Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Craig Gallek]

On Thu, Jul 7, 2016 at 4:36 PM, Jiri Kosina <jikos@kernel.org> wrote:
> From: Jiri Kosina <jkosina@suse.cz>
>
> Convert the per-device linked list into a hashtable. The primary
> motivation for this change is that currently, we're not tracking all the
> qdiscs in hierarchy (e.g. excluding default qdiscs), as the lookup
> performed over the linked list by qdisc_match_from_root() is rather
> expensive.
>
> The ultimate goal is to get rid of hidden qdiscs completely, which will
> bring much more determinism in user experience.
>
> As we're adding hashtable.h include into generic netdevice.h, we have to make
> sure HASH_SIZE macro is now non-conflicting with local definitions.
>
> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
> ---
>
> v1 -> v2:       fix up RCU hastable usage wrt. rtnl
>                 fix compilation of .c files which define their own
>                  HASH_SIZE that now oncflicts with the one from
>                  hashtable.h (newly included via netdevice.h)
This sort of seems like it's just side-stepping the problem.  Given
that the size of this hash table is fixed, the lookup time of this
operation is still going to approach linear as the number of qdiscs
increases.  I took a quick pass at trying to use rhashtable for this
purpose a few weeks ago but dropped it when I realized many of the
table operations (which can trigger resize events) need to happen
while holding the rtnl lock.  I still think it would be possible to
use a dynamically sizable datastructure for this purpose, but it will
be a fair amount of work to change the current locking semantics to
make it work...

Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Jiri Kosina]

On Thu, 7 Jul 2016, Craig Gallek wrote:

> This sort of seems like it's just side-stepping the problem.  Given
> that the size of this hash table is fixed, the lookup time of this
> operation is still going to approach linear as the number of qdiscs
> increases.  

That's true; however the primary goal here is not to actually ultimately 
improve speed of qdisc lookup per se, but rather to make it possible to 
unhide the qdiscs which are currently omitted as the linked list takes too 
long to walk. The static hashtable is going help here.

Thanks,

-- 
Jiri Kosina
SUSE Labs

Deleting child qdisc doesn't reset parent to default qdisc?

[Jiri Kosina]

Hi,

I've came across the behavior where adding a child qdisc and then deleting 
it again makes the networking dysfunctional (I guess that's because all of 
a sudden there is absolutely no working qdisc on the device, although 
there originally was a default one in the parent).

In a nutshell, is this expected behavior or bug?

=====
jikos:~ # tc qdisc show
qdisc tbf 10: dev eth0 root refcnt 2 rate 800Mbit burst 131000b lat 1.0ms 
jikos:~ # ping -c 1 nix.cz | head -2
PING nix.cz (195.47.235.3) 56(84) bytes of data.
64 bytes from info.nix.cz (195.47.235.3): icmp_seq=1 ttl=89 time=1.59 ms

jikos:~ # tc qdisc add dev eth0 parent 10:1 sfq
jikos:~ # tc qdisc show
qdisc tbf 10: dev eth0 root refcnt 2 rate 800Mbit burst 131000b lat 1.0ms 
qdisc sfq 8008: dev eth0 parent 10:1 limit 127p quantum 1514b depth 127 divisor 1024 

jikos:~ # ping -c 1 nix.cz | head -2
PING nix.cz (195.47.235.3) 56(84) bytes of data.
64 bytes from info.nix.cz (195.47.235.3): icmp_seq=1 ttl=89 time=1.67 ms

jikos:~ # tc qdisc del dev eth0 parent 10:1 sfq
jikos:~ # tc qdisc show
qdisc tbf 10: dev eth0 root refcnt 2 rate 800Mbit burst 131000b lat 1.0ms 
jikos:~ # ping -c 1 nix.cz | head -2
PING nix.cz (195.47.235.3) 56(84) bytes of data.
	[ ... nothing happens ... ]
^C
jikos:~ # tc qdisc add dev eth0 parent 10:1 sfq
jikos:~ # ping -c 1 nix.cz | head -2
PING nix.cz (195.47.235.3) 56(84) bytes of data.
64 bytes from info.nix.cz (195.47.235.3): icmp_seq=1 ttl=89 time=1.66 ms
=====

Thanks,

-- 
Jiri Kosina

Re: Deleting child qdisc doesn't reset parent to default qdisc?

[Eric Dumazet]

On Thu, 2016-04-14 at 16:44 +0200, Jiri Kosina wrote:
> Hi,
> 
> I've came across the behavior where adding a child qdisc and then deleting 
> it again makes the networking dysfunctional (I guess that's because all of 
> a sudden there is absolutely no working qdisc on the device, although 
> there originally was a default one in the parent).
> 
> In a nutshell, is this expected behavior or bug?

This is the expected behavior.

If the kernel was suddenly doing a 'replace' when you ask a delete,
then the scripts doing a delete , than a add would break.

tc users are skilled admins ;)

Re: Deleting child qdisc doesn't reset parent to default qdisc?

