net: ipv4: emulate READ_ONCE() on ->hdrincl bit-field in raw_sendmsg()
authorNicolai Stange <nstange@suse.de>
Mon, 8 Jan 2018 14:54:44 +0000 (15:54 +0100)
committerDavid S. Miller <davem@davemloft.net>
Tue, 9 Jan 2018 16:59:16 +0000 (11:59 -0500)
commit20b50d79974ea3192e8c3ab7faf4e536e5f14d8f
treec25a3de78f24e225e238e1ffe3eebb193f5da3f0
parent3dc2fa47549aca71773afdd12a78d31802bb22b4
net: ipv4: emulate READ_ONCE() on ->hdrincl bit-field in raw_sendmsg()

Commit 8f659a03a0ba ("net: ipv4: fix for a race condition in
raw_sendmsg") fixed the issue of possibly inconsistent ->hdrincl handling
due to concurrent updates by reading this bit-field member into a local
variable and using the thus stabilized value in subsequent tests.

However, aforementioned commit also adds the (correct) comment that

  /* hdrincl should be READ_ONCE(inet->hdrincl)
   * but READ_ONCE() doesn't work with bit fields
   */

because as it stands, the compiler is free to shortcut or even eliminate
the local variable at its will.

Note that I have not seen anything like this happening in reality and thus,
the concern is a theoretical one.

However, in order to be on the safe side, emulate a READ_ONCE() on the
bit-field by doing it on the local 'hdrincl' variable itself:

int hdrincl = inet->hdrincl;
hdrincl = READ_ONCE(hdrincl);

This breaks the chain in the sense that the compiler is not allowed
to replace subsequent reads from hdrincl with reloads from inet->hdrincl.

Fixes: 8f659a03a0ba ("net: ipv4: fix for a race condition in raw_sendmsg")
Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/raw.c