[PATCH v2 1/3] erofs: iomap support for non-tailpacking DIO

[Gao Xiang]

Hi Chao,

On Wed, Aug 04, 2021 at 10:57:08AM +0800, Chao Yu wrote:
> On 2021/7/31 3:46, Gao Xiang wrote:

...

> >   }
> > +static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
> > +		unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
> > +{
> > +	int ret;
> > +	struct erofs_map_blocks map;
> > +
> > +	map.m_la = offset;
> > +	map.m_llen = length;
> > +
> > +	ret = erofs_map_blocks_flatmode(inode, &map, EROFS_GET_BLOCKS_RAW);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	iomap->bdev = inode->i_sb->s_bdev;
> > +	iomap->offset = map.m_la;
> > +	iomap->length = map.m_llen;
> > +	iomap->flags = 0;
> > +
> > +	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
> > +		iomap->type = IOMAP_HOLE;
> > +		iomap->addr = IOMAP_NULL_ADDR;
> > +		if (!iomap->length)
> > +			iomap->length = length;
> 
> This only happens for the case offset exceeds isize?

Thanks for the review.

Yeah, this is a convention (length 0 with !EROFS_MAP_MAPPED) for post-EOF
in erofs_map_blocks_flatmode(), need to follow iomap rule as well.

> 
> > +		return 0;
> > +	}
> > +
> > +	/* that shouldn't happen for now */
> > +	if (map.m_flags & EROFS_MAP_META) {
> > +		DBG_BUGON(1);
> > +		return -ENOTBLK;
> > +	}
> > +	iomap->type = IOMAP_MAPPED;
> > +	iomap->addr = map.m_pa;
> > +	return 0;
> > +}
> > +
> > +const struct iomap_ops erofs_iomap_ops = {
> > +	.iomap_begin = erofs_iomap_begin,
> > +};
> > +
> > +static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
> > +{
> > +	struct inode *inode = file_inode(iocb->ki_filp);
> > +	loff_t align = iocb->ki_pos | iov_iter_count(to) |
> > +		iov_iter_alignment(to);
> > +	struct block_device *bdev = inode->i_sb->s_bdev;
> > +	unsigned int blksize_mask;
> > +
> > +	if (bdev)
> > +		blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1;
> > +	else
> > +		blksize_mask = (1 << inode->i_blkbits) - 1;
> > +
> > +	if (align & blksize_mask)
> > +		return -EINVAL;
> > +
> > +	/*
> > +	 * Temporarily fall back tail-packing inline to buffered I/O instead
> > +	 * since tail-packing inline support relies on an iomap core update.
> > +	 */
> > +	if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE &&
> > +	    iocb->ki_pos + iov_iter_count(to) >
> > +			rounddown(inode->i_size, EROFS_BLKSIZ))
> > +		return 1;
> > +	return 0;
> > +}
> > +
> > +static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
> > +{
> > +	/* no need taking (shared) inode lock since it's a ro filesystem */
> > +	if (!iov_iter_count(to))
> > +		return 0;
> > +
> > +	if (iocb->ki_flags & IOCB_DIRECT) {
> > +		int err = erofs_prepare_dio(iocb, to);
> > +
> > +		if (!err)
> > +			return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
> > +					    NULL, 0);
> > +		if (err < 0)
> > +			return err;
> > +		/*
> > +		 * Fallback to buffered I/O if the operation being performed on
> > +		 * the inode is not supported by direct I/O. The IOCB_DIRECT
> > +		 * flag needs to be cleared here in order to ensure that the
> > +		 * direct I/O path within generic_file_read_iter() is not
> > +		 * taken.
> > +		 */
> > +		iocb->ki_flags &= ~IOCB_DIRECT;
> > +	}
> > +	return generic_file_read_iter(iocb, to);
> 
> It looks it's fine to call filemap_read() directly since above codes have
> covered DIO case, then we don't need to change iocb->ki_flags flag, it's
> minor though.

Yeah, we could use filemap_read() here instead. yet IMO, it might be
better to drop IOCB_DIRECT too to keep iocb consistent with the real
semantics (even it's not used internally.)

> 
> > +}
> > +
> >   /* for uncompressed (aligned) files and raw access for other files */
> >   const struct address_space_operations erofs_raw_access_aops = {
> >   	.readpage = erofs_raw_access_readpage,
> >   	.readahead = erofs_raw_access_readahead,
> >   	.bmap = erofs_bmap,
> > +	.direct_IO = noop_direct_IO,
> > +};
> > +
> > +const struct file_operations erofs_file_fops = {
> > +	.llseek		= generic_file_llseek,
> > +	.read_iter	= erofs_file_read_iter,
> > +	.mmap		= generic_file_readonly_mmap,
> > +	.splice_read	= generic_file_splice_read,
> >   };
> > diff --git a/fs/erofs/inode.c b/fs/erofs/inode.c
> > index aa8a0d770ba3..00edb7562fea 100644
> > --- a/fs/erofs/inode.c
> > +++ b/fs/erofs/inode.c
> > @@ -247,7 +247,10 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
> >   	switch (inode->i_mode & S_IFMT) {
> >   	case S_IFREG:
> >   		inode->i_op = &erofs_generic_iops;
> > -		inode->i_fop = &generic_ro_fops;
> > +		if (!erofs_inode_is_data_compressed(vi->datalayout))
> > +			inode->i_fop = &erofs_file_fops;
> > +		else
> > +			inode->i_fop = &generic_ro_fops;
> 
> if (erofs_inode_is_data_compressed(vi->datalayout))
> 	inode->i_fop = &generic_ro_fops;
> else
> 	inode->i_fop = &erofs_file_fops;
> 
> Otherwise, it looks good to me.

ok, will fix in the next version.

Thanks,
Gao Xiang

> 
> Reviewed-by: Chao Yu <chao at kernel.org>
> 
> Thanks
> 
> >   		break;
> >   	case S_IFDIR:
> >   		inode->i_op = &erofs_dir_iops;
> > diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
> > index 543c2ff97d30..2669c785d548 100644
> > --- a/fs/erofs/internal.h
> > +++ b/fs/erofs/internal.h
> > @@ -371,6 +371,7 @@ static inline int z_erofs_map_blocks_iter(struct inode *inode,
> >   #endif	/* !CONFIG_EROFS_FS_ZIP */
> >   /* data.c */
> > +extern const struct file_operations erofs_file_fops;
> >   struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
> >   /* inode.c */
> >