Skip to contentSkip to conversation

net: sched: remove redundant 'rtnl_held' argument

PATCH In review Vlad Buslov 2 participants 7 messages Nov 27, 2020 – Dec 2, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: sched: remove redundant 'rtnl_held' argument

Functions tfilter_notify_chain() and tcf_get_next_proto() are always called with rtnl lock held in current implementation. Moreover, attempting to call them without rtnl lock would cause a warning down the call chain in function __tcf_get_next_proto() that requires the lock to be held by callers. Remove the 'rtnl_held' argument in order to simplify the code and make rtnl lock requirement explicit.

---
 include/net/pkt_cls.h |  2 +-
 net/sched/cls_api.c   | 18 ++++++++----------
 net/sched/sch_api.c   |  4 ++--
 3 files changed, 11 insertions(+), 13 deletions(-)
include/net/pkt_cls.h +1 -1
index 133f9ad4d4f9..0f2a9c44171c 100644
@@ -48,7 +48,7 @@ void tcf_chain_put_by_act(struct tcf_chain *chain);
 struct tcf_chain *tcf_get_next_chain(struct tcf_block *block,
 				     struct tcf_chain *chain);
 struct tcf_proto *tcf_get_next_proto(struct tcf_chain *chain,
-				     struct tcf_proto *tp, bool rtnl_held);
+				     struct tcf_proto *tp);
 void tcf_block_netif_keep_dst(struct tcf_block *block);
 int tcf_block_get(struct tcf_block **p_block,
 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
net/sched/cls_api.c +8 -10
index ff3e943febaa..37b77bd30974 100644
@@ -991,13 +991,12 @@ __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
  */
 
 struct tcf_proto *
-tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
-		   bool rtnl_held)
+tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
 {
 	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
 
 	if (tp)
-		tcf_proto_put(tp, rtnl_held, NULL);
+		tcf_proto_put(tp, true, NULL);
 
 	return tp_next;
 }
@@ -1924,15 +1923,14 @@ static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
 				 struct tcf_block *block, struct Qdisc *q,
 				 u32 parent, struct nlmsghdr *n,
-				 struct tcf_chain *chain, int event,
-				 bool rtnl_held)
+				 struct tcf_chain *chain, int event)
 {
 	struct tcf_proto *tp;
 
-	for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
-	     tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
+	for (tp = tcf_get_next_proto(chain, NULL);
+	     tp; tp = tcf_get_next_proto(chain, tp))
 		tfilter_notify(net, oskb, n, tp, block,
-			       q, parent, NULL, event, false, rtnl_held);
+			       q, parent, NULL, event, false, true);
 }
 
 static void tfilter_put(struct tcf_proto *tp, void *fh)
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;
@@ -2895,7 +2893,7 @@ static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
 		break;
 	case RTM_DELCHAIN:
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, true);
+				     chain, RTM_DELTFILTER);
 		/* Flush the chain first as the user requested chain removal. */
 		tcf_chain_flush(chain, true);
 		/* In case the chain was successfully deleted, put a reference
net/sched/sch_api.c +2 -2
index 1a2d2471b078..51cb553e4317 100644
@@ -1943,8 +1943,8 @@ static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
 	     chain = tcf_get_next_chain(block, chain)) {
 		struct tcf_proto *tp;
 
-		for (tp = tcf_get_next_proto(chain, NULL, true);
-		     tp; tp = tcf_get_next_proto(chain, tp, true)) {
+		for (tp = tcf_get_next_proto(chain, NULL);
+		     tp; tp = tcf_get_next_proto(chain, tp)) {
 			struct tcf_bind_args arg = {};
 
 			arg.w.fn = tcf_node_bind;
Signature · 1 line
2.21.0
maintainer
#1 lore raw

net/sched/cls_api.c static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, · line 2266

Quoted 9 lines · On Fri, 27 Nov 2020 17:12:05 +0200 Vlad Buslov wrote:
net/sched/cls_api.c +1 -1
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;

Hum. This looks off.

#2 lore raw
Quoted 12 lines · Jakub Kicinski

net/sched/cls_api.c static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, · line 2266

Quoted 9 lines · On Fri, 27 Nov 2020 17:12:05 +0200 Vlad Buslov wrote:
net/sched/cls_api.c +1 -1
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;

Hum. This looks off.

