> And this thread is inside the fallocate handler. > > Can you try if the hack below makes the problem go away? > > + current->flags |= PF_FSTRANS; that seems to do the trick. I will test it some more, but so far it has already survived a workload that usually triggered the deadlock quite reliably.