Date:   Tue, 4 Jan 2022 17:02:00 +0800
From:   Gao Xiang <hsiangkao@linux.alibaba.com>
To:     Chao Yu <chao@kernel.org>
Cc:     linux-erofs@lists.ozlabs.org, Liu Bo <bo.liu@linux.alibaba.com>,
        LKML <linux-kernel@vger.kernel.org>, Yue Hu <huyue2@yulong.com>
Subject: Re: [PATCH v2 5/5] erofs: use meta buffers for zmap operations
Message-ID: <YdQNCJzQULVxC2QC@B-P7TQMD6M-0146.local>
References: <20220102040017.51352-1-hsiangkao@linux.alibaba.com>
 <20220102040017.51352-6-hsiangkao@linux.alibaba.com>
 <5ed798da-4f01-17d4-cba2-dda50728bd25@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <5ed798da-4f01-17d4-cba2-dda50728bd25@kernel.org>
Precedence: bulk

Hi Chao,

On Tue, Jan 04, 2022 at 04:15:05PM +0800, Chao Yu wrote:
> On 2022/1/2 12:00, Gao Xiang wrote:

...

> > --- a/fs/erofs/zdata.c
> > +++ b/fs/erofs/zdata.c
> > @@ -698,20 +698,18 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
> >   		goto err_out;
> >   	if (z_erofs_is_inline_pcluster(clt->pcl)) {
> > -		struct page *mpage;
> > +		void *mp;
> > -		mpage = erofs_get_meta_page(inode->i_sb,
> > -					    erofs_blknr(map->m_pa));
> > -		if (IS_ERR(mpage)) {
> > -			err = PTR_ERR(mpage);
> > +		mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb,
> > +					erofs_blknr(map->m_pa), EROFS_NO_KMAP);
> > +		if (IS_ERR(mp)) {
> > +			err = PTR_ERR(mp);
> >   			erofs_err(inode->i_sb,
> >   				  "failed to get inline page, err %d", err);
> >   			goto err_out;
> >   		}
> > -		/* TODO: new subpage feature will get rid of it */
> > -		unlock_page(mpage);
> > -
> > -		WRITE_ONCE(clt->pcl->compressed_pages[0], mpage);
> > +		get_page(fe->map.buf.page);
> 
> Comparing to previous implementation, it adds an extra reference on the page, why?

Thanks for the question. Previously, erofs_get_meta_page was called
independently without reusing zmap mpage, so the page refcount had no
relationship with zmap mpage.

However, now we reuse zmap metabuf instead(fe->map.buf), so an extra
page refcount is needed since zmap metabuf will be released at the end
of readpage or readahead...

Thanks,
Gao Xiang