Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/cactus/go-camo

A secure image proxy server
https://github.com/cactus/go-camo

camo go golang golang-application image-proxy mixed-content mixed-content-error proxy-server

Last synced: about 2 months ago
JSON representation

A secure image proxy server

Awesome Lists containing this project

README

        

= go-camo
:toc: macro
:hide-uri-scheme:
ifdef::env-github[]
:toc-title:
:tip-caption: :bulb:
:note-caption: :bulb:
:important-caption: :heavy_exclamation_mark:
:caution-caption: :fire:
:warning-caption: :warning:
endif::[]

// some links
:link-atmos-camo: https://github.com/atmos/camo[camo]
:link-hmac: https://en.wikipedia.org/wiki/HMAC[HMAC]
:link-hsts: https://en.wikipedia.org/wiki/HTTP_Strict_Transport_Security[HSTS]
:link-asciidoctor: https://asciidoctor.org[asciidoctor]
:link-damontools: https://cr.yp.to/daemontools.html[daemontools]
:link-runit: https://en.wikipedia.org/wiki/Runit[runit]
:link-upstart: https://en.wikipedia.org/wiki/Upstart_(software)[upstart]
:link-launchd: https://en.wikipedia.org/wiki/Launchd[launchd]
:link-systemd: https://www.freedesktop.org/wiki/Software/systemd/[systemd]
:link-daemontools-why: https://cr.yp.to/daemontools/faq/create.html#why[daemontools/why]
:link-heroku-buildpack-go: https://github.com/kr/heroku-buildpack-go
:link-dns-rebinding: https://en.wikipedia.org/wiki/DNS_rebinding[dns rebinding]
:link-ip6-special-addresses: https://en.wikipedia.org/wiki/IPv6_address#Special_addresses
:link-docker-containers: https://hub.docker.com/r/cactus4docker/go-camo[docker hub]
:link-github-containers: https://github.com/cactus/go-camo/pkgs/container/go-camo[github packages]
:link-releases: https://github.com/cactus/go-camo/releases[binary releases]
:link-mit-license: https://www.opensource.org/licenses/mit-license.php[MIT license]
:link-mrsaints: https://github.com/MrSaints[MrSaints]
:link-arachnys-fork: https://github.com/arachnys/go-camo[fork]

image:https://img.shields.io/github/release/cactus/go-camo.svg[Current Release,link=http://github.com/cactus/go-camo/releases]
image:https://img.shields.io/docker/image-size/cactus4docker/go-camo?label=container%20size[Container Image Size (latest by date),link=https://hub.docker.com/r/cactus4docker/go-camo]
image:https://github.com/cactus/go-camo/workflows/unit-tests/badge.svg[BuildStatus]
image:.github/img-readme-license.svg[License,link=https://github.com/cactus/go-camo/blob/master/LICENSE.adoc]
// image:https://circleci.com/gh/cactus/go-camo.svg?style=svg[CircleCI,link=https://circleci.com/gh/cactus/go-camo]

ifdef::env-github[]
[discrete]
== Contents
endif::[]
toc::[]

== About

go-camo is a go version of a https://github.com/atmos/camo[camo] server.

A camo server is a special type of image proxy
that proxies non-secure images over SSL/TLS,
in order to prevent mixed content warnings on secure pages.
The server works in conjunction with back-end code
that rewrites image URLs and signs them with an {link-hmac}.

== How it works

The general steps are as follows:

* A client requests a page from the web app.
* The original URL in the content is parsed.
* An HMAC signature of the URL is generated.
* The URL and hmac are encoded.
* The encoded URL and hmac are placed into the expected format,
creating the signed URL.
* The signed URL replaces the original image URL.
* The web app returns the content to the client.
* The client requests the signed URL from Go-Camo.
* Go-Camo validates the HMAC, decodes the URL,
then requests the content from the origin server and streams it to the client.

[source,text]
----
+----------+ request +-------------+
| |----------------------------->| |
| | | |
| | | web-app |
| | img src=https://go-camo/url | |
| |<-----------------------------| |
| | +-------------+
| client |
| | https://go-camo/url +-------------+ http://some/img
| |----------------------------->| |--------------->
| | | |
| | | go-camo |
| | img data | | img data
| |<-----------------------------| |<---------------
| | +-------------+
+----------+
----

Go-Camo supports both hex and base64 encoded urls at the same time.

[%header%autowidth.stretch]
|===
| encoding | tradeoffs
| hex | longer, case insensitive, slightly faster
| base64 | shorter, case sensitive, slightly slower
|===

Benchmark results with go1.12.7:

[source,text]
----
BenchmarkHexEncoder-4 1000000 1364 ns/op
BenchmarkB64Encoder-4 1000000 1447 ns/op
BenchmarkHexDecoder-4 1000000 1312 ns/op
BenchmarkB64Decoder-4 1000000 1379 ns/op
----

For examples of url generation, see the link:examples/[examples] directory.

