Re: [PATCH -next RFC 01/14] block: add some bdev apis

[Matthew Wilcox <willy@xxxxxxxxxxxxx>]

On Tue, Dec 05, 2023 at 08:37:15PM +0800, Yu Kuai wrote:
> +struct folio *bdev_read_folio(struct block_device *bdev, pgoff_t index)
> +{
> +     return read_mapping_folio(bdev->bd_inode->i_mapping, index, NULL);
> +}
> +EXPORT_SYMBOL_GPL(bdev_read_folio);

I'm coming to the opinion that 'index' is the wrong parameter here.
Looking through all the callers of bdev_read_folio() in this patchset,
they all have a position in bytes, and they all convert it to
index for this call.  The API should probably be:

struct folio *bdev_read_folio(struct block_device *bdev, loff_t pos)
{
        return read_mapping_folio(bdev->bd_inode->i_mapping,
                        pos / PAGE_SIZE, NULL);
}

... and at some point, we'll get round to converting read_mapping_folio()
to take its argument in loff_t.

Similiarly for these two APIs:

> +struct folio *bdev_read_folio_gfp(struct block_device *bdev, pgoff_t index,
> +                               gfp_t gfp)
> +struct folio *bdev_get_folio(struct block_device *bdev, pgoff_t index)

> +struct folio *bdev_find_or_create_folio(struct block_device *bdev,
> +                                     pgoff_t index, gfp_t gfp)
> +{
> +     return __filemap_get_folio(bdev->bd_inode->i_mapping, index,
> +                                FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
> +}
> +EXPORT_SYMBOL_GPL(bdev_find_or_create_folio);

This one probably shouldn't exist.  I've been converting callers of
find_or_create_page() to call __filemap_get_folio; I suspect we
should expose a __bdev_get_folio and have the callers use the FGP
arguments directly, but I'm open to other opinions here.

> +void bdev_sync_readahead(struct block_device *bdev, struct file_ra_state *ra,
> +                      struct file *file, pgoff_t index,
> +                      unsigned long req_count)
> +{
> +     struct file_ra_state tmp_ra = {};
> +
> +     if (!ra) {
> +             ra = &tmp_ra;
> +             file_ra_state_init(ra, bdev->bd_inode->i_mapping);
> +     }
> +     page_cache_sync_readahead(bdev->bd_inode->i_mapping, ra, file, index,
> +                               req_count);
> +}

I think the caller should always be passing in a valid file_ra_state.
It's only cramfs that doesn't have one, and it really should!
Not entirely sure about the arguments here; part of me says "bytes",
but this is weird enough to maybe take arguments in pages.