--host 0.0.0.0 Not working

[hooraygith]

I run the webpack-dev-server with params --host 0.0.0.0, then visit the page through my ip, the page shows "Invalid Host header".
I look into the code, it seems the Server.prototype.checkHost in file lib/Server.js is wrong, it don't deal with the params of --host 0.0.0.0

[fielding]

I ran in to "Invalid Host header" as well today after upgrading to webpack-dev-server 2.4.3. My specific instance involved some shenanigans behind nginx reverse proxy to a another ssh tunneled reverse proxy (handles my vhosts for any dev projects with out the man telling me I can't serve up specific ports)... so after spending a short time actually looking over my config/headers, I just rolled ack to 2.4.2 and it fixed it right up with no other changes.

Doubt that is any news/that helpful, but thought I'd chime in that it seems isolated to 2.4.3/went away when I downgraded. Oh, and I also using 0.0.0.0 specifically for my host settings.

[sokra]

https://github.com/webpack/webpack-dev-server/releases/tag/v2.4.3

best add --public your-host:8080 to fix it.

[sokra]

btw. it's intended to not work when just using --host 0.0.0.0. This is a security risk.

[sokra]

behind proxy you can use the disableHostCheck: true flag.

[fielding]

I should have pulled my head out of my ass and read the v2.4.3 release. Right on, appreciate the attention to the potential vulnerability, and I dig the disableHostCheck ability being put right in even more. Thanks =)

[hooraygith]

@sokra thx for reply.
I tried disableHostCheck，and there is a problem that disableHostCheck not in the lib/optionsSchema.json

[sokra]

@hooraygith I added it and released a new version. Try to update.

[bdwain]

hi @sokra . Thanks for fixing the security issue. It seems like the disable host check option is not available on CLI. Is it possible to get that added? I'm happy to make a PR if that's an ok change.

also, in general I think these changes could use more documentation. I'm pretty confused and not totally sure I need the disable option, but I haven't found a good alternative yet from the release notes or the issues that are filed because I need to hit my local dev server from a tablet. Also, I have custom hosts in my etc/hosts which my app uses to determine things. like foo-local.com vs bar-local.com, both of which are aliased to 127.0.0.1

Either way, I think it'd be good to expose the disable option in the CLI for people who are using it.

[phairoh]

@sokra I understand the security fix and am grateful as always for all the work done on webpack but I would like to know why this was included in a patch and not a major version. Presumably this security risk has been present for quite some time and as such I don't see the urgency.

I assume that webpack tries to follow semver which is pretty clear that almost any breaking change should be a major version bump, especially if what you're versioning has a huge audience, which certainly webpack does. A major patch bump is also much better for communicating to users that there are breaking changes which would have saved at least a few people on this thread and presumably more elsewhere hours of lost work. As it is, I'm still trying to figure out how to make this change work with my team's somewhat unique configuration (that's actually somewhat similar to @bdwain's). My guess is that I'll end up rolling back the version but that certainly can't be a permanent solution.

While we could have avoided this by locking down the versions in our package.json we are still in the development and as such haven't done that yet but if packages followed semver it shouldn't be too much of a problem.

[edmorley]

@sokra, perhaps in addition to whitelisting localhost, if the Host header contains any IP, then the request can be accepted? If the request was made to a specific IP (and not a DNS rebound malicious domain), then this cannot have been an attack as made via DNS rebinding.

The only concern with this approach would be if a domain could be registered that matched the regex for an IP. However RFC 3696 section 2 seems to suggest that the top level domain can never contain all numeric characters, so a hostname can never overlap with an IPv4 IP.

[bdwain]

@edmorley i also have another use case that seems to have been broken by this. Not sure if the current implementation has a solution for this.

For development, i have a few aliases to 127.0.0.1 in my etc/hosts file. like

127.0.0.1 foo-local.com bar-local.com

and my app uses the url format to determine certain things about how it behaves. I noticed i could no longer hit foo-local.com:PORT after this change though.

[edmorley]

I've filed a retrospective GitHub issue with the original private disclosure email wording, which should hopefully make things a bit clearer: #887 - happy to answer any additional questions.

For development, i have a few aliases to 127.0.0.1 in my etc/hosts file. like

127.0.0.1 foo-local.com bar-local.com

For this case, I think we need a way to whitelist multiple hostnames, similar to Django's ALLOWED_HOSTS setting:
https://docs.djangoproject.com/en/1.11/ref/settings/#allowed-hosts