[Phil Sutter]

On Thu, Apr 14, 2016 at 08:01:39AM -0700, Eric Dumazet wrote:
> On Thu, 2016-04-14 at 16:44 +0200, Jiri Kosina wrote:
> > Hi,
> > 
> > I've came across the behavior where adding a child qdisc and then deleting 
> > it again makes the networking dysfunctional (I guess that's because all of 
> > a sudden there is absolutely no working qdisc on the device, although 
> > there originally was a default one in the parent).
> > 
> > In a nutshell, is this expected behavior or bug?
> 
> This is the expected behavior.

OTOH some qdiscs (CBQ, DRR, DSMARK, HFSC, HTB, QFQ) assign the default
one upon deletion instead of noop_qdisc, hence I would describe
the situation using the words 'inconsistent' and 'accident' rather than
'expected'. :)

Anyhow, the problem with skilled admins is they accept quirks too easily
and just build their scripts around them - the same scripts we have to
keep compatible to then.

Cheers, Phil

Re: Deleting child qdisc doesn't reset parent to default qdisc?

[Jiri Kosina]

On Thu, 14 Apr 2016, Phil Sutter wrote:

> > > I've came across the behavior where adding a child qdisc and then deleting 
> > > it again makes the networking dysfunctional (I guess that's because all of 
> > > a sudden there is absolutely no working qdisc on the device, although 
> > > there originally was a default one in the parent).
> > > 
> > > In a nutshell, is this expected behavior or bug?
> > 
> > This is the expected behavior.
> 
> OTOH some qdiscs (CBQ, DRR, DSMARK, HFSC, HTB, QFQ) assign the default
> one upon deletion instead of noop_qdisc, hence I would describe
> the situation using the words 'inconsistent' and 'accident' rather than
> 'expected'. :)

Would a patch that'd unify this in a sense that all qdiscs would assign 
the default one upon deletion acceptable?

Thanks,

-- 
Jiri Kosina
SUSE Labs

Re: Deleting child qdisc doesn't reset parent to default qdisc?

[Eric Dumazet]

On Thu, 2016-04-14 at 18:08 +0200, Jiri Kosina wrote:
> On Thu, 14 Apr 2016, Phil Sutter wrote:
> 
> > > > I've came across the behavior where adding a child qdisc and then deleting 
> > > > it again makes the networking dysfunctional (I guess that's because all of 
> > > > a sudden there is absolutely no working qdisc on the device, although 
> > > > there originally was a default one in the parent).
> > > > 
> > > > In a nutshell, is this expected behavior or bug?
> > > 
> > > This is the expected behavior.
> > 
> > OTOH some qdiscs (CBQ, DRR, DSMARK, HFSC, HTB, QFQ) assign the default
> > one upon deletion instead of noop_qdisc, hence I would describe
> > the situation using the words 'inconsistent' and 'accident' rather than
> > 'expected'. :)
> 
> Would a patch that'd unify this in a sense that all qdiscs would assign 
> the default one upon deletion acceptable?
> 

And what would be the chosen behavior ?

Relying on TBF installing a bfifo for you at delete would be hazardous.

For example CBQ got it differently than HFSC

If qdisc_create_dflt() fails in CBQ, we fail the 'delete', while HFSC
falls back to noop_qdisc, without warning the user :(

At least always using noop_qdisc is consistent. No magic there.

Doing 'unification' right now would break existing scripts.

This is too late, I am afraid.

Re: Deleting child qdisc doesn't reset parent to default qdisc?

[Jamal Hadi Salim]

On 16-04-14 01:49 PM, Eric Dumazet wrote:

> And what would be the chosen behavior ?
>

TBF is probably a bad example because it started life as
a classless qdisc. There was only one built-in fifo queue
that was shaped. Then someone made it classful and changed
this behavior. To me it sounds reasonable to have the
default behavior restored. At minimal consistency.

> Relying on TBF installing a bfifo for you at delete would be hazardous.
>
> For example CBQ got it differently than HFSC
>
> If qdisc_create_dflt() fails in CBQ, we fail the 'delete', while HFSC
> falls back to noop_qdisc, without warning the user :(
>
> At least always using noop_qdisc is consistent. No magic there.
>
> Doing 'unification' right now would break existing scripts.
>
> This is too late, I am afraid.


Sigh. So rant:
IMO, we should let any new APIs and API updates stay longer
in discussion. Or better mark them as unstable for sometime.
The excuse that "it is out in the wild therefore cant be changed"
is harmful because the timeline is "forever" whereas
patches are applied after a short period of posting
and discussions and sometimes not involving the right people.
It is like having a jury issuing a death sentence after 1 week of
deliberation. You cant take it back after execution.

cheers,
jamal

Re: Deleting child qdisc doesn't reset parent to default qdisc?

[Eric Dumazet]

On Fri, 2016-04-15 at 08:42 -0400, Jamal Hadi Salim wrote:
> On 16-04-14 01:49 PM, Eric Dumazet wrote:
> 
> > And what would be the chosen behavior ?
> >
> 
> TBF is probably a bad example because it started life as
> a classless qdisc. There was only one built-in fifo queue
> that was shaped. Then someone made it classful and changed
> this behavior. To me it sounds reasonable to have the
> default behavior restored. At minimal consistency.


