From: Vlad Buslov <vladbu@mellanox.com>
To: Jiri Pirko <jiri@resnulli.us>
Cc: netdev@vger.kernel.org, davem@davemloft.net, jhs@mojatatu.com,
xiyou.wangcong@gmail.com, pablo@netfilter.org,
kadlec@blackhole.kfki.hu, fw@strlen.de, ast@kernel.org,
daniel@iogearbox.net, edumazet@google.com, keescook@chromium.org,
linux-kernel@vger.kernel.org, netfilter-devel@vger.kernel.org,
coreteam@netfilter.org, kliteyn@mellanox.com
Subject: Re: [PATCH 06/14] net: sched: implement reference counted action release
Date: Tue, 15 May 2018 12:16:10 +0300 [thread overview]
Message-ID: <vbfmux1mg85.fsf@reg-r-vrt-018-180.mtr.labs.mlnx> (raw)
In-Reply-To: <20180515090357.GK2134@nanopsycho.orion>
On Tue 15 May 2018 at 09:03, Jiri Pirko <jiri@resnulli.us> wrote:
> Mon, May 14, 2018 at 09:07:06PM CEST, vladbu@mellanox.com wrote:
>>
>>On Mon 14 May 2018 at 16:47, Jiri Pirko <jiri@resnulli.us> wrote:
>>> Mon, May 14, 2018 at 04:27:07PM CEST, vladbu@mellanox.com wrote:
>>>
>>> [...]
>>>
>>>
>>>>+static int tcf_action_del_1(struct net *net, char *kind, u32 index,
>>>>+ struct netlink_ext_ack *extack)
>>>>+{
>>>>+ const struct tc_action_ops *ops;
>>>>+ int err = -EINVAL;
>>>>+
>>>>+ ops = tc_lookup_action_n(kind);
>>>>+ if (!ops) {
>>>
>>> How this can happen? Apparently you hold reference to the module since
>>> you put it couple lines below. In that case, I don't really understand
>>> why you need to lookup and just don't use "ops" pointer you have in
>>> tcf_action_delete().
>>
>>The problem is that I cant really delete action while holding reference
>
> Wait a sec. I was talking about a "module reference" (module_put())
I misunderstood your question. Yes, I guess I can just save return value
of tcf_action_put to variable, continue using ops pointer and only
call module_put after delete.
>
>
>>to it. I can try to decrement reference twice, however that might result
>>race condition if another task tries to delete that action at the same
>>time.
>>
>>Imagine situation:
>>1. Action is in actions idr, refcount==1
>>2. Task tries to delete action, takes reference while working with it,
>>refcount==2
>>3. Another task tries to delete same action, takes reference,
>>refcount==3
>>4. First task decrements refcount twice, refcount==1
>>5. At the same time second task decrements refcount twice, refcount==-1
>>
>>My solution is to save action index, but release the reference. Then try
>>to lookup action again and delete it if it is still in idr. (was not
>>concurrently deleted)
>>
>>Now same potential race condition with this patch:
>>1. Action is in actions idr, refcount==1
>>2. Task tries to delete action, takes reference while working with it,
>>refcount==2
>>3. Another task tries to delete same action, takes reference,
>>refcount==3
>>4. First task releases reference and deletes actions from idr, which
>>results another refcount decrement, refcount==1
>>5. At the same time second task releases reference to action,
>>refcount==0, action is deleted
>>6. When task tries to lookup-delete action from idr by index, action is
>>not found. This case is handled correctly by code and no negative
>>overflow of refcount happens.
