Message-ID: <20240130104605.2i6mmdncuhwwwfin@quack3>
Date: Tue, 30 Jan 2024 11:46:05 +0100
From: Jan Kara <jack@...e.cz>
To: Christoph Hellwig <hch@....de>
Cc: linux-mm@...ck.org, Matthew Wilcox <willy@...radead.org>,
	Jan Kara <jack@...e.com>, David Howells <dhowells@...hat.com>,
	Brian Foster <bfoster@...hat.com>,
	Christian Brauner <brauner@...nel.org>,
	"Darrick J. Wong" <djwong@...nel.org>, linux-xfs@...r.kernel.org,
	linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 19/19] writeback: simplify writeback iteration

On Thu 25-01-24 09:57:58, Christoph Hellwig wrote:
> Based on the feedback from Jan I've tried to figure out how to
> avoid the error magic in the for_each_writeback_folio.  This patch
> tries to implement this by switching to an open while loop over a
> single writeback_iter() function:
> 
> 	while ((folio = writeback_iter(mapping, wbc, folio, &error))) {
> 		...
> 	}
> 
> the twist here is that the error value is passed by reference, so that
> the iterator can restore it when breaking out of the loop.
> 
> Additionally it moves the AOP_WRITEPAGE_ACTIVATE out of the iterator
> and into the callers, in preparation for eventually killing it off
> with the phase out of write_cache_pages().
> 
> To me this form of the loop feels easier to follow, and it has the
> added advantage that writeback_iter() can actually be nicely used in
> nested loops, which should help with further iterizing the iomap
> writeback code.
> 
> Signed-off-by: Christoph Hellwig <hch@....de>

Looking at it now I'm thinking whether we would not be better off to
completely dump the 'error' argument of writeback_iter() /
writeback_iter_next() and just make all .writepage implementations set
wbc->err directly. But that means touching all the ~20 writepage
implementations we still have...

Couple of comments regarding this implementation below (overall I agree it
seems somewhat easier to follow the code).

> +/**
> + * writepage_iter - iterate folio of a mapping for writeback
> + * @mapping: address space structure to write
> + * @wbc: writeback context
> + * @folio: previously iterated folio (%NULL to start)
> + * @error: in-out pointer for writeback errors (see below)
> + *
> + * This function should be called in a while loop in the ->writepages
> + * implementation and returns the next folio for the writeback operation
> + * described by @wbc on @mapping.
> + *
> + * To start writeback @folio should be passed as NULL, for every following
> + * iteration the folio returned by this function previously should be passed.
> + * @error should contain the error from the previous writeback iteration when
> + * calling writeback_iter.
> + *
> + * Once the writeback described in @wbc has finished, this function will return
> + * %NULL and if there was an error in any iteration restore it to @error.
> + *
> + * Note: callers should not manually break out of the loop using break or goto.
> + */
> +struct folio *writeback_iter(struct address_space *mapping,
> +		struct writeback_control *wbc, struct folio *folio, int *error)
>  {
> -	unsigned long nr = folio_nr_pages(folio);
> +	if (folio) {
> +		wbc->nr_to_write -= folio_nr_pages(folio);
> +		if (*error && !wbc->err)
> +			wbc->err = *error;
>  
> -	wbc->nr_to_write -= nr;
> -
> -	/*
> -	 * Handle the legacy AOP_WRITEPAGE_ACTIVATE magic return value.
> -	 * Eventually all instances should just unlock the folio themselves and
> -	 * return 0;
> -	 */
> -	if (error == AOP_WRITEPAGE_ACTIVATE) {
> -		folio_unlock(folio);
> -		error = 0;
> +		/*
> +		 * For integrity sync  we have to keep going until we have
> +		 * written all the folios we tagged for writeback prior to
> +		 * entering the writeback loop, even if we run past
> +		 * wbc->nr_to_write or encounter errors.
> +		 *
> +		 * This is because the file system may still have state to clear
> +		 * for each folio.  We'll eventually return the first error
> +		 * encountered.
> +		 *
> +		 * For background writeback just push done_index past this folio
> +		 * so that we can just restart where we left off and media
> +		 * errors won't choke writeout for the entire file.
> +		 */
> +		if (wbc->sync_mode == WB_SYNC_NONE &&
> +		    (wbc->err || wbc->nr_to_write <= 0))
> +			goto finish;

