Log in
Sebastiaan Andeweg
Seblog.nl

Seblog.nl

Day 30: private webmentions (special valentines edition)

Now that I support both private posts and webmention, it is time to put those two together and send private webmentions. In short, this is a way to notify URLs you mention in your posts, while still maintaining the privacy of the post.

Following the description on Indieweb.org, my private posts now send a HTTP Link header to my new token endpoint. When I send a private mention, I also send a code: a JWT token with the target and source URLs of the webmention and an expiration of 2 minutes. The other side can discover my token endpoint by looking at the Link header, and send the code in a POST request.

The code is then exchanged for an access token, but now with a me, which is set to the URL of the mentioned post, a page, which is my page, and a nonce to mess things up a little. This access token expires in an hour, which is plenty for automated fetching, but might be a bit short for people verifying manual. Normal people can still try and login via IndieAuth though!

However, the mentioned URLs and the whitelisted people are different fields, so they can be different. In fact, they should different: if I mention someone.com/a-post, I might to give someone.com access to the post by logging in, but only if the owner of someone.com/a-post is actually someone.com. In case of twitter.com/url-to-tweet, I don’t want to give twitter.com access. I’m not sure how to solve this yet, so I will do it manual until it hurts.

Also note that I only implemented sending private webmentions today. Please don’t private-mention me now. I will do that part tomorrow.

Now for the Special Valentines Edition™: I just created a post with a u-like-of pointing to a personal site / profile of someone I like. The post is of course private and only I and that person can see it when we log in. I have sent a private webmention for it. The only sad thing is that the other side did not support private webmention and nothing got sent, so I guess it does not really count as my first private mention. Tragic Valentines Day for me so far.

Day 29: responsible disclosures

Yesterday, I found a weakness in Quill, so I notified Aaron and he fixed it today. (Actually, I discovered it today at 1:30, and he fixed it yesterday at 20:30. Timezones are magic!)

On Day 15 of my 100days, I found a vulnerability on my own site. I promised to blog about it, but I actually waited, to give Bastian Allgeier the opportunity to fix his site too. Then I just postponed the writing some more, because life happens. I finally wrote about it today.

This brings my count of responsible disclosures to 2, and I’m a bit proud of that.

How I hacked my own site by feeding it a profile picture via webmention

Two weeks ago I wrote that I hacked my own site. I think it’s important to share how I did it, to make people more aware of possible vulnerabilities, so they can find them too. If others didn’t write about their findings, I wouldn’t have found this one.

I did my best to reach out to people using the same code. If you are using the Kirby Webmentions plugin or my fork of it, please make sure to update!

Webmentions

As some of you may know, my site supports webmentions. In short, this enables me to show replies underneath my posts, that are written by people on their own site. If you write a reply, link to me, mark it up with Microformats and send a webmention, my site fetches your post and shows it as a reply. I use a service called Bridgy to also receive comments from Twitter and Instagram. All of this is automated and very cool.

However, while very cool, it is also potentially dangerous to show external content on your site. The vulnerability I found is an example of what can go wrong.

If you look around on my site, you see I do not only show the content of the reply, but also a picture of the author, if provided. This is especially nice when showing likes:

This nice overview of likes comes from the Kirby Webmentions plugin by Bastian Allgeier, which I modified a bit.

My server takes the fall

In order to protect visitors of my site from other security issues, the plugin downloads the images and shows those downloaded ones. This way my visitors only deal with my server, and not with the servers of everyone who liked my post. It’s a nice service, but it also means that I move the problem: I now have to handle those images with care on my side. My server takes the fall for my visitors.

The problem is: my server just downloads whatever image you give it. In most cases, this will be a nice avatar I can display for my friendly visitors. But one can think of a case where a not-so-friendly visitor feeds my site something else than an image. The plugin of course checks if it’s an image and rejects files that are not a image, but it’s still worth a try.

So, what did you feed it?

Since my server runs on PHP, the nicest thing for an attacker to feed my server is a PHP-file. That way, you can run whatever code you want on my server, doing all kinds of evil things. However, just straight off feeding my site a PHP-file did not work. The plugin is not crazy. It checked wether the MIME of the file was an image of type jpeg, png or gif. It rejected an image.php file like this:

echo "hi!";

Using image.jpg as filename would fail too: the plugin saw that the file had no MIME of an image, so it did not download it. This was the point where I went to bed with a feeling of security: my site was safe and I could not get a php-script in.

The next day, however, I had second thoughts. I needed a real image for my new plan, so I took a screenshot of a smiley. I then opened it in notepad and added the following to the bottom of the file:

<?php mail(‘my.email@gmail.com’, ‘Seb’, ‘hi’);

I then renamed the file to image.php, because you need the PHP-extension in your file to let the server run your code. The last step was disabling PHP on my test-server, to prevent the test-server from executing the code and send mail me. The code just appeared at the end of the image.

I then made a test-post with a u-like-of set to the URL of a post on my blog, and a p-author h-card with an <img src="/photo.php">. It was a like, with an author and an my bad image.

And it worked.