While Go-Camo will support proxying HTTPS images as well,
for performance reasons you may choose to filter HTTPS requests out from proxying,
and let the client simply fetch those as they are.
The linked code examples do this.

Note that it is recommended to front Go-Camo with a CDN when possible.

== Differences from Camo

* Go-Camo supports 'Path Format' url format only.
Camo's "Query String Format" is not supported.
* Go-Camo supports some optional "allow/deny" origin filters.
* Go-Camo supports client http keep-alives.
* Go-Camo provides native SSL support.
* Go-Camo provides native HTTP/2 support
* Go-Camo supports using more than one os thread (via GOMAXPROCS)
without the need of multiple instances or additional proxying.
* Go-Camo builds to a static binary.
This makes deploying to large numbers of servers a snap.
* Go-Camo supports both Hex and Base64 urls.
Base64 urls are smaller, but case sensitive.
* Go-Camo supports HTTP HEAD requests.
* Go-Camo allows custom default headers to be added --
useful for things like adding {link-hsts} headers.

== Installing pre-built binaries

Download the tarball appropriate for your OS/ARCH from {link-releases}. +
Extract, and copy files to desired locations.

== Building

Building requires:

* make
* posix compatible shell (sh)
* git
* go (latest version recommended. At least version >= 1.21)
* {link-asciidoctor} (for building man pages only)

Building:

[source,text]
----
# first clone the repo
$ git clone [email protected]:cactus/go-camo
$ cd go-camo

# show make targets
$ make
Available targets:
help this help
clean clean up
all build binaries and man pages
check run checks and validators
test run tests
cover run tests with cover output
build build all binaries
man build all man pages
tar build release tarball
cross-tar cross compile and build release tarballs

# build all binaries (into ./bin/) and man pages (into ./man/)
# strips debug symbols by default
$ make all

# do not strip debug symbols
$ make all GOBUILD_LDFLAGS=""
----

== Running

[source,text]
----
$ go-camo -k "somekey"
# run the gc less frequently (a bit better performance, uses more memory)
$ env GOGC=300 go-camo -k "somekey"
----

Go-Camo does not daemonize on its own.
For production usage,
it is recommended to launch in a process supervisor,
and drop privileges as appropriate.

Examples of supervisors include:
{link-damontools}, {link-runit}, {link-upstart},
{link-launchd}, {link-systemd},
and many more.

For the reasoning behind lack of daemonization, see {link-daemontools-why}.
In addition, the code is much simpler because of it.

== Running on Heroku

In order to use this on Heroku with the provided Procfile, you need to:

* Create an app specifying the {link-heroku-buildpack-go} buildpack
* Set `GOCAMO_HMAC` to the key you are using

== Securing an installation

go-camo will generally do what you tell it to with regard to fetching signed urls.
There is some limited support for trying to prevent {link-dns-rebinding} attacks.

go-camo will attempt to reject any address matching an rfc1918 network block,
or a private scope ipv6 address, be it in the url or via resulting hostname
resolution.

Please note, however, that this does not provide protection for a network that
uses public address space (ipv4 or ipv6), or some of the
{link-ip6-special-addresses}[more exotic] ipv6 addresses.

The list of networks rejected includes...

[%header%autowidth.stretch]
|===
| Network | Description

| `127.0.0.0/8`
| loopback

| `169.254.0.0/16`
| ipv4 link local

| `10.0.0.0/8`
| rfc1918

| `172.16.0.0/12`
| rfc1918

| `192.168.0.0/16`
| rfc1918

| `::1/128`
| ipv6 loopback

| `fe80::/10`
| ipv6 link local

| `fec0::/10`
| deprecated ipv6 site-local

| `fc00::/7`
| ipv6 ULA

| `::ffff:0:0/96`
| IPv4-mapped IPv6 address
|===

More generally, it is recommended to either:

* Run go-camo on an isolated instance (physical, vlans, firewall rules, etc).
* Run a local resolver for go-camo that returns NXDOMAIN responses
for addresses in deny-listed ranges (e.g. unbound's `private-address` functionality).
This is also useful to help prevent dns rebinding in general.

== Configuring

=== Environment Vars

* `GOCAMO_HMAC` - HMAC key to use.
* `HTTPS_PROXY` - Configure an outbound proxy for HTTPS requests. +
Either a complete URL or a `host[:port]`, in which case an HTTP scheme
is assumed. See <> notes for more information.
* `HTTP_PROXY` - Configure an outbound proxy for HTTP requests. +
Either a complete URL or a `host[:port]`, in which case an HTTP scheme
is assumed. See <> notes for more information.

