On Fri, Jan 24, 2025 at 10:22 AM Bernd Schubert
<bernd.schubert@xxxxxxxxxxx> wrote:
>
> Hi Joanne,
>
> On 1/24/25 12:30, Bernd Schubert wrote:
> > Hmm, would only need to check head? Oh I see it, we need to use
> > list_move_tail().
>
>
> how about the attached updated patch, which uses
> list_first_entry_or_null()? It also changes from list_move()
> to list_move_tail() so that oldest entry is always on top.
> I didn't give it any testing, though.
Woah that's cool, I didn't know you could send attachments over the
mailing list.
Ah I didn't realize list_move doesn't already by default add to the
tail of the list - thanks for catching that, yes those should be
list_move_tail() then.
In t he attached patch, I think we still need the original
ent_list_request_expired() logic:
static bool ent_list_request_expired(struct fuse_conn *fc, struct
list_head *list)
{
struct fuse_ring_ent *ent;
struct fuse_req *req;
list_for_each_entry(ent, list, list) {
req = ent->fuse_req;
if (req)
return time_is_before_jiffies(req->create_time +
fc->timeout.req_timeout);
}
return false;
}
and we can't assume req is non-NULL. For entries that have been
committed, their ->req is set to NULL but they are still on the
ent_commit_queue.
Thanks,
Joanne
>
> This is on top of the other fuse-uring updates I had sent
> out about an hour ago.
>
>
> Here is the corresponding branch
> https://github.com/bsbernd/linux/tree/optimize-fuse-uring-req-timeouts
>
>
> Thanks,
> Bernd
