Re: question about map_read_chunks()

["J. Bruce Fields" <bfields@fieldses.org>]

On Fri, Sep 27, 2013 at 10:23:10AM -0500, Tom Tucker wrote:
> Hi Dan,
> 
> On 9/27/13 7:21 AM, Dan Carpenter wrote:
> >I have looked at this again, and I still worry that it looks like a bug.
> >(remote security related blah blah blah).
> >
> >regards,
> >dan carpenter
> >
> >On Mon, Feb 20, 2012 at 12:50:19PM +0300, Dan Carpenter wrote:
> >>I had a couple questions about some map_read_chunks().
> >>
> >>net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
> >>
> >>    150          ch_bytes = ntohl(ch->rc_target.rs_length);
> >>                 ^^^^^^^^
> >>It look like this is 32 bits from the network?
> >>
> >>    151          head->arg.head[0] = rqstp->rq_arg.head[0];
> >>    152          head->arg.tail[0] = rqstp->rq_arg.tail[0];
> >>    153          head->arg.pages = &head->pages[head->count];
> >>    154          head->hdr_count = head->count; /* save count of hdr pages */
> >>    155          head->arg.page_base = 0;
> >>    156          head->arg.page_len = ch_bytes;
> >>    157          head->arg.len = rqstp->rq_arg.len + ch_bytes;
> >>                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >>Can overflow.
> >>    158          head->arg.buflen = rqstp->rq_arg.buflen + ch_bytes;
> agreed.
> >>                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >>Same.  I didn't follow it through to see if an overflow matters.  Does
> >>it?
> >>
> >>    159          head->count++;
> >>    160          chl_map->ch[0].start = 0;
> >>    161          while (byte_count) {
> >>    162                  rpl_map->sge[sge_no].iov_base =
> >>    163                          page_address(rqstp->rq_arg.pages[page_no]) + page_off;
> >>    164                  sge_bytes = min_t(int, PAGE_SIZE-page_off, ch_bytes);
> >>                                           ^^^
> >>This is the wrong cast to use.  A large ch_bytes would be counted as a
> >>negative value and get around the cap here.
> True, but if we validate the wire data like we should, that's
> probably not an issue.

Well, my last attempt to do rdma nfs io resulted in an instant server
reboot, so I can take a look at this but it may be the least of our
problems....

--b.