Then you need to save the initial qdisc (bfifo for TBF) in a special
place, to make sure the delete operation is guaranteed to succeed.

Or fail the delete if the bfifo can not be allocated.

I can tell that determinism if far more interesting than usability for
some users occasionally playing with tc.

Surely the silent fallback to noop_qdisc is wrong.

Anyway, we probably need to improve our ability to understand qdisc
hierarchies. Having some hidden qdiscs is the real problem here.

We need to add some hash table so that qdisc_match_from_root() does not
have to scan hundred of qdiscs.

[RFC PATCH] net: sched: convert qdisc linked list to hashtable (was Re: Deleting child qdisc doesn't reset parent to default qdisc?)

[Jiri Kosina]

On Fri, 15 Apr 2016, Eric Dumazet wrote:

> Anyway, we probably need to improve our ability to understand qdisc 
> hierarchies. Having some hidden qdiscs is the real problem here.
> 
> We need to add some hash table so that qdisc_match_from_root() does not 
> have to scan hundred of qdiscs.

So how about something like the patch below? I already have preliminary 
patches on top which unhide the default qdiscs, but let's make this one 
step after the other.

Thanks.




From: Jiri Kosina <jkosina@suse.cz>
Subject: [PATCH] net: sched: convert qdisc linked list to hashtable

Convert the per-device linked list into a hashtable. The primary motivation
for this change is that currently, we're not tracking all the qdiscs in
hierarchy (e.g. excluding default qdiscs), as the lookup performed over the
linked list by qdisc_match_from_root() is rather expensive.

The ultimate goal is to get rid of hidden qdiscs completely, which will bring
much more determinism in user experience.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
---
 include/linux/netdevice.h |  2 ++
 include/net/pkt_sched.h   |  4 ++--
 include/net/sch_generic.h |  2 +-
 net/core/dev.c            |  1 +
 net/sched/sch_api.c       | 23 +++++++++++++----------
 net/sched/sch_generic.c   |  6 +++---
 net/sched/sch_mq.c        |  2 +-
 net/sched/sch_mqprio.c    |  2 +-
 8 files changed, 24 insertions(+), 18 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index f45929c..630838e 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -52,6 +52,7 @@
 #include <uapi/linux/netdevice.h>
 #include <uapi/linux/if_bonding.h>
 #include <uapi/linux/pkt_cls.h>
+#include <linux/hashtable.h>
 
 struct netpoll_info;
 struct device;
@@ -1778,6 +1779,7 @@ struct net_device {
 	unsigned int		num_tx_queues;
 	unsigned int		real_num_tx_queues;
 	struct Qdisc		*qdisc;
+	DECLARE_HASHTABLE	(qdisc_hash, 16);
 	unsigned long		tx_queue_len;
 	spinlock_t		tx_global_lock;
 	int			watchdog_timeo;
diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
index fea53f4..8ba11b4 100644
--- a/include/net/pkt_sched.h
+++ b/include/net/pkt_sched.h
@@ -90,8 +90,8 @@ int unregister_qdisc(struct Qdisc_ops *qops);
 void qdisc_get_default(char *id, size_t len);
 int qdisc_set_default(const char *id);
 
-void qdisc_list_add(struct Qdisc *q);
-void qdisc_list_del(struct Qdisc *q);
+void qdisc_hash_add(struct Qdisc *q);
+void qdisc_hash_del(struct Qdisc *q);
 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
 struct Qdisc *qdisc_lookup_class(struct net_device *dev, u32 handle);
 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 62d5531..26f5cb3 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -67,7 +67,7 @@ struct Qdisc {
 	u32			limit;
 	const struct Qdisc_ops	*ops;
 	struct qdisc_size_table	__rcu *stab;
-	struct list_head	list;
+	struct hlist_node       hash;
 	u32			handle;
 	u32			parent;
 	int			(*reshape_fail)(struct sk_buff *skb,
diff --git a/net/core/dev.c b/net/core/dev.c
index 904ff43..edc1617 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -7511,6 +7511,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
 	INIT_LIST_HEAD(&dev->all_adj_list.lower);
 	INIT_LIST_HEAD(&dev->ptype_all);
 	INIT_LIST_HEAD(&dev->ptype_specific);
+	hash_init(dev->qdisc_hash);
 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
 	setup(dev);
 
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index ddf047d..82953cb 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -29,6 +29,7 @@
 #include <linux/hrtimer.h>
 #include <linux/lockdep.h>
 #include <linux/slab.h>
+#include <linux/hashtable.h>
 
 #include <net/net_namespace.h>
 #include <net/sock.h>
@@ -265,33 +266,33 @@ static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
 	    root->handle == handle)
 		return root;
 
-	list_for_each_entry_rcu(q, &root->list, list) {
+	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
 		if (q->handle == handle)
 			return q;
 	}
 	return NULL;
 }
 
-void qdisc_list_add(struct Qdisc *q)
+void qdisc_hash_add(struct Qdisc *q)
 {
 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 		struct Qdisc *root = qdisc_dev(q)->qdisc;
 
