[Devel] Re: [PATCH 1/4] fuse: add basic support of iovec[] to fuse_req

[Miklos Szeredi]

Maxim Patlasov <mpatlasov at parallels.com> writes:

> The patch allows fuse_req to refer to array of iovec-s describing
> layout of user-data over req->pages. fuse_copy_pages() is re-worked to
> support both cased: former layout where pages[] corresponded to <buf, len>
> and newer one where pages[] corresponds to iovec[].
>
> Signed-off-by: Maxim Patlasov <mpatlasov at parallels.com>
> ---
>  fs/fuse/dev.c    |   52 +++++++++++++++++++++++++++++++++++++++++++++++++---
>  fs/fuse/fuse_i.h |   12 ++++++++++--
>  2 files changed, 59 insertions(+), 5 deletions(-)
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 7df2b5e..cdae525 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -850,9 +850,9 @@ static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
>  	return 0;
>  }
>  
> -/* Copy pages in the request to/from userspace buffer */
> -static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
> -			   int zeroing)
> +/* Start from addr(pages[0]) + page_offset. No holes in the middle. */
> +static int fuse_copy_pages_for_buf(struct fuse_copy_state *cs, unsigned nbytes,
> +				   int zeroing)
>  {
>  	unsigned i;
>  	struct fuse_req *req = cs->req;
> @@ -874,6 +874,52 @@ static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
>  	return 0;
>  }
>  
> +/* Take iov_offset as offset in iovec[0]. Iterate based on iovec[].iov_len */
> +static int fuse_copy_pages_for_iovec(struct fuse_copy_state *cs,
> +				     unsigned nbytes, int zeroing)
> +{
> +	unsigned i;
> +	struct fuse_req *req = cs->req;
> +	const struct iovec *iov = req->iovec;
> +	unsigned iov_offset = req->iov_offset;
> +
> +	for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
> +		int err;
> +		unsigned long user_addr = (unsigned long)iov->iov_base +
> +					  iov_offset;
> +		unsigned offset = user_addr & ~PAGE_MASK;
> +		unsigned count = min_t(size_t, PAGE_SIZE - offset,
> +				       iov->iov_len - iov_offset);

It would be much cleaner if we didn't have to deal with the original
iovec here, but only offset and length relative to the page.

I understand that that would mean allocating an array for these.  In the
other thread I mentioned the possibility of allocating the page array.
Instead of a page array, we could have an array of (pageptr, offset,
len) which would simplify the whole thing.

Thanks,
Miklos