallow": 1, "full clone": 0} {
+		t.Run(name, func(t *testing.T) {
+			dir := t.TempDir()
+			require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+				URL: remoteDir, Ref: pinned, Dir: dir, Depth: depth,
+			})(t.Context()))
+			assert.Equal(t, pinned, gitRevParse(t, dir, "HEAD"))
+		})
+	}
+}
+
+type countingTransport struct {
+	transport.Transport
+	sessions atomic.Int64
+}
+
+func (c *countingTransport) NewUploadPackSession(ep *transport.Endpoint, auth transport.AuthMethod) (transport.UploadPackSession, error) {
+	c.sessions.Add(1)
+	return c.Transport.NewUploadPackSession(ep, auth)
+}
+
+func installCountingTransport(t *testing.T) *countingTransport {
+	t.Helper()
+	counter := &countingTransport{Transport: gogitfile.DefaultClient}
+	gogitclient.InstallProtocol("file", counter)
+	t.Cleanup(func() { gogitclient.InstallProtocol("file", gogitfile.DefaultClient) })
+
+	return counter
+}
+
 func gitRevParse(t *testing.T, dir, rev string) string {
 	t.Helper()
 	out, err := exec.Command("git", "-C", dir, "rev-parse", rev).Output()
@@ -568,12 +692,56 @@ func TestAcquireCloneLock(t *testing.T) {
 			t.Fatal("acquire on a different directory must not block")
 		}
 	})
+}
 
-	t.Run("same directory reuses the same mutex", func(t *testing.T) {
-		dir := t.TempDir()
+// An unresponsive remote must not pin a job: the refresh has to be interruptible.
+func TestNewGitCloneExecutorFetchHonoursContext(t *testing.T) {
+	block := make(chan struct{})
+	reached := make(chan struct{})
+	var once sync.Once
+	server := httptest.NewServer(http.HandlerFunc(func(_ http.ResponseWriter, _ *http.Request) {
+		once.Do(func() { close(reached) })
+		<-block
+	}))
+	t.Cleanup(func() {
+		close(block)
+		server.Close()
+	})
+
+	dir := filepath.Join(t.TempDir(), "cached-action")
+	repo, err := gogit.PlainInit(dir, false)
+	require.NoError(t, err)
+	_, err = repo.CreateRemote(&gogitconfig.RemoteConfig{Name: "origin", URLs: []string{server.URL}})
+	require.NoError(t, err)
+
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+
+	done := make(chan error, 1)
+	go func() {
+		done <- NewGitCloneExecutor(NewGitCloneExecutorInput{URL: server.URL, Ref: "main", Dir: dir})(ctx)
+	}()
 
-		v1, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
-		v2, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
-		require.Same(t, v1, v2)
-	})
+	select {
+	case <-reached:
+	case <-time.After(10 * time.Second):
+		t.Fatal("the executor never reached the remote")
+	}
+	cancel()
+
+	select {
+	case err := <-done:
+		require.Error(t, err)
+	case <-time.After(10 * time.Second):
+		t.Fatal("fetch ignored context cancellation")
+	}
+}
+
+func TestStaleRefreshErr(t *testing.T) {
+	ctx, cancel := context.WithCancel(t.Context())
+	require.NoError(t, staleRefreshErr(ctx, errors.New("remote hung up")))
+
+	cancel()
+	require.ErrorIs(t, staleRefreshErr(ctx, errors.New("remote hung up")), context.Canceled)
+	require.NoError(t, staleRefreshErr(ctx, gogit.NoErrAlreadyUpToDate))
 }

act/common/line_writer.go

@@ -46,14 +46,14 @@ func (lw *lineWriter) Write(p []byte) (n int, err error) {
 		line, err := pBuf.ReadString('\n')
 		w, _ := lw.buffer.WriteString(line)
 		written += w
-		if err == nil {
-			lw.handleLine(lw.buffer.String())
-			lw.buffer.Reset()
-		} else if err == io.EOF {
-			break
-		} else {
+		if err != nil {
+			if err == io.EOF {
+				break
+			}
 			return written, err
 		}
+		lw.handleLine(lw.buffer.String())
+		lw.buffer.Reset()
 	}
 
 	return written, nil

act/common/outbound_ip.go

@@ -19,7 +19,9 @@ func GetOutboundIP() net.IP {
 	conn, err := net.Dial("udp", "8.8.8.8:80")
 	if err == nil {
 		defer conn.Close()
-		return conn.LocalAddr().(*net.UDPAddr).IP
+		if addr, ok := conn.LocalAddr().(*net.UDPAddr); ok {
+			return addr.IP
+		}
 	}
 
 	// So the machine cannot access the internet. Pick an IP address from network interfaces.

act/container/container_types.go

@@ -6,12 +6,15 @@ package container
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
+	"sync"
 
 	"gitea.com/gitea/runner/act/common"
 
 	"github.com/docker/go-connections/nat"
+	"github.com/moby/moby/api/types/container"
 )
 
 // ExitCodeError reports a non-zero process exit code from a container command.
@@ -23,26 +26,27 @@ func (e ExitCodeError) Error() string {
 
 // NewContainerInput the input for the New function
 type NewContainerInput struct {
-	Image          string
-	Username       string
-	Password       string
-	Entrypoint     []string
-	Cmd            []string
-	WorkingDir     string
-	Env            []string
-	Binds          []string
-	Mounts         map[string]string
-	Name           string
-	Stdout         io.Writer
-	Stderr         io.Writer
-	NetworkMode    string
-	Privileged     bool
-	UsernsMode     string
-	Platform       string
-	Options        string
-	NetworkAliases []string
-	ExposedPorts   nat.PortSet
-	PortBindings   nat.PortMap
+	Image           string
+	Username        string
+	Password        string
+	Entrypoint      []string
+	Cmd             []string
+	WorkingDir      string
+	Env             []string
+	Binds           []string
+	Mounts          map[string]string
+	Name            string
+	Stdout          io.Writer
+	Stderr          io.Writer
+	NetworkMode     string
+	Privileged      bool
+	UsernsMode      string
+	Platform        string
+	RunnerOptions   string // container options the runner was configured with, trusted
+	WorkflowOptions string // container options the workflow asked for, untrusted
+	NetworkAliases  []string
+	ExposedPorts    nat.PortSet
+	PortBindings    nat.PortMap
 
 	// Gitea specific
 	AutoRemove   bool
@@ -57,14 +61,54 @@ type FileEntry struct {
 	Body string
 }
 
+// Container and healthcheck states, as plain strings so a caller of Info needs no docker
+// SDK of its own.
+const (
+	StateRunning = string(container.StateRunning)
+
+	HealthNone      = string(container.NoHealthcheck)
+	HealthStarting  = string(container.Starting)
+	HealthHealthy   = string(container.Healthy)
+	HealthUnhealthy = string(container.Unhealthy)
+)
+
+// ErrContainerNotFound reports a container the daemon no longer knows. Its text is a
+// fragment, missingContainerError composes it into the message every operation shares.
+var ErrContainerNotFound = errors.New("does not exist")
+
+// DockerProxy is a job's docker socket, fronting the daemon's for the job's lifetime.
+type DockerProxy struct {
+	Socket    string
+	close     func(context.Context) error
+	closeOnce sync.Once
+	closeErr  error
+}
+
+func (p *DockerProxy) Close(ctx context.Context) error {
+	p.closeOnce.Do(func() { p.closeErr = p.close(ctx) })
+	return p.closeErr
+}
+
+// Info is a snapshot of a container, as of one inspect.
+type Info struct {
+	ID       string
+	State    string // the docker container state: "created", "running", "exited", ...
+	ExitCode int
+	Health   string // one of the Health* constants
+	// HealthOutput is the last healthcheck probe's output.
+	HealthOutput string
+	Ports        map[string]string // container port ("5432") to the host port it is published on
+	Mounts       map[string]string // container path to its source on the daemon
+}
+
 // Container for managing docker run containers
 type Container interface {
 	Create(capAdd, capDrop []string) common.Executor
-	ConnectToNetwork(name string) common.Executor
 	Copy(destPath string, files ...*FileEntry) common.Executor
-	CopyTarStream(ctx context.Context, destPath string, tarStream io.Reader) error
-	CopyDir(destPath, srcPath string, useGitIgnore bool) common.Executor
+	CopyDir(destPath, srcPath string, useGitIgnore, skipGitDir bool) common.Executor
 	GetContainerArchive(ctx context.Context, srcPath string) (io.ReadCloser, error)
+	Inspect(ctx context.Context) (*Info, error)
+	DumpLogs(ctx context.Context) error
 	Pull(forcePull bool) common.Executor
 	Start(attach bool) common.Executor
 	Exec(command []string, env map[string]string, user, workdir string) common.Executor
@@ -82,12 +126,14 @@ type NewDockerBuildExecutorInput struct {
 	BuildContext io.Reader
 	ImageTag     string
 	Platform     string
+	BuildArgs    map[string]*string
 }
 