 		WARN_ON_ONCE(root == &noop_qdisc);
 		ASSERT_RTNL();
-		list_add_tail_rcu(&q->list, &root->list);
+		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
 	}
 }
-EXPORT_SYMBOL(qdisc_list_add);
+EXPORT_SYMBOL(qdisc_hash_add);
 
-void qdisc_list_del(struct Qdisc *q)
+void qdisc_hash_del(struct Qdisc *q)
 {
 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 		ASSERT_RTNL();
-		list_del_rcu(&q->list);
+		hash_del_rcu(&q->hash);
 	}
 }
-EXPORT_SYMBOL(qdisc_list_del);
+EXPORT_SYMBOL(qdisc_hash_del);
 
 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
 {
@@ -1004,7 +1005,7 @@ qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
 				goto err_out4;
 		}
 
-		qdisc_list_add(sch);
+		qdisc_hash_add(sch);
 
 		return sch;
 	}
@@ -1440,6 +1441,7 @@ static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
 {
 	int ret = 0, q_idx = *q_idx_p;
 	struct Qdisc *q;
+	int b;
 
 	if (!root)
 		return 0;
@@ -1454,7 +1456,7 @@ static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
 			goto done;
 		q_idx++;
 	}
-	list_for_each_entry(q, &root->list, list) {
+	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
 		if (q_idx < s_q_idx) {
 			q_idx++;
 			continue;
@@ -1771,6 +1773,7 @@ static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
 			       int *t_p, int s_t)
 {
 	struct Qdisc *q;
+	int b;
 
 	if (!root)
 		return 0;
@@ -1778,7 +1781,7 @@ static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
 		return -1;
 
-	list_for_each_entry(q, &root->list, list) {
+	hash_for_each_rcu(qdisc_dev(root)->qdisc_hash, b, q, hash) {
 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
 			return -1;
 	}
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index f9e0e9c..7efc923 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -378,7 +378,6 @@ struct Qdisc noop_qdisc = {
 	.dequeue	=	noop_dequeue,
 	.flags		=	TCQ_F_BUILTIN,
 	.ops		=	&noop_qdisc_ops,
-	.list		=	LIST_HEAD_INIT(noop_qdisc.list),
 	.q.lock		=	__SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
 	.dev_queue	=	&noop_netdev_queue,
 	.busylock	=	__SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
@@ -565,7 +564,6 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 		sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
 		sch->padded = (char *) sch - (char *) p;
 	}
-	INIT_LIST_HEAD(&sch->list);
 	skb_queue_head_init(&sch->q);
 
 	spin_lock_init(&sch->busylock);
@@ -645,7 +643,7 @@ void qdisc_destroy(struct Qdisc *qdisc)
 		return;
 
 #ifdef CONFIG_NET_SCHED
-	qdisc_list_del(qdisc);
+	qdisc_hash_del(qdisc);
 
 	qdisc_put_stab(rtnl_dereference(qdisc->stab));
 #endif
@@ -732,6 +730,8 @@ static void attach_default_qdiscs(struct net_device *dev)
 			qdisc->ops->attach(qdisc);
 		}
 	}
+	if (dev->qdisc)
+		qdisc_hash_add(dev->qdisc);
 }
 
 static void transition_one_qdisc(struct net_device *dev,
diff --git a/net/sched/sch_mq.c b/net/sched/sch_mq.c
index 56a77b8..3bee15d 100644
--- a/net/sched/sch_mq.c
+++ b/net/sched/sch_mq.c
@@ -88,7 +88,7 @@ static void mq_attach(struct Qdisc *sch)
 			qdisc_destroy(old);
 #ifdef CONFIG_NET_SCHED
 		if (ntx < dev->real_num_tx_queues)
-			qdisc_list_add(qdisc);
+			qdisc_hash_add(qdisc);
 #endif
 
 	}
diff --git a/net/sched/sch_mqprio.c b/net/sched/sch_mqprio.c
index b8002ce..dbfb3a5 100644
--- a/net/sched/sch_mqprio.c
+++ b/net/sched/sch_mqprio.c
@@ -182,7 +182,7 @@ static void mqprio_attach(struct Qdisc *sch)
 		if (old)
 			qdisc_destroy(old);
 		if (ntx < dev->real_num_tx_queues)
-			qdisc_list_add(qdisc);
+			qdisc_hash_add(qdisc);
 	}
 	kfree(priv->qdiscs);
 	priv->qdiscs = NULL;

-- 
Jiri Kosina
SUSE Labs

[RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Jiri Kosina]

From: Jiri Kosina <jkosina@suse.cz>

Convert the per-device linked list into a hashtable. The primary 
motivation for this change is that currently, we're not tracking all the 
qdiscs in hierarchy (e.g. excluding default qdiscs), as the lookup 
performed over the linked list by qdisc_match_from_root() is rather 
expensive.

The ultimate goal is to get rid of hidden qdiscs completely, which will 
bring much more determinism in user experience.

