Re: [GIT PULL] MM updates for 6.3-rc1

From: David Howells <dhowells@redhat.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: dhowells@redhat.com, Vishal Moola <vishal.moola@gmail.com>,
	Steve French <stfrench@microsoft.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Jan Kara <jack@suse.cz>, Paulo Alcantara <pc@cjr.nz>,
	Matthew Wilcox <willy@infradead.org>,
	Huang Ying <ying.huang@intel.com>,
	Baolin Wang <baolin.wang@linux.alibaba.com>,
	Xin Hao <xhao@linux.alibaba.com>,
	linux-mm@kvack.org, mm-commits@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [GIT PULL] MM updates for 6.3-rc1
Date: Fri, 24 Feb 2023 12:12:18 +0000	[thread overview]
Message-ID: <2134430.1677240738@warthog.procyon.org.uk> (raw)
In-Reply-To: <2009825.1677229488@warthog.procyon.org.uk>

The problem appears to be here:

			if (folio_mapping(folio) != mapping ||
			    !folio_test_dirty(folio)) {
				folio_unlock(folio);
				goto skip_write;
			}

in my version it's:

			if (folio->mapping != mapping ||
			    !folio_test_dirty(folio)) {
				start += folio_size(folio);
				folio_unlock(folio);
				continue;
			}

In your version, skip_write doesn't advance start if it too many skips occur.
Changing your version to match fixes the problem - or, at least, the
symptoms.  I'm not sure exactly why it's occurring, but the -EBADF (-9) is
coming from cifs_get_writable_file() not finding an open file.  I think this
is a Steve question.

With Vishal's change to use filemap_get_folios_tag() instead of
find_get_pages_range_tag(), the most common file write case (open, write,
write, ..., write, close) in which all the data is added to the pagecache in
one contiguous lump without seeking, hits this a lot.

Unfortunately, unlike find_get_pages_range_tag(), filemap_get_folios_tag()
doesn't allow you to set a limit on the number of pages it will return.

David