 // NewDockerNetworkCreateExecutorInput the input for the NewDockerNetworkCreateExecutor function
 type NewDockerNetworkCreateExecutorInput struct {
 	EnableIPv4 *bool
 	EnableIPv6 *bool
+	RunnerUUID string
 }
 
 // NewDockerPullExecutorInput the input for the NewDockerPullExecutor function

act/container/docker_build.go

@@ -49,6 +49,7 @@ func NewDockerBuildExecutor(input NewDockerBuildExecutorInput) common.Executor {
 			Remove:      true,
 			AuthConfigs: LoadDockerAuthConfigs(ctx),
 			Dockerfile:  input.Dockerfile,
+			BuildArgs:   input.BuildArgs,
 		}
 		platform, err := parsePlatform(input.Platform)
 		if err != nil {

act/container/docker_cli.go

@@ -4,8 +4,10 @@
 
 //go:build !(WITHOUT_DOCKER || !(linux || darwin || windows || netbsd))
 
-// This file is exact copy of https://github.com/docker/cli/blob/9a471180cb7d39c236d090399a9d362c3f5a8ebd/cli/command/container/opts.go
-// appended with license information.
+// This file is exact copy of https://github.com/docker/cli/blob/dfc4efb1e2ab8c06d70d2a1366ad448d2f917e90/cli/command/container/opts.go with:
+// * appended with license information
+// * regexp and loader.ParseVolume in place of the import-restricted internal/lazyregexp and internal/volumespec
+// * invalidParameter from the package's errors.go, and convertPortSet/convertPortMap for the callers in docker_run.go
 //
 // docker/cli is licensed under the Apache License, Version 2.0.
 // See DOCKER_LICENSE for the full license text.
@@ -15,8 +17,7 @@
 package container
 
 import (
-	"bytes"
-	"encoding/json"
+	"encoding/json/jsontext"
 	"errors"
 	"fmt"
 	"net"
@@ -30,6 +31,7 @@ import (
 	"strings"
 	"time"
 
+	cerrdefs "github.com/containerd/errdefs"
 	"github.com/docker/cli/cli/compose/loader"
 	"github.com/docker/cli/opts"
 	"github.com/docker/go-connections/nat"
@@ -348,7 +350,7 @@ type containerConfig struct {
 // parse parses the args for the specified command and generates a Config,
 // a HostConfig and returns them with the specified command.
 // If the specified args are not valid, it will return an error.
-func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*containerConfig, error) { //nolint:gocyclo // verbatim copy from docker/cli
+func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*containerConfig, error) { //nolint:gocyclo,unparam // verbatim copy from docker/cli
 	var (
 		attachStdin  = copts.attach.Get("stdin")
 		attachStdout = copts.attach.Get("stdout")
@@ -380,7 +382,7 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 	var binds []string
 	volumes := copts.volumes.GetMap()
 	// add any bind targets to the list of container volumes
-	for bind := range copts.volumes.GetMap() {
+	for bind := range volumes {
 		parsed, err := loader.ParseVolume(bind)
 		if err != nil {
 			return nil, err
@@ -515,13 +517,13 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 	// collect all the environment variables for the container
 	envVariables, err := opts.ReadKVEnvStrings(copts.envFile.GetSlice(), copts.env.GetSlice())
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("--env-file: %w", err)
 	}
 
 	// collect all the labels for the container
 	labels, err := opts.ReadKVStrings(copts.labelsFile.GetSlice(), copts.labels.GetSlice())
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("--label-file: %w", err)
 	}
 
 	pidMode := container.PidMode(copts.pidMode)
@@ -956,11 +958,11 @@ func parseSecurityOpts(securityOpts []string) ([]string, error) {
 				if err != nil {
 					return securityOpts, fmt.Errorf("opening seccomp profile (%s) failed: %w", v, err)
 				}
-				var b bytes.Buffer
-				if err := json.Compact(&b, f); err != nil {
+				profile := jsontext.Value(f)
+				if err := profile.Compact(); err != nil {
 					return securityOpts, fmt.Errorf("compacting json for seccomp profile (%s) failed: %w", v, err)
 				}
-				securityOpts[key] = "seccomp=" + b.String()
+				securityOpts[key] = "seccomp=" + string(profile)
 			}
 		}
 	}
@@ -1164,16 +1166,17 @@ func toNetipAddrSlice(ips []string) []netip.Addr {
 }
 
 // invalidParameter wraps an error to indicate it was caused by invalid input.
-// This is a local replacement for docker/docker/errdefs.InvalidParameter.
-type invalidParameterError struct{ error }
+// This is a copy of docker/cli's cli/command/container/errors.go, which is not importable.
+type invalidParameterErr struct{ error }
 
-func (e invalidParameterError) InvalidParameter() {}
+func (invalidParameterErr) InvalidParameter() {}
+func (e invalidParameterErr) Unwrap() error   { return e.error }
 
 func invalidParameter(err error) error {
-	if err == nil {
-		return nil
+	if err == nil || cerrdefs.IsInvalidArgument(err) {
+		return err
 	}
-	return invalidParameterError{err}
+	return invalidParameterErr{err}
 }
 
 func convertPortSet(ports nat.PortSet) (network.PortSet, error) {

act/container/docker_create_flags.go

@@ -0,0 +1,113 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !(WITHOUT_DOCKER || !(linux || darwin || windows || netbsd))
+
+package container
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"slices"
+	"strings"
+
+	"github.com/docker/cli/opts"
+	"github.com/kballard/go-shellquote"
+	"github.com/spf13/pflag"
+)
+
+const (
+	pullPolicyAlways  = "always"
+	pullPolicyMissing = "missing"
+	pullPolicyNever   = "never"
+)
+
+var pullPolicies = []string{pullPolicyAlways, pullPolicyMissing, pullPolicyNever}
+
+// createFlags are the flags docker/cli registers on the `create` and `run` commands
+// instead of in addFlags, so they are not part of containerOptions.
+type createFlags struct {
+	platform     string
+	pull         string
+	name         string
+	useAPISocket bool
+}
+
+func registerCreateFlags(flags *pflag.FlagSet) *createFlags {
+	cf := new(createFlags)
+	flags.StringVar(&cf.platform, "platform", "", "Set platform if server is multi-platform capable")
+	flags.StringVar(&cf.pull, "pull", pullPolicyMissing, `Pull image before creating ("always", "missing", "never")`)
+	flags.StringVar(&cf.name, "name", "", "Assign a name to the container")
+	flags.BoolVar(&cf.useAPISocket, "use-api-socket", false, "Bind mount Docker API socket and required auth")
+	// Accepted without effect: pull progress is only logged at debug level, and docker
+	// no longer implements content trust.
+	flags.BoolP("quiet", "q", false, "Suppress the pull output")
+	flags.Bool("disable-content-trust", true, "Skip image verification (deprecated)")
+	return cf
+}
+
+// parseContainerOptions parses a container options string. The flags are returned even
+// on error, holding whatever was read before the failure.
+func parseContainerOptions(options string) (*pflag.FlagSet, *containerOptions, *createFlags, error) {
+	flags := pflag.NewFlagSet("container_flags", pflag.ContinueOnError)
+	flags.SetOutput(io.Discard)
+	copts := addFlags(flags)
+	copts.env = opts.NewListOpts(validateEnv) // addFlags registered this field's address, so the swap takes effect
+	cf := registerCreateFlags(flags)
+
+	args, err := shellquote.Split(options)
+	if err != nil {
+		return flags, copts, cf, fmt.Errorf("cannot split container options: '%s': '%w'", options, err)
+	}
+
+	if err := flags.Parse(args); err != nil {
+		return flags, copts, cf, fmt.Errorf("cannot parse container options: '%s': '%w'", options, err)
+	}
+
+	return flags, copts, cf, nil
+}
+
+// createFlagsFromOptions reads the create-level flags that have to be known before the
+// container is created. Malformed options keep the defaults here and are reported by
+// mergeContainerConfigs at create time.
+func createFlagsFromOptions(options string) *createFlags {
+	_, _, cf, _ := parseContainerOptions(options)
+	return cf
+}
+
+// validateEnv is opts.ValidateEnv without its lookup of a bare name in the runner's environment.
+func validateEnv(val string) (string, error) {
+	if name, _, _ := strings.Cut(val, "="); name == "" {
+		return "", errors.New("invalid environment variable: " + val)
+	}
+	return val, nil
+}
+
+// rejectHostReadingOptions refuses the flags naming files that are read here, on the
+// runner, rather than in the container.
+func rejectHostReadingOptions(options string) error {
+	flags, _, _, err := parseContainerOptions(options)
+	if err != nil {
+		return err
+	}
+
+	for _, name := range []string{"env-file", "label-file"} {
+		if flags.Changed(name) {
+			return fmt.Errorf("container option --%s reads files from the runner and is not allowed in a workflow", name)
+		}
+	}
+	return nil
+}
+
+func (cf *createFlags) validate() error {
+	if !slices.Contains(pullPolicies, cf.pull) {
+		return fmt.Errorf("invalid --pull option %q: must be one of %q", cf.pull, pullPolicies)
+	}
+
+	if cf.useAPISocket {
+		return errors.New("--use-api-socket is not supported, use the runner's container.docker_host setting to expose a docker socket")
+	}
+
+	return nil
+}