As we're adding hashtable.h include into generic netdevice.h, we have to make
sure HASH_SIZE macro is now non-conflicting with local definitions.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
---

v1 -> v2: 	fix up RCU hastable usage wrt. rtnl
	   	fix compilation of .c files which define their own 
		 HASH_SIZE that now oncflicts with the one from 
		 hashtable.h (newly included via netdevice.h)

 include/linux/netdevice.h |  2 ++
 include/net/pkt_sched.h   |  4 ++--
 include/net/sch_generic.h |  2 +-
 net/core/dev.c            |  1 +
 net/ipv6/ip6_gre.c        |  8 ++++----
 net/ipv6/ip6_tunnel.c     |  6 +++---
 net/ipv6/ip6_vti.c        |  6 +++---
 net/ipv6/sit.c            | 10 +++++-----
 net/sched/sch_api.c       | 23 +++++++++++++----------
 net/sched/sch_generic.c   |  6 +++---
 net/sched/sch_mq.c        |  2 +-
 net/sched/sch_mqprio.c    |  2 +-
 12 files changed, 39 insertions(+), 33 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index f45929c..630838e 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -52,6 +52,7 @@
 #include <uapi/linux/netdevice.h>
 #include <uapi/linux/if_bonding.h>
 #include <uapi/linux/pkt_cls.h>
+#include <linux/hashtable.h>
 
 struct netpoll_info;
 struct device;
@@ -1778,6 +1779,7 @@ struct net_device {
 	unsigned int		num_tx_queues;
 	unsigned int		real_num_tx_queues;
 	struct Qdisc		*qdisc;
+	DECLARE_HASHTABLE	(qdisc_hash, 16);
 	unsigned long		tx_queue_len;
 	spinlock_t		tx_global_lock;
 	int			watchdog_timeo;
diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
index fea53f4..8ba11b4 100644
--- a/include/net/pkt_sched.h
+++ b/include/net/pkt_sched.h
@@ -90,8 +90,8 @@ int unregister_qdisc(struct Qdisc_ops *qops);
 void qdisc_get_default(char *id, size_t len);
 int qdisc_set_default(const char *id);
 
-void qdisc_list_add(struct Qdisc *q);
-void qdisc_list_del(struct Qdisc *q);
+void qdisc_hash_add(struct Qdisc *q);
+void qdisc_hash_del(struct Qdisc *q);
 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
 struct Qdisc *qdisc_lookup_class(struct net_device *dev, u32 handle);
 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 62d5531..26f5cb3 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -67,7 +67,7 @@ struct Qdisc {
 	u32			limit;
 	const struct Qdisc_ops	*ops;
 	struct qdisc_size_table	__rcu *stab;
-	struct list_head	list;
+	struct hlist_node       hash;
 	u32			handle;
 	u32			parent;
 	int			(*reshape_fail)(struct sk_buff *skb,
diff --git a/net/core/dev.c b/net/core/dev.c
index 904ff43..edc1617 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -7511,6 +7511,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
 	INIT_LIST_HEAD(&dev->all_adj_list.lower);
 	INIT_LIST_HEAD(&dev->ptype_all);
 	INIT_LIST_HEAD(&dev->ptype_specific);
+	hash_init(dev->qdisc_hash);
 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
 	setup(dev);
 
diff --git a/net/ipv6/ip6_gre.c b/net/ipv6/ip6_gre.c
index fdc9de2..0f70ecc 100644
--- a/net/ipv6/ip6_gre.c
+++ b/net/ipv6/ip6_gre.c
@@ -62,11 +62,11 @@ module_param(log_ecn_error, bool, 0644);
 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
 
 #define HASH_SIZE_SHIFT  5
-#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
+#define __HASH_SIZE (1 << HASH_SIZE_SHIFT)
 
 static int ip6gre_net_id __read_mostly;
 struct ip6gre_net {
-	struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
+	struct ip6_tnl __rcu *tunnels[4][__HASH_SIZE];
 
 	struct net_device *fb_tunnel_dev;
 };
@@ -96,7 +96,7 @@ static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
    will match fallback tunnel.
  */
 
-#define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
+#define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(__HASH_SIZE - 1))
 static u32 HASH_ADDR(const struct in6_addr *addr)
 {
 	u32 hash = ipv6_addr_hash(addr);
@@ -1089,7 +1089,7 @@ static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
 
 	for (prio = 0; prio < 4; prio++) {
 		int h;
-		for (h = 0; h < HASH_SIZE; h++) {
+		for (h = 0; h < __HASH_SIZE; h++) {
 			struct ip6_tnl *t;
 
 			t = rtnl_dereference(ign->tunnels[prio][h]);
diff --git a/net/ipv6/ip6_tunnel.c b/net/ipv6/ip6_tunnel.c
index 7b0481e..a9da620 100644
--- a/net/ipv6/ip6_tunnel.c
+++ b/net/ipv6/ip6_tunnel.c
@@ -65,7 +65,7 @@ MODULE_ALIAS_RTNL_LINK("ip6tnl");
 MODULE_ALIAS_NETDEV("ip6tnl0");
 
 #define HASH_SIZE_SHIFT  5
-#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
+#define __HASH_SIZE (1 << HASH_SIZE_SHIFT)
 
 static bool log_ecn_error = true;
 module_param(log_ecn_error, bool, 0644);
@@ -87,7 +87,7 @@ struct ip6_tnl_net {
 	/* the IPv6 tunnel fallback device */
 	struct net_device *fb_tnl_dev;
 	/* lists for storing tunnels in use */
-	struct ip6_tnl __rcu *tnls_r_l[HASH_SIZE];
+	struct ip6_tnl __rcu *tnls_r_l[__HASH_SIZE];
 	struct ip6_tnl __rcu *tnls_wc[1];
 	struct ip6_tnl __rcu **tnls[2];
 };
@@ -2031,7 +2031,7 @@ static void __net_exit ip6_tnl_destroy_tunnels(struct net *net)
 		if (dev->rtnl_link_ops == &ip6_link_ops)
 			unregister_netdevice_queue(dev, &list);
 
-	for (h = 0; h < HASH_SIZE; h++) {
+	for (h = 0; h < __HASH_SIZE; h++) {
 		t = rtnl_dereference(ip6n->tnls_r_l[h]);
 		while (t) {
 			/* If dev is in the same netns, it has already
diff --git a/net/ipv6/ip6_vti.c b/net/ipv6/ip6_vti.c
index d90a11f..2d192af 100644
--- a/net/ipv6/ip6_vti.c
+++ b/net/ipv6/ip6_vti.c
@@ -51,7 +51,7 @@
 #include <net/netns/generic.h>
 
 #define HASH_SIZE_SHIFT  5
-#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
+#define __HASH_SIZE (1 << HASH_SIZE_SHIFT)
 
 static u32 HASH(const struct in6_addr *addr1, const struct in6_addr *addr2)
 {
@@ -69,7 +69,7 @@ struct vti6_net {
 	/* the vti6 tunnel fallback device */
 	struct net_device *fb_tnl_dev;
 	/* lists for storing tunnels in use */
-	struct ip6_tnl __rcu *tnls_r_l[HASH_SIZE];
+	struct ip6_tnl __rcu *tnls_r_l[__HASH_SIZE];
 	struct ip6_tnl __rcu *tnls_wc[1];
 	struct ip6_tnl __rcu **tnls[2];
 };
@@ -1040,7 +1040,7 @@ static void __net_exit vti6_destroy_tunnels(struct vti6_net *ip6n)
 	struct ip6_tnl *t;
 	LIST_HEAD(list);
 
-	for (h = 0; h < HASH_SIZE; h++) {
+	for (h = 0; h < __HASH_SIZE; h++) {
 		t = rtnl_dereference(ip6n->tnls_r_l[h]);
 		while (t) {
 			unregister_netdevice_queue(t->dev, &list);
diff --git a/net/ipv6/sit.c b/net/ipv6/sit.c
index 0a5a255..9f776d7 100644
--- a/net/ipv6/sit.c
+++ b/net/ipv6/sit.c
@@ -62,7 +62,7 @@
    For comments look at net/ipv4/ip_gre.c --ANK
  */
 
-#define HASH_SIZE  16
+#define __HASH_SIZE  16
 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
 
 static bool log_ecn_error = true;
@@ -78,9 +78,9 @@ static struct rtnl_link_ops sit_link_ops __read_mostly;
 
 static int sit_net_id __read_mostly;
 struct sit_net {
-	struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
-	struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
-	struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
+	struct ip_tunnel __rcu *tunnels_r_l[__HASH_SIZE];
+	struct ip_tunnel __rcu *tunnels_r[__HASH_SIZE];
+	struct ip_tunnel __rcu *tunnels_l[__HASH_SIZE];
 	struct ip_tunnel __rcu *tunnels_wc[1];
 	struct ip_tunnel __rcu **tunnels[4];
 
@@ -1773,7 +1773,7 @@ static void __net_exit sit_destroy_tunnels(struct net *net,
 
 	for (prio = 1; prio < 4; prio++) {
 		int h;
-		for (h = 0; h < HASH_SIZE; h++) {
+		for (h = 0; h < __HASH_SIZE; h++) {
 			struct ip_tunnel *t;
 
 			t = rtnl_dereference(sitn->tunnels[prio][h]);
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index ddf047d..c093d32 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -29,6 +29,7 @@
 #include <linux/hrtimer.h>
 #include <linux/lockdep.h>
 #include <linux/slab.h>
+#include <linux/hashtable.h>
 
 #include <net/net_namespace.h>
 #include <net/sock.h>
@@ -265,33 +266,33 @@ static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
 	    root->handle == handle)
 		return root;
 
-	list_for_each_entry_rcu(q, &root->list, list) {
+	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
 		if (q->handle == handle)
 			return q;
 	}
 	return NULL;
 }
 
-void qdisc_list_add(struct Qdisc *q)
+void qdisc_hash_add(struct Qdisc *q)
 {
 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 		struct Qdisc *root = qdisc_dev(q)->qdisc;
 
 		WARN_ON_ONCE(root == &noop_qdisc);
 		ASSERT_RTNL();
-		list_add_tail_rcu(&q->list, &root->list);
+		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
 	}
 }
-EXPORT_SYMBOL(qdisc_list_add);
+EXPORT_SYMBOL(qdisc_hash_add);
 
