True though you could also query an API in front of Solr to a stand still pretty easily. DoSing is a pretty easy thing to do to anything that needs to be open to the public.
The biggest issue with the proxy approach is an attacker with Solr knowledge that doesn't need to DoS, just send a handful of really slow queries to Solr. This is something that can be mitigated, but the more skilled the attacker the more interesting the slow queries get. I should note this is a problem to mitigate with any system that handles user queries by passing then directly to edismax. Solr sort of encourages you to talk straight to edismax, and most systems don't prepare or escape the query. Instead they want to support the full range of query operations. An attacker can still put nasty function queries in the query box enough times to make a Solr server crawl. Doug On Saturday, December 26, 2015, GW <thegeofo...@gmail.com> wrote: > Yes, your proxy seems to work. > > The only thing that bothers me is anyone can query your Solr installation. > > The world is not a nice place and I can't tell you how many DOS attacks > I've fended off in the last 30 years. > > If I thought you were an a-hole I could set up a few machines and query > your server to a standstill. > > About ten years ago I was working on a contract. The competitor that lost > the bid did a email DOS attack on me after they took out a whole bunch car > adds (hot deals) in the local paper. My email was f###'d and my phone was > ringing off the hook. > > Cheers, > > GW > > > On 25 December 2015 at 21:55, Doug Turnbull < > dturnb...@opensourceconnections.com <javascript:;>> wrote: > > > Yeah I prefer a whitelist of locked down query request handlers via a > > proxy that are reasonably well protected. I would never expose update to > > the web or allow any updating over a public interface. > > > > If you want an example, you can checkout > > > > > > > http://solr.quepid.com/solr/statedecoded/select?q=*:*&qt=update&stream.body= > > <delete><query>*:*</query></delete>&commit=true > > > > http://solr.quepid.com/solr/statedecoded/update?stream.body= > > <delete><query>*:*</query></delete>&commit=true > > > > But still get search results back: > > http://solr.quepid.com/solr/statedecoded/select?q=*:* > > > > Click all those all day long. And do let me know if you find holes... I'm > > sure there's room for improvement > > > > Cheers, > > -Doug > > > > On Friday, December 25, 2015, GW <thegeofo...@gmail.com <javascript:;>> > wrote: > > > > > If you are using Linux a simple one liner in IP tables > > > > > > iptables -I INPUT \! --src www.yourwebserver.com -m tcp -p tcp --dport > > > 8983 -j DROP > > > > > > > > > If windows, you can do something similar > > > > > > otherwise it is very easy for anyone to delete all your documents with > > > > > > http://yoursolrserver.com:8983/solr/your-core/update?stream.body= > > > <delete><query>*:*</query></delete>&commit=true > > > > > > > > > > > > > > > On 25 December 2015 at 20:42, Doug Turnbull < > > > dturnb...@opensourceconnections.com <javascript:;> <javascript:;>> > wrote: > > > > > > > Hi Shawn > > > > > > > > Maybe I should have qualified the parameters of scenarios this make > me > > > > comfortable just proxying Solr directly w/o an API > > > > > > > > These situations include: > > > > > > > > 1. I've got no qualms about giving the whole world access to every > > > document > > > > in the index. There's nothing protected about anything. > > > > 2. The content can be easily rebuilt , it's not very large. (I can > > easily > > > > push a button and make a new one) > > > > > > > > Sure you can denial of service Solr, and I might lose my search > index. > > > But > > > > you can denial of service anything. This includes just about anything > > you > > > > put in front of Solr. Moreover, the added complexity of a > > > > Drupal/Wordpress/your API might only add to the security problems > with > > > > their own security issues. I'd rather keep it simple and have fewer > > > moving > > > > parts. > > > > > > > > Cases where I would want an API in front of Solr (these are just the > > > > security ones): > > > > - I want to protect the content (ie based on some notion of a "user" > or > > > > other permissions) > > > > - Rebuilding the content would be very hard and time consuming > > > > > > > > I would also say to expose Solr directly to everyone you probably > > should > > > > know about Solr's bugaboos: > > > > - the lovely qt parameter and the request dispatcher (the nginx proxy > > > below > > > > disallows qt) > > > > - deep paging (prevented by the nginx proxy) > > > > - how to lock down a request handler fairly robustly, how to use > > > invariants > > > > - mitigating intentionally malicious queries (such as the lovely > > "sleep" > > > > function query). > > > > > > > > I'm also curious to hear what the websolr people do, or anyone else > > that > > > > hosts Solr for the JavaScript app development crowd. > > > > > > > > Cheers > > > > -Doug > > > > > > > > > > > > On Friday, December 25, 2015, Shawn Heisey <apa...@elyograg.org > <javascript:;> > > > <javascript:;>> wrote: > > > > > > > > > On 12/25/2015 12:17 PM, Eric Dain wrote: > > > > > > Does allowing javascript direct access to SolrCloud raise > security > > > > > concern? > > > > > > should I build a REST service in between? > > > > > > > > > > > > I need to provide async search capability to web pages. the pages > > > will > > > > be > > > > > > public with no authentication. > > > > > > > > > > End users should never have access to Solr. Access to Solr from > the > > > > > end-user machine is required if you want to accept Solr responses > > > > directly. > > > > > > > > > > In one of the other replies that you received, Doug has given you > an > > > > > nginx config for proxying access to Solr -- indirect access. This > > can > > > > > protect against *changes* to the index, and it has protection > against > > > > > high start/rows values, but there are many other ways that an > > attacker > > > > > can construct denial of service queries, which this proxy config > will > > > > > not prevent. > > > > > > > > > > I think that indirect access (through a proxy) should not be > allowed > > > > > either, unless you can trust all the people that will have access. > > > > > > > > > > If Solr is open to a sufficiently wide audience (especially the > > > > > Internet), someone will find a way to abuse the service even with a > > > > > proxy, either to cause harm or to learn things they shouldn't know. > > > > > > > > > > The most secure option is to only allow the webservers and trusted > > > > > administrators to access Solr. All end user (Internet) access to > > Solr > > > > > should be handled through a custom web application. This might be > > > > > something that you find and install (such as wordpress, drupal, > etc), > > > or > > > > > one that you write yourself. > > > > > > > > > > You can still do AJAX while maintaining security. You'll need to > > write > > > > > something in a server-side web programming language like PHP, Java, > > > etc. > > > > > This code will need to accept the AJAX requests from your > > client-side > > > > > javascript code, validate the request parameters to make sure > they're > > > > > sane, get a response from Solr, and return relevant data. If the > > > > > parameters don't validate, return an error, and handle that error > > > > > appropriately in the javascript code. > > > > > > > > > > Thanks, > > > > > Shawn > > > > > > > > > > > > > > > > > > -- > > > > *Doug Turnbull **| *Search Relevance Consultant | OpenSource > > Connections > > > > <http://opensourceconnections.com>, LLC | 240.476.9983 > > > > Author: Relevant Search <http://manning.com/turnbull> > > > > This e-mail and all contents, including attachments, is considered to > > be > > > > Company Confidential unless explicitly stated otherwise, regardless > > > > of whether attachments are marked as such. > > > > > > > > > > > > > -- > > *Doug Turnbull **| *Search Relevance Consultant | OpenSource Connections > > <http://opensourceconnections.com>, LLC | 240.476.9983 > > Author: Relevant Search <http://manning.com/turnbull> > > This e-mail and all contents, including attachments, is considered to be > > Company Confidential unless explicitly stated otherwise, regardless > > of whether attachments are marked as such. > > > -- *Doug Turnbull **| *Search Relevance Consultant | OpenSource Connections <http://opensourceconnections.com>, LLC | 240.476.9983 Author: Relevant Search <http://manning.com/turnbull> This e-mail and all contents, including attachments, is considered to be Company Confidential unless explicitly stated otherwise, regardless of whether attachments are marked as such.