Hi Jakub,

Prio==0 means user requests to flush whole chain. In such case rtnl lock is obtained earlier in tc_del_tfilter():

	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
	 * found), qdisc is not unlocked, classifier type is not specified,
	 * classifier is not unlocked.
	 */
	if (!prio ||
	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
	    !tcf_proto_is_unlocked(name)) {
		rtnl_held = true;
		rtnl_lock();
	}
maintainer
#3 lore raw
Quoted 19 lines · On Tue, 1 Dec 2020 09:55:37 +0200 Vlad Buslov wrote:
Quoted 12 lines · Jakub Kicinski

net/sched/cls_api.c static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, · line 2266

Quoted 9 lines · On Fri, 27 Nov 2020 17:12:05 +0200 Vlad Buslov wrote:
net/sched/cls_api.c +1 -1
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;  

Hum. This looks off.

Hi Jakub,

Prio==0 means user requests to flush whole chain. In such case rtnl lock is obtained earlier in tc_del_tfilter():

Quoted 11 lines
	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
	 * found), qdisc is not unlocked, classifier type is not specified,
	 * classifier is not unlocked.
	 */
	if (!prio ||
	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
	    !tcf_proto_is_unlocked(name)) {
		rtnl_held = true;
		rtnl_lock();
	}

Makes sense, although seems a little fragile. Why not put a true in there, in that case?

Do you have a larger plan here? The motivation seems a little unclear if I'm completely honest. Are you dropping the rtnl_held from all callers of __tcf_get_next_proto() just to save the extra argument / typing? That's nice but there's also value in the API being consistent.

#4 lore raw
Quoted 32 lines · Jakub Kicinski
Quoted 19 lines · On Tue, 1 Dec 2020 09:55:37 +0200 Vlad Buslov wrote:
Quoted 12 lines · Jakub Kicinski

net/sched/cls_api.c static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, · line 2266

Quoted 9 lines · On Fri, 27 Nov 2020 17:12:05 +0200 Vlad Buslov wrote:
net/sched/cls_api.c +1 -1
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;  

Hum. This looks off.

Hi Jakub,

Prio==0 means user requests to flush whole chain. In such case rtnl lock is obtained earlier in tc_del_tfilter():

Quoted 11 lines
	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
	 * found), qdisc is not unlocked, classifier type is not specified,
	 * classifier is not unlocked.
	 */
	if (!prio ||
	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
	    !tcf_proto_is_unlocked(name)) {
		rtnl_held = true;
		rtnl_lock();
	}
Quoted 2 lines

Makes sense, although seems a little fragile. Why not put a true in there, in that case?

Because, as I described in commit message, the function will trigger an assertion if called without rtnl lock, so passing rtnl_held==false argument makes no sense and is confusing for the reader.

Quoted 4 lines

Do you have a larger plan here? The motivation seems a little unclear if I'm completely honest. Are you dropping the rtnl_held from all callers of __tcf_get_next_proto() just to save the extra argument / typing?

The plan is to have 'rtnl_held' arg for functions that can be called without rtnl lock and not have such argument for functions that require caller to hold rtnl :)

To elaborate further regarding motivation for this patch: some time ago I received an email asking why I have rtnl_held arg in function that has ASSERT_RTNL() in one of its dependencies. I re-read the code and determined that it was a leftover from earlier version and is not needed in code that was eventually upstreamed. Removing the argument was an easy decision since Jiri hates those and repeatedly asked me to minimize usage of such function arguments, so I didn't expect it to be controversial.

Quoted 1 line

That's nice but there's also value in the API being consistent.

Cls_api has multiple functions that don't have 'rtnl_held' argument. Only functions that can work without rtnl lock have it. Why do you suggest it is inconsistent to remove it here?

maintainer
#5 lore raw
Quoted 33 lines · On Tue, 1 Dec 2020 20:39:16 +0200 Vlad Buslov wrote:
Quoted 32 lines · Jakub Kicinski
Quoted 30 lines · On Tue, 1 Dec 2020 09:55:37 +0200 Vlad Buslov wrote:
Quoted 12 lines · Jakub Kicinski

net/sched/cls_api.c static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, · line 2266

Quoted 9 lines · On Fri, 27 Nov 2020 17:12:05 +0200 Vlad Buslov wrote:
net/sched/cls_api.c +1 -1
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;    

Hum. This looks off.

Hi Jakub,

Prio==0 means user requests to flush whole chain. In such case rtnl lock is obtained earlier in tc_del_tfilter():

	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
	 * found), qdisc is not unlocked, classifier type is not specified,
	 * classifier is not unlocked.
	 */
	if (!prio ||
	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
	    !tcf_proto_is_unlocked(name)) {
		rtnl_held = true;
		rtnl_lock();
	}
Quoted 6 lines
Quoted 2 lines

Makes sense, although seems a little fragile. Why not put a true in there, in that case?

Because, as I described in commit message, the function will trigger an assertion if called without rtnl lock, so passing rtnl_held==false argument makes no sense and is confusing for the reader.

The assumption being that tcf_ functions without the arg must hold the lock?

