|
1 | 1 | # Security
|
2 | 2 |
|
3 |
| -The OpenTelemetry Collector defaults to operating in a secure manner, but is |
| 3 | +The OpenTelemetry Collector defaults to operating in a secure manner but is |
4 | 4 | configuration driven. This document captures important security aspects and
|
5 |
| -considerations for the Collector. This document is intended for both end-users |
6 |
| -and component developers. It assumes at least a basic understanding of the |
7 |
| -Collector architecture and functionality. |
| 5 | +considerations for the Collector. This document is intended for component |
| 6 | +developers. It assumes at least a basic understanding of the Collector |
| 7 | +architecture and functionality. |
8 | 8 |
|
9 |
| -> Note: Please review the [configuration |
10 |
| -> documentation](https://opentelemetry.io/docs/collector/configuration/) |
| 9 | +> Note: Please review the |
| 10 | +> [configuration documentation](https://opentelemetry.io/docs/collector/configuration/) |
11 | 11 | > prior to this security document.
|
12 | 12 |
|
13 |
| -## TL;DR |
14 |
| - |
15 |
| -### End-users |
16 |
| - |
17 |
| -- Configuration |
18 |
| - - SHOULD only enable the minimum required components |
19 |
| - - SHOULD ensure sensitive configuration information is stored securely |
20 |
| -- Permissions |
21 |
| - - SHOULD not run Collector as root/admin user |
22 |
| - - MAY require privileged access for some components |
23 |
| -- Receivers/Exporters |
24 |
| - - SHOULD use encryption and authentication |
25 |
| - - SHOULD limit exposure of servers to authorized users |
26 |
| - - MAY pose a security risk if configuration parameters are modified improperly |
27 |
| -- Processors |
28 |
| - - SHOULD configure obfuscation/scrubbing of sensitive metadata |
29 |
| - - SHOULD configure recommended processors |
30 |
| -- Extensions |
31 |
| - - SHOULD NOT expose sensitive health or telemetry data |
| 13 | +Security documentation for end users can be found on the OpenTelemetry |
| 14 | +documentation website: |
32 | 15 |
|
33 |
| -> For more information about securing the OpenTelemetry Collector, see |
34 |
| -> [this](https://medium.com/opentelemetry/securing-your-opentelemetry-collector-1a4f9fa5bd6f) |
35 |
| -> blog post. |
| 16 | +- [Collector configuration best practices](https://opentelemetry.io/docs/security/config-best-practices/) |
| 17 | +- [Collector hosting best practices](https://opentelemetry.io/docs/security/hosting-best-practices/) |
36 | 18 |
|
37 |
| -### Component Developers |
| 19 | +## TL;DR |
38 | 20 |
|
39 | 21 | - Configuration
|
40 | 22 | - MUST come from the central configuration file
|
41 | 23 | - SHOULD use configuration helpers
|
42 | 24 | - Permissions
|
43 | 25 | - SHOULD minimize privileged access
|
44 |
| - - MUST document what required privileged access and why |
| 26 | + - MUST document what requires privileged access and why |
45 | 27 | - Receivers/Exporters
|
46 | 28 | - MUST default to encrypted connections
|
47 | 29 | - SHOULD leverage helper functions
|
48 | 30 | - Extensions
|
49 | 31 | - SHOULD NOT expose sensitive health or telemetry data by default
|
50 | 32 |
|
| 33 | +> For more information about securing the OpenTelemetry Collector, see |
| 34 | +> [this blog post](https://medium.com/opentelemetry/securing-your-opentelemetry-collector-1a4f9fa5bd6f). |
| 35 | +
|
51 | 36 | ## Configuration
|
52 | 37 |
|
53 | 38 | The Collector binary does not contain an embedded or default configuration and
|
54 | 39 | MUST NOT start without a configuration file being specified. The configuration
|
55 |
| -file passed to the Collector MUST be validated prior to be loaded. If an |
| 40 | +file passed to the Collector MUST be validated prior to being loaded. If an |
56 | 41 | invalid configuration is detected, the Collector MUST fail to start as a
|
57 | 42 | protective mechanism.