>>
>>>
>>>
>>>>+ NL_SET_ERR_MSG(extack, "Specified TC action not found");
>>>>+ goto err_out;
>>>>+ }
>>>>+
>>>>+ if (ops->delete)
>>>>+ err = ops->delete(net, index);
>>>>+
>>>>+ module_put(ops->owner);
>>>>+err_out:
>>>>+ return err;
>>>>+}
>>>>+
>>>> static int tca_action_flush(struct net *net, struct nlattr *nla,
>>>> struct nlmsghdr *n, u32 portid,
>>>> struct netlink_ext_ack *extack)
>>>>@@ -1052,6 +1088,36 @@ static int tca_action_flush(struct net *net, struct nlattr *nla,
>>>> return err;
>>>> }
>>>>
>>>>+static int tcf_action_delete(struct net *net, struct list_head *actions,
>>>>+ struct netlink_ext_ack *extack)
>>>>+{
>>>>+ int ret;
>>>>+ struct tc_action *a, *tmp;
>>>>+ char kind[IFNAMSIZ];
>>>>+ u32 act_index;
>>>>+
>>>>+ list_for_each_entry_safe(a, tmp, actions, list) {
>>>>+ const struct tc_action_ops *ops = a->ops;
>>>>+
>>>>+ /* Actions can be deleted concurrently
>>>>+ * so we must save their type and id to search again
>>>>+ * after reference is released.
>>>>+ */
>>>>+ strncpy(kind, a->ops->kind, sizeof(kind) - 1);
>>>>+ act_index = a->tcfa_index;
>>>>+
>>>>+ list_del(&a->list);
>>>>+ if (tcf_action_put(a))
>>>>+ module_put(ops->owner);
>>>>+
>>>>+ /* now do the delete */
>>>>+ ret = tcf_action_del_1(net, kind, act_index, extack);
>>>>+ if (ret < 0)
>>>>+ return ret;
>>>>+ }
>>>>+ return 0;
>>>>+}
>>>>+
>>>
>>> [...]
>>
next prev parent reply other threads:[~2018-05-15 9:16 UTC|newest]
Thread overview: 86+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-05-14 14:27 [PATCH 00/14] Modify action API for implementing lockless actions Vlad Buslov
2018-05-14 14:27 ` [PATCH 01/14] net: sched: use rcu for action cookie update Vlad Buslov
2018-05-14 15:10 ` Jiri Pirko
2018-05-14 23:39 ` kbuild test robot
2018-05-14 14:27 ` [PATCH 02/14] net: sched: change type of reference and bind counters Vlad Buslov
2018-05-14 15:11 ` Jiri Pirko
2018-05-19 21:04 ` Marcelo Ricardo Leitner
2018-05-20 10:55 ` Vlad Buslov
2018-05-14 14:27 ` [PATCH 03/14] net: sched: add 'delete' function to action ops Vlad Buslov
2018-05-14 15:12 ` Jiri Pirko
2018-05-14 16:30 ` Jiri Pirko
2018-05-14 14:27 ` [PATCH 04/14] net: sched: implement unlocked action init API Vlad Buslov
2018-05-14 15:16 ` Jiri Pirko
2018-05-19 21:11 ` Marcelo Ricardo Leitner
2018-05-14 14:27 ` [PATCH 05/14] net: sched: always take reference to action Vlad Buslov
2018-05-14 16:23 ` Jiri Pirko
2018-05-14 18:49 ` Vlad Buslov
2018-05-15 8:58 ` Jiri Pirko
2018-05-15 11:52 ` Vlad Buslov
2018-05-15 1:38 ` kbuild test robot
2018-05-15 1:38 ` [RFC PATCH] net: sched: __tcf_idr_check() can be static kbuild test robot
2018-05-14 14:27 ` [PATCH 06/14] net: sched: implement reference counted action release Vlad Buslov
2018-05-14 16:28 ` Jiri Pirko
2018-05-14 16:47 ` Jiri Pirko
2018-05-14 19:07 ` Vlad Buslov
2018-05-15 9:03 ` Jiri Pirko
2018-05-15 9:16 ` Vlad Buslov [this message]
2018-05-19 21:43 ` Marcelo Ricardo Leitner
2018-05-20 6:22 ` Jiri Pirko
2018-05-20 10:59 ` Vlad Buslov
