Date: Sun, 5 Apr 2015 22:54:46 +0100
From: Al Viro <viro@ZenIV.linux.org.uk>
To: Dmitry Monakhov <dmonakhov@openvz.org>
Cc: linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH 02/16] vfs: check kiocb->ki_flags instead filp->fl_flags
Message-ID: <20150405215446.GK889@ZenIV.linux.org.uk>
References: <1428174805-853-1-git-send-email-dmonakhov@openvz.org>
 <1428174805-853-3-git-send-email-dmonakhov@openvz.org>
 <20150404213621.GH889@ZenIV.linux.org.uk>
 <871tjy96rp.fsf@openvz.org>
 <20150405181145.GJ889@ZenIV.linux.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20150405181145.GJ889@ZenIV.linux.org.uk>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2025
Lines: 45

On Sun, Apr 05, 2015 at 07:11:45PM +0100, Al Viro wrote:
> On Sun, Apr 05, 2015 at 02:03:22PM +0300, Dmitry Monakhov wrote:
> 
> > I'm not sure I have get your point about ocfs2 because it does
> > iov_iter_truncate() right after generic_write_checks()
> 
> This
>         ret = ocfs2_prepare_inode_for_write(file, ppos, count, appending,
>                                             &can_do_direct, &has_refcount);
> being done before generic_write_checks().  It actually duplicates some
> parts of generic_write_checks() inside (O_APPEND-related, and AFAICS
> they _are_ triggered twice that way).

XFS seems to be buggered as well:
        /* DIO must be aligned to device logical sector size */
        if ((pos | count) & target->bt_logical_sectormask)
                return -EINVAL;

        /* "unaligned" here means not aligned to a filesystem block */
        if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
                unaligned_io = 1;
...
        ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);

now, play with rlimit() and suddenly the alignment checks above have nothing
to do with what'll actually happen after that sucker - it's calling
generic_write_checks(), so...

Incidentally, we want the result of alignment check to decide how to take
the lock that protects the file size, so simply lifting O_APPEND treatment
above those won't do.  I suspect that in case of lock taken shared we
need to redo alignment checks and treat "it became unaligned" as "unlock
and redo it with lock taken exclusive".

BTW, xfs_break_layouts() having dropped and regained lock would invalidate
the O_APPEND treatment in generic_write_checks() just prior (both in
xfs_file_aio_write_checks())...

Al "really not fond of xfs_rw_ilock()" Viro...

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