|
58 | 43 |
|
59 |
| -> Note: Issue |
60 |
| -> [#886](https://github.com/open-telemetry/opentelemetry-collector/issues/886) |
61 |
| -> proposes adding a default configuration to the binary. |
62 |
| -
|
63 |
| -The configuration drives the Collector's behavior and care should be taken to |
64 |
| -ensure the configuration only enables the minimum set of capabilities and as |
65 |
| -such exposes the minimum set of required ports. In addition, any incoming or |
66 |
| -outgoing communication SHOULD leverage TLS and authentication. |
67 |
| - |
68 |
| -The Collector keeps the configuration in memory, but where the configuration is |
69 |
| -loaded from at start time depends on the packaging used. For example, in |
70 |
| -Kubernetes secrets and configmaps CAN be leveraged. In comparison, the Docker |
71 |
| -image embeds the configuration in the container where is it not stored in an |
72 |
| -encrypted manner by default. |
73 |
| - |
74 |
| -The configuration MAY contain sensitive information including: |
75 |
| - |
76 |
| -- Authentication information such as API tokens |
77 |
| -- TLS certificates including private keys |
| 44 | +Component developers MUST get configuration information from the Collector's |
| 45 | +configuration file. Component developers SHOULD leverage |
| 46 | +[configuration helper functions](https://github.com/open-telemetry/opentelemetry-collector/tree/main/config). |
78 | 47 |
|
79 |
| -Sensitive information SHOULD be stored securely such as on an encrypted |
80 |
| -filesystem or secret store. Environment variables CAN be used to handle |
81 |
| -sensitive and non-sensitive data as the Collector MUST support environment |
82 |
| -variable expansion. |
| 48 | +When defining Go structs for configuration data that may contain sensitive |
| 49 | +information, use the `configopaque` package to define fields with the |
| 50 | +`configopaque.String` type. This ensures that the data is masked when serialized |
| 51 | +to prevent accidental exposure. |
83 | 52 |
|
84 |
| -> For more information on environment variable expansion, see |
85 |
| -> [this](https://opentelemetry.io/docs/collector/configuration/#environment-variables) |
| 53 | +> For more information, see the |
| 54 | +> [configopaque](https://pkg.go.dev/go.opentelemetry.io/collector/config/configopaque) |
86 | 55 | > documentation.
|
87 | 56 |
|
88 |
| -When defining Go structs for configuration data that may contain sensitive information, use the `configopaque` package to define fields with the `configopaque.String` type. This ensures that the data is masked when serialized to prevent accidental exposure. |
89 |
| - |
90 |
| -> For more information, see the [configopaque](https://pkg.go.dev/go.opentelemetry.io/collector/config/configopaque) documentation. |
91 |
| -
|
92 |
| -Component developers MUST get configuration information from the Collector's |
93 |
| -configuration file. Component developers SHOULD leverage [configuration helper |
94 |
| -functions](https://github.com/open-telemetry/opentelemetry-collector/tree/main/config). |
95 |
| - |
96 |
| -More information about configuration is provided in the following sections. |
97 |
| - |
98 | 57 | ## Permissions
|
99 | 58 |
|
100 | 59 | The Collector supports running as a custom user and SHOULD NOT be run as a
|
101 | 60 | root/admin user. For the majority of use-cases, the Collector SHOULD NOT require
|
102 |
| -privileged access to function. Some components MAY require privileged access |
103 |
| -and care should be taken before enabling these components. Collector components |
104 |
| -MAY require external permissions including network access or RBAC. |
| 61 | +privileged access to function. Some components MAY require privileged access or |
| 62 | +external permissions, including network access or RBAC. |
105 | 63 |
|
106 | 64 | Component developers SHOULD minimize privileged access requirements and MUST
|
107 | 65 | document what requires privileged access and why.
