non-responsive [simple] mount; How to debug?

[Thomas131]

Hi!
I set up a bindfs a few days ago; nothing special. This code on nixos:

  fileSystems."/home/[my2nduser]/mnt" = {
    device = "/";
    fsType = "fuse.bindfs";
    options = [ "map=1000/1001" ];
  };

Which translates to this fstab entry:

/ /home/[my2nduser]/mnt fuse.bindfs map=1000/1001 0 2

My system froze today. I had some unrelated technical difficulties, but I didn't change technical stuff since the last reboot, so that should be unrelated I guess. (Oh for some reason, the root-Folder ("/") was owned by uid 1000 - I hope bindfs wasn't responsible for that.)

On my 2nd user:

• ls ~ froze
• lsof froze
• (sudo) umount froze
• ls --exclude=mnt worked

I found nothing interesting related to fuse or bind in journalctl or dmesg.

I killed the fuser process; made an sudo umount and a mount -a afterwards and things work again.

How to debug such behavior if it happens again so I can do a proper bug report in case it happens again? Thanks!
Thomas131

[mpartel]

It's probably the recursion, leading to a deadlock. To verify, try ls /home/[my2nduser]/mnt/home/[my2nduser]/mnt/

If bindfs ends up "calling itself" like that, it will deadlock, as it can't even begin handling the 2nd request before finishing the 1st one. (And if it didn't immediately deadlock, it would exhaust some kind of active call or memory limit with infinite recursion.)

Bindfs has a basic recursion check when resolving symlinks (--resolve-symlinks, documented on the man page), but I should probably add a warning or something if the mounted directory is inside the source directory 🤔

[Thomas131]

Thanks for pointing that out! I guess that was the error. I leave the issue open for now in case you want to make this limitation more clear or prevent recursive mounts completely.