On Mon, Sep 23, 2019 at 8:44 AM Eric Dumazet <eric.duma...@gmail.com> wrote: > > > > On 9/21/19 7:08 PM, Jakub Kicinski wrote: > > On Wed, 18 Sep 2019 14:37:21 -0700, Cong Wang wrote: > >> On Wed, Sep 18, 2019 at 12:57 PM 'Eric Dumazet' via syzkaller > >> <syzkal...@googlegroups.com> wrote: > >>> > >>> If the allocation done in tcf_exts_init() failed, > >>> we end up with a NULL pointer in exts->actions. > >> ... > >>> diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c > >>> index > >>> efd3cfb80a2ad775dc8ab3c4900bd73d52c7aaad..9aef93300f1c11791acbb9262dfe77996872eafe > >>> 100644 > >>> --- a/net/sched/cls_api.c > >>> +++ b/net/sched/cls_api.c > >>> @@ -3027,8 +3027,10 @@ static int tc_dump_chain(struct sk_buff *skb, > >>> struct netlink_callback *cb) > >>> void tcf_exts_destroy(struct tcf_exts *exts) > >>> { > >>> #ifdef CONFIG_NET_CLS_ACT > >>> - tcf_action_destroy(exts->actions, TCA_ACT_UNBIND); > >>> - kfree(exts->actions); > >>> + if (exts->actions) { > >> > >> I think it is _slightly_ better to check exts->nr_actions!=0 here, > >> as it would help exts->actions!=NULL&& exts->nr_actions==0 > >> cases too. > >> > >> What do you think? > > > > Alternatively, since tcf_exts_destroy() now takes NULL, and so > > obviously does kfree() - perhaps tcf_action_destroy() should > > return early if actions are NULL? > > > > I do not have any preference really, this is slow path and was trying to > fix a crash. > > tcf_action_destroy() makes me nervous, since it seems to be able to break its > loop > in case __tcf_idr_release() returns an error. This means that some actions > will > never be release.
Good point. Seems we can just continue the loop even when -EPERM is returned, there is in fact no harm to leave those still bound to filters there until the filers release them. Not sure if we should still propagate -EPERM to users in this partially failure case.