|
108 | 66 |
|
109 |
| -More information about permissions is provided in the following sections. |
110 |
| - |
111 | 67 | ## Receivers and Exporters
|
112 | 68 |
|
113 | 69 | Receivers and Exporters can be either push or pull-based. In either case, the
|
114 | 70 | connection established SHOULD be over a secure and authenticated channel.
|
115 |
| -Unused receivers and exporters SHOULD be disabled to minimize the attack vector |
116 |
| -of the Collector. |
117 |
| - |
118 |
| -Receivers and Exporters MAY expose buffer, queue, payload, and/or worker |
119 |
| -settings via configuration parameters. If these settings are available, |
120 |
| -end-users should proceed with caution before modifying the default values. |
121 |
| -Improperly setting these values may expose the Collector to additional attack |
122 |
| -vectors including resource exhaustion. |
123 | 71 |
|
124 |
| -> It is possible that a receiver MAY require the Collector run in a privileged |
125 |
| -> mode in order to operate, which could be a security concern, but today this |
126 |
| -> is not the case. |
127 |
| -
|
128 |
| -Component developers MUST default to encrypted connections (via the `insecure: |
129 |
| -false` configuration setting) and SHOULD leverage |
| 72 | +Component developers MUST default to encrypted connections (using the |
| 73 | +`insecure: false` configuration setting) and SHOULD leverage |
130 | 74 | [gRPC](https://github.com/open-telemetry/opentelemetry-collector/tree/main/config/configgrpc)
|
131 | 75 | and
|
132 | 76 | [http](https://github.com/open-telemetry/opentelemetry-collector/tree/main/config/confighttp)
|
133 | 77 | helper functions.
|
134 | 78 |
|
135 |
| -### Safeguards against denial of service attacks |
136 |
| - |
137 |
| -Users SHOULD bind receivers' servers to addresses that limit connections to authorized users. |
138 |
| -For example, if the OTLP receiver OTLP/gRPC server only has local clients, the `endpoint` setting SHOULD be bound to `localhost`: |
139 |
| - |
140 |
| -```yaml |
141 |
| -receivers: |
142 |
| - otlp: |
143 |
| - protocols: |
144 |
| - grpc: |
145 |
| - endpoint: localhost:4317 |
146 |
| -``` |
147 |
| -
|
148 |
| -Generally, `localhost`-like addresses should be preferred over the 0.0.0.0 address. |
149 |
| -For more information, see [CWE-1327](https://cwe.mitre.org/data/definitions/1327.html). |
150 |
| - |
151 |
| -If `localhost` resolves to a different IP due to your DNS settings then explicitly use the loopback IP instead: `127.0.0.1` for IPv4 or `::1` for IPv6. In IPv6 setups, ensure your system supports both IPv4 and IPv6 loopback addresses to avoid issues. |
152 |
| - |
153 |
| -Using `localhost` may not work in environments like Docker, Kubernetes, and other environments that have non-standard networking setups. We've documented a few working example setups for the OTLP receiver gRPC endpoint below, but other receivers and other Collector components may need similar configuration. |
154 |
| - |
155 |
| -#### Docker |
156 |
| -You can run the Collector in Docker by binding to the correct address. An OTLP exporter in Docker might look something like this: |
| 79 | +## Safeguards against denial of service attacks |
157 | 80 |
|
158 |
| -Collector config file |
159 |
| - |
160 |
| -`config.yaml`: |
161 |
| -```yaml |
162 |
| -receivers: |
163 |
| - otlp: |
164 |
| - protocols: |
165 |
| - grpc: |
166 |
| - endpoint: my-hostname:4317 # the same hostname from your docker run command |
167 |
| -``` |
168 |
| -Docker run command: |
169 |
| -`docker run --hostname my-hostname --name container-name -p 127.0.0.1:4567:4317 otel/opentelemetry-collector:0.104.0` |
