Talk:Cross-site request forgery
|This article is of interest to the following WikiProjects:|
|This article is substantially duplicated by a piece in an external publication. Please do not flag this article as a copyright violation of the following source:
- 1 High-Impact CSRF Examples
- 2 GET vs POST
- 3 CAPTCHAs
- 4 Gmail exploit
- 5 HTTPS
- 6 Name Fragmentation / Dilution
- 7 Prevention by validating referrers?
- 8 Alice vs. Eve
- 9 Browsers restricting content to same IP/Domain
- 10 Semiprotection
- 11 Ajax only?
- 12 Help citing
- 13 External links
- 14 Today on Slashdot
- 15 "known ... since the 1990s"?
- 16 What about this?
- 17 Example and Prevention merit elaboration
- 18 Use different browsers
- 19 Intra-site request forgery
- 20 Django?
- 22 Incorrect Example
- 23 Origin header safety
High-Impact CSRF Examples
Hey guys. In 2008 I published a technical report describing CSRF vulnerabilities in four popular sites (ING Direct, YouTube, The NYTimes and Metafilter). The ING Direct attack was particularly interesting because it allowed money to be transferred out of a bank account using a pure CSRF attack (which is the first such attack of which I'm aware). This CSRF article is strong as-is, but might be stronger if the main example (which uses a bank) were a real example and not a toy example. Regardless, I felt uncomfortable changing the text of the article to include content I've written, so I've decided to simply point to the paper and blog post here--use it as you wish.
GET vs POST
"Also, making the server accept POST requests only instead of GET requests will make such attacks harder."
NeilFraser 23:27, 5 April 2006 (UTC)
Olekasper 08:10, 10 August 2006 (UTC)
If you close the XSS hole on your site then either one of the above methods will close the XSRF hole.
I'm sure that this needs to be cited in some source somewhere. Anybody know of one?
Please correct me if I'm wrong.
CAPTCHAs create major usability issues and depending on the implementation, may still be vulnerable to CSRF attacks using iframes. CAPTCHAs are better at solving other problems, like bots and spam. Hidden form fields are the standard CSRF defense, and very effective assuming the browser is free of vulnerabilities and the site is free of cross-site scripting vulnerabilities. (These assumptions are reasonable because if they were not true, the attacker is unlikely to bother with CSRF as there are much easier attacks.)
Rulesdoc 20:57, 18 October 2006 (UTC)
"A common approach is to include a secret, user-specific token in forms that is verified in addition to the cookie. This approach doesn't save you against the most malicious of attackers, as the attacker can use XMLHttpRequest to grab the one-time token in the background and then stuff it into the POST."
- As written, this is misleading. XHR is subject to the same-domain security model, therefore this particular scenario is only possible in combination with XSS.
Is the Gmail exploit from 1 January 2007 an accurate example of a cross-site request forgery? It doesn't match the description of CSRF from this article (there are no side effects of showing contacts list, the user is not authorizing an action). It doesn't fit neatly into the XSS category either. I'm removing the reference for now but please add it back if I'm mistaken. Rulesdoc 07:14, 2 January 2007 (UTC)
I think it is CSRF, because it's clearly not XSS, and surely theft of a contact list is a side effect. You are right that it doesn't change server state, but I think that like XSS, CSRF has a number of different twists and turns. I've added back in the later paragraphs which had nothing to do with GMail and where about an alternate protection method (double-submit-cookies). Should I add the GMail stuff back in again? JoeWalker 22:48, 2 January 2007 (UTC)
Sorry about reverting away your other edits. I'd like to get some more opinions on the name for the Gmail exploit. It seems like this is an example of exploiting an inadequate JSON authentication scheme, neither CSRF or XSS. Normally in CSRF, you don't have access to the response (and don't need access to it), it's the request that matters -- but for the Gmail exploit, the response is really important. If the contacts list data were in XML instead of JSON, there would be no attack. Rulesdoc 08:19, 5 January 2007 (UTC)
184.108.40.206, while it's true that HTTPS is a useful tool for web applications, the problems that it solves are orthogonal to CSRF attacks. HTTPS is unnecessary or impossible for many applications (e.g. router web administration tools), yet these are often the apps most vulnerable to CSRF and easily protected by the techniques described in this article. Discussion of HTTPS would be better suited for an article on general best practices for web security. Rulesdoc 21:44, 5 January 2007 (UTC)
Name Fragmentation / Dilution
(Update: This change has been made.)
Prevention by validating referrers?
