On Tue, Jan 9, 2018 at 7:00 PM, Jason Wang <jasow...@redhat.com> wrote: > > > On 2018年01月10日 08:07, Cong Wang wrote: >> >> tfile->tun could be detached before we close the tun fd, >> via tun_detach_all(), so it should not be used to check for >> tfile->tx_array. >> >> Use the same logic as in tun_attach(), just test !tfile->deatched. >> >> Reported-by: Dmitry Vyukov <dvyu...@google.com> >> Fixes: 1576d9860599 ("tun: switch to use skb array for tx") >> Cc: Jason Wang <jasow...@redhat.com> >> Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com> >> --- >> drivers/net/tun.c | 3 ++- >> 1 file changed, 2 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/net/tun.c b/drivers/net/tun.c >> index 4f4a842a1c9c..1a1f834440a6 100644 >> --- a/drivers/net/tun.c >> +++ b/drivers/net/tun.c >> @@ -613,6 +613,7 @@ static void tun_queue_purge(struct tun_file *tfile) >> static void __tun_detach(struct tun_file *tfile, bool clean) >> { >> + bool clean_tx_array = !tfile->detached; >> struct tun_file *ntfile; >> struct tun_struct *tun; >> @@ -657,7 +658,7 @@ static void __tun_detach(struct tun_file *tfile, >> bool clean) >> tun->dev->reg_state == NETREG_REGISTERED) >> unregister_netdevice(tun->dev); >> } >> - if (tun) >> + if (clean_tx_array) >> skb_array_cleanup(&tfile->tx_array); >> sock_put(&tfile->sk); >> } > > > Looks like we may still leak if we do > > open > attach > detach > close
Good catch. > > Should we do cleanup unconditionally? It doesn't look like we can, because tfile is not zero'ed at least, it is allocated by sk_alloc() so tfile->tx_array could be some random value which prevents skb_array_cleanup() functioning correctly. Zero'ing tfile->tx_array in open() is not sufficient, as spinlocks are not yet initialized. I think we probably need to test tfile->tx_array.ring.queue, it is ugly but I don't see there is an API for it.