-
Notifications
You must be signed in to change notification settings - Fork 30.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Integrity checks via policy #23834
Integrity checks via policy #23834
Conversation
cc @nodejs/security-wg |
@@ -0,0 +1,42 @@ | |||
# Policies |
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Good work! I would an explicit experimental warning if the option is used.
lib/internal/modules/cjs/loader.js
Outdated
@@ -22,8 +22,9 @@ | |||
'use strict'; | |||
|
|||
const { NativeModule } = require('internal/bootstrap/loaders'); | |||
const util = require('util'); | |||
const { manifest } = require('internal/process/policy'); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would prefer it this file was loaded only if policies are in use. Every module we add the longer it gets Node.js to start.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not convinced this has significant impact. Is there some metric we are trying to keep within? If startup time is more critical we could also take other approaches than avoiding loading code such as taking snapshots. Is this a nit or something that you wish to block on?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
+1 on code caches, but it's something that is kind of far in the future as far as I understand.
Is there some metric we are trying to keep within?
@jasnell or @BridgeAR might have some number. However every file adds some millis to startup time.
Considering that it takes very little effort to load it at runtime, why not doing it?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
extra complexity from my point of view to do so for little gain from my understanding. even just bundling core into a single source text would have bigger savings than this I would think. the less complex the system via conditional branching the easier i find it to reason about and maintain.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This PR gaves us 10%: #20567. Most of those where added with the same reasoning, but as Node.js gets more feature, those pile up.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I am definitely for lazy loading this! It is nothing that is required all the time. Adding snapshots is great and we want to do that on top of this. The current startup time is way to high and adding things as we used to doesn't make things any better. Instead we should always lazy load most files. This will also reduce the memory overhead for a simple Node.js application that does not use these features.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
lazy loaded now, rebased though
lib/internal/policy/sri.js
Outdated
return entries; | ||
}; | ||
|
||
module.exports = Object.freeze({ |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why are you freezing? this is internal.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Defensive programming practice, habit / no real downside. I'd just leave it as is unless you really want it to change.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would prefer it changed, this likely sit on a hot path for startup, and as far as I know V8 didn't really like freezed object when optimizing.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
changed
doc/api/policy.md
Outdated
} | ||
``` | ||
|
||
In order to generate integrity strings, a script such as `printf "sha384-$(cat checked.js | openssl dgst -sha384 -binary | base64)"` can be used. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is not really friendly for people on Windows. We should provide a tool on npm to automate this.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I can agree to that but a generalized manifest tool seems something that we aren't ready to make a design for while we still are working out the design of manifests themselves.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is not really a blocker, but something that we should put into the roadmap for this to get out of experimental.
Memo to not forget) For https://github.com/nodejs/node/blob/master/doc/api/cli.md |
lib/internal/process/policy.js
Outdated
return o; | ||
}); | ||
manifest = new Manifest(json); | ||
} catch (e) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think we should not lose this error.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
do we have a recommended practice for nesting errors, we don't to my knowledge
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We don't, but I think a user would like to know where the JSON is malformed, reading the file errors or what else could happen.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
i can add the error.message but don't know where i should put the error.code, adding both would be a bit confusing.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
How about not wrapping at all?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
seems fine, removed the wrapper
lib/internal/modules/cjs/loader.js
Outdated
@@ -733,6 +748,12 @@ Module._extensions['.json'] = function(module, filename) { | |||
|
|||
// Native extension for .node | |||
Module._extensions['.node'] = function(module, filename) { | |||
var content = fs.readFileSync(filename); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
these two lines should definitely be inside the condition
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
fixed
lib/internal/modules/cjs/loader.js
Outdated
@@ -722,6 +731,12 @@ Module._extensions['.js'] = function(module, filename) { | |||
// Native extension for .json | |||
Module._extensions['.json'] = function(module, filename) { | |||
var content = fs.readFileSync(filename, 'utf8'); | |||
|
|||
const moduleURL = pathToFileURL(filename); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
should seemingly be inside condition
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
fixed
lib/internal/modules/cjs/loader.js
Outdated
@@ -662,6 +663,11 @@ function normalizeReferrerURL(referrer) { | |||
// Returns exception, if any. | |||
Module.prototype._compile = function(content, filename) { | |||
|
|||
const moduleURL = pathToFileURL(filename); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
same here
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
fixed
doc/api/policy.md
Outdated
Once this has been set, all modules must conform to a policy file passed to the flag: | ||
|
||
```sh | ||
node --experimental-policy policy.json app.js |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'd prefer syntax like --experimental-policy=policy.json app.js
(note the =
)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
using =
doesn't mesh with other cli options like --require
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
changed doc, though autocomplete won't work with the = form, curious why the preference but not really anything of note. I do have some concerns and need to think of if we ever want separate flags for policies if this occupies the = space.