act/container/docker_create_flags_test.go

@@ -0,0 +1,62 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package container
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCreateFlagsFromOptions(t *testing.T) {
+	for _, tc := range []struct {
+		options  string
+		platform string
+		pull     string
+	}{
+		{"", "", pullPolicyMissing},
+		{"-v /a:/b --platform=linux/arm64 --pull always", "linux/arm64", pullPolicyAlways},
+		{"--platform linux/arm/v7 --pull never", "linux/arm/v7", pullPolicyNever},
+		{`--platform "linux/amd64`, "", pullPolicyMissing}, // malformed, defaults kept
+	} {
+		t.Run(tc.options, func(t *testing.T) {
+			cf := createFlagsFromOptions(tc.options)
+			assert.Equal(t, tc.platform, cf.platform)
+			assert.Equal(t, tc.pull, cf.pull)
+		})
+	}
+}
+
+func TestCreateFlagsValidate(t *testing.T) {
+	for _, tc := range []struct {
+		options string
+		wantErr string
+	}{
+		{"--quiet --disable-content-trust --name mine", ""},
+		{"--pull sometimes", `invalid --pull option "sometimes"`},
+		{"--use-api-socket", "--use-api-socket is not supported"},
+	} {
+		t.Run(tc.options, func(t *testing.T) {
+			err := createFlagsFromOptions(tc.options).validate()
+			if tc.wantErr == "" {
+				require.NoError(t, err)
+				return
+			}
+			require.ErrorContains(t, err, tc.wantErr)
+		})
+	}
+}
+
+func TestNewContainerAppliesCreateFlags(t *testing.T) {
+	input := &NewContainerInput{Platform: "linux/amd64", RunnerOptions: "--pull never", WorkflowOptions: "--platform linux/arm64"}
+	cr, ok := NewContainer(input).(*containerReference)
+	require.True(t, ok)
+	assert.Equal(t, "linux/arm64", input.Platform)
+	assert.Equal(t, pullPolicyNever, cr.pullPolicy)
+
+	kept := &NewContainerInput{Platform: "linux/amd64", RunnerOptions: "--privileged"}
+	NewContainer(kept)
+	assert.Equal(t, "linux/amd64", kept.Platform)
+}

act/container/docker_logger.go

@@ -8,7 +8,7 @@ package container
 
 import (
 	"bufio"
-	"encoding/json"
+	"encoding/json/v2"
 	"errors"
 	"io"
 
@@ -20,8 +20,8 @@ type dockerMessage struct {
 	Stream      string `json:"stream"`
 	Error       string `json:"error"`
 	ErrorDetail struct {
-		Message string
-	}
+		Message string `json:"message"`
+	} `json:"errorDetail"`
 	Status   string `json:"status"`
 	Progress string `json:"progress"`
 }
@@ -57,18 +57,19 @@ func logDockerResponse(logger logrus.FieldLogger, dockerResponse io.ReadCloser,
 
 		if msg.ErrorDetail.Message != "" {
 			writeLog(logger, isError, "%s", msg.ErrorDetail.Message)
-			return errors.New(msg.Error)
+			return errors.New(msg.ErrorDetail.Message)
 		}
 
-		if msg.Status != "" {
+		switch {
+		case msg.Status != "":
 			if msg.Progress != "" {
 				writeLog(logger, isError, "%s :: %s :: %s\n", msg.Status, msg.ID, msg.Progress)
 			} else {
 				writeLog(logger, isError, "%s :: %s\n", msg.Status, msg.ID)
 			}
-		} else if msg.Stream != "" {
+		case msg.Stream != "":
 			writeLog(logger, isError, "%s", msg.Stream)
-		} else {
+		default:
 			writeLog(logger, false, "Unable to handle line: %s", string(line))
 		}
 	}

act/container/docker_network.go

@@ -8,12 +8,69 @@ package container
 
 import (
 	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
 
 	"gitea.com/gitea/runner/act/common"
 
 	"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 {
 	return func(ctx context.Context) error {
 		cli, err := GetDockerClient(ctx)
@@ -36,18 +93,45 @@ func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExec
 			}
 		}
 
-		_, err = cli.NetworkCreate(ctx, name, client.NetworkCreateOptions{
-			Driver:     "bridge",
-			Scope:      "local",
-			EnableIPv4: opts.EnableIPv4,
-			EnableIPv6: opts.EnableIPv6,
-		})
-		if err != nil {
-			return err
+		for i := range networkCreateAttempts {
+			if i > 0 {
+				common.Logger(ctx).Infof("Waiting for a free docker address pool to create network %s", name)
+				select {
+				case <-ctx.Done():
+					return ctx.Err()
+				case <-time.After(time.Duration(i) * networkCreateRetryDelay):
+				}
+			}
+			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 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 {
@@ -66,6 +150,7 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
 		}
 		// For Gitea, reduce log noise
 		// common.Logger(ctx).Debugf("%v", networks)
+		var errs []error
 		for _, n := range networks.Items {
 			if n.Name == name {
 				result, err := cli.NetworkInspect(ctx, n.ID, client.NetworkInspectOptions{})
@@ -73,16 +158,17 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
 					return err
 				}
 
-				if len(result.Network.Containers) == 0 {
-					if _, err = cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil {
-						common.Logger(ctx).Debugf("%v", err)
-					}
-				} else {
-					common.Logger(ctx).Debugf("Refusing to remove network %v because it still has active endpoints", name)
+				// it holds a subnet out of the daemon's pool until something reclaims it
+				if len(result.Network.Containers) != 0 {
+					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
+				}
+				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...)
 	}
 }

act/container/docker_network_test.go

@@ -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{
+		{ID: "orphan"},
+		{ID: "busy"},
+		{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)
+}

