linux-kernel - Re: [PATCH 6/6] kill-the-BKL/reiserfs: release the write lock on flush_commit

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 1 May 2009 15:13:41 +0200
From:	Frederic Weisbecker <fweisbec@...il.com>
To:	Ingo Molnar <mingo@...e.hu>
Cc:	LKML <linux-kernel@...r.kernel.org>, Jeff Mahoney <jeffm@...e.com>,
	ReiserFS Development List <reiserfs-devel@...r.kernel.org>,
	Chris Mason <chris.mason@...cle.com>,
	Alexander Beregalov <a.beregalov@...il.com>,
	Alessio Igor Bogani <abogani@...ware.it>,
	Jonathan Corbet <corbet@....net>,
	Alexander Viro <viro@...iv.linux.org.uk>
Subject: Re: [PATCH 6/6] kill-the-BKL/reiserfs: release the write lock on
	flush_commit_list()

On Fri, May 01, 2009 at 07:42:47AM +0200, Ingo Molnar wrote:
> 
> * Frederic Weisbecker <fweisbec@...il.com> wrote:
> 
> > flush_commit_list() uses ll_rw_block() to commit the pending log blocks.
> > ll_rw_block() might sleep, and the bkl was released at this point. Then
> > we can also relax the write lock at this point.
> > 
> > [ Impact: release the reiserfs write lock when it is not needed ]
> > 
> > Cc: Jeff Mahoney <jeffm@...e.com>
> > Cc: Chris Mason <chris.mason@...cle.com>
> > Cc: Alexander Beregalov <a.beregalov@...il.com>
> > Signed-off-by: Frederic Weisbecker <fweisbec@...il.com>
> > ---
> >  fs/reiserfs/journal.c |    7 +++++--
> >  1 files changed, 5 insertions(+), 2 deletions(-)
> > 
> > diff --git a/fs/reiserfs/journal.c b/fs/reiserfs/journal.c
> > index 373d080..b1ebd5a 100644
> > --- a/fs/reiserfs/journal.c
> > +++ b/fs/reiserfs/journal.c
> > @@ -1120,8 +1120,11 @@ static int flush_commit_list(struct super_block *s,
> >  		    SB_ONDISK_JOURNAL_SIZE(s);
> >  		tbh = journal_find_get_block(s, bn);
> >  		if (tbh) {
> > -			if (buffer_dirty(tbh))
> > -			    ll_rw_block(WRITE, 1, &tbh) ;
> > +			if (buffer_dirty(tbh)) {
> > +		            reiserfs_write_unlock(s);
> > +			    ll_rw_block(WRITE, 1, &tbh);
> > +			    reiserfs_write_lock(s);
> > +			}
> >  			put_bh(tbh) ;
> >  		}
> >  	}
> 
> there's 7 other instances of ll_rw_block():
> 
> fs/reiserfs/journal.c-			spin_unlock(lock);
> fs/reiserfs/journal.c:			ll_rw_block(WRITE, 1, &bh);
> fs/reiserfs/journal.c-			spin_lock(lock);
> --
> fs/reiserfs/journal.c-		            reiserfs_write_unlock(s);
> fs/reiserfs/journal.c:			    ll_rw_block(WRITE, 1, &tbh);
> fs/reiserfs/journal.c-			    reiserfs_write_lock(s);
> --
> fs/reiserfs/journal.c-	/* read in the log blocks, memcpy to the corresponding real block */
> fs/reiserfs/journal.c:	ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
> fs/reiserfs/journal.c-	for (i = 0; i < get_desc_trans_len(desc); i++) {
> --
> fs/reiserfs/journal.c-		set_buffer_dirty(real_blocks[i]);
> fs/reiserfs/journal.c:		ll_rw_block(SWRITE, 1, real_blocks + i);
> fs/reiserfs/journal.c-	}
> --
> fs/reiserfs/journal.c-	}
> fs/reiserfs/journal.c:	ll_rw_block(READ, j, bhlist);
> fs/reiserfs/journal.c-	for (i = 1; i < j; i++)
> --
> fs/reiserfs/stree.c-		if (!buffer_uptodate(bh[j]))
> fs/reiserfs/stree.c:			ll_rw_block(READA, 1, bh + j);
> fs/reiserfs/stree.c-		brelse(bh[j]);
> --
> fs/reiserfs/stree.c-						    reada_blocks, reada_count);
> fs/reiserfs/stree.c:			ll_rw_block(READ, 1, &bh);
> fs/reiserfs/stree.c-			reiserfs_write_unlock(sb);
> --
> fs/reiserfs/super.c-{
> fs/reiserfs/super.c:	ll_rw_block(READ, 1, &(SB_BUFFER_WITH_SB(s)));
> fs/reiserfs/super.c-	reiserfs_write_unlock(s);
> 
> in particular the second stree.c one and the super.c has a 
> write-unlock straight before the lock-drop.
> 
> I think the stree.c unlock could be moved to before the 
> ll_rw_block() call straight away.


Indeed.


> 
> The super.c one needs more care: first put &(SB_BUFFER_WITH_SB(s)) 
> into a local variable, then unlock the wite-lock, then call 
> ll_rw_block(). (This is important because &(SB_BUFFER_WITH_SB(s)) is 
> global filesystem state that has to be read with the lock held.)


Indeed &(SB_BUFFER_WITH_SB(s)) is a pointer to blocks that
reflect the state of the filesystem but it was already not
safe on the old code.

ll_rw_block() may sleep, and wait_on_buffer() too. And this
pointer could have changed already during these sleeps.

If we put it in a local variable, it prevents from a change of the
pointer value, but not from its content, like in the older scheme.

And I guess this pointer is unlikely to change, this is about the superblock
and the bitmap...

But I'm not sure. I guess it's indeed better to put it in a local variable.

> ll_rw_block() generally always has a chance to block (especially on 
> READ) - so the other places could be converted to drop the 
> write-lock too. Most seem straightforward - some need similar 
> local-variable treatment as super.c.


Ok, thanks!


> 	Ingo

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/