On Tue, Aug 16, 2022 at 9:11 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> filldir_t instances (directory iterators callbacks) used to return 0 for
> "OK, keep going" or -E... for "stop".
Yeah, and it was really confusing wrt what the reported user space error was.
> So let's just return bool ("should we keep going?") - it's less confusing
> that way.
Ack. And the "true means keep going" that you picked is, I think, the
more natural model with the conceptual model being
while (dir_emit()) ..
and this makes the filldir return value match that dir_emit() one.
That said, our filldir code is still confusing as hell. And I would
really like to see that "shared vs non-shared" iterator thing go away,
with everybody using the shared one - and filesystems that can't deal
with it using their own lock.
But that's a completely independent wart in our complicated filldir saga.
But if somebody were to look at that iterate-vs-iterate_shared, that
would be lovely. A quick grep shows that we don't have *that* many of
the non-shared cases left:
git grep '\.iterate\>.*='
seems to imply that converting them to a "use my own load" wouldn't be
_too_ bad.
And some of them might actually be perfectly ok with the shared
semantics (ie inode->i_rwsem held just for reading) and they just were
never converted originally.
Anybody?
Linus
