Re: [PATCH] nvmet-tcp: Don't kmap() pages which can't come from HIGHMEM

[Sagi Grimberg]

> > @@ -329,7 +312,6 @@ static void nvmet_tcp_map_pdu_iovec(struct nvmet_tcp_cmd *cmd)
> >   	u32 length, offset, sg_offset;
> >   
> >   	length = cmd->pdu_len;
> > -	cmd->nr_mapped = DIV_ROUND_UP(length, PAGE_SIZE);
> >   	offset = cmd->rbytes_done;
> >   	cmd->sg_idx = offset / PAGE_SIZE;
> >   	sg_offset = offset % PAGE_SIZE;
> > @@ -338,7 +320,7 @@ static void nvmet_tcp_map_pdu_iovec(struct nvmet_tcp_cmd *cmd)
> >   	while (length) {
> >   		u32 iov_len = min_t(u32, length, sg->length - sg_offset);
> >   
> > -		iov->iov_base = kmap(sg_page(sg)) + sg->offset + sg_offset;
> > +		iov->iov_base = page_address(sg_page(sg)) + sg->offset + sg_offset;
> >   		iov->iov_len = iov_len;
> >   
> >   		length -= iov_len;
> > @@ -347,8 +329,7 @@ static void nvmet_tcp_map_pdu_iovec(struct nvmet_tcp_cmd *cmd)
> >   		sg_offset = 0;
> >   	}
> >   
> > -	iov_iter_kvec(&cmd->recv_msg.msg_iter, READ, cmd->iov,
> > -		cmd->nr_mapped, cmd->pdu_len);
> > +	iov_iter_kvec(&cmd->recv_msg.msg_iter, READ, cmd->iov, 0, cmd->pdu_len);
> >   }
>
> I earlier meant just use a local variable for 'nr_mapped' rather than replace
> it with '0'. I don't think that 0 segments would result in usable kvec.

yes, that is wrong. Need a local int nr_mapped variable.

> I'm not even sure the existing code is correct, though. The sg->length can be
> higher order than a PAGE_SIZE, so it may be over-reporting nr_segs. It's just
> supposed to be the number of initialized iov's.

The sg is allocated with sgl_alloc so it is for sure 0-order.