-void qdisc_list_del(struct Qdisc *q)
+void qdisc_hash_del(struct Qdisc *q)
 {
 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 		ASSERT_RTNL();
-		list_del_rcu(&q->list);
+		hash_del_rcu(&q->hash);
 	}
 }
-EXPORT_SYMBOL(qdisc_list_del);
+EXPORT_SYMBOL(qdisc_hash_del);
 
 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
 {
@@ -1004,7 +1005,7 @@ qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
 				goto err_out4;
 		}
 
-		qdisc_list_add(sch);
+		qdisc_hash_add(sch);
 
 		return sch;
 	}
@@ -1440,6 +1441,7 @@ static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
 {
 	int ret = 0, q_idx = *q_idx_p;
 	struct Qdisc *q;
+	int b;
 
 	if (!root)
 		return 0;
@@ -1454,7 +1456,7 @@ static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
 			goto done;
 		q_idx++;
 	}
-	list_for_each_entry(q, &root->list, list) {
+	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
 		if (q_idx < s_q_idx) {
 			q_idx++;
 			continue;
@@ -1771,6 +1773,7 @@ static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
 			       int *t_p, int s_t)
 {
 	struct Qdisc *q;
+	int b;
 
 	if (!root)
 		return 0;
@@ -1778,7 +1781,7 @@ static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
 		return -1;
 
-	list_for_each_entry(q, &root->list, list) {
+	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
 			return -1;
 	}
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index f9e0e9c..7efc923 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -378,7 +378,6 @@ struct Qdisc noop_qdisc = {
 	.dequeue	=	noop_dequeue,
 	.flags		=	TCQ_F_BUILTIN,
 	.ops		=	&noop_qdisc_ops,
-	.list		=	LIST_HEAD_INIT(noop_qdisc.list),
 	.q.lock		=	__SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
 	.dev_queue	=	&noop_netdev_queue,
 	.busylock	=	__SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
@@ -565,7 +564,6 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 		sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
 		sch->padded = (char *) sch - (char *) p;
 	}
-	INIT_LIST_HEAD(&sch->list);
 	skb_queue_head_init(&sch->q);
 
 	spin_lock_init(&sch->busylock);
@@ -645,7 +643,7 @@ void qdisc_destroy(struct Qdisc *qdisc)
 		return;
 
 #ifdef CONFIG_NET_SCHED
-	qdisc_list_del(qdisc);
+	qdisc_hash_del(qdisc);
 
 	qdisc_put_stab(rtnl_dereference(qdisc->stab));
 #endif
@@ -732,6 +730,8 @@ static void attach_default_qdiscs(struct net_device *dev)
 			qdisc->ops->attach(qdisc);
 		}
 	}
+	if (dev->qdisc)
+		qdisc_hash_add(dev->qdisc);
 }
 
 static void transition_one_qdisc(struct net_device *dev,
diff --git a/net/sched/sch_mq.c b/net/sched/sch_mq.c
index 56a77b8..3bee15d 100644
--- a/net/sched/sch_mq.c
+++ b/net/sched/sch_mq.c
@@ -88,7 +88,7 @@ static void mq_attach(struct Qdisc *sch)
 			qdisc_destroy(old);
 #ifdef CONFIG_NET_SCHED
 		if (ntx < dev->real_num_tx_queues)
-			qdisc_list_add(qdisc);
+			qdisc_hash_add(qdisc);
 #endif
 
 	}
diff --git a/net/sched/sch_mqprio.c b/net/sched/sch_mqprio.c
index b8002ce..dbfb3a5 100644
--- a/net/sched/sch_mqprio.c
+++ b/net/sched/sch_mqprio.c
@@ -182,7 +182,7 @@ static void mqprio_attach(struct Qdisc *sch)
 		if (old)
 			qdisc_destroy(old);
 		if (ntx < dev->real_num_tx_queues)
-			qdisc_list_add(qdisc);
+			qdisc_hash_add(qdisc);
 	}
 	kfree(priv->qdiscs);
 	priv->qdiscs = NULL;
-- 
Jiri Kosina
SUSE Labs

Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Eric Dumazet]

On Thu, 2016-07-07 at 22:36 +0200, Jiri Kosina wrote:
> From: Jiri Kosina <jkosina@suse.cz>
> 
> Convert the per-device linked list into a hashtable. The primary 
> motivation for this change is that currently, we're not tracking all the 
> qdiscs in hierarchy (e.g. excluding default qdiscs), as the lookup 
> performed over the linked list by qdisc_match_from_root() is rather 
> expensive.
> 
> The ultimate goal is to get rid of hidden qdiscs completely, which will 
> bring much more determinism in user experience.
> 
> As we're adding hashtable.h include into generic netdevice.h, we have to make
> sure HASH_SIZE macro is now non-conflicting with local definitions.
> 
> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
> ---


> diff --git a/net/ipv6/ip6_gre.c b/net/ipv6/ip6_gre.c
> index fdc9de2..0f70ecc 100644
> --- a/net/ipv6/ip6_gre.c
> +++ b/net/ipv6/ip6_gre.c
> @@ -62,11 +62,11 @@ module_param(log_ecn_error, bool, 0644);
>  MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
>  
>  #define HASH_SIZE_SHIFT  5
> -#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
> +#define __HASH_SIZE (1 << HASH_SIZE_SHIFT)

__ prefix is mostly used for functions having some kind of
shells/helpers.

I would rather use IP6_GRE_HASH_SIZE or something which has lower
chances of being used elsewhere.

