[rds-devel] [Security] TIPC security issues

Andy Grover andy.grover at oracle.com
Thu Oct 28 11:45:04 PDT 2010

On 10/28/2010 08:32 AM, Linus Torvalds wrote:
> Heh. We apparently have _another_ iovec overflow in networking. This time rds.
> Reported by Thomas Pollet<thomas.pollet at gmail.com>: look at
> net/rds/rdma.c around line 490. It doesn't use the regular iovec code,
> instead it cooks its own, and has a few problems with overflow.
> It gathers the number of pages into an "unsigned int", and for each
> entry in its own rds_iovec it will check that the size is<  UINT_MAX,
> and then generate the number of pages for that entry. With the whole
> "unaligned address adds one" logic, it means that each entry can get
> (UINT_MAX>>  PAGE_SHIFT)+1 pages.

FWIW both the signed issue and not checking the iovec changed were 
correct in 2.6.36, and only added in ff87e97.

> And how many entries can we have? Apparently that is capped to
> UINT_MAX too. So add all those up, and they can easily overflow the
> unsigned int page counter.
> So this time fixing verify_iovec() doesn't help, because rds just
> cooks its own, and this is using a totally different interface: it
> seems to hook into sendmsg, but it looks like it uses the ancillary
> data objects and passes in its own magical iovec rather than use any
> "normal" iovec thing. I don't know the code, I may be totally off.

Yes that's right, it's to map a memory region that will be the target of 
an RDMA operation. I don't know why struct rds_iovec was used instead of 
struct iovec, but I think we're stuck, since it's part of our socket API.

I'll send DaveM patches to fix those two immediately-identified problems 
today, and we'll take a good long look at the rest of the code for 
further issues.

Regards -- Andy

More information about the rds-devel mailing list