170 |
| - |
171 |
| -The key here is using the `--hostname` argument - that allows the collector to bind to the `my-hostname` address. |
172 |
| -You could access it from outside that Docker network (for example on a regular program running on the host) by connecting to `127.0.0.1:4567`. |
173 |
| - |
174 |
| -#### Docker Compose |
175 |
| -Similarly to plain Docker, you can run the Collector in Docker by binding to the correct address. |
176 |
| - |
177 |
| -`compose.yaml`: |
178 |
| -```yaml |
179 |
| -services: |
180 |
| - otel-collector: |
181 |
| - image: otel/opentelemetry-collector-contrib:0.104.0 |
182 |
| - ports: |
183 |
| - - "4567:4317" |
184 |
| -``` |
185 |
| - |
186 |
| -Collector config file: |
187 |
| - |
188 |
| -`config.yaml`: |
189 |
| -```yaml |
190 |
| -receivers: |
191 |
| - otlp: |
192 |
| - protocols: |
193 |
| - grpc: |
194 |
| - endpoint: otel-collector:4317 # Using the service name from your Docker compose file |
195 |
| -``` |
196 |
| - |
197 |
| -You can connect to this Collector from another Docker container running in the same network by connecting to `otel-collector:4317`. You could access it from outside that Docker network (for example on a regular program running on the host) by connecting to `127.0.0.1:4567`. |
198 |
| - |
199 |
| -#### Kubernetes |
200 |
| -If you run the Collector as a `Daemonset`, you can use a configuration like below: |
201 |
| -```yaml |
202 |
| -apiVersion: apps/v1 |
203 |
| -kind: DaemonSet |
204 |
| -metadata: |
205 |
| - name: collector |
206 |
| -spec: |
207 |
| - selector: |
208 |
| - matchLabels: |
209 |
| - name: collector |
210 |
| - template: |
211 |
| - metadata: |
212 |
| - labels: |
213 |
| - name: collector |
214 |
| - spec: |
215 |
| - containers: |
216 |
| - - name: collector |
217 |
| - image: otel/opentelemetry-collector:0.104.0 |
218 |
| - ports: |
219 |
| - - containerPort: 4317 |
220 |
| - hostPort: 4317 |
221 |
| - protocol: TCP |
222 |
| - name: otlp-grpc |
223 |
| - - containerPort: 4318 |
224 |
| - hostPort: 4318 |
225 |
| - protocol: TCP |
226 |
| - name: otlp-http |
227 |
| - env: |
228 |
| - - name: MY_POD_IP |
229 |
| - valueFrom: |
230 |
| - fieldRef: |
231 |
| - fieldPath: status.podIP |
232 |
| -
|
233 |
| -``` |
234 |
| -In this example, we use the [Kubernetes Downward API](https://kubernetes.io/docs/concepts/workloads/pods/downward-api/) to get your own Pod IP, then bind to that network interface. Then, we use the `hostPort` option to ensure that the Collector is exposed on the host. The Collector's config should look something like: |
235 |
| - |
236 |
| -```yaml |
237 |
| -receivers: |
238 |
| - otlp: |
239 |
| - protocols: |
240 |
| - grpc: |
241 |
| - endpoint: ${env:MY_POD_IP}:4317 |
242 |
| - http: |
243 |
| - endpoint: ${env:MY_POD_IP}:4318 |
244 |
| -``` |
245 |
| - |
246 |
| -You can send OTLP data to this Collector from any Pod on the Node by accessing `${MY_HOST_IP}:4317` to send OTLP over gRPC and `${MY_HOST_IP}:4318` to send OTLP over HTTP, where `MY_HOST_IP` is the Node's IP address. You can get this IP from the Downwards API: |
247 |
| - |
248 |
| -```yaml |
249 |
| -env: |
250 |
| - - name: MY_HOST_IP |
251 |
| - valueFrom: |
252 |
| - fieldRef: |
253 |
| - fieldPath: status.hostIP |
254 |
| -``` |
255 |
| - |
256 |
| -## Processors |
257 |
| - |
258 |
| -Processors sit between receivers and exporters. They are responsible for |
259 |
| -processing the data in some way. From a security perspective, they are useful |
260 |
| -in a couple ways. |
261 |
| - |
262 |
| -### Scrubbing sensitive data |
263 |
| - |
264 |
| -It is common for a Collector to be used to scrub sensitive data before |
