The article doesn't reveal much, but judging from the list of vulnerable coders and the fact that the Imagemagick forum post references the delegates.xml file, I imagine the exploit allows users to create their own delegate [0], then upload an image that causes that delegate to be called. My guess is that the exploit is two parts: 1) appending a line to the delegates.xml file by exploiting a vulnerable coder, and 2) exploiting the custom delegate you create.
Delegates seem very dangerous and there's all sorts of ways to exploit a custom delegate.
I'm surprised we haven't seen more Imagemagick vulnerabilities... it's a really old library with a massive codebase supporting tons of file formats and arbitrary command chaining. And it's often called by higher level languages via functions like `system()` or `shell()` where developers could easily neglect to "properly sanitize" user inputs, since what those "inputs" can be is so wide ranging.
"Sanitize input filename for http / https delegates"
So presumably the attack is to set the "filename" property of a file such that when including '%f' in a command, the filename breaks the shell command and inserts its own arbitrary code.
1) Create file with name e.g. evilserver.com/funnycat';rm-rf/;.gif hosted on remote server
2) Somehow instruct vulnerable imagemagick install to download evilserver.com/funnycat';rm-rf/;.gif
3) Assume 99% chance that the render command on server will include %f, which is unsanitized, and will cause rm -rf / to execute
I'm a bit confused on step (2)... how do you tell imagemagick to download a file? Will this only affect installations that are explicitly downloading untrusted URLs from users? Like if imgur "upload via URL" fed the URL directly to imagemagick, it would be vulnerable...