Re: [PATCH] rxrpc_find_service_conn_rcu: use read_seqbegin() rather than read_seqbegin_or_lock()

[Oleg Nesterov]

On 11/01, Al Viro wrote:
>
> On Wed, Nov 01, 2023 at 09:23:03PM +0100, Oleg Nesterov wrote:
>
> > Yes this is confusing. Again, even the documentation is wrong! That is why
> > I am trying to remove the misuse of read_seqbegin_or_lock(), then I am going
> > to change the semantics of need_seqretry() to enforce the locking on the 2nd
> > pass.
>
> What for?  Sure, documentation needs to be fixed,

So do you agree that the current usage of read_seqbegin_or_lock() in
rxrpc_find_service_conn_rcu() is misleading ? Do you agree it can use
read_seqbegin/read_seqretry without changing the current behaviour?

> but *not* in direction you
> suggested in that patch.

Hmm. then how do you think the doc should be changed? To describe the
current behaviour.

> Why would you want to force that "switch to locked on the second pass" policy
> on every possible caller?

Because this is what (I think) read_seqbegin_or_lock() is supposed to do.
It should take the lock for writing if the lockless access failed. At least
according to the documentation.

This needs another discussion and perhaps this makes no sense. But I'd
like to turn need_seqretry(seq) into something like

	static inline int need_seqretry(seqlock_t *lock, int *seq)
	{
		int ret = !(*seq & 1) && read_seqretry(lock, *seq);

		if (ret)
			*seq = 1; /* make this counter odd */

		return ret;
	}

and update the users which actually want read_seqlock_excl() on the 2nd pass.
thread_group_cputime(), fs/d_path.c and fs/dcache.c.

For example, __dentry_path()

	--- a/fs/d_path.c
	+++ b/fs/d_path.c
	@@ -349,10 +349,9 @@ static char *__dentry_path(const struct dentry *d, struct prepend_buffer *p)
		}
		if (!(seq & 1))
			rcu_read_unlock();
	-	if (need_seqretry(&rename_lock, seq)) {
	-		seq = 1;
	+	if (need_seqretry(&rename_lock, &seq))
			goto restart;
	-	}
	+
		done_seqretry(&rename_lock, seq);
		if (b.len == p->len)
			prepend_char(&b, '/');


but again, this need another discussion.

Oleg.