I think it would be a bit more comprehensible if we replace the goto with:
			folio_batch_release(&wbc->fbatch);
			if (wbc->range_cyclic)
				mapping->writeback_index =
					folio->index + folio_nr_pages(folio);
			*error = wbc->err;
			return NULL;

> +	} else {
> +		if (wbc->range_cyclic)
> +			wbc->index = mapping->writeback_index; /* prev offset */
> +		else
> +			wbc->index = wbc->range_start >> PAGE_SHIFT;
> +		if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
> +			tag_pages_for_writeback(mapping, wbc->index,
> +					wbc_end(wbc));
> +		folio_batch_init(&wbc->fbatch);
> +		wbc->err = 0;
>  	}
>  
> -	if (error && !wbc->err)
> -		wbc->err = error;
> +	folio = writeback_get_folio(mapping, wbc);
> +	if (!folio)
> +		goto finish;

And here we just need to do:
		if (wbc->range_cyclic)
			mapping->writeback_index = 0;
		*error = wbc->err;
		return NULL;

> +	return folio;
> +
> +finish:
> +	folio_batch_release(&wbc->fbatch);
>  
>  	/*
> -	 * For integrity sync  we have to keep going until we have written all
> -	 * the folios we tagged for writeback prior to entering the writeback
> -	 * loop, even if we run past wbc->nr_to_write or encounter errors.
> -	 * This is because the file system may still have state to clear for
> -	 * each folio.   We'll eventually return the first error encountered.
> +	 * For range cyclic writeback we need to remember where we stopped so
> +	 * that we can continue there next time we are called.  If  we hit the
> +	 * last page and there is more work to be done, wrap back to the start
> +	 * of the file.
>  	 *
> -	 * For background writeback just push done_index past this folio so that
> -	 * we can just restart where we left off and media errors won't choke
> -	 * writeout for the entire file.
> +	 * For non-cyclic writeback we always start looking up at the beginning
> +	 * of the file if we are called again, which can only happen due to
> +	 * -ENOMEM from the file system.
>  	 */
> -	if (wbc->sync_mode == WB_SYNC_NONE &&
> -	    (wbc->err || wbc->nr_to_write <= 0)) {
> -		writeback_finish(mapping, wbc, folio->index + nr);
> -		return NULL;
> +	if (wbc->range_cyclic) {
> +		WARN_ON_ONCE(wbc->sync_mode != WB_SYNC_NONE);
> +		if (wbc->err || wbc->nr_to_write <= 0)
> +			mapping->writeback_index =
> +				folio->index + folio_nr_pages(folio);
> +		else
> +			mapping->writeback_index = 0;
>  	}
> -
> -	return writeback_get_folio(mapping, wbc);
> +	*error = wbc->err;
> +	return NULL;
>  }
>  
>  /**
> @@ -2563,13 +2575,17 @@ static int writeback_use_writepage(struct address_space *mapping,
>  		struct writeback_control *wbc)
>  {
>  	struct blk_plug plug;
> -	struct folio *folio;
> -	int err;
> +	struct folio *folio = 0;
			     ^^ NULL please


> +	int err = 0;
>  
>  	blk_start_plug(&plug);
> -	for_each_writeback_folio(mapping, wbc, folio, err) {
> +	while ((folio = writeback_iter(mapping, wbc, folio, &err))) {
>  		err = mapping->a_ops->writepage(&folio->page, wbc);
>  		mapping_set_error(mapping, err);
> +		if (err == AOP_WRITEPAGE_ACTIVATE) {
> +			folio_unlock(folio);
> +			err = 0;
> +		}
>  	}
>  	blk_finish_plug(&plug);
>  
> @@ -2590,6 +2606,8 @@ int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
>  			ret = mapping->a_ops->writepages(mapping, wbc);
>  		} else if (mapping->a_ops->writepage) {
>  			ret = writeback_use_writepage(mapping, wbc);
> +			if (!ret)
> +				ret = wbc->err;

AFAICT this should not be needed as writeback_iter() made sure wbc->err is
returned when set?

								Honza
-- 
Jan Kara <jack@...e.com>
SUSE Labs, CR

Powered by blists - more mailing lists