On Wed, Jan 31, 2024 at 05:02:25PM +1100, Dave Chinner wrote: > This may not be true for other filesystems, but I don't think we > can really say "page faults in any filesystem transaction are unsafe > and should be banned".... I think the point is page faults with current->journal_info set is unsafe, as the can recurse into another file system using it. If we don't set current->journal_info (and your ideas for that sound sensible to me), the question of page faults in transactions is decoupled from that. We just need to ensure we never recurse into a transaction in the same or a dependent fs, which ot me suggest we'd better avoid it if we easily can.