act/container/docker_proxy.go

@@ -0,0 +1,395 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !(WITHOUT_DOCKER || !(linux || darwin || windows || netbsd))
+
+package container
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"maps"
+	"mime"
+	"net"
+	"net/http"
+	"net/http/httputil"
+	"os"
+	"path/filepath"
+	"regexp"
+	"runtime"
+	"strings"
+	"sync"
+	"time"
+
+	"gitea.com/gitea/runner/act/common"
+
+	cerrdefs "github.com/containerd/errdefs"
+	"github.com/moby/moby/api/types/container"
+	"github.com/moby/moby/api/types/mount"
+	"github.com/moby/moby/client"
+)
+
+const (
+	jobLabel      = "com.gitea.runner.job"
+	maxCreateBody = 8 << 20
+
+	dockerProxyProbeTimeout = 5 * time.Second
+)
+
+var (
+	createPath    = regexp.MustCompile(`^(/v[0-9.]+)?/(containers|networks|volumes)/create$`)
+	rawStreamPath = regexp.MustCompile(`^(/v[0-9.]+)?/(containers/[^/]+/attach|exec/[^/]+/start)$`)
+)
+
+func NewDockerProxy(ctx context.Context, job string) *DockerProxy {
+	if host := os.Getenv("DOCKER_HOST"); runtime.GOOS != "linux" || host != "" && !strings.HasPrefix(host, "unix://") {
+		return nil
+	}
+	probeCtx, cancel := context.WithTimeout(ctx, dockerProxyProbeTimeout)
+	defer cancel()
+	cli, err := GetDockerClient(probeCtx)
+	if err != nil {
+		return nil
+	}
+	defer cli.Close()
+	daemonSocket, ok := strings.CutPrefix(cli.DaemonHost(), "unix://")
+	if !ok {
+		return nil
+	}
+	if info, err := os.Stat(daemonSocket); err != nil || info.Mode()&os.ModeSocket == 0 {
+		return nil
+	}
+	dir, err := filepath.Abs(os.TempDir())
+	if err != nil {
+		common.Logger(ctx).Infof("docker proxy probe failed, jobs get the daemon socket directly: %v", err)
+		return nil
+	}
+	seen, err := daemonSeesDir(probeCtx, cli, dir)
+	if err != nil {
+		common.Logger(ctx).Infof("docker proxy probe failed, jobs get the daemon socket directly: %v", err)
+		return nil
+	}
+	if !seen {
+		common.Logger(ctx).Infof("the docker daemon cannot reach the runner's temporary filesystem, jobs get the daemon socket directly")
+		return nil
+	}
+	if ctx.Err() != nil {
+		return nil
+	}
+	proxy, err := StartDockerProxy(daemonSocket, dir, job)
+	if err != nil {
+		common.Logger(ctx).Warnf("docker proxy not started, the job gets the daemon socket directly: %v", err)
+	}
+	return proxy
+}
+
+// daemonSeesDir reports whether the daemon opens the files the runner writes in dir,
+// which is what a job's proxy socket mounted from there needs.
+func daemonSeesDir(ctx context.Context, cli client.APIClient, dir string) (bool, error) {
+	marker, err := os.CreateTemp(dir, "gitea-runner-probe-")
+	if err != nil {
+		return false, err
+	}
+	defer func() {
+		if err := os.Remove(marker.Name()); err != nil {
+			common.Logger(ctx).Warnf("removing the docker proxy probe marker failed: %v", err)
+		}
+	}()
+	if err := marker.Close(); err != nil {
+		return false, err
+	}
+	images, err := cli.ImageList(ctx, client.ImageListOptions{})
+	if err != nil {
+		return false, err
+	}
+	if len(images.Items) == 0 {
+		return false, errors.New("no image available for the docker proxy probe")
+	}
+	created, err := cli.ContainerCreate(ctx, client.ContainerCreateOptions{
+		Config: &container.Config{Image: images.Items[0].ID, Cmd: []string{"true"}},
+		HostConfig: &container.HostConfig{Mounts: []mount.Mount{
+			{Type: mount.TypeBind, Source: marker.Name(), Target: "/gitea-runner-probe", ReadOnly: true},
+		}},
+	})
+	if cerrdefs.IsInvalidArgument(err) {
+		return false, nil
+	}
+	if err != nil {
+		return false, err
+	}
+	cleanupCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), dockerProxyProbeTimeout)
+	defer cancel()
+	if _, err := cli.ContainerRemove(cleanupCtx, created.ID, client.ContainerRemoveOptions{Force: true, RemoveVolumes: true}); err != nil {
+		return false, fmt.Errorf("removing the docker proxy probe container failed: %w", err)
+	}
+	return true, nil
+}
+
+// StartDockerProxy serves a job's docker socket in dir, labelling what the job creates through it.
+func StartDockerProxy(daemonSocket, dir, job string) (*DockerProxy, error) {
+	info, err := os.Stat(daemonSocket)
+	if err != nil {
+		return nil, err
+	}
+	if info.Mode()&os.ModeSocket == 0 {
+		return nil, errors.New("docker daemon path is not a Unix socket")
+	}
+	if err := os.MkdirAll(dir, 0o700); err != nil {
+		return nil, err
+	}
+	instance, err := os.MkdirTemp(dir, "p-")
+	if err != nil {
+		return nil, err
+	}
+	socket := filepath.Join(instance, "docker.sock")
+	listener, err := net.Listen("unix", socket)
+	if err != nil {
+		return nil, errors.Join(err, os.RemoveAll(instance))
+	}
+	if err := copyDockerSocketPermissions(socket, info); err != nil {
+		return nil, errors.Join(err, listener.Close(), os.RemoveAll(instance))
+	}
+	dial := func(ctx context.Context, _, _ string) (net.Conn, error) {
+		return (&net.Dialer{}).DialContext(ctx, "unix", daemonSocket)
+	}
+	transport := &http.Transport{DialContext: dial}
+	forward := &httputil.ReverseProxy{
+		Rewrite: func(r *httputil.ProxyRequest) {
+			r.Out.URL.Scheme = "http"
+			r.Out.URL.Host = "docker"
+		},
+		Transport: transport,
+	}
+	streams, cancelStreams := context.WithCancel(context.Background())
+	creates, cancelCreates := context.WithCancel(context.Background())
+	var admission sync.Mutex
+	var handlers sync.WaitGroup
+	server := &http.Server{ReadHeaderTimeout: 30 * time.Second, ConnContext: func(ctx context.Context, conn net.Conn) context.Context {
+		return context.WithValue(ctx, dockerProxyConnKey{}, conn)
+	}, Handler: http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		admission.Lock()
+		if streams.Err() != nil {
+			admission.Unlock()
+			http.Error(w, "docker proxy is closing", http.StatusServiceUnavailable)
+			return
+		}
+		handlers.Add(1)
+		admission.Unlock()
+		defer handlers.Done()
+		creating := r.Method == http.MethodPost && createPath.MatchString(r.URL.Path)
+		parent, lifetime := r.Context(), streams
+		if creating {
+			parent, lifetime = context.WithoutCancel(parent), creates
+		}
+		ctx, cancel := context.WithCancel(parent)
+		defer cancel()
+		stop := context.AfterFunc(lifetime, func() {
+			cancel()
+			if !creating {
+				if conn, ok := parent.Value(dockerProxyConnKey{}).(net.Conn); ok {
+					_ = conn.Close()
+				}
+			}
+		})
+		defer stop()
+		r = r.WithContext(ctx)
+		if creating {
+			r.Body = http.MaxBytesReader(w, r.Body, maxCreateBody)
