From: Jan Kara <jack@suse.cz>
Subject: Re: [PATCH RFC] jbd: don't wake kjournald unnecessarily
Date: Wed, 19 Dec 2012 18:11:18 +0100
Message-ID: <20121219171118.GA28042@quack.suse.cz>
References: <50D0A1FD.7040203@redhat.com>
 <20121219012710.GF5987@quack.suse.cz>
 <20121219020526.GG5987@quack.suse.cz>
 <50D12FC3.6090209@redhat.com>
 <20121219081334.GB20163@quack.suse.cz>
 <50D1E16B.6000703@redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Jan Kara <jack@suse.cz>,
	ext4 development <linux-ext4@vger.kernel.org>
To: Eric Sandeen <sandeen@redhat.com>
Content-Disposition: inline
In-Reply-To: <50D1E16B.6000703@redhat.com>
Sender: linux-ext4-owner@vger.kernel.org

On Wed 19-12-12 09:46:51, Eric Sandeen wrote:
> On 12/19/12 2:13 AM, Jan Kara wrote:
> > On Tue 18-12-12 21:08:51, Eric Sandeen wrote:
> >> On 12/18/12 8:05 PM, Jan Kara wrote:
> >>> On Wed 19-12-12 02:27:10, Jan Kara wrote:
> >>>>> With a u8 tid_t, the "else" clause from commit d9b0193 fires
> >>>>> frequently; I really think the underlying problem is that tid_geq()
> >>>>> etc does not properly handle wraparounds - if, say, target is 255
> >>>>> and j_commit_request is 0, we don't know if j_commit_request
> >>>>> is 255 tids behind, or 1 tid ahead.  I have to think about that
> >>>>> some more, unless it's obvious to someone else.
> >>>>   Well, there's no way to handle wraps better AFAICT. Tids eventually wrap
> >>>> and if someone has stored away tid of a transaction he wants committed and
> >>>> keeps it for a long time before using it, it can end up being anywhere
> >>>> before / after current j_commit_request. The hope was that it takes long
> >>>> enough to wrap around 32-bit tids. If this happens often in practice we may
> >>>> have to switch to 64-bit tids (in memory, on disk 32-bit tids are enough
> >>>> because of limited journal size).
> >>
> >> I was wondering if, since the tid_g*() functions only work if the
> >> distance is half the unsigned int space, we can force a commit at some
> >> point if j_transaction_sequence has gotten too far ahead?  I'm not sure
> >> where or if that could be done...
> >   I don't quiete understand. If someone stores tid = transaction->t_tid and
> > in two weeks calls log_start_commit(tid), I don't see how any forcing of
> > commits could solve that tid may now look ahead of the log...
> 
> I'm probably missing something, but I was thinking we could compare
> j_commit_sequence to j_transaction_sequence and force a commit up to at least
> j_commit_sequence if it's too "stale" - but I'm only handwaving. :)
  You are probably missing the fact that
j_transaction_sequence - 2 <= j_commit_sequence <= j_transaction_sequence.
  I.e., we have always one running transaction and at most one committing
transaction which is the previous one. j_commit_sequence is TID of the
transaction which successfully finished commit.

The warnings we are seeing are caused by TIDs stored in
EXT3_I(inode)->i_datasync_tid (and i_sync_tid). Those can get rather old
before they are used.

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR