]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
vxlan: test dev->flags & IFF_UP before calling gro_cells_receive()
authorEric Dumazet <edumazet@google.com>
Sun, 10 Mar 2019 17:36:40 +0000 (10:36 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 19 Mar 2019 12:11:54 +0000 (13:11 +0100)
commitb1ea40a610868339dcb156245973b12c6e44e922
tree38eb9fc3c08d71707219afbda7c846ff5d950a1e
parent72ad13ea5a4f20f5fbc559f4991bd1e51a176a8c
vxlan: test dev->flags & IFF_UP before calling gro_cells_receive()

[ Upstream commit 59cbf56fcd98ba2a715b6e97c4e43f773f956393 ]

Same reasons than the ones explained in commit 4179cb5a4c92
("vxlan: test dev->flags & IFF_UP before calling netif_rx()")

netif_rx() or gro_cells_receive() must be called under a strict contract.

At device dismantle phase, core networking clears IFF_UP
and flush_all_backlogs() is called after rcu grace period
to make sure no incoming packet might be in a cpu backlog
and still referencing the device.

A similar protocol is used for gro_cells infrastructure, as
gro_cells_destroy() will be called only after a full rcu
grace period is observed after IFF_UP has been cleared.

Most drivers call netif_rx() from their interrupt handler,
and since the interrupts are disabled at device dismantle,
netif_rx() does not have to check dev->flags & IFF_UP

Virtual drivers do not have this guarantee, and must
therefore make the check themselves.

Otherwise we risk use-after-free and/or crashes.

Fixes: d342894c5d2f ("vxlan: virtual extensible lan")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/vxlan.c