netdev - Re: [PATCH][net-next][v2] net: limit each hash list length to MAX_GRO

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Sun, 08 Jul 2018 10:32:07 +0900 (KST)
From:   David Miller <davem@...emloft.net>
To:     eric.dumazet@...il.com
Cc:     lirongqing@...du.com, netdev@...r.kernel.org
Subject: Re: [PATCH][net-next][v2] net: limit each hash list length to
 MAX_GRO_SKBS

From: Eric Dumazet <eric.dumazet@...il.com>
Date: Sat, 7 Jul 2018 17:22:01 -0700

> Maybe gro_count should be replaced by a bitmask, so that we can
> speed up napi_gro_flush(), since it now has to use 3 cache lines
> (gro_hash[] size is 192 bytes)

Something like this?

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index b683971e500d..a4d859a7e9de 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -310,7 +310,7 @@ struct gro_list {
 /*
  * Structure for NAPI scheduling similar to tasklet but with weighting
  */
-#define GRO_HASH_BUCKETS	8
+#define GRO_HASH_BUCKETS	8	/* Must be <= 32 due to gro_mask */
 struct napi_struct {
 	/* The poll_list must only be managed by the entity which
 	 * changes the state of the NAPI_STATE_SCHED bit.  This means
@@ -322,7 +322,7 @@ struct napi_struct {
 
 	unsigned long		state;
 	int			weight;
-	unsigned int		gro_count;
+	unsigned int		gro_hash_mask;
 	int			(*poll)(struct napi_struct *, int);
 #ifdef CONFIG_NETPOLL
 	int			poll_owner;
diff --git a/net/core/dev.c b/net/core/dev.c
index 89825c1eccdc..0dfb84a82586 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -5158,12 +5158,14 @@ static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
 
 	list_for_each_entry_safe_reverse(skb, p, head, list) {
 		if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
-			return;
+			goto out;
 		list_del_init(&skb->list);
 		napi_gro_complete(skb);
-		napi->gro_count--;
 		napi->gro_hash[index].count--;
 	}
+out:
+	if (list_empty(head))
+		napi->gro_hash_mask &= ~(1 << index);
 }
 
 /* napi->gro_hash[].list contains packets ordered by age.
@@ -5174,8 +5176,10 @@ void napi_gro_flush(struct napi_struct *napi, bool flush_old)
 {
 	u32 i;
 
-	for (i = 0; i < GRO_HASH_BUCKETS; i++)
-		__napi_gro_flush_chain(napi, i, flush_old);
+	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
+		if (napi->gro_hash_mask & (1 << i))
+			__napi_gro_flush_chain(napi, i, flush_old);
+	}
 }
 EXPORT_SYMBOL(napi_gro_flush);
 
@@ -5267,8 +5271,8 @@ static void gro_flush_oldest(struct list_head *head)
 	if (WARN_ON_ONCE(!oldest))
 		return;
 
-	/* Do not adjust napi->gro_count, caller is adding a new SKB to
-	 * the chain.
+	/* Do not adjust napi->gro_hash_mask, caller is adding a new
+	 * SKB to the chain.
 	 */
 	list_del(&oldest->list);
 	napi_gro_complete(oldest);
@@ -5342,8 +5346,9 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
 	if (pp) {
 		list_del_init(&pp->list);
 		napi_gro_complete(pp);
-		napi->gro_count--;
 		napi->gro_hash[hash].count--;
+		if (list_empty(&napi->gro_hash[hash].list))
+			napi->gro_hash_mask &= ~(1 << hash);
