Re: [PATCH v2 3/4] btrfs: fix error handling of fallbacked uncompress write

[Filipe Manana <fdmanana@kernel.org>]

On Wed, Jun 22, 2022 at 12:56:25AM +0000, Naohiro Aota wrote:
> On Tue, Jun 21, 2022 at 04:04:14PM +0100, Filipe Manana wrote:
> > On Tue, Jun 21, 2022 at 03:41:01PM +0900, Naohiro Aota wrote:
> > > When cow_file_range() fails in the middle of the allocation loop, it
> > > unlocks the pages but leaves the ordered extents intact. Thus, we need to
> > > call btrfs_cleanup_ordered_extents() to finish the created ordered extents.
> > > 
> > > Also, we need to call end_extent_writepage() if locked_page is available
> > > because btrfs_cleanup_ordered_extents() never process the region on the
> > > locked_page.
> > > 
> > > Furthermore, we need to set the mapping as error if locked_page is
> > > unavailable before unlocking the pages, so that the errno is properly
> > > propagated to the userland.
> > > 
> > > CC: stable@vger.kernel.org # 5.18+
> > > Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
> > > ---
> > > I choose 5.18+ as the target as they are after refactoring and we can apply
> > > the series cleanly. Technically, older versions potentially have the same
> > > issue, but it might not happen actually. So, let's choose easy targets to
> > > apply.
> > > ---
> > >  fs/btrfs/inode.c | 17 +++++++++++++++--
> > >  1 file changed, 15 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> > > index 326150552e57..38d8e6d78e77 100644
> > > --- a/fs/btrfs/inode.c
> > > +++ b/fs/btrfs/inode.c
> > > @@ -933,8 +933,18 @@ static int submit_uncompressed_range(struct btrfs_inode *inode,
> > >  		goto out;
> > >  	}
> > >  	if (ret < 0) {
> > > -		if (locked_page)
> > > +		btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
> > > +		if (locked_page) {
> > > +			u64 page_start = page_offset(locked_page);
> > > +			u64 page_end = page_start + PAGE_SIZE - 1;
> > > +
> > > +			btrfs_page_set_error(inode->root->fs_info, locked_page,
> > > +					     page_start, PAGE_SIZE);
> > > +			set_page_writeback(locked_page);
> > > +			end_page_writeback(locked_page);
> > > +			end_extent_writepage(locked_page, ret, page_start, page_end);
> > >  			unlock_page(locked_page);
> > > +		}
> > >  		goto out;
> > >  	}
> > >  
> > > @@ -1383,9 +1393,12 @@ static noinline int cow_file_range(struct btrfs_inode *inode,
> > >  	 * However, in case of unlock == 0, we still need to unlock the pages
> > >  	 * (except @locked_page) to ensure all the pages are unlocked.
> > >  	 */
> > > -	if (!unlock && orig_start < start)
> > > +	if (!unlock && orig_start < start) {
> > > +		if (!locked_page)
> > > +			mapping_set_error(inode->vfs_inode.i_mapping, ret);
> > 
> > Instead of this we can pass PAGE_SET_ERROR in page_ops, which will result in
> > setting the error in the inode's mapping.
> 
> PAGE_SET_ERROR always set the error as EIO, so it cannot propagate other
> errors returned by cow_file_range() to the userland. Thus, I choose
> mapping_set_error() here.

Ok, actually only EIO and ENOSPC are used as mapping errors. mapping_set_error()
converts any error other than ENOSPC to AS_EIO. And it's been like for many years
as far as I remember.

> 
> > 
> > In fact we currently only mark the locked page with error (at writepage_delalloc()).
> 
> Yes and at submit_uncompressed_range() with this series applied.
> 
> > However all pages before and after it are still locked and we don't set the error on
> > them, I think we should - I don't see why not.
> 
> No, we unlock all pages except locked_pages in the error case, and yes, we

Yes, but by passing PAGE_SET_ERROR, we will end up setting the error bit on the
pages before unlocking them.

> set the error only on the locked_page. So, that will make the other pages
> to be writebacked later again (and make tons of the
> btrfs_writepage_fixup_work). That is common among unlock = 0 case and
> unlock = 1 case. Should we change that?

I don't see why not. But that can be done later in a separate patch, no worries.
Just something I think I didn't notice before.

Ok, the whole patchset looks good to me, thanks.

> 
> Setting all the pages as error would be good to avoid the tons of fixup in
> this case.
> 
> > Did I miss something (again)?
> > 
> > Sorry I only noticed this now.
> > 
> > Thanks.
> > 
> > >  		extent_clear_unlock_delalloc(inode, orig_start, start - 1,
> > >  					     locked_page, 0, page_ops);
> > > +	}
> > >  
> > >  	/*
> > >  	 * For the range (2). If we reserved an extent for our delalloc range
> > > -- 
> > > 2.35.1
> > >