Quoted 17 lines
Quoted 3 lines

Do you have a larger plan here? The motivation seems a little unclear if I'm completely honest. Are you dropping the rtnl_held from all callers of __tcf_get_next_proto() just to save the extra argument / typing?

The plan is to have 'rtnl_held' arg for functions that can be called without rtnl lock and not have such argument for functions that require caller to hold rtnl :)

To elaborate further regarding motivation for this patch: some time ago I received an email asking why I have rtnl_held arg in function that has ASSERT_RTNL() in one of its dependencies. I re-read the code and determined that it was a leftover from earlier version and is not needed in code that was eventually upstreamed. Removing the argument was an easy decision since Jiri hates those and repeatedly asked me to minimize usage of such function arguments, so I didn't expect it to be controversial.

Quoted 5 lines
Quoted 1 line

That's nice but there's also value in the API being consistent.

Cls_api has multiple functions that don't have 'rtnl_held' argument. Only functions that can work without rtnl lock have it. Why do you suggest it is inconsistent to remove it here?

I see. I was just trying to figure out if you have a plan for larger restructuring to improve the situation. I also dislike to arguments being passed around in a seemingly random fashion. Removing or adding them to a single function does not move the needle much, IMO.

But since the patch is correct I'll apply it now, thanks!

#6 lore raw
Quoted 37 lines · Jakub Kicinski
Quoted 36 lines · On Tue, 1 Dec 2020 20:39:16 +0200 Vlad Buslov wrote:
Quoted 34 lines · Jakub Kicinski
Quoted 19 lines · On Tue, 1 Dec 2020 09:55:37 +0200 Vlad Buslov wrote:
Quoted 12 lines · Jakub Kicinski

net/sched/cls_api.c static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n, · line 2266

Quoted 9 lines · On Fri, 27 Nov 2020 17:12:05 +0200 Vlad Buslov wrote:
net/sched/cls_api.c +1 -1
@@ -2262,7 +2260,7 @@ static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
 
 	if (prio == 0) {
 		tfilter_notify_chain(net, skb, block, q, parent, n,
-				     chain, RTM_DELTFILTER, rtnl_held);
+				     chain, RTM_DELTFILTER);
 		tcf_chain_flush(chain, rtnl_held);
 		err = 0;
 		goto errout;    

Hum. This looks off.

Hi Jakub,

Prio==0 means user requests to flush whole chain. In such case rtnl lock is obtained earlier in tc_del_tfilter():

Quoted 11 lines
	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
	 * found), qdisc is not unlocked, classifier type is not specified,
	 * classifier is not unlocked.
	 */
	if (!prio ||
	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
	    !tcf_proto_is_unlocked(name)) {
		rtnl_held = true;
		rtnl_lock();
	}

Makes sense, although seems a little fragile. Why not put a true in there, in that case?

Quoted 6 lines
Quoted 3 lines

Because, as I described in commit message, the function will trigger an assertion if called without rtnl lock, so passing rtnl_held==false argument makes no sense and is confusing for the reader.

The assumption being that tcf_ functions without the arg must hold the lock?

Yes.

Quoted 20 lines
Quoted 20 lines
Quoted 4 lines

Do you have a larger plan here? The motivation seems a little unclear if I'm completely honest. Are you dropping the rtnl_held from all callers of __tcf_get_next_proto() just to save the extra argument / typing?

The plan is to have 'rtnl_held' arg for functions that can be called without rtnl lock and not have such argument for functions that require caller to hold rtnl :)

To elaborate further regarding motivation for this patch: some time ago I received an email asking why I have rtnl_held arg in function that has ASSERT_RTNL() in one of its dependencies. I re-read the code and determined that it was a leftover from earlier version and is not needed in code that was eventually upstreamed. Removing the argument was an easy decision since Jiri hates those and repeatedly asked me to minimize usage of such function arguments, so I didn't expect it to be controversial.

Quoted 1 line

That's nice but there's also value in the API being consistent.

Quoted 8 lines
Quoted 3 lines

Cls_api has multiple functions that don't have 'rtnl_held' argument. Only functions that can work without rtnl lock have it. Why do you suggest it is inconsistent to remove it here?

I see. I was just trying to figure out if you have a plan for larger restructuring to improve the situation. I also dislike to arguments being passed around in a seemingly random fashion. Removing or adding them to a single function does not move the needle much, IMO.

No, this is not part of larger effort. I would like to stop passing 'rtnl_held' everywhere, but for that I need other drivers that implement TC offload to stop requiring rtnl lock, which would allow removing rtnl_held from tcf_proto_ops callbacks.

Quoted 2 lines

But since the patch is correct I'll apply it now, thanks!

Thank you!