Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Sat, 25 Jan 2020 11:09:51 -0800
From:   Matthew Wilcox <willy@infradead.org>
To:     Gao Xiang <hsiangkao@aol.com>
Cc:     linux-fsdevel@vger.kernel.org, linux-mm@kvack.org,
        linux-erofs@lists.ozlabs.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 07/12] erofs: Convert uncompressed files from readpages
 to readahead
Message-ID: <20200125190951.GN4675@bombadil.infradead.org>
References: <20200125013553.24899-1-willy@infradead.org>
 <20200125013553.24899-8-willy@infradead.org>
 <20200125015323.GA9918@hsiangkao-HP-ZHAN-66-Pro-G1>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200125015323.GA9918@hsiangkao-HP-ZHAN-66-Pro-G1>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Jan 25, 2020 at 09:53:29AM +0800, Gao Xiang wrote:
> > +		/* all the page errors are ignored when readahead */
> > +		if (IS_ERR(bio)) {
> > +			pr_err("%s, readahead error at page %lu of nid %llu\n",
> > +			       __func__, page->index,
> > +			       EROFS_I(mapping->host)->nid);
> >  
> > -				bio = NULL;
> > -			}
> > +			bio = NULL;
> > +			put_page(page);
> 
> Out of curiously, some little question... Why we need put_page(page) twice
> if erofs_read_raw_page returns with error...
> 
> One put_page(page) is used as a temporary reference count for this request,
> we could put_page(page) in advance since pages are still locked before endio.
> 
> Another put_page(page) is used for page cache xarray. I think in this case
> the page has been successfully inserted to the page cache anyway, after erroring
> out it will trigger .readpage again... so probably we need to keep this
> refcount count for page cache xarray?
> 
> If I'm missing something, kindly correct me if I'm wrong....

You're quite right.  After readahead has completed, the page should have
a refcount of 1 and be unlocked.  If we hit an error, the page should
be !uptodate.  It doesn't matter whether we set PageError or not in this
case; filemap_fault() will ClearPageError() before retrying if the page
is !uptodate.  This extra put_page() is wrong, and I'll remove it from
the next version.  Thanks!