net: fix rds_iovec page count overflow
commit33b1b26d48822cbe04d14ba255672ed5ecadb3da
authorLinus Torvalds <torvalds@linux-foundation.org>
Thu, 28 Oct 2010 15:40:55 +0000 (28 15:40 +0000)
committerAndi Kleen <ak@linux.intel.com>
Thu, 28 Apr 2011 15:21:02 +0000 (28 08:21 -0700)
tree0c2e99b71f65bf2b32cecfbe611860cd33105a21
parent4d9566dc782ea4c60c44f3d6ee8e982061967c90
net: fix rds_iovec page count overflow

commit 1b1f693d7ad6d193862dcb1118540a030c5e761f upstream.

As reported by Thomas Pollet, the rdma page counting can overflow.  We
get the rdma sizes in 64-bit unsigned entities, but then limit it to
UINT_MAX bytes and shift them down to pages (so with a possible "+1" for
an unaligned address).

So each individual page count fits comfortably in an 'unsigned int' (not
even close to overflowing into signed), but as they are added up, they
might end up resulting in a signed return value. Which would be wrong.

Catch the case of tot_pages turning negative, and return the appropriate
error code.

Reported-by: Thomas Pollet <thomas.pollet@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
[v2: nr is unsigned in the old code]
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Acked-by: Tim Gardner <tim.gardner@canonical.com>
Acked-by: Brad Figg <brad.figg@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
net/rds/rdma.c