Or maybe you could use new HASH_SIZE(name), providing proper 'name'

@@ -732,6 +730,8 @@ static void attach_default_qdiscs(struct net_device *dev)
>  			qdisc->ops->attach(qdisc);
>  		}
>  	}
> +	if (dev->qdisc)
> +		qdisc_hash_add(dev->qdisc);
>  }
>  

I do not understand this addition, could you comment on it ?

Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Jiri Kosina]

On Fri, 8 Jul 2016, Eric Dumazet wrote:

> > diff --git a/net/ipv6/ip6_gre.c b/net/ipv6/ip6_gre.c
> > index fdc9de2..0f70ecc 100644
> > --- a/net/ipv6/ip6_gre.c
> > +++ b/net/ipv6/ip6_gre.c
> > @@ -62,11 +62,11 @@ module_param(log_ecn_error, bool, 0644);
> >  MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
> >  
> >  #define HASH_SIZE_SHIFT  5
> > -#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
> > +#define __HASH_SIZE (1 << HASH_SIZE_SHIFT)
> 
> __ prefix is mostly used for functions having some kind of
> shells/helpers.
> 
> I would rather use IP6_GRE_HASH_SIZE or something which has lower
> chances of being used elsewhere.

Alright, makes sense, will do this in v3.

> @@ -732,6 +730,8 @@ static void attach_default_qdiscs(struct net_device *dev)
> >  			qdisc->ops->attach(qdisc);
> >  		}
> >  	}
> > +	if (dev->qdisc)
> > +		qdisc_hash_add(dev->qdisc);
> >  }
> >  
> 
> I do not understand this addition, could you comment on it ?

With linked lists, assigning to struct net_device's Qdisc pointer is 
enough to "initialize" the linked list and have it contain one (root) 
item. With hashtable, this is not the case, it needs to be explicitly 
added.

Hmm, dev_init_scheduler() (and perhaps also dev_shutdown()) would possibly 
need similar treatment in order to have accurate data there 100% of the 
time even during initialization.

-- 
Jiri Kosina
SUSE Labs

Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Thomas Graf]

On 07/07/16 at 10:36pm, Jiri Kosina wrote:
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index f45929c..630838e 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -52,6 +52,7 @@
>  #include <uapi/linux/netdevice.h>
>  #include <uapi/linux/if_bonding.h>
>  #include <uapi/linux/pkt_cls.h>
> +#include <linux/hashtable.h>
>  
>  struct netpoll_info;
>  struct device;
> @@ -1778,6 +1779,7 @@ struct net_device {
>  	unsigned int		num_tx_queues;
>  	unsigned int		real_num_tx_queues;
>  	struct Qdisc		*qdisc;
> +	DECLARE_HASHTABLE	(qdisc_hash, 16);

This blows up net_device to an insane size: 64K * sizeof(struct
hlist_head). Can we allocate this on demand for net_devices where
it is actually needed? The majority of virtual devices won't need
this. Doesn't have to be rhashtable, can still be fixed size but
at least allocate it.

Re: [RFC PATCH v2] net: sched: convert qdisc linked list to hashtable

[Eric Dumazet]

On Fri, 2016-07-08 at 13:07 +0200, Thomas Graf wrote:
> On 07/07/16 at 10:36pm, Jiri Kosina wrote:
> > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> > index f45929c..630838e 100644
> > --- a/include/linux/netdevice.h
> > +++ b/include/linux/netdevice.h
> > @@ -52,6 +52,7 @@
> >  #include <uapi/linux/netdevice.h>
> >  #include <uapi/linux/if_bonding.h>
> >  #include <uapi/linux/pkt_cls.h>
> > +#include <linux/hashtable.h>
> >  
> >  struct netpoll_info;
> >  struct device;
> > @@ -1778,6 +1779,7 @@ struct net_device {
> >  	unsigned int		num_tx_queues;
> >  	unsigned int		real_num_tx_queues;
> >  	struct Qdisc		*qdisc;
> > +	DECLARE_HASHTABLE	(qdisc_hash, 16);
> 
> This blows up net_device to an insane size: 64K * sizeof(struct
> hlist_head). Can we allocate this on demand for net_devices where
> it is actually needed? The majority of virtual devices won't need
> this. Doesn't have to be rhashtable, can still be fixed size but
> at least allocate it.


Jiri probably misread the API and should have used :

DECLARE_HASHTABLE	(qdisc_hash, 4);

Google has a very similar patch with 16 buckets, and it is 'good enough',
although we do not hit the qdisc_tree_reduce_backlog() penalty.