doc/api/policy.md
Outdated
|
||
### Integrity Checks | ||
|
||
Policy files must use integrity checks with Subresource Integrity strings compatible with the [integrity attribute as present in browsers](https://www.w3.org/TR/SRI/#the-integrity-attribute) associated with absolute URLs. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nit: line wrap at <= 80 chars
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
fixed, had to reword a bit
I've expressed my concerns about this on patch on the security-wg slack. Will reach out to chat via hangouts more about this and will circle back here to explain the concerns/resolutions (have other things on my plate now). |
@bmeck Can you speak more on use cases? Who would use this when and why? |
@bnoordhuis the main intent here is part of a larger story of forming an idea of an auditable level of trust in code being run on Node.js as laid out in the drive document in the description. As such, this PR addresses a singular concern of the goals laid out. By introducing the concept of integrity to files, we open a way for users to run code without fear of running code that was not audited (either by tooling or human). The main use cases for this are for running command line applications, not library concerns. When running an application installed via some tool or through an image Node.js does not have a mechanism to check if the source of an application has been tampered with (intentional or not). This PR introduces a means for applications to be run that prevents alteration of source code from what was installed from being loaded through the module system. Anytime an unaudited application runs it can open the client to malicious behavior as we saw with eslint earlier this year. In the case of eslint, the behavior was limited to The majority of installations of CLI applications and usage of 3rd party modules could benefit from this feature by preventing some malicious behaviors from altering files on disk. In particular a focus should be placed on how files can replace existing parts of an application like OWASP's Unrestricted File Upload concerns for HTTP servers. However, anytime arbitrary writes could occur within an application directory this also is of concern even without file uploads. The means by which these are generated is left open for now while experimentation can occur. Tools related to generating these files need to have a high level of trust. They might wish to defer to extra information like if a known bad integrity exists due to a CVE and audit if that integrity is anywhere in a manifest, similar to checks of package versions. Unlike package version checks which are only concerned with installation time guarantees this PR would be checked whenever a module is loaded, even if someone managed to alter the contents on disk via some persistent attack. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great work! Look forward to seeing the policy supported through the package.json.
Re the URLs, can we support relative URLs in the manfiest, taken to be relative to the URL of the manifest itself? That is basically what package maps do. |
lib/internal/policy/sri.js
Outdated
Object: { | ||
defineProperty | ||
} | ||
} = global; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
These double destructures can get a bit weird… wouldn’t const { defineProperty } = Object;
do the trick as well?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
weird? that works and I can simplify but double destructuring should be understandable.
We should not give the actual policy the ability to be placed in the application directory. It is an easy misconfiguration problem if the application directory is writable by the user running the application. The policy should not be able to be mutated by running a program:
In addition, ideally policies could be used across multiple applications and things like https://github.com/yarnpkg/pnp-sample-app and https://github.com/npm/tink would be placing the dependencies outside of the application directory and be shared integrities amongst all the places running things. The ability to request permissions should ideally be placed in a location that tooling such as package managers can pick it up, but it should be a point that is tamper resistant or encourages designs that are less prone to attack. We should not completely remove the ability to modify a policy from a node application though. The ability to run prompts to modify a policy is a powerful tool that should be looked into (not doable through this PR). Things like prompts about integrity mismatch and missing privileges that can prompt to update permissions of a policy are powerful and help the user understand why something is failing and if they wish to proceed anyway. |
So to understand your expectation is that the policy would be in a trusted file without write permissions on the system, separate to the application folder? Perhaps this should be documented somewhere as it wasn't clear to me that was a strong constraint of the security model being advocated here. I don't quite follow your argument though as I imagine in most cases the policy would always be in the application folder - and ideally without write permissions, but the folder location seems a separate concern and therefore not a difference in security to me. |
The location isn't a constraint, but this is following the fact that most applications have write permissions to their working directory or could chmod files within them to get permissions. Design here is trying to encourage best practice not force it as I don't see a reasonable way to enforce such practice from Node.