=== Command line flags

[source,text]
----
$ go-camo -h
Usage: go-camo [flags]

An image proxy that proxies non-secure images over SSL/TLS

Flags:
-h, --help Show context-sensitive help.
-k, --key=STRING HMAC key
-H, --header=HEADER,... Add additional header to each response.
This option can be used multiple times to add
multiple headers.
--listen="0.0.0.0:8080" Address:Port to bind to for HTTP
--ssl-listen=HOST_PORT Address:Port to bind to for HTTPS/SSL/TLS
--socket-listen=PATH Path for unix domain socket to bind to for HTTP
--quic Enable http3/quic. Binds to the same port
number as ssl-listen but udp+quic.
--automaxprocs Set GOMAXPROCS automatically to match Linux
container CPU quota/limits.
--ssl-key=PATH ssl private key (key.pem) path
--ssl-cert=PATH ssl cert (cert.pem) path
--max-size=INT Max allowed response size (KB)
--timeout=4s Upstream request timeout
--max-redirects=3 Maximum number of redirects to follow
--max-size-redirect=URL redirect to URL when max-size is exceeded
--metrics Enable Prometheus compatible metrics endpoint
--no-debug-vars Disable the /debug/vars/ metrics endpoint.
This option has no effects when the metrics are
not enabled.
--no-log-ts Do not add a timestamp to logging
--prof Enable go http profiler endpoint
--log-json Log in JSON format
--no-fk Disable frontend http keep-alive support
--no-bk Disable backend http keep-alive support
--allow-content-video Additionally allow 'video/*' content
--allow-content-audio Additionally allow 'audio/*' content
--allow-credential-urls Allow urls to contain user/pass credentials
--filter-ruleset=PATH Text file containing filtering rules (one per
line)
--server-name="go-camo" Value to use for the HTTP server field
--expose-server-version Include the server version in the HTTP server
response header
--enable-xfwd4 Enable x-forwarded-for passthrough/generation
-v, --verbose Show verbose (debug) log level output
-V, --version Print version and exit; specify twice to show
license information.
----

A few notes about specific flags:

* `--filter-ruleset`
+
--
If a `filter-ruleset` file is defined,
that file is read and each line is converted into a filter rule.
See link:man/go-camo-filtering.5.adoc[`go-camo-filtering(5)`]
for more information regarding the format for the filter file itself.

Regarding evaluation: The ruleset is NOT evaluated in-order.
The rules process in two phases: "allow rule phase" where the allow rules are evaluated,
and the "deny rule phase" where the deny rules are evaluated.
First match in each phase "wins" that phase.

In the "allow phase", an origin request must match at least one allow rule.
The first rule to match "wins" and the request moves on to the next phase.
If there are no allow rules supplied, this phase is skipped.

In the deny rule phase, any rule that matches results in a rejection.
The first match "wins" and the request is failed.
If there are no deny rules supplied, this phase is skipped.

[NOTE]
====
It is always preferable to do filtering at the point of URL generation and signing.
The `filter-ruleset` functionality (both allow and deny) is supplied
predominantly as a fallback safety measure,
for cases where you have previously generated a URL and you need a quick temporary fix,
or where rolling keys takes a while and/or is difficult.
====
--

* `--max-size`
+
--
The `--max-size` value is defined in KB.
Set to `0` to disable size restriction.
The default is `0`.
--

* `--metrics`
+
--
If the `metrics` flag is provided, then the service will expose a Prometheus
`/metrics` endpoint and a `/debug/vars` endpoint from the go `expvar` package.
--

* `--no-debug-vars`
+
--
If the `no-debug-vars` flag is provided along with the `metrics` flag, the
`/debug/vars` endpoint is removed.
--

* `-k`, `--key`
+
--
If the HMAC key is provided on the command line,
it will override (if present),
an HMAC key set in the environment var.
--

* `-H, --header`
+
--
Additional default headers (sent on every response) can also be set.
This argument may be specified many times.

The list of default headers sent are:

[source,text]
----
X-Content-Type-Options: nosniff
X-XSS-Protection: 1; mode=block
Content-Security-Policy: default-src 'none'; img-src data:; style-src 'unsafe-inline'
----

As an example, if you wanted to return a `Strict-Transport-Security` header
by default, you could add this to the command line:

[source,text]
----
-H "Strict-Transport-Security: max-age=16070400"
----
--

* `--allow-content-video` and `--allow-content-audio`
+
--
By default only `image/*` content-types are accepted and proxied, all other
content-types are rejected.

Add the `--allow-content-video` argument to addtionally allow `video/*` content
types.