The server sees the image and checks for the MIME, which was image/jpeg, because it was an image. It then downloaded it, including the un-executed PHP string in the bottom of it. It changed the name of the image into the SHA1-hash of the original image-url, but then it appended the extension of the original file, which was .php!

My server then had a file called a266d629bb26d74752080bb1b95bbd0a488bea53.php, which was linked as an image in my post. Every time I refreshed the page, the snippet of code in the bottom of the file got executed, so it sent an e-mail to me.

In this example, I sent an e-mail, but it could’ve been anything.

How to solve?

First off: check your input! And then check it again. A crucial thing for PHP-files is that they get executed if they have the .php extension, so you should not rely on user input for that. Change the filename and change the extension.

Bastian updated the plugin, so now it does not only check for MIME, but also only accepts files with the extensions jpg, jpeg, png and gif. Only if it has a correct extension, it downloads the file, and it checks MIME twice, both before and after the download. I think it’s locked down pretty well, although it still feels a bit scary.

Aaron Parecki, who did this way of showing likes first, uses an external service for his webmention images, and that’s not a bad idea either. If someone manages to get in something bad, it’s not on your the same server as your site. It could also be a good idea to turn off PHP for your upload folder, if you have that kind of access to your server.

Final words

I really like this webmention plugin! It’s thanks to this plugin that I know IndieWeb and all the wonderful things it brings.

But while the plugin and IndieWeb are nice, it’s also good to keep and eye on security. At this moment, webmention is relatively safe because not many people know about it or use it. Although it can be a lot of fun to have a post of a friend automatically show up beneath your post, we have to be aware of the risks of showing content of external parties.

So, be warned, and have fun.

Day 28: Twitter login

Taking yesterdays private post adventure one step further: it’s now also possible to log in to Seblog with Twitter. I still recommend having your own site and use IndieAuth, but I have to be pragmatic: a lot of my friends do not have their own site. (If you want your own site, like this, just give a shout, I can help.)

To make this login adventure a bit more visible, I also added a ‘Log in’ link in the top right corner of every page.

Unfortunately, there is not much going on for people who logged in via Twitter. The private post of yesterday is still only visible for those who were on IRC-people yesterday. But it’s a step!

Gister het gedempte knarsen van voetstappen in de sneeuw. Vandaag het geluid van dooi, als een beekje door het bos. (hè get wat poëtisch)

Day 27: private posts

A few months ago I wrote that I made 'privéstukjes' on my site. The implementation was as simple as one field called 'private' that I would give a value (true for example). If the field has a value, my site returns a 401 Unauthorized header and a page explaining that there is nothing to find there. I did not implement a way to log in, for I only used it for (two) drafts / pieces of bad writing I wanted to keep for myself.

But what is the fun of a private post when nobody can see it?

Today, I implemented a login for my private posts. It was a bit more work than I thought, getting IndieAuth to work and thinking out all the different states of a post, but I made it.

Posts can now also have an 'audience' field, where I keep a comma separated list of URLs that have access to the post. I plan on expanding or replacing that with predefined lists (friends, family, etc.), but for now this works fine.

If one is not logged in, the private post page returns a 401 with a (Dutch) explanation of what is going on. There is also a field to put in your personal URL for an IndieAuth login. If you login, but are not on whitelisted to view the post, you will get a 403 Forbidden, and a Dutch explanation. (Translating my site based on visitor language is still a to do.) If you are on the list, you will see the post.

I have made a test post, so you can try out for yourself. All URLs on IndieWeb's IRC-people page, at the time of writing this, are whitelisted. Have a try at it if you're on that list!

Day 26: RSVP context

Oh no! I almost forgot to work on something IndieWeb today. This was partially the plan, because I also want to focus on some writing (which did not really worked out), but I still wanted to do something small in the evening to keep this thing going. And that part I almost forgot.

So here is a really small change: Tantek was talking about reply contexts for RSVPs, and I said I only show the name of the event, and asked if that counts. And yeah, in a way it counts.

I wanted to add a bit more though, so now I also show the date of the event:

A very minor thing, but hey, it's a thing! I don't really want to add more details to my RSVPs, because I like them small and people are one click away of knowing more about the event anyway.

Day 25: 'better' RSS feed

A blast from the past: I have an RSS feed and it's available at seblog.nl/feed.rss. (I also have seblog.nl/feed, which is actually the h-feed version of it.) I made the RSS feed a while ago for Martijn, who I believe is my only subscriber at the moment.

Back when I made the feed, I didn't want to put in much work for it, because I already have a couple of rich h-feeds. So I made it so that "if a post has a title, give that, else, just call it 'post'" and "if a post has content, give that, else, just say 'A post on Seblog'". Very lazy. It turns out I actually have a lot of posts that have neither a title nor content. Things like my likes, reads and bookmarks all have no title nor content, because it is stored in different fields. So my RSS feed ends up being filled with contentless posts.

I am still lazy about my RSS feed, but I improved it a bit. I now just call the same snippet that produces the HTML for the frontpage for the content of the feed item. This means the whole post is in the RSS feed. Probably not the best way to do it, but hey, who uses RSS nowadays anyway?

@martijn, let me know how it looks!

Meer laden