On Tue, 24 Jan 2017 14:24:23 -0800 (PST)
Stefano Stabellini <sstabell...@kernel.org> wrote:
> On Tue, 24 Jan 2017, Greg Kurz wrote:
> > On Mon, 23 Jan 2017 12:20:57 -0800 (PST)
> > Stefano Stabellini <sstabell...@kernel.org> wrote:
> >
> > > On Sat, 21 Jan 2017, Greg Kurz wrote:
> > > > The current code tries to copy `read_count' bytes starting at offset
> > > > `offset' from a `read_count`-sized iovec. This causes v9fs_pack() to
> > > > fail with ENOBUFS.
> > > >
> > > > Since the PDU iovec is already partially filled with `offset' bytes,
> > > > let's skip them when creating `qiov_full' and have v9fs_pack() to
> > > > copy the whole of it. Moreover, this is consistent with the other
> > > > places where v9fs_init_qiov_from_pdu() is called.
> > > >
> > > > This fixes commit "bcb8998fac16 9pfs: call v9fs_init_qiov_from_pdu
> > > > before v9fs_pack".
> > >
> > > Sorry about that!
> > >
> > >
> > > > Signed-off-by: Greg Kurz <gr...@kaod.org>
> > > > ---
> > > > hw/9pfs/9p.c | 4 ++--
> > > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/hw/9pfs/9p.c b/hw/9pfs/9p.c
> > > > index fa58877570f6..f0eef1a3ef53 100644
> > > > --- a/hw/9pfs/9p.c
> > > > +++ b/hw/9pfs/9p.c
> > > > @@ -1685,8 +1685,8 @@ static int v9fs_xattr_read(V9fsState *s, V9fsPDU
> > > > *pdu, V9fsFidState *fidp,
> > > > }
> > > > offset += err;
> > > >
> > > > - v9fs_init_qiov_from_pdu(&qiov_full, pdu, 0, read_count, false);
> > > > - err = v9fs_pack(qiov_full.iov, qiov_full.niov, offset,
> > > > + v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count,
> > > > false);
> > >
> > > v9fs_init_qiov_from_pdu calls init_in_iov_from_pdu passing read_count as
> > > size argument. offset is not passed to init_in_iov_from_pdu, it is only
> > > used to initialized qiov_full.
> > >
> > > In other words, don't we need to:
> > >
> > > v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count + offset,
> > > false);
> > >
> > > To make sure that qiov_full has "read_count + offset" bytes in it, and
> > > qiov_full is initialized skipping the first "offset" bytes?
> > >
> >
> > If we do that then qemu_iovec_concat() will skip offset bytes and then copy
> > read_count + offset bytes or am I missing something ?
>
> You are right, sorry. qemu_iovec_concat didn't do what I thought it
> would. In that case, I think we need to change v9fs_init_qiov_from_pdu
> (in addition to the changes already in this patch):
>
> @@ -1659,12 +1659,14 @@ static void v9fs_init_qiov_from_pdu(QEMUIOVector
> *qiov, V9fsPDU *pdu,
> if (is_write) {
> pdu->s->transport->init_out_iov_from_pdu(pdu, &iov, &niov);
> } else {
> - pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size);
> + pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size +
> skip);
> }
>
> The reason is the same as before: iov needs to have offset+read_count
> bytes in it, for qemu_iovec_concat to skip offset and iov still have
> read_count bytes left.
>
Ah! I had missed that since virtio_init_in_iov_from_pdu doesn't use
the size argument, but I now understand :)
Thanks for the clarification.
>
> > > > + err = v9fs_pack(qiov_full.iov, qiov_full.niov, 0,
> > > > ((char *)fidp->fs.xattr.value) + off,
> > > > read_count);
> > > > qemu_iovec_destroy(&qiov_full);
> > > >
> >
> >