Add the `--allow-content-audio` argument to addtionally allow `audio/*` content
types.
--

== Upstream Http Proxying

Care should be taken when using upstream http proxy support. go-camo has
several protections against SSRF vectors, for example:

* Checking http redirect chains against rfc1918 addresses.
* Limits to maximum number of redirects.
* Protection against self-redirect loops.
* Various other protections.

The use of an upstream http proxy may subvert several of these protections, as
go-camo will be required to offload certain operations to the upstream http proxy.

Some examples (list is not exhaustive):

* The upstream http proxy itself may be responsible for following redirects
(depending on configuration). As such, go-camo may not have visibility into
the redirect chain. This could result in resource exhaustion (redirect
loops), or SSRF (redirects to internal URLs).
* The upstream http proxy itself will be responsible for connecting to external
servers, and would need to be configured for any request size limits.
While go-camo would still limit request sizes based on its own configuration,
the upstream http proxy may still fetch the content before handoff.
* There may be other chances for "configuration confusion" -- where two
services are configured together in such a way, that introduces
issues not possible when operating standalone.

Proper configuration of the upstream http proxy may mitigate these issues. +
Test your configurations and monitor carefully!

== Monitoring

=== Metrics

When the `--metrics` flag is used,
the service will expose a Prometheus-compatible `/metrics` endpoint.
This can be used by monitoring systems to gather data.

The endpoint includes all of the default `go_` and `process_`.
In addition, a number of custom metrics.

[%header%autowidth.stretch]
|===
| Name | Type | Help

| camo_response_duration_seconds | Histogram
| A histogram of latencies for proxy responses.

| camo_response_size_bytes | Histogram
| A histogram of sizes for proxy responses.

| camo_proxy_content_length_exceeded_total | Counter
| The number of requests where the content length was exceeded.

| camo_proxy_reponses_failed_total | Counter
| The number of responses that failed to send to the client.

| camo_proxy_reponses_truncated_total | Counter
| The number of responses that were too large to send.

| camo_responses_total | Counter
| Total HTTP requests processed by the go-camo, excluding scrapes.
|===

It also includes a `camo_build_info` metric that exposes the version.
In addition, you can expose some extra data to metrics via env vars, if desired:

* Revision via `APP_INFO_REVISION`
* Branch via `APP_INFO_BRANCH`
* BuildDate via `APP_INFO_BUILD_DATE`
* You can also override the version by setting `APP_INFO_VERSION`

A `/debug/vars` endpoint is also included with `--metrics` by default.
This endpoint returns memstats and some additional data. This endpoint can be
disabled by additionally supplying the `--no-debug-vars` flag.

== Additional tools

Go-Camo includes a couple of additional tools.

=== url-tool

The `url-tool` utility provides a simple way to generate signed URLs
from the command line.

[source,text]
----
$ url-tool -h
Usage:
url-tool [OPTIONS]

Application Options:
-k, --key= HMAC key
-p, --prefix= Optional URL prefix used by encode output

Help Options:
-h, --help Show this help message

Available commands:
decode Decode a URL and print result
encode Encode a URL and print result
----

Example usage:

[source,text]
----
# hex
$ url-tool -k "test" encode -p "https://img.example.org" "http://golang.org/doc/gopher/frontpage.png"
https://img.example.org/0f6def1cb147b0e84f39cbddc5ea10c80253a6f3/687474703a2f2f676f6c616e672e6f72672f646f632f676f706865722f66726f6e74706167652e706e67

$ url-tool -k "test" decode "https://img.example.org/0f6def1cb147b0e84f39cbddc5ea10c80253a6f3/687474703a2f2f676f6c616e672e6f72672f646f632f676f706865722f66726f6e74706167652e706e67"
http://golang.org/doc/gopher/frontpage.png

# base64
$ url-tool -k "test" encode -b base64 -p "https://img.example.org" "http://golang.org/doc/gopher/frontpage.png"
https://img.example.org/D23vHLFHsOhPOcvdxeoQyAJTpvM/aHR0cDovL2dvbGFuZy5vcmcvZG9jL2dvcGhlci9mcm9udHBhZ2UucG5n

$ url-tool -k "test" decode "https://img.example.org/D23vHLFHsOhPOcvdxeoQyAJTpvM/aHR0cDovL2dvbGFuZy5vcmcvZG9jL2dvcGhlci9mcm9udHBhZ2UucG5n"
http://golang.org/doc/gopher/frontpage.png
----

== Containers

There are containers built automatically from version tags, pushed to both {link-docker-containers} and {link-github-containers}.

These containers are untested and provided only for those
with specific containerization requirements.
When in doubt, prefer the statically compiled {link-releases},
unless you specifically need a container.

== Changelog

See xref:CHANGELOG.adoc[CHANGELOG].

== License

Released under the {link-mit-license}.
See xref:LICENSE.adoc[LICENSE] file for details.