NOTE: These docs are in-progress. Feel free to direct any questions / feedback / improvements to your private channel on the Semgrep slack!
The Semgrep Network Broker facilitates secure access between Semgrep and a private network.
The broker accomplishes this by establishing a Wireguard VPN tunnel with the Semgrep backend, and then proxying inbound (Semgrep --> customer) HTTP requests through this tunnel. This approach allows Semgrep to interact with on-prem resources without having to expose them to the public internet.
Examples of inbound traffic include:
- Pull Request comments
- JIRA integrations
- Webhooks
- Run
make build
to generate thesemgrep-network-broker
binary - Run
make docker
to generate a docker image - Docker images are also published to ghcr.io/semgrep/semgrep-network-broker
The broker requires a Wireguard keypair in order to establish a secure connection.
semgrep-network-broker genkey
generates a random private key in base64 and prints it to stdoutsemgrep-network-broker pubkey
reads a base64 private key from stdin and prints the corresponding base64 public key to stdout
> semgrep-network-broker genkey
some_private_key
> echo "some_private_key" | semgrep-network-broker pubkey
some_public_key
Your public key is safe to share. Do not share your private key with anyone (including Semgrep).
Semgrep will help you create a configuration file tailored to your Semgrep deployment.
Do not alter the wireguard
and heartbeat
sections.
Do not share the value of inbound.wireguard.privateKey
. This is your organization's private key. Reach out to Semgrep on Slack if you need to rotate your Wireguard keys.
Example:
inbound:
wireguard:
localAddress: ...
privateKey: ...
peers:
- publicKey: ...
endpoint: ...
allowedIps: ...
heartbeat:
url: ...
allowlist: [...]
The httpClient
configuration section modifies the HTTP client used for proxying requests.
Example:
inbound:
httpClient:
additionalCACerts:
- /path/to/custom/cert.pem
The github
configuration section simplifies granting Semgrep access to leave PR comments.
Example:
inbound:
github:
baseUrl: https://github.example.com/api/v3
token: ...
allowCodeAccess: false # default is false, set to true to allow Semgrep to read file contents
Under the hood, this config adds these allowlist items:
- GET
https://github.example.com/api/v3/repos/:owner/:repo
- GET
https://github.example.com/api/v3/repos/:owner/:repo/pulls
- GET
https://github.example.com/api/v3/orgs/:org/installation
- GET
https://github.example.com/api/v3/orgs/:org/repos
- GET
https://github.example.com/api/v3/users/:user/installation
- GET
https://github.example.com/api/v3/users/:user/installation/repositories
- GET
https://github.example.com/api/v3/app
- POST
https://github.example.com/api/v3/app/installations/:id/access_tokens
- POST
https://github.example.com/api/v3/app-manifests/:code/conversions
- POST
https://github.example.com/api/v3/repos/:owner/:repo/pulls/:number/comments
- POST
https://github.example.com/api/v3/repos/:owner/:repo/issues/:number/comments
And if allowCodeAccess
is set, additionally:
- GET
https://github.example.com/api/v3/repos/:repo/contents/:filepath
- GET
https://github.example.com/api/v3/repos/:repo/commits
Similarly, the gitlab
configuration section grants Semgrep access to leave MR comments.
Example:
inbound:
gitlab:
baseUrl: https://gitlab.example.com/api/v4
token: ...