2018-05-14 14:27 ` [PATCH 07/14] net: sched: use reference counting action init Vlad Buslov
2018-05-15 11:24 ` Jiri Pirko
2018-05-15 11:32 ` Vlad Buslov
2018-05-15 11:39 ` Jiri Pirko
2018-05-15 11:41 ` Vlad Buslov
2018-05-15 11:57 ` Jiri Pirko
2018-05-15 12:00 ` Vlad Buslov
2018-05-14 14:27 ` [PATCH 08/14] net: sched: account for temporary action reference Vlad Buslov
2018-05-16 7:12 ` Jiri Pirko
2018-05-14 14:27 ` [PATCH 09/14] net: sched: don't release reference on action overwrite Vlad Buslov
2018-05-16 7:43 ` Jiri Pirko
2018-05-16 7:47 ` Vlad Buslov
2018-05-16 7:50 ` Jiri Pirko
2018-05-19 21:52 ` Marcelo Ricardo Leitner
2018-05-20 18:42 ` Vlad Buslov
2018-05-14 14:27 ` [PATCH 10/14] net: sched: extend act API for lockless actions Vlad Buslov
2018-05-16 7:50 ` Jiri Pirko
2018-05-16 8:16 ` Vlad Buslov
2018-05-16 8:56 ` Jiri Pirko
2018-05-16 9:39 ` Vlad Buslov
2018-05-19 22:17 ` Marcelo Ricardo Leitner
2018-05-14 14:27 ` [PATCH 11/14] net: core: add new/replace rate estimator lock parameter Vlad Buslov
2018-05-16 9:54 ` Jiri Pirko
2018-05-16 10:00 ` Vlad Buslov
2018-05-16 10:11 ` Jiri Pirko
2018-05-14 14:27 ` [PATCH 12/14] net: sched: retry action check-insert on concurrent modification Vlad Buslov
2018-05-16 9:59 ` Jiri Pirko
2018-05-16 11:55 ` Vlad Buslov
2018-05-16 12:26 ` Jiri Pirko
2018-05-16 12:43 ` Vlad Buslov
2018-05-16 13:21 ` Jiri Pirko
2018-05-16 13:52 ` Vlad Buslov
2018-05-16 14:13 ` Jiri Pirko
2018-05-16 14:26 ` Vlad Buslov
2018-05-16 14:55 ` Jiri Pirko
2018-05-19 22:35 ` Marcelo Ricardo Leitner
2018-05-14 14:27 ` [PATCH 13/14] net: sched: use unique idr insert function in unlocked actions Vlad Buslov
2018-05-16 9:50 ` Jiri Pirko
2018-05-16 9:54 ` Vlad Buslov
2018-05-19 22:20 ` Marcelo Ricardo Leitner
2018-05-20 21:13 ` Or Gerlitz
2018-05-20 21:33 ` Marcelo Ricardo Leitner
2018-05-20 21:40 ` Or Gerlitz
2018-05-20 22:43 ` Marcelo Ricardo Leitner
2018-05-14 14:27 ` [PATCH 14/14] net: sched: implement delete for all actions Vlad Buslov
2018-05-16 9:48 ` Jiri Pirko
2018-05-16 9:58 ` Vlad Buslov
2018-05-19 22:45 ` Marcelo Ricardo Leitner
2018-05-14 18:03 ` [PATCH 00/14] Modify action API for implementing lockless actions Jamal Hadi Salim
2018-05-14 20:46 ` Vlad Buslov
2018-05-15 18:25 ` Jamal Hadi Salim
2018-05-15 21:21 ` Vlad Buslov
2018-05-15 8:20 ` Jiri Pirko
2018-05-24 23:34 ` Cong Wang
2018-05-25 20:39 ` Vlad Buslov
2018-05-25 21:40 ` Cong Wang
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=vbfmux1mg85.fsf@reg-r-vrt-018-180.mtr.labs.mlnx \
--to=vladbu@mellanox.com \
--cc=ast@kernel.org \
--cc=coreteam@netfilter.org \
--cc=daniel@iogearbox.net \
--cc=davem@davemloft.net \
--cc=edumazet@google.com \
--cc=fw@strlen.de \
--cc=jhs@mojatatu.com \
--cc=jiri@resnulli.us \
--cc=kadlec@blackhole.kfki.hu \
--cc=keescook@chromium.org \
--cc=kliteyn@mellanox.com \
--cc=linux-kernel@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=netfilter-devel@vger.kernel.org \
--cc=pablo@netfilter.org \
--cc=xiyou.wangcong@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).