Created attachment 2046 [details]
python poc for first variant
A malicious or faulty IMAP server can crash claws-mail when it lets the server traverse into indefinitely many subdirectories during rebuild folder tree.
The source of this is relatively obvious: imap_scan_tree_recursive() will call itself recursively without any limit set, which eventually will crash. I recomend to set a reasonable limit of recursion depth (not sure how crazy people plausibly go with imap structures, but I guess a limit at 500 should handle all possibly legit needs).
However while trying to create a reproducer for this I noticed that when terminating the connection after some iterations (I tried with 1000) it will be unresponsive for a while and also cause a stack overflow, however a different one. It will crash somewhere in glib. I haven't analyzed that in more detail, but it seems the rebuild folder tree functionality doesn't detect the connection termination.
I'm attaching test scripts, these are written in python and open an imap server on localhost. Configure an imap account to localhost without tls and do rightclick->"Rebuild folder tree" to reproduce. I'm also attaching ASAN stack traces for both bugs.
Created attachment 2047 [details]
python poc for second variant
Created attachment 2048 [details]
asan crash dump variant 1
Created attachment 2049 [details]
asan crash dump variant 2
And also as Debian bug https://bugs.debian.org/966630
Created attachment 2080 [details]
patch candidate wwp-rev0
Here's a patch candidate, that limits the IMAP rebuild folder tree recursion to 256 (hardcoded, arbitrary value).
Created attachment 2081 [details]
patch candidate wwp-rev1
Created attachment 2082 [details]
patch candidate wwp-rev2
this time with a hidden pref, depth limit set to 64
I have tested the patch and I still get somewhat undesired behavior.
I can confirm that it no longer crashes directly if I run against the poc.
*however* I did some testing where I first ran claws without this patch and then ran an asan build of the patched claws. This resulted in a stack overflow which looks like the variant 2 stack overflow.
What I suspect is happening here is that traversing through the locally cached copy of the folder tree can also lead to a stack overflow.
Now given the patch applied such a situation should no longer be possible to occur. However I guess it would also be desirable to avoid crashing on bad local imap data.
You can also cause a stack overflow by using deeply nested local directories. Simply create a deeply nested dir structure in a local mail directory:
for i in $(seq 1 1000); do mkdir a; cd a; done
Then do a "Rebuild Folder Tree".
So, is 3.17.7 accepted as a fix for CVE-2020-16094 or not?
See the Status of this bug.
IOW: yes. (but who is doing the accepting in your question?)