+			if err := addLabel(r, job); err != nil {
+				status := http.StatusBadRequest
+				if _, ok := errors.AsType[*http.MaxBytesError](err); ok {
+					status = http.StatusRequestEntityTooLarge
+				}
+				http.Error(w, err.Error(), status)
+				return
+			}
+		} else if r.Method == http.MethodPost && rawStreamPath.MatchString(r.URL.Path) {
+			tunnel(w, r, dial, forward)
+			return
+		}
+		forward.ServeHTTP(w, r)
+	})}
+	served := make(chan struct{})
+	go func() {
+		defer close(served)
+		_ = server.Serve(listener)
+	}()
+	return &DockerProxy{Socket: socket, close: func(ctx context.Context) error {
+		ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
+		defer cancel()
+		admission.Lock()
+		listenerErr := listener.Close()
+		cancelStreams()
+		admission.Unlock()
+		<-served
+		shutdownErr := server.Shutdown(ctx)
+		cancelCreates()
+		serverErr := server.Close()
+		handlers.Wait()
+		transport.CloseIdleConnections()
+		return errors.Join(ctx.Err(), listenerErr, shutdownErr, serverErr, os.RemoveAll(instance))
+	}}, nil
+}
+
+func addLabel(r *http.Request, job string) error {
+	body, err := io.ReadAll(r.Body)
+	if err != nil {
+		return err
+	}
+	if len(bytes.TrimSpace(body)) == 0 {
+		body = []byte("{}")
+	}
+	var fields map[string]json.RawMessage
+	var config struct{ Labels map[string]string }
+	if err := json.Unmarshal(body, &fields); err != nil {
+		return fmt.Errorf("invalid create request: %w", err)
+	}
+	if err := json.Unmarshal(body, &config); err != nil {
+		return fmt.Errorf("invalid create labels: %w", err)
+	}
+	if fields == nil {
+		fields = make(map[string]json.RawMessage)
+	}
+	maps.DeleteFunc(fields, func(name string, _ json.RawMessage) bool {
+		return strings.EqualFold(name, "Labels")
+	})
+	if config.Labels == nil {
+		config.Labels = make(map[string]string)
+	}
+	config.Labels[jobLabel] = job
+	if fields["Labels"], err = json.Marshal(config.Labels); err != nil {
+		return err
+	}
+	if body, err = json.Marshal(fields); err != nil {
+		return err
+	}
+	r.Body = io.NopCloser(bytes.NewReader(body))
+	r.ContentLength = int64(len(body))
+	r.TransferEncoding = nil
+	return nil
+}
+
+type dockerProxyConnKey struct{}
+
+type dockerProxyResponse struct {
+	response *http.Response
+}
+
+func (r dockerProxyResponse) RoundTrip(_ *http.Request) (*http.Response, error) {
+	return r.response, nil
+}
+
+// tunnel splices attach and exec streams, which the daemon hijacks with or without an HTTP upgrade
+func tunnel(w http.ResponseWriter, r *http.Request, dial func(context.Context, string, string) (net.Conn, error), forward *httputil.ReverseProxy) {
+	upstream, err := dial(r.Context(), "", "")
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusBadGateway)
+		return
+	}
+	defer upstream.Close()
+	stop := context.AfterFunc(r.Context(), func() { _ = upstream.Close() })
+	defer stop()
+	if err := r.Write(upstream); err != nil {
+		http.Error(w, err.Error(), http.StatusBadGateway)
+		return
+	}
+	reader := bufio.NewReader(upstream)
+	var response *http.Response
+	for {
+		response, err = http.ReadResponse(reader, r)
+		if err != nil {
+			http.Error(w, err.Error(), http.StatusBadGateway)
+			return
+		}
+		if response.StatusCode >= 200 || response.StatusCode == http.StatusSwitchingProtocols {
+			break
+		}
+		maps.Copy(w.Header(), response.Header)
+		w.WriteHeader(response.StatusCode)
+		clear(w.Header())
+		_ = response.Body.Close()
+	}
+	defer func() {
+		_ = upstream.Close()
+		_ = response.Body.Close()
+	}()
+	mediaType, _, _ := mime.ParseMediaType(response.Header.Get("Content-Type"))
+	if response.StatusCode != http.StatusSwitchingProtocols && (response.StatusCode != http.StatusOK || mediaType != "application/vnd.docker.raw-stream") {
+		ordinary := *forward
+		ordinary.Transport = dockerProxyResponse{response: response}
+		ordinary.ServeHTTP(w, r)
+		return
+	}
+	downstream, buffered, err := http.NewResponseController(w).Hijack()
+	if err != nil {
+		return
+	}
+	defer downstream.Close()
+	if _, err := fmt.Fprintf(buffered, "%s %s\r\n", response.Proto, response.Status); err != nil {
+		return
+	}
+	if err := response.Header.Write(buffered); err != nil {
+		return
+	}
+	if _, err := buffered.WriteString("\r\n"); err != nil {
+		return
+	}
+	if err := buffered.Flush(); err != nil {
+		return
+	}
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		if _, err := io.Copy(upstream, io.MultiReader(io.LimitReader(buffered, int64(buffered.Reader.Buffered())), downstream)); err != nil { // Bypass net/http after the prefix so stdin EOF preserves output.
+			_ = upstream.Close()
+		} else if writer, ok := upstream.(interface{ CloseWrite() error }); ok {
+			_ = writer.CloseWrite()
+		} else {
+			_ = upstream.Close()
+		}
+	}()
+	_, _ = io.Copy(downstream, reader)
+	_ = downstream.Close()
+	_ = upstream.Close()
+	<-done
+}
+
+func RemoveDockerJobResources(ctx context.Context, job string) error {
+	cli, err := GetDockerClient(ctx)
+	if err != nil {
+		return err
+	}
+	defer cli.Close()
+	return removeLabelled(ctx, cli, job)
+}
+
+func removeLabelled(ctx context.Context, cli client.APIClient, job string) error {
+	logger := common.Logger(ctx)
+	filters := make(client.Filters).Add("label", jobLabel+"="+job)
+	containers, err := cli.ContainerList(ctx, client.ContainerListOptions{All: true, Filters: filters})
+	errs := []error{err}
+	for _, c := range containers.Items {
+		logger.Infof("removing container %s the job left behind", strings.TrimPrefix(strings.Join(c.Names, ","), "/"))
+		errs = append(errs, (&containerReference{cli: cli, id: c.ID}).remove()(ctx))
+	}
+	networks, err := cli.NetworkList(ctx, client.NetworkListOptions{Filters: filters})
+	errs = append(errs, err)
+	for _, n := range networks.Items {
+		if _, err := cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil && !cerrdefs.IsNotFound(err) {
+			errs = append(errs, fmt.Errorf("failed to remove network %s: %w", n.Name, err))
+		}
+	}
+	volumes, err := cli.VolumeList(ctx, client.VolumeListOptions{Filters: filters})
+	errs = append(errs, err)
+	for _, v := range volumes.Items {
+		if _, err := cli.VolumeRemove(ctx, v.Name, client.VolumeRemoveOptions{}); err != nil && !cerrdefs.IsNotFound(err) {
+			errs = append(errs, fmt.Errorf("failed to remove volume %s: %w", v.Name, err))
+		}
+	}
+	return errors.Join(errs...)
+}

