From: Jan Kara Subject: Re: [PATCH] ext4: Return error if we fail to allocate block in noalloc_get_block_write Date: Tue, 7 Jul 2009 19:18:07 +0200 Message-ID: <20090707171807.GD24457@duck.suse.cz> References: <1246441575-20311-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com> <20090701095636.GB20540@skywalker> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: cmm@us.ibm.com, tytso@mit.edu, sandeen@redhat.com, Jan Kara , linux-ext4@vger.kernel.org To: "Aneesh Kumar K.V" Return-path: Received: from cantor.suse.de ([195.135.220.2]:42938 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754909AbZGGRSM (ORCPT ); Tue, 7 Jul 2009 13:18:12 -0400 Content-Disposition: inline In-Reply-To: <20090701095636.GB20540@skywalker> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Wed 01-07-09 15:26:36, Aneesh Kumar K.V wrote: > On Wed, Jul 01, 2009 at 03:16:15PM +0530, Aneesh Kumar K.V wrote: > > block_write_full_page consider a zero return from get_block as success. > > noalloc_get_block_write returned zero even if we failed to find a mapping > > blocks. Returning non zero ensures we fallback to the error handling path > > of block_write_full_page which would properly redirty the page after > > the below patch is applied. > > > > http://article.gmane.org/gmane.linux.file-systems/33145 > > > > Signed-off-by: Aneesh Kumar K.V > > --- > > > > NOTE: I am not sure whether -EGAIN is the right error to be returned error. > > This patch should enable us to push the pending ext4 patches in the patch > > queue without depending on the full series from Jan. Will reply to this > > email with patch ordering. > > > > fs/ext4/inode.c | 5 ++++- > > 1 files changed, 4 insertions(+), 1 deletions(-) > > > > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c > > index 25638bc..6c814af 100644 > > --- a/fs/ext4/inode.c > > +++ b/fs/ext4/inode.c > > @@ -2514,7 +2514,10 @@ static int noalloc_get_block_write(struct inode *inode, sector_t iblock, > > if (ret > 0) { > > bh_result->b_size = (ret << inode->i_blkbits); > > ret = 0; > > - } > > + } else if (create && ret == 0) > > + /* write request on unmapped buffer head. */ > > + ret = -EAGAIN; > > + > > return ret; > > } > > > > The patch ordering I tested this with was > + stable-boundary > + stable-boundary-undo.patch > + jan-kara-dont-clear-dirty-bits-in-block_write_full_page > + dont-look-at-buffer_heads-outside-i_size > + fix-mmap-truncate-race-with-subpage-blocksize > + 01.patch <=============================================== This is the above patch > + fix-mmap-truncate-race-with-nondelalloc > + move-__ext4_journaled-writepage-function > + add-WARN_ON-with-unmapped-dirty-bh-in-writepage > + vfs-jan-kara-page_mkwrite-infrastructure > + allocate-blocks-correctly-with-subpage-blocksize > > jan-kara-unmap-underlying-metadata-of-new-buffers-only-when-mapped > > Now with jan-kara-dont-clear-dirty-bits-in-block_write_full_page > block_write_full_page will redirty the page when get_block fails. That > would ensure that we don't drop the pages if we find unmapped > buffer_heads in the writepage callback. The patch > fix-mmap-truncate-race-with-subpage-blocksize actually pass pages with > unmapped buffer heads to block_write_full_page. > > This also means we can push patches upto > move-__ext4_journaled-writepage-function to upstream in 2.6.31 > > I am yet to run the full patch set on ABAT. So the patches had limited > testing. Yes, this looks as a good solution to me since Nick had some objections to later patches in my patchset so they probably won't get merged for 2.6.31... Honza -- Jan Kara SUSE Labs, CR