265 |
| -exporting it to a backend. This is especially important when sending the data |
266 |
| -to a third-party backend. The Collector SHOULD be configured to obfuscate or |
267 |
| -scrub sensitive data before exporting. |
268 |
| - |
269 |
| -> Note: Issue |
270 |
| -> [#2466](https://github.com/open-telemetry/opentelemetry-collector/issues/2466) |
271 |
| -> proposes adding default obfuscation or scrubbing of known sensitive metadata. |
272 |
| - |
273 |
| -### Safeguards around resource utilization |
274 |
| - |
275 |
| -In addition, processors offer safeguards around resource utilization. The |
276 |
| -`batch` and especially `memory_limiter` processor help ensure that the |
277 |
| -Collector is resource efficient and does not run out of memory when overloaded. At |
278 |
| -least these two processors SHOULD be enabled on every defined pipeline. |
279 |
| - |
280 |
| -> For more information on recommended processors and order, see |
281 |
| -> [this](https://github.com/open-telemetry/opentelemetry-collector/tree/main/processor) |
282 |
| -> documentation. |
| 81 | +See the [Collector configuration security documentation](https://opentelemetry.io/docs/security/config-best-practices/#protect-against-denial-of-service-attacks) to learn how to safeguard against denial of service attacks. |
283 | 82 |
|
284 | 83 | ## Extensions
|
285 | 84 |
|
286 |
| -While receivers, processors, and exporters handle telemetry data directly, |
287 |
| -extensions typical serve different needs. |
288 |
| - |
289 |
| -### Health and Telemetry |
290 |
| - |
291 |
| -The initial extensions provided health check information, Collector metrics and |
292 |
| -traces, and the ability to generate and collect profiling data. When enabled |
293 |
| -with their default settings, all of these extensions except the health check |
294 |
| -extension are only accessibly locally to the Collector. Care should be taken |
295 |
| -when configuring these extensions for remote access as sensitive information |
296 |
| -may be exposed as a result. |
297 |
| - |
298 | 85 | Component developers SHOULD NOT expose health or telemetry data outside the
|
299 | 86 | Collector by default.
|
300 |
| - |
301 |
| -### Forwarding |
302 |
| - |
303 |
| -A forwarding extension is typically used when some telemetry data not natively |
304 |
| -supported by the Collector needs to be collected. For example, the |
305 |
| -`http_forwarder` extension can receive and forward HTTP payloads. Forwarding |
306 |
| -extensions are similar to receivers and exporters so the same security |
307 |
| -considerations apply. |
308 |
| - |
309 |
| -### Observers |
310 |
| - |
311 |
| -An observer is capable of performing service discovery of endpoints. Other |
312 |
| -components of the collector such as receivers MAY subscribe to these extensions |
313 |
| -to be notified of endpoints coming or going. Observers MAY require certain |
314 |
| -permissions in order to perform service discovery. For example, the |
315 |
| -`k8s_observer` requires certain RBAC permissions in Kubernetes, while the |
316 |
| -`host_observer` requires the Collector to run in privileged mode. |
317 |
| - |
318 |
| -### Subprocesses |
319 |
| - |
320 |
| -Extensions may also be used to run subprocesses. This can be useful when |
321 |
| -collection mechanisms that cannot natively be run by the Collector (e.g. |
322 |
| -FluentBit). Subprocesses expose a completely separate attack vector that would |
323 |
| -depend on the subprocess itself. In general, care should be taken before |
324 |
| -running any subprocesses alongside the Collector. |
0 commit comments