act/container/docker_proxy_test.go

@@ -0,0 +1,469 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !(WITHOUT_DOCKER || !(linux || darwin || windows || netbsd))
+
+package container
+
+import (
+	"archive/tar"
+	"bufio"
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+
+	cerrdefs "github.com/containerd/errdefs"
+	"github.com/moby/moby/api/types/container"
+	"github.com/moby/moby/api/types/image"
+	"github.com/moby/moby/api/types/mount"
+	"github.com/moby/moby/api/types/volume"
+	mobyclient "github.com/moby/moby/client"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/mock"
+	"github.com/stretchr/testify/require"
+)
+
+func (m *mockDockerClient) VolumeList(ctx context.Context, opts mobyclient.VolumeListOptions) (mobyclient.VolumeListResult, error) {
+	args := m.Called(ctx, opts)
+	return args.Get(0).(mobyclient.VolumeListResult), args.Error(1)
+}
+
+func (m *mockDockerClient) VolumeRemove(ctx context.Context, id string, opts mobyclient.VolumeRemoveOptions) (mobyclient.VolumeRemoveResult, error) {
+	args := m.Called(ctx, id, opts)
+	return args.Get(0).(mobyclient.VolumeRemoveResult), args.Error(1)
+}
+
+// unix socket paths are limited to about a hundred bytes, which the default macOS TMPDIR exceeds
+func shortTempDir(t testing.TB) string {
+	t.Helper()
+	t.Setenv("TMPDIR", "/tmp")
+	return t.TempDir()
+}
+
+func daemonSocketPath(t testing.TB, cli mobyclient.APIClient) string {
+	t.Helper()
+	host := cli.DaemonHost()
+	if !strings.HasPrefix(host, "unix://") {
+		t.Skipf("skipping: daemon at %s is not a unix socket", host)
+	}
+	return strings.TrimPrefix(host, "unix://")
+}
+
+func TestDockerProxy(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("Unix socket ownership is unavailable on Windows")
+	}
+	bodies := make(chan []byte, 8)
+	createStarted, releaseCreate := make(chan struct{}), make(chan struct{})
+	streamsDone := make(chan struct{}, 2)
+	daemonSocket := filepath.Join(shortTempDir(t), "d.sock")
+	listener, err := net.Listen("unix", daemonSocket)
+	require.NoError(t, err)
+	daemon := &http.Server{ReadHeaderTimeout: time.Second, Handler: http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch {
+		case strings.HasSuffix(r.URL.Path, "/create"):
+			body, _ := io.ReadAll(r.Body)
+			if r.URL.RawQuery == "wait" {
+				close(createStarted)
+				<-releaseCreate
+			}
+			bodies <- body
+			w.WriteHeader(http.StatusCreated)
+		case r.URL.Path == "/exec/detached/start" || r.URL.Path == "/exec/error/start":
+			if r.URL.Path == "/exec/error/start" {
+				w.Header().Set("Content-Type", "application/vnd.docker.raw-stream")
+				w.WriteHeader(http.StatusBadRequest)
+			}
+			_, _ = w.Write([]byte("ordinary"))
+		case rawStreamPath.MatchString(r.URL.Path) || r.URL.Path == "/session":
+			_, _ = io.Copy(io.Discard, r.Body)
+			conn, buffered, err := w.(http.Hijacker).Hijack()
+			if err != nil {
+				return
+			}
+			defer conn.Close()
+			if r.Header.Get("Upgrade") != "" {
+				_, _ = conn.Write([]byte("HTTP/1.1 101 Switching Protocols\r\nConnection: Upgrade\r\nUpgrade: tcp\r\n\r\nready\n"))
+			} else {
+				_, _ = conn.Write([]byte("HTTP/1.1 200 OK\r\nContent-Type: application/vnd.docker.raw-stream\r\n\r\nready\n"))
+			}
+			input, _ := io.ReadAll(buffered)
+			_, _ = conn.Write(append([]byte("final:"), input...))
+			streamsDone <- struct{}{}
+		default:
+			w.Header().Set("Api-Version", "1.47")
+			_, _ = w.Write([]byte("OK " + r.Method + " " + r.URL.Path))
+		}
+	})}
+	go func() { _ = daemon.Serve(listener) }()
+	t.Cleanup(func() { _ = daemon.Close() })
+	proxy, err := StartDockerProxy(daemonSocket, shortTempDir(t), "job-1")
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = proxy.Close(context.Background()) })
+	client := &http.Client{Timeout: 5 * time.Second, Transport: &http.Transport{DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) {
+		return (&net.Dialer{}).DialContext(ctx, "unix", proxy.Socket)
+	}}}
+	t.Cleanup(client.CloseIdleConnections)
+	dialProxy := func(t *testing.T) *net.UnixConn {
+		t.Helper()
+		conn, err := net.DialUnix("unix", nil, &net.UnixAddr{Name: proxy.Socket, Net: "unix"})
+		require.NoError(t, err)
+		t.Cleanup(func() { _ = conn.Close() })
+		require.NoError(t, conn.SetDeadline(time.Now().Add(5*time.Second)))
+		return conn
+	}
+	startStream := func(t *testing.T, path, upgrade string) (*net.UnixConn, *bufio.Reader) {
+		t.Helper()
+		conn := dialProxy(t)
+		_, err := fmt.Fprintf(conn, "POST %s HTTP/1.1\r\nHost: docker\r\n%sContent-Length: 2\r\n\r\n{}input", path, upgrade)
+		require.NoError(t, err)
+		reader := bufio.NewReader(conn)
+		resp, err := http.ReadResponse(reader, nil)
+		require.NoError(t, err)
+		t.Cleanup(func() {
+			_ = conn.Close()
+			_ = resp.Body.Close()
+		})
+		prefix, err := reader.ReadString('\n')
+		require.NoError(t, err)
+		require.Equal(t, "ready\n", prefix)
+		return conn, reader
+	}
+
+	t.Run("labels creates", func(t *testing.T) {
+		for _, testCase := range []struct{ path, body, want string }{
+			{
+				"/v1.47/containers/create", `{"Image":"alpine","Unknown":{"enabled":true},"Labels":{"own":"1","com.gitea.runner.job":"other"}}`,
+				`{"Image":"alpine","Unknown":{"enabled":true},"Labels":{"own":"1","com.gitea.runner.job":"job-1"}}`,
+			},
+			{
+				"/networks/create", `{"Name":"n"}`,
+				`{"Name":"n","Labels":{"com.gitea.runner.job":"job-1"}}`,
+			},
+			{
+				"/volumes/create", `{"Name":"v","labels":null}`,
+				`{"Name":"v","Labels":{"com.gitea.runner.job":"job-1"}}`,
+			},
+			{
+				"/volumes/create", "",
+				`{"Labels":{"com.gitea.runner.job":"job-1"}}`,
+			},
+			{
+				"/volumes/create", "null",
+				`{"Labels":{"com.gitea.runner.job":"job-1"}}`,
+			},
+			{
+				"/volumes/create", `{"Labels":{"discard":"1"},"labels":null,"LABELS":{"own":"1"},"LABELS":{"extra":"2"}}`,
+				`{"Labels":{"own":"1","extra":"2","com.gitea.runner.job":"job-1"}}`,
+			},
+		} {
+			resp, err := client.Post("http://docker"+testCase.path, "application/json", strings.NewReader(testCase.body))
+			require.NoError(t, err)
+			resp.Body.Close()
+			require.Equal(t, http.StatusCreated, resp.StatusCode)
+			assert.JSONEq(t, testCase.want, string(<-bodies), testCase.body)
+		}
+	})
+
+	t.Run("passes other requests through", func(t *testing.T) {
+		resp, err := client.Get("http://docker/v1.47/_ping")
+		require.NoError(t, err)
+		defer resp.Body.Close()
+		body, _ := io.ReadAll(resp.Body)
+		assert.Equal(t, "1.47", resp.Header.Get("Api-Version"))
+		assert.Equal(t, "OK GET /v1.47/_ping", string(body))
+	})
+
+	t.Run("tunnels raw streams through stdin EOF", func(t *testing.T) {
+		for _, stream := range []struct{ path, upgrade string }{
+			{path: "/v1.47/exec/abc/start"},
+			{path: "/containers/abc/attach", upgrade: "Connection: Upgrade\r\nUpgrade: tcp\r\n"},
+		} {
+			conn, reader := startStream(t, stream.path, stream.upgrade)
+			require.NoError(t, conn.CloseWrite())
+			output, err := io.ReadAll(reader)
+			require.NoError(t, err)
+			assert.Equal(t, "final:input", string(output))
+			<-streamsDone
+		}
+	})
+
+	t.Run("detached and error responses retain keepalive label injection", func(t *testing.T) {
+		conn := dialProxy(t)
+		reader := bufio.NewReader(conn)
+		for endpoint, status := range map[string]int{"detached": http.StatusOK, "error": http.StatusBadRequest} {
+			_, err := fmt.Fprintf(conn, "POST /exec/%s/start HTTP/1.1\r\nHost: docker\r\nContent-Length: 2\r\n\r\n{}", endpoint)
+			require.NoError(t, err)
+			resp, err := http.ReadResponse(reader, nil)
+			require.NoError(t, err)
+			body, err := io.ReadAll(resp.Body)
+			require.NoError(t, err)
+			require.NoError(t, resp.Body.Close())
+			assert.Equal(t, status, resp.StatusCode)
+			assert.Equal(t, "ordinary", string(body))
+			_, err = io.WriteString(conn, "POST /volumes/create HTTP/1.1\r\nHost: docker\r\nContent-Length: 0\r\n\r\n")
+			require.NoError(t, err)
+			resp, err = http.ReadResponse(reader, nil)
+			require.NoError(t, err)
+			require.NoError(t, resp.Body.Close())
+			require.Equal(t, http.StatusCreated, resp.StatusCode)
+			assert.JSONEq(t, `{"Labels":{"com.gitea.runner.job":"job-1"}}`, string(<-bodies))
+		}
+	})
+
+	t.Run("close joins streams and admitted create", func(t *testing.T) {
+		t.Cleanup(func() { close(releaseCreate) })
+		_, raw := startStream(t, "/exec/live/start", "")
+		_, upgraded := startStream(t, "/session", "Connection: Upgrade\r\nUpgrade: tcp\r\n")
+		creator := dialProxy(t)
+		_, err := io.WriteString(creator, "POST /volumes/create?wait HTTP/1.1\r\nHost: docker\r\nContent-Length: 2\r\n\r\n{}")
+		require.NoError(t, err)
+		<-createStarted
+		require.NoError(t, creator.CloseWrite())
+		closed := make(chan error, 1)
+		ctx, cancel := context.WithTimeout(t.Context(), 5*time.Second)
+		defer cancel()
+		go func() { closed <- proxy.Close(ctx) }()
+		for _, reader := range []*bufio.Reader{raw, upgraded} {
+			_, err := reader.ReadByte()
+			require.ErrorIs(t, err, io.EOF)
+			<-streamsDone
+		}
+		select {
+		case err := <-closed:
+			t.Fatalf("Close returned before admitted create settled: %v", err)
+		default:
+		}
+		releaseCreate <- struct{}{}
+		resp, err := http.ReadResponse(bufio.NewReader(creator), nil)
+		require.NoError(t, err)
+		require.NoError(t, resp.Body.Close())
+		require.Equal(t, http.StatusCreated, resp.StatusCode)
+		assert.JSONEq(t, `{"Labels":{"com.gitea.runner.job":"job-1"}}`, string(<-bodies))
+		require.NoError(t, <-closed)
+	})
+}
+
+func TestRemoveLabelledRemovesContainersNetworksAndVolumes(t *testing.T) {
+	containerFailure := errors.New("container removal failed")
+	listFailure := errors.New("listing failed")
+	volumeFailure := errors.New("volume removal failed")
+	ctx := context.Background()
+	filters := make(mobyclient.Filters).Add("label", jobLabel+"=job-1")
+	cli := &mockDockerClient{}
+	cli.On("ContainerList", ctx, mobyclient.ContainerListOptions{All: true, Filters: filters}).
+		Return(mobyclient.ContainerListResult{Items: []container.Summary{{ID: "c1", Names: []string{"/app"}}}}, nil).Once()
+	cli.On("ContainerKill", ctx, "c1", mock.Anything).Return(mobyclient.ContainerKillResult{}, nil).Once()
+	cli.On("ContainerRemove", ctx, "c1", mobyclient.ContainerRemoveOptions{RemoveVolumes: true, Force: true}).
+		Return(mobyclient.ContainerRemoveResult{}, containerFailure).Once()
+	cli.On("NetworkList", ctx, mobyclient.NetworkListOptions{Filters: filters}).Return(mobyclient.NetworkListResult{}, listFailure).Once()
+	cli.On("VolumeList", ctx, mobyclient.VolumeListOptions{Filters: filters}).
+		Return(mobyclient.VolumeListResult{Items: []volume.Volume{{Name: "app_data"}}}, nil).Once()
+	cli.On("VolumeRemove", ctx, "app_data", mobyclient.VolumeRemoveOptions{}).Return(mobyclient.VolumeRemoveResult{}, volumeFailure).Once()
+
+	err := removeLabelled(ctx, cli, "job-1")
+	require.ErrorIs(t, err, containerFailure)
+	require.ErrorIs(t, err, listFailure)
+	require.ErrorIs(t, err, volumeFailure)
+	require.ErrorContains(t, err, "failed to remove container c1")
+	require.ErrorContains(t, err, "failed to remove volume app_data")
+	cli.AssertExpectations(t)
+}
+
+func TestDockerProxyWithDaemon(t *testing.T) {
+	requireDocker(t)
+	ctx := context.Background()
+	require.NoError(t, NewDockerPullExecutor(NewDockerPullExecutorInput{Image: "alpine"})(ctx))
+	direct, err := GetDockerClient(ctx)
+	require.NoError(t, err)
+	defer direct.Close()
+	dir := shortTempDir(t)
+	seen, err := daemonSeesDir(ctx, direct, dir)
+	require.NoError(t, err)
+	t.Logf("daemon sees the runner's filesystem: %v", seen)
+
+	job := "proxy-test-" + strconv.FormatInt(time.Now().UnixNano(), 36)
+	proxy, err := StartDockerProxy(daemonSocketPath(t, direct), dir, job)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = proxy.Close(context.Background()) })
+	viaProxy, err := mobyclient.New(mobyclient.WithHost("unix://" + proxy.Socket))
+	require.NoError(t, err)
+	defer viaProxy.Close()
+
+	created, err := viaProxy.ContainerCreate(ctx, mobyclient.ContainerCreateOptions{
+		Config: &container.Config{Image: "alpine", Cmd: []string{"sleep", "300"}, Labels: map[string]string{"own": "1"}},
+	})
+	require.NoError(t, err)
+	t.Cleanup(func() { _, _ = direct.ContainerRemove(ctx, created.ID, mobyclient.ContainerRemoveOptions{Force: true}) })
+	_, err = viaProxy.ContainerStart(ctx, created.ID, mobyclient.ContainerStartOptions{})
+	require.NoError(t, err)
+	net, err := viaProxy.NetworkCreate(ctx, job, mobyclient.NetworkCreateOptions{})
+	require.NoError(t, err)
+	t.Cleanup(func() { _, _ = direct.NetworkRemove(ctx, net.ID, mobyclient.NetworkRemoveOptions{}) })
+	_, err = viaProxy.VolumeCreate(ctx, mobyclient.VolumeCreateOptions{Name: job})
+	require.NoError(t, err)
+	t.Cleanup(func() { _, _ = direct.VolumeRemove(ctx, job, mobyclient.VolumeRemoveOptions{Force: true}) })
+
+	inspected, err := direct.ContainerInspect(ctx, created.ID, mobyclient.ContainerInspectOptions{})
+	require.NoError(t, err)
+	assert.Equal(t, map[string]string{"own": "1", jobLabel: job}, inspected.Container.Config.Labels)
+	vol, err := direct.VolumeInspect(ctx, job, mobyclient.VolumeInspectOptions{})
+	require.NoError(t, err)
+	assert.Equal(t, job, vol.Volume.Labels[jobLabel])
+
+	exec, err := viaProxy.ExecCreate(ctx, created.ID, mobyclient.ExecCreateOptions{Cmd: []string{"cat"}, AttachStdin: true, AttachStdout: true, TTY: true})
+	require.NoError(t, err)
+	attached, err := viaProxy.ExecAttach(ctx, exec.ID, mobyclient.ExecAttachOptions{TTY: true})
+	require.NoError(t, err)
+	_, err = attached.Conn.Write([]byte("hello\n"))
+	require.NoError(t, err)
+	echoed, err := attached.Reader.ReadString('\n')
+	require.NoError(t, err)
+	assert.Equal(t, "hello\r\n", echoed)
+	attached.Close()
+
+	require.NoError(t, proxy.Close(ctx))
+	require.NoError(t, RemoveDockerJobResources(ctx, job))
+	_, err = direct.ContainerInspect(ctx, created.ID, mobyclient.ContainerInspectOptions{})
+	assert.True(t, cerrdefs.IsNotFound(err))
+	_, err = direct.NetworkInspect(ctx, net.ID, mobyclient.NetworkInspectOptions{})
+	assert.True(t, cerrdefs.IsNotFound(err))
+	_, err = direct.VolumeInspect(ctx, job, mobyclient.VolumeInspectOptions{})
+	assert.True(t, cerrdefs.IsNotFound(err))
+}
+
+func BenchmarkDockerProxy(b *testing.B) {
+	ctx := context.Background()
+	direct, err := GetDockerClient(ctx)
+	require.NoError(b, err)
+	defer direct.Close()
+	if _, err := direct.Ping(ctx, mobyclient.PingOptions{}); err != nil {
+		b.Skipf("docker daemon unreachable: %v", err)
+	}
+	proxy, err := StartDockerProxy(daemonSocketPath(b, direct), shortTempDir(b), "bench")
+	require.NoError(b, err)
+	defer func() { _ = proxy.Close(ctx) }()
+	viaProxy, err := mobyclient.New(mobyclient.WithHost("unix://" + proxy.Socket))
+	require.NoError(b, err)
+	defer viaProxy.Close()
+
+	require.NoError(b, NewDockerPullExecutor(NewDockerPullExecutorInput{Image: "alpine"})(ctx))
+	created, err := direct.ContainerCreate(ctx, mobyclient.ContainerCreateOptions{Config: &container.Config{Image: "alpine", Cmd: []string{"sleep", "600"}}})
+	require.NoError(b, err)
+	defer func() { _, _ = direct.ContainerRemove(ctx, created.ID, mobyclient.ContainerRemoveOptions{Force: true}) }()
+	_, err = direct.ContainerStart(ctx, created.ID, mobyclient.ContainerStartOptions{})
+	require.NoError(b, err)
+
+	var archive bytes.Buffer
+	writer := tar.NewWriter(&archive)
+	payload := make([]byte, 64<<20)
+	require.NoError(b, writer.WriteHeader(&tar.Header{Name: "blob", Mode: 0o600, Size: int64(len(payload))}))
+	_, _ = writer.Write(payload)
+	require.NoError(b, writer.Close())
+
+	for name, cli := range map[string]mobyclient.APIClient{"direct": direct, "proxy": viaProxy} {
+		b.Run("ping/"+name, func(b *testing.B) {
+			for b.Loop() {
+				if _, err := cli.Ping(ctx, mobyclient.PingOptions{}); err != nil {
+					b.Fatal(err)
+				}
+			}
+		})
+		b.Run("copy64MiB/"+name, func(b *testing.B) {
+			b.SetBytes(int64(archive.Len()))
+			for b.Loop() {
+				_, err := cli.CopyToContainer(ctx, created.ID, mobyclient.CopyToContainerOptions{DestinationPath: "/tmp", Content: bytes.NewReader(archive.Bytes())})
+				if err != nil {
+					b.Fatal(err)
+				}
+			}
+		})
+	}
+}
+
+type probeClient struct {
+	mobyclient.APIClient
+	create func(mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error)
+	remove func(context.Context, string, mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error)
+}
+
+func (c *probeClient) ImageList(context.Context, mobyclient.ImageListOptions) (mobyclient.ImageListResult, error) {
+	return mobyclient.ImageListResult{Items: []image.Summary{{ID: "probe-image"}}}, nil
+}
+
+func (c *probeClient) ContainerCreate(_ context.Context, opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
+	return c.create(opts)
+}
+
+func (c *probeClient) ContainerRemove(ctx context.Context, id string, opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
+	return c.remove(ctx, id, opts)
+}
+
+func TestDaemonSeesDir(t *testing.T) {
+	dir := t.TempDir()
+	markers := make(map[string]bool)
+	for _, testCase := range []struct {
+		name      string
+		private   bool
+		removeErr error
+	}{
+		{name: "cleanup after cancellation"},
+		{name: "private filesystem with stale marker", private: true},
+		{name: "cleanup error after cancellation", removeErr: errors.New("cleanup failed")},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			ctx, cancel := context.WithCancel(t.Context())
+			defer cancel()
+			daemonDir := dir
+			if testCase.private {
+				daemonDir = t.TempDir()
+				require.NoError(t, os.WriteFile(filepath.Join(daemonDir, "probe"), []byte("stale"), 0o600))
+			}
+			removed := false
+			cli := &probeClient{
+				create: func(opts mobyclient.ContainerCreateOptions) (mobyclient.ContainerCreateResult, error) {
+					require.Len(t, opts.HostConfig.Mounts, 1)
+					marker := opts.HostConfig.Mounts[0].Source
+					assert.Equal(t, mount.Mount{Type: mount.TypeBind, Source: marker, Target: "/gitea-runner-probe", ReadOnly: true}, opts.HostConfig.Mounts[0])
+					assert.False(t, markers[marker])
+					markers[marker] = true
+					info, err := os.Stat(marker)
+					require.NoError(t, err)
+					assert.True(t, info.Mode().IsRegular())
+					cancel()
+					if _, err := os.Stat(filepath.Join(daemonDir, filepath.Base(marker))); errors.Is(err, os.ErrNotExist) {
+						return mobyclient.ContainerCreateResult{}, cerrdefs.ErrInvalidArgument
+					}
+					return mobyclient.ContainerCreateResult{ID: "probe"}, nil
+				},
+				remove: func(ctx context.Context, id string, opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
+					removed = true
+					assert.Equal(t, "probe", id)
+					assert.Equal(t, mobyclient.ContainerRemoveOptions{Force: true, RemoveVolumes: true}, opts)
+					require.NoError(t, ctx.Err())
+					_, bounded := ctx.Deadline()
+					assert.True(t, bounded)
+					return mobyclient.ContainerRemoveResult{}, testCase.removeErr
+				},
+			}
+			seen, err := daemonSeesDir(ctx, cli, dir)
+			require.ErrorIs(t, err, testCase.removeErr)
+			assert.Equal(t, !testCase.private && testCase.removeErr == nil, seen)
+			assert.Equal(t, !testCase.private, removed)
+			entries, err := os.ReadDir(dir)
+			require.NoError(t, err)
+			assert.Empty(t, entries)
+		})
+	}
+}