allowCodeAccess: false # default is false, set to true to allow Semgrep to read file contents
Under the hood, this config adds these allowlist items:
- DELETE
https://gitlab.example.com/api/v4/groups/:namespace/hooks/:hook
- DELETE
https://gitlab.example.com/api/v4/projects/:project/hooks/:hook
- GET
https://gitlab.example.com/api/v4/groups/:namespace/hooks
- GET
https://gitlab.example.com/api/v4/namespaces/:namespace
- GET
https://gitlab.example.com/api/v4/projects/:project
- GET
https://gitlab.example.com/api/v4/projects/:project/members/all/:user
- GET
https://gitlab.example.com/api/v4/projects/:project/merge_requests
- GET
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/versions
- GET
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/discussions
- GET
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/discussions/:discussion/notes/:note/award_emoji
- GET
https://gitlab.example.com/api/v4/projects/:project/repository/branches
- GET
https://gitlab.example.com/api/v4/:entity_type/:namespace/projects
- POST
https://gitlab.example.com/api/v4/groups/:namespace/hooks
- POST
https://gitlab.example.com/api/v4/projects/:project/hooks
- POST
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/discussions
- POST
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/discussions/:discussion/notes
- PUT
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/discussions/:discussion/notes/:note
- PUT
https://gitlab.example.com/api/v4/projects/:project/merge_requests/:number/discussions/:discussion
And if allowCodeAccess
is set, additionally:
- GET
https://gitlab.example.com/api/v4/projects/:project/repository/files/:filepath
- GET
https://gitlab.example.com/api/v4/projects/:project/repository/commits
- GET
https://gitlab.example.com/api/v4/projects/:project/repository/compare
- POST
https://gitlab.example.com/api/v4/projects/:project/statuses/:commit
Similarly, the bitbucket
configuration section grants Semgrep access to leave MR comments.
inbound:
bitbucket:
baseUrl: https://bitbucket.example.com/rest/api/latest
token: ...
Under the hood, this config adds these allowlist items:
- GET
https://bitbucket.example.com/rest/api/latest/projects/:project
- GET
https://bitbucket.example.com/rest/api/latest/projects/:project/repos
- GET
https://bitbucket.example.com/rest/api/latest/projects/:project/repo/:repo
- GET
https://bitbucket.example.com/rest/api/latest/projects/:project/repos/:repo/default-branch
- GET
https://bitbucket.example.com/rest/api/latest/projects/:project/:repo/pull-requests
- POST
https://bitbucket.example.com/rest/api/latest/projects/:project/repos/:repo/pull-requests/:number/comments
- POST
https://bitbucket.example.com/rest/api/latest/projects/:project/repos/:repo/pull-requests/:number/blocker-comments
The allowlist
configuration section provides finer-grained control over what HTTP requests are allowed to be forwarded out of the broker. The first matching allowlist item is used. No allowlist match means the request will not be proxied.
Examples:
inbound:
allowlist:
# allow GET requests from http://example.com/foo (exact URL match)
- url: http://example.com/foo
methods: [GET]
# allow GET or POST requests from any path on http://example.com
- url: http://example.com/*
methods: [GET, POST]
# allow GET requests from a URL that looks like a GitHub Enterprise review comments URL, and add a bearer token to the request
- url: http://example.com/api/v3/repos/:owner/:repo/pulls/:number/comments
methods: [GET]
setRequestHeaders:
Authorization: "Bearer ...snip..."
Here's an example of allowing PR comments for a GitHub Enterprise instance hosted on https://git.example.com. Replace <GH TOKEN>
with a GitHub PAT.
allowlist:
- url: https://git.example.com/api/v3/repos/:owner/:repo
methods: [GET]
setRequestHeaders:
Authorization: "Bearer <GH TOKEN>"
- url: https://git.example.com/api/v3/repos/:owner/:repo/pulls
methods: [GET]
setRequestHeaders:
Authorization: "Bearer <GH TOKEN>"
- url: https://git.example.com/api/v3/repos/:owner/:repo/pulls/:number/comments
methods: [POST]
setRequestHeaders:
Authorization: "Bearer <GH TOKEN>"
- url: https://git.example.com/api/v3/repos/:owner/:repo/issues/:number/comments
methods: [POST]
setRequestHeaders:
Authorization: "Bearer <GH TOKEN>"
The logging
configuration section allows you to set additional logging options for requests that are proxied through the broker.
