]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/commit
UBUNTU: SAUCE: can: bcm: delay release of struct bcm_op after synchronize_rcu
authorThadeu Lima de Souza Cascardo <cascardo@canonical.com>
Fri, 11 Jun 2021 19:14:08 +0000 (16:14 -0300)
committerThadeu Lima de Souza Cascardo <cascardo@canonical.com>
Wed, 16 Jun 2021 22:48:43 +0000 (19:48 -0300)
commit5499bae10fe4f7f5c33ed14d7a6a6d8953ec15e6
tree287c20a0d72ed7bf529bc46c76e6de3ab9f6d392
parent173ef0b1826d328327ff308e78a8a55cbebc349b
UBUNTU: SAUCE: can: bcm: delay release of struct bcm_op after synchronize_rcu

BugLink: https://bugs.launchpad.net/bugs/1931855
can_rx_register callbacks may be called concurrently to the call to
can_rx_unregister. The callbacks and callback data, though, are protected by
RCU and the struct sock reference count.

So the callback data is really attached to the life of sk, meaning that it
should be released on sk_destruct. However, bcm_remove_op calls tasklet_kill,
and RCU callbacks may be called under RCU softirq, so that cannot be used on
kernels before the introduction of HRTIMER_MODE_SOFT.

However, bcm_rx_handler is called under RCU protection, so after calling
can_rx_unregister, we may call synchronize_rcu in order to wait for any RCU
read-side critical sections to finish. That is, bcm_rx_handler won't be called
anymore for those ops. So, we only free them, after we do that synchronize_rcu.

Reported-by: syzbot+0f7e7e5e2f4f40fa89c0@syzkaller.appspotmail.com
Reported-by: Norbert Slusarek <nslusarek@gmx.net>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Acked-by: Benjamin M Romer <benjamin.romer@canonical.com>
Acked-by: Ian May <ian.may@canonical.com>
net/can/bcm.c