act/container/docker_proxy_unix.go

@@ -0,0 +1,27 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !WITHOUT_DOCKER && (linux || darwin || netbsd)
+
+package container
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"syscall"
+)
+
+func copyDockerSocketPermissions(socket string, info os.FileInfo) error {
+	stat, ok := info.Sys().(*syscall.Stat_t)
+	if !ok {
+		return errors.New("docker socket ownership is unavailable")
+	}
+	if err := os.Chown(socket, int(stat.Uid), int(stat.Gid)); err != nil {
+		return err
+	}
+	if err := os.Chown(filepath.Dir(socket), int(stat.Uid), -1); err != nil {
+		return err
+	}
+	return os.Chmod(socket, info.Mode().Perm())
+}

act/container/docker_proxy_windows.go

@@ -0,0 +1,15 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !WITHOUT_DOCKER
+
+package container
+
+import (
+	"errors"
+	"os"
+)
+
+func copyDockerSocketPermissions(_ string, _ os.FileInfo) error {
+	return errors.New("docker socket ownership cannot be preserved on Windows")
+}

act/container/docker_pull.go

@@ -30,23 +30,19 @@ func NewDockerPullExecutor(input NewDockerPullExecutorInput) common.Executor {
 			return nil
 		}
 
-		pull := input.ForcePull
-		if !pull {
+		// skip the pull when the image is already here: either none was forced, or a digest
+		// 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)
 			logger.Debugf("Image exists? %v", imageExists)
 			if err != nil {
 				return fmt.Errorf("unable to determine if image already exists for image '%s' (%s): %w", input.Image, input.Platform, err)
 			}
-
-			if !imageExists {
-				pull = true
+			if imageExists {
+				return nil
 			}
 		}
 
-		if !pull {
-			return nil
-		}
-
 		imageRef := cleanImage(ctx, input.Image)
 		logger.Debugf("pulling image '%v' (%s)", imageRef, input.Platform)
 
@@ -61,22 +57,32 @@ func NewDockerPullExecutor(input NewDockerPullExecutorInput) common.Executor {
 			return err
 		}
 
-		reader, err := cli.ImagePull(ctx, imageRef, imagePullOptions)
-
-		_ = logDockerResponse(logger, reader, err != nil)
-		if err != nil {
-			if 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 = ""
-				reader, err = cli.ImagePull(ctx, imageRef, imagePullOptions)
-
-				_ = logDockerResponse(logger, reader, err != nil)
-			}
+		// 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
+		pullOnce := func(opts client.ImagePullOptions) error {
+			reader, err := cli.ImagePull(ctx, imageRef, opts)
+			streamErr := logDockerResponse(logger, reader, 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
 }
 
+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 {
 	ref, err := reference.ParseAnyReference(imageName)
 	if err != nil {

act/container/docker_pull_test.go

@@ -6,11 +6,15 @@ package container
 
 import (
 	"context"
+	"io"
+	"strings"
 	"testing"
 
 	"github.com/docker/cli/cli/config"
 	log "github.com/sirupsen/logrus"
+	"github.com/sirupsen/logrus/hooks/test"
 	assert "github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 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.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"}`))
+}

act/container/docker_socket.go

@@ -1,138 +0,0 @@
-// Copyright 2024 The Gitea Authors. All rights reserved.
-// Copyright 2024 The nektos/act Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package container
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"strings"
-
-	log "github.com/sirupsen/logrus"
-)
-
-var CommonSocketLocations = []string{
-	"/var/run/docker.sock",
-	"/run/podman/podman.sock",
-	"$HOME/.colima/docker.sock",
-	"$XDG_RUNTIME_DIR/docker.sock",
-	"$XDG_RUNTIME_DIR/podman/podman.sock",
-	`\\.\pipe\docker_engine`,
-	"$HOME/.docker/run/docker.sock",
-}
-
-// returns socket URI or false if not found any
-func socketLocation() (string, bool) {
-	if dockerHost, exists := os.LookupEnv("DOCKER_HOST"); exists {
-		return dockerHost, true
-	}
-
-	for _, p := range CommonSocketLocations {
-		if _, err := os.Lstat(os.ExpandEnv(p)); err == nil {
-			if strings.HasPrefix(p, `\\.\`) {
-				return "npipe://" + filepath.ToSlash(os.ExpandEnv(p)), true
-			}
-			return "unix://" + filepath.ToSlash(os.ExpandEnv(p)), true
-		}
-	}
-
-	return "", false
-}
-
-// This function, `isDockerHostURI`, takes a string argument `daemonPath`. It checks if the
-// `daemonPath` is a valid Docker host URI. It does this by checking if the scheme of the URI (the
-// part before "://") contains only alphabetic characters. If it does, the function returns true,
-// indicating that the `daemonPath` is a Docker host URI. If it doesn't, or if the "://" delimiter
-// is not found in the `daemonPath`, the function returns false.
-func isDockerHostURI(daemonPath string) bool {
-	if before, _, ok := strings.Cut(daemonPath, "://"); ok {
-		scheme := before
-		if strings.IndexFunc(scheme, func(r rune) bool {
-			return (r < 'a' || r > 'z') && (r < 'A' || r > 'Z')
-		}) == -1 {
-			return true
-		}
-	}
-	return false
-}
-
-type SocketAndHost struct {
-	Socket string
-	Host   string
-}
-
-func GetSocketAndHost(containerSocket string) (SocketAndHost, error) {
-	log.Debugf("Handling container host and socket")
-
-	// Prefer DOCKER_HOST, don't override it
-	dockerHost, hasDockerHost := socketLocation()
-	socketHost := SocketAndHost{Socket: containerSocket, Host: dockerHost}
-
-	// ** socketHost.Socket cases **
-	// Case 1: User does _not_ want to mount a daemon socket (passes a dash)
-	// Case 2: User passes a filepath to the socket; is that even valid?
-	// Case 3: User passes a valid socket; do nothing
-	// Case 4: User omitted the flag; set a sane default
-
-	// ** DOCKER_HOST cases **
-	// Case A: DOCKER_HOST is set; use it, i.e. do nothing
-	// Case B: DOCKER_HOST is empty; use sane defaults
-
-	// Set host for sanity's sake, when the socket isn't useful
-	if !hasDockerHost && (socketHost.Socket == "-" || !isDockerHostURI(socketHost.Socket) || socketHost.Socket == "") {
-		// Cases: 1B, 2B, 4B
-		socket, found := socketLocation()
-		socketHost.Host = socket
-		hasDockerHost = found
-	}
-
-	// A - (dash) in socketHost.Socket means don't mount, preserve this value
-	// otherwise if socketHost.Socket is a filepath don't use it as socket
-	// Exit early if we're in an invalid state (e.g. when no DOCKER_HOST and user supplied "-", a dash or omitted)
-	if !hasDockerHost && socketHost.Socket != "" && !isDockerHostURI(socketHost.Socket) {
-		// Cases: 1B, 2B
-		// Should we early-exit here, since there is no host nor socket to talk to?
-		return SocketAndHost{}, fmt.Errorf("DOCKER_HOST was not set, couldn't be found in the usual locations, and the container daemon socket ('%s') is invalid", socketHost.Socket)
-	}
-
-	// Default to DOCKER_HOST if set
-	if socketHost.Socket == "" && hasDockerHost {
-		// Cases: 4A
-		log.Debugf("Defaulting container socket to DOCKER_HOST")
-		socketHost.Socket = socketHost.Host
-	}
-	// Set sane default socket location if user omitted it
-	if socketHost.Socket == "" {
-		// Cases: 4B
-		socket, _ := socketLocation()
-		// socket is empty if it isn't found, so assignment here is at worst a no-op
-		log.Debugf("Defaulting container socket to default '%s'", socket)
-		socketHost.Socket = socket
-	}
-
-	// Exit if both the DOCKER_HOST and socket are fulfilled
-	if hasDockerHost {
-		// Cases: 1A, 2A, 3A, 4A
-		if !isDockerHostURI(socketHost.Socket) {
-			// Cases: 1A, 2A
-			log.Debugf("DOCKER_HOST is set, but socket is invalid '%s'", socketHost.Socket)
-		}
-		return socketHost, nil
-	}
-
-	// Set a sane DOCKER_HOST default if we can
-	if isDockerHostURI(socketHost.Socket) {
-		// Cases: 3B
-		log.Debugf("Setting DOCKER_HOST to container socket '%s'", socketHost.Socket)
-		socketHost.Host = socketHost.Socket
-		// Both DOCKER_HOST and container socket are valid; short-circuit exit
-		return socketHost, nil
-	}
-
-	// Here there is no DOCKER_HOST _and_ the supplied container socket is not a valid URI (either invalid or a file path)
-	// Cases: 2B <- but is already handled at the top
-	// I.e. this path should never be taken
-	return SocketAndHost{}, fmt.Errorf("no DOCKER_HOST and an invalid container socket '%s'", socketHost.Socket)
-}