inbound:
logging:
logRequestBody: false # If true, the contents of any proxied HTTP request matching the allowlist will be logged in the request_body field in the proxy.request event
logResponseBody: false # If true, the contents of any proxied HTTP response will be logged in the response_body field in the proxy.response event
Here's an example log output of curl -X POST -H "Content-Type: application/json" "https://httpbin.org/anything" -d '{"foo": "bar"}'
being proxied through the network broker:
INFO[0006] request.start client_ip="::1" id=1 method=POST path="/proxy/https://httpbin.org/anything" query= user_agent=curl/8.2.1
INFO[0006] proxy.request allowlist_match="https://httpbin.org/*" client_ip="::1" destinationUrl="https://httpbin.org/anything" id=1 method=POST path="/proxy/https://httpbin.org/anything" query= request_body="{\"foo\": \"bar\"}" user_agent=curl/8.2.1
INFO[0006] proxy.response allowlist_match="https://httpbin.org/*" client_ip="::1" destinationUrl="https://httpbin.org/anything" id=1 method=POST path="/proxy/https://httpbin.org/anything" query= response_body="{\n \"args\": {}, \n \"data\": \"{\\\"foo\\\": \\\"bar\\\"}\", \n \"files\": {}, \n \"form\": {}, \n \"headers\": {\n \"Accept\": \"*/*\", \n \"Accept-Encoding\": \"gzip\", \n \"Content-Length\": \"14\", \n \"Content-Type\": \"application/json\", \n \"Host\": \"httpbin.org\", \n \"User-Agent\": \"curl/8.2.1\", \n \"X-Amzn-Trace-Id\": \"Root=1-650469a8-0032596526902b563d7e5ebc\"\n }, \n \"json\": {\n \"foo\": \"bar\"\n }, \n \"method\": \"POST\", \n \"origin\": \"::1, ...snip..., ...snip...\", \n \"url\": \"https://httpbin.org/anything\"\n}\n" user_agent=curl/8.2.1
INFO[0006] request.response body_size=511 client_ip="::1" id=1 latency=341.905708ms method=POST path="/proxy/https://httpbin.org/anything" query= status_code=200 user_agent=curl/8.2.1
logRequestBody
and logResponseBody
can also be set on a per-allowlist basis:
inbound:
allowlist:
- url: https://httpbin.org/*
methods: [GET, POST, DELETE]
logRequestBody: true
logResponseBody: true
The broker can be run in Kubernetes, as a bare Docker container, or simply as a standalone binary on a machine. If more than one instance of the broker is run at a time to manage availability, you may see some noise in the logs as the broker is not yet architected with this specific configuration in mind. However, it should still perform correctly without duplicating requests.
Config file(s) are passed to the app with -c
:
semgrep-network-broker -c config.yaml
Multiple config files can be overlaid on top of each other by passing multiple -c
args (ex. semgrep-network-broker -c config1.yaml -c config2.yaml -c config3.yaml
). Note that while maps will be merged together, arrays will be replaced.
Requirements:
- internet access to
wireguard.semgrep.dev
on UDP port 51820
semgrep-network-broker dump
dumps the current config. This is useful to see what the result of multiple configurations overlays would result in
semgrep-network-broker genkey
generates a base64 private key to stdout
semgrep-network-broker pubkey
generates a base64 public key for a given private key (via stdin)
semgrep-network-broker relay
launches an HTTP server that relays request that match a certain rule.
outbound:
listenPort: 8080
relay:
test:
destinationUrl: https://httpbin.org/anything
jsonPath: "$.foo"
equals:
- bar
would result in requests addressed to http://localhost:8080/relay/test being relayed to https://httpbin.org/anything as long as the result of the jsonpath query $.foo
executed on the request body results in the string bar
.
Check out an example here for how to use the relay for GitHub PR comments.
You can also define additional relay mappings via the additionalConfigs
field:
outbound:
listenPort: 8080
relay:
test:
destinationUrl: https://httpbin.org/anything
jsonPath: "$.foo"
equals:
- bar
additionalConfigs:
- destinationUrl: htttps://example.com/fallback
The example above would relay traffic to https://httpbin.org/anything if the request body contains {"foo": "bar"}
, otherwise, it'd relay traffic to htttps://example.com/fallback
.