The difference is 2 fold.
|
0526d9c
to
e2199a1
Compare
All comments appear resolved to me, if anything is still missing let me know, otherwise going to merge tomorrow morning. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
@bmeck I see the docs now, not sure what I was doing when I looked before sorry for any confusion. |
And to confirm, it was intended to be informative to the TSC, not blocking so at this point you should be good to go. |
This enables code loaded via the module system to be checked for integrity to ensure the code loaded matches expectations.
7cb3a5e
to
404e229
Compare
Need to debug why Windows is failing. |
CI: https://ci.nodejs.org/job/node-test-pull-request/20149/ should be passing now |
This enables code loaded via the module system to be checked for integrity to ensure the code loaded matches expectations. PR-URL: #23834 Reviewed-By: Guy Bedford <guybedford@gmail.com> Reviewed-By: Vladimir de Turckheim <vlad2t@hotmail.com> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Landed in 9d5fbeb. |
For something that has pretty far reaching implications, this PR only got three sign-offs, even though there have been plenty (partial?) reviewers. Am I the only one who feels somewhat uncomfortable about that? Did it really get all the scrutiny and discussion it needed? |
@bnoordhuis It was up for quite some time, got tagged for TSC and security, got some TSC comments/review, and was talked about in the slack channel for the security WG, the Realms calls for TC39, and node-dev IRC. If there is something more we could have done, we can continue to iterate on what is required to land pull requests. As it stands, it only is getting scrutiny after landing, even though I tried to advocate in common communication channels for several weeks before landing. If there are objections to the PR we can revert it, and if there are security concerns we certainly should fix them or document them. All in all, landing this seemed appropriate given the time and amount of publicity it had. |
This enables code loaded via the module system to be checked for integrity to ensure the code loaded matches expectations. PR-URL: #23834 Reviewed-By: Guy Bedford <guybedford@gmail.com> Reviewed-By: Vladimir de Turckheim <vlad2t@hotmail.com> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Notable Changes: * events: * For unhandled `error` events with an argument that is not an `Error` object, the resulting exeption will have more information about the argument. #25621 * child_process: * When the `maxBuffer` option is passed, `stdout` and `stderr` will be truncated rather than unavailable in case of an error. #24951 * policy: * Experimental support for module integrity checks through a manifest file is implemented now. #23834 * n-api: * The `napi_threadsafe_function` feature is now stable. #25556 * report: * An experimental diagnostic API for capturing process state is available as `process.report` and through command line flags. #22712 * tls: * `tls.connect()` takes a `timeout` option analogous to the `net.connect()` one. #25517 * worker: * `process.umask()` is available as a read-only function inside Worker threads now. #25526 * An `execArgv` option that supports a subset of Node.js command line options is supported now. #25467 PR-URL: #25687
Notable Changes: * events: * For unhandled `error` events with an argument that is not an `Error` object, the resulting exeption will have more information about the argument. #25621 * child_process: * When the `maxBuffer` option is passed, `stdout` and `stderr` will be truncated rather than unavailable in case of an error. #24951 * policy: * Experimental support for module integrity checks through a manifest file is implemented now. #23834 * n-api: * The `napi_threadsafe_function` feature is now stable. #25556 * report: * An experimental diagnostic API for capturing process state is available as `process.report` and through command line flags. #22712 * tls: * `tls.connect()` takes a `timeout` option analogous to the `net.connect()` one. #25517 * worker: * `process.umask()` is available as a read-only function inside Worker threads now. #25526 * An `execArgv` option that supports a subset of Node.js command line options is supported now. #25467 PR-URL: #25687
For others searching to understand the context of this better, some of the PR discussion ocurred in slack, starting with this comment |
This is an initial effort to introduce integrity checks into code loading paths for Node.js . It is meant to cover the requirements of Subresource Integrity Checks as laid out previously. I have left the generation of manifests outside of the scope of Node.js for this PR and think this is the minimal viable way to introduce integrity checks in a sufficient manner.
There have been concerns about using absolute paths for file integrity, but I have not found a suitable alternative as of yet and feel that this matches Subresource Integrity where location is related to integrity. A separate mechanism for location-less integrity could be introduced falling more under the style of script-src and is not under the scope of this PR. In particular I lay out why I find starting with location based integrity more conservative and less likely to introduce the same surface as location less with a couple examples.
Checklist
make -j4 test
(UNIX), orvcbuild test
(Windows) passes