Couldn't this issue be prevented by validating that the rerrer is from the same site? (i know the referrer can be faked, but you would have to get your victim to do this...)
- Forging a victim's headers is possible with Flash: http://webappsec.org/lists/websecurity/archive/2006-07/msg00069.html —The preceding unsigned comment was added by Shiflett (talk • contribs) 13:09, 17 April 2007 (UTC).
In response to the above comment by 220.127.116.11: HTTP_REFERER is a purely optional header, so requiring it to be present by a web application is going to result in false positives for some legitimate users. For example it is sometimes blocked by firewalls as part of a privacy protection policy. --Andrew.urquhart 19:00, 28 June 2007 (UTC)
Referrers get suppressed for all sorts of legitimate reasons (e.g. connections over HTTPS). Why use referrers when hidden fields are known to work reliably? Rulesdoc 00:21, 30 July 2007 (UTC)
I'd like to question whether the statement "The attacker must target either a site that doesn't check the “Referer” header (which is common) or a victim with a browser or plugin bug that allows referrer spoofing (which is rare)." should be there at all. It is totally irrelevant, because referrer header checking is flawed, due to the above reasons, particularly as the referrer header is never submitted with HTTPS requests, so any site that was secure enough to support HTTPS, if it checked the referrer, would block all requests. —Preceding unsigned comment added by 18.104.22.168 (talk) 11:33, 17 June 2010 (UTC)
- I'd like to question this statement as well. Referer checking is a terrible defense at best unless you're going to block requests with no referer as well which would break a very significant number of legitimate clients. The only practical and meaningful answer is to actually fix it on your side with a secret token or re-prompt for credentials for every meaningful action. Anyone object to removing this limitation line? — Preceding unsigned comment added by 22.214.171.124 (talk) 23:00, 1 July 2011 (UTC)
- If this statement is kept, it should be modified to clarify whether it's common for the attacker to target a site or for a site to not check the "Referrer" header. Same goes for clarifying what is rare in the second part of the sentence. Rdfiasco (talk) 20:06, 22 May 2013 (UTC)
Alice vs. Eve
Isn't the name of the attacker usually Eve and not Alice, like it appears to be in the first couple of paragraphs? Wppds 16:29, 3 April 2007 (UTC)
- See Alice and Bob for standard security archetypes. You are right in that Alice is usually the victim, not attacker. However Eve is typically a passive attacker (observing only). So in this case a different player such as Mallory or Dave may be more appropriate. - Dmeranda 18:43, 3 April 2007 (UTC)
Browsers restricting content to same IP/Domain
From the article:
It is also suggested that browsers should isolate content on a page to other content from the same IP, or at least the same domain, by default. This would break very little of the web and solve the problem.
I take this comment to task, specifically the "break very little of the web" part: many sites use advertisements served from other IPs and domains. If browsers disabled this, these sites would have to go back to using frames to include ads from an ad server.
I am not saying whether advertisements are good or bad, just that many sites use them and browsers that "isolate content on a page to other content from the same IP or...the same domain" would break more than "very little of the web".
Also, many sites such as blogs don't host blogger provided images...they force the users to use image hosting sites such as Photo Bucket and include an img tag pointing to the outside site.
A better solution (than blocking content from outside sites) would be for the browser to not include cookies on requests to other domains (or IP addresses) while collecting content to display on this page. This stops the sending of the user's identity/authentication to the outside site, without stopping the site the user is visiting from including content from an outside site. Sorry, I don't have a Wikipedia account - Corey 01:14, 3 July 2007 (UTC)
This is starting to sound like original research (see WP:NOR). Wikipedia is not the right place for new ideas about browser redesign to make their initial debut. I've taken the discussion of browser changes out of the main article until there is a citable source for this proposal. Rulesdoc 03:21, 30 July 2007 (UTC)
We don't need to use TOR we use the IP address of each and every surfer that visit our testpage. You know stumbleupon? good, they just sent me 100K. All I need to do is redirect those stumblers to an iframe embedded form and Wikipedia will be littered with fake edits...
Article: "An alternate method is to "double submit" cookies. This method only works with Ajax based requests"
Hi. Cross-site scripting was tagged "refimprove" in February and recently cited top to bottom and needs only a couple more refs. This article looks easier to do if only because it is shorter. I am not an expert on the topic but unless there are objections am going to try, and if anyone would like to help the more the merrier. I think it is worthwhile and both articles are sources for a W3C Working Group Note which in turn is a reference for this note and work in progress. Apologies in advance if this one also turns into a long series of tiny edits by the way (XSS took about two weeks, twice what I estimated). —SusanLesch (talk) 07:45, 11 June 2008 (UTC)
Today on Slashdot
"known ... since the 1990s"?
The footnote for the phrase "known and in some cases exploited since the 1990s" mentions no attacks prior to 2005. The phrase was coined in 2001 (http://www.tux.org/~peterw/csrf.txt. http://seclists.org/bugtraq/2001/Jun/0217.html, etc.), so I don't see how that footnote supports the contention that CSRF was known and exploited since the 1990s. Sure, the problem has been around since the 1990s, but I'm not aware of anyone recognizing it before 2001. —Preceding unsigned comment added by 126.96.36.199 (talk) 13:45, 30 September 2008 (UTC)
What about this?
1. Attacker loads www.server.com/page1.php to a hidden iframe. [with the GET method]
2. The page contains a form WITH THE SECRET, that can be submitted to the server.
Step 2 can be reiterated to go through all the secret-protected forms.
I'd appreciate a correction if I am mistaken. If I'm not, however, it might be a good idea to indicate this behavior on the page.
Wadim.grasza (talk) 17:45, 21 October 2008 (UTC)
Example and Prevention merit elaboration
I have been trying to grok CSRF for a while now; the first time I tried to learn it from this page, I failed. Somebody just explained it to me, so I went back to check whether his explanation agrees with this article. It does, but the article is too brief. It suffers from the "we omit the subscript when it's clear" problem, for me. I hope to elaborate the Example and Prevention sections to be clear about which parties with which motivations take which actions. See also How to evaluate Web Applications security designs? written in preparation for a meeting next week. DanConnolly (talk) 16:36, 5 December 2008 (UTC)
Use different browsers
I use Firefox bundled with NoScript and CookieSafe to do general browsing and Galeon to do my banking - note that Galeon is configured to go through a proxy (Squid) that allows only a limited number of trusted sites to be accessed (allow list) - when I'm done banking I close the browser. Since CSRF risk is high and there is little we can do right now, I prefer this approach just to be on the safe side. —Preceding unsigned comment added by 188.8.131.52 (talk) 16:18, 30 June 2009 (UTC)
Intra-site request forgery
The article says:
Verifying that the request's header contains a X-Requested-With. This protection used by Ruby on Rails and Django (Web framework) has been proven unsecure under a combination of browser plugins and redirects which can allow an attacker to provide custom HTTP headers on a request to any website, hence allow a forged requests
Somebody who understands this stuff better than I do should check me on this, but I think it's correct to say:
...This protection used by Ruby on Rails and Django (Web framework) prior to version 1.2.5, has been proven unsecure ...
In the "Background" section, the following is used as an example of a CSRF:
"Customers of a bank in Mexico were attacked in early 2008 with an image tag in email. The link in the image tag changed the DNS entry for the bank in their ADSL router to point to a malicious website impersonating the bank."
This is in fact not a CSRF attack, but a DNS cache poisoning attack. I didn't want to outright delete a cited "example," but this should probably be taken out and replaced with a correct and more detailed example.
Alternatively, it's possible that the malicious website which was injected into the DNS cache may have used a CSRF attack by forwarding users' credentials to the real site when they "log in" on the fake one, then submitting a malicious request once users were known to be logged in for real. If this is the case, this aspect of the attack should be made more clear. Right now the entry is confusing and unrelated to the topic. — Preceding unsigned comment added by 184.108.40.206 (talk) 14:43, 21 June 2012 (UTC)
Origin header safety
On 11 September 2012, Hydrox has changed
"Verifying that the request's header contains a X-Requested-With […] has been proven unsecure"
"Verifying that the request's header contains a X-Requested-With or checking the HTTP Referer header and/or HTTP Origin […] has been proven unsecure"
But i don't see in the 2 links provided where the Origin header is mentioned. And other good sites on security like https://code.google.com/p/browsersec/wiki/Part3#Origin_headers and https://code.google.com/p/html5security/wiki/CrossOriginRequestSecurity doesn't say anything about a Origin header spoofing.
Is there a real source to substantiate this claim ? Or is it infered from the sentence of the django blog "can allow an attacker to provide custom HTTP headers on a request to any website". Is this case, it depends how you define "custom". Is it "you can customize all http headers" or "you can add a custom http header", meaning headers not handled by the browsers like all the X-... (X-Requested-With).