On Mon 27-04-09 12:38:25, Andrew Morton wrote:
> On Mon, 27 Apr 2009 16:43:48 +0200
> Jan Kara <jack@xxxxxxx> wrote:
> > So far, do_sync() called:
> > sync_inodes(0);
> > sync_supers();
> > sync_filesystems(0);
> > sync_filesystems(1);
> > sync_inodes(1);
>
> The description has me all confused.
>
> > This ordering makes it kind of hard for filesystems as sync_inodes(0) need not
> > submit all the IO (for example it skips inodes with I_SYNC set) so e.g. forcing
> > transaction to disk in ->sync_fs() is not really enough.
>
> Is not really enough for what?
>
> sync_fs(wait==0) is not supposed to be reliable - it's an advice to the
> fs that it should push as much "easy" writeback into the queue as
> possible. We'll do the real sync later, with sync_fs(wait==1).
Yes, but note that after sync_fs(wait==1) we do sync_inodes(wait==1) and
only this last sync_inodes() call is guaranteed to get all the inode data
to disk. So sync_fs() is called *before* all the dirty data are actually
written. That is against expectation of sync_fs() implementation of most
filesystems...
> > Therefore sys_sync has
> > not been completely reliable on some filesystems (ext3, ext4, reiserfs, ocfs2
> > and others are hit by this) when racing e.g. with background writeback.
>
> No sync can ever be reliable in the presence of concurrent write
> activity, unless we freeze userspace.
Of course, but it should be reliable in the presence of pdflush()
flushing dirty data. And it was not currently because even background
writeback sets I_SYNC flag of the inode and sync_inodes(wait==0) skips
these inodes.
This is the real bug this patch is trying to fix, but generally it tries
to make the code more robust so that the reliability of sys_sync() does not
depend on the exact behavior of WB_SYNC_NONE writeback done by
sync_inodes(wait==0).
> > A
> > similar problem hits also other filesystems (e.g. ext2) because of
> > write_supers() being called before the sync_inodes(1).
> >
> > Change the ordering of calls in do_sync() - this requires a new function
> > sync_blkdevs() to preserve the property that block devices are always synced
> > after write_super() / sync_fs() call.
> >
> > The same issue is fixed in __fsync_super() function used on umount /
> > remount read-only.
>
> So it's all a bit unclear (to me) what this patch is trying to fix?
Hopefully explained above ;).
Honza
>
>
> > Signed-off-by: Jan Kara <jack@xxxxxxx>
> > ---
> > fs/super.c | 27 ++++++++++++++++++++++++++-
> > fs/sync.c | 3 ++-
> > include/linux/fs.h | 2 ++
> > 3 files changed, 30 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/super.c b/fs/super.c
> > index 786fe7d..4826540 100644
> > --- a/fs/super.c
> > +++ b/fs/super.c
> > @@ -267,6 +267,7 @@ void __fsync_super(struct super_block *sb)
> > {
> > sync_inodes_sb(sb, 0);
> > vfs_dq_sync(sb);
> > + sync_inodes_sb(sb, 1);
> > lock_super(sb);
> > if (sb->s_dirt && sb->s_op->write_super)
> > sb->s_op->write_super(sb);
> > @@ -274,7 +275,6 @@ void __fsync_super(struct super_block *sb)
> > if (sb->s_op->sync_fs)
> > sb->s_op->sync_fs(sb, 1);
> > sync_blockdev(sb->s_bdev);
> > - sync_inodes_sb(sb, 1);
> > }
> >
> > /*
> > @@ -502,6 +502,31 @@ restart:
> > mutex_unlock(&mutex);
> > }
> >
> > +/*
> > + * Sync all block devices underlying some superblock
> > + */
> > +void sync_blockdevs(void)
> > +{
> > + struct super_block *sb;
> > +
> > + spin_lock(&sb_lock);
> > +restart:
> > + list_for_each_entry(sb, &super_blocks, s_list) {
> > + if (!sb->s_bdev)
> > + continue;
> > + sb->s_count++;
> > + spin_unlock(&sb_lock);
> > + down_read(&sb->s_umount);
> > + if (sb->s_root)
> > + sync_blockdev(sb->s_bdev);
> > + up_read(&sb->s_umount);
> > + spin_lock(&sb_lock);
> > + if (__put_super_and_need_restart(sb))
> > + goto restart;
> > + }
> > + spin_unlock(&sb_lock);
> > +}
>
> The comment doesn't match the implementation. This function syncs all
> blockdevs underlying _all_ superblocks.
>
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
