On Tue, May 20, 2025 at 1:30 PM Stanislav Fomichev <stfomic...@gmail.com> wrote:
>
> sendmsg() with a single iov becomes ITER_UBUF, sendmsg() with multiple
> iovs becomes ITER_IOVEC. iter_iov_len does not return correct
> value for UBUF, so teach to treat UBUF differently.
>
> Cc: Al Viro <v...@zeniv.linux.org.uk>
> Cc: Pavel Begunkov <asml.sile...@gmail.com>
> Cc: Mina Almasry <almasrym...@google.com>
> Fixes: bd61848900bf ("net: devmem: Implement TX path")
> Signed-off-by: Stanislav Fomichev <stfomic...@gmail.com>
> ---
> include/linux/uio.h | 8 +++++++-
> net/core/datagram.c | 3 ++-
> 2 files changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/uio.h b/include/linux/uio.h
> index 49ece9e1888f..393d0622cc28 100644
> --- a/include/linux/uio.h
> +++ b/include/linux/uio.h
> @@ -99,7 +99,13 @@ static inline const struct iovec *iter_iov(const struct
> iov_iter *iter)
> }
>
> #define iter_iov_addr(iter) (iter_iov(iter)->iov_base +
> (iter)->iov_offset)
> -#define iter_iov_len(iter) (iter_iov(iter)->iov_len - (iter)->iov_offset)
> +
> +static inline size_t iter_iov_len(const struct iov_iter *i)
> +{
> + if (i->iter_type == ITER_UBUF)
> + return i->count;
> + return iter_iov(i)->iov_len - i->iov_offset;
> +}
>
This change looks good to me from devmem perspective, but aren't you
potentially breaking all these existing callers to iter_iov_len?
ackc -i iter_iov_len
fs/read_write.c
846: iter_iov_len(iter), ppos);
849: iter_iov_len(iter), ppos);
858: if (nr != iter_iov_len(iter))
mm/madvise.c
1808: size_t len_in = iter_iov_len(iter);
1838: iov_iter_advance(iter, iter_iov_len(iter));
io_uring/rw.c
710: len = iter_iov_len(iter);
Or are you confident this change is compatible with these callers for
some reason?
Maybe better to handle this locally in zerocopy_fill_skb_from_devmem,
and then follow up with a more ambitious change that streamlines how
all the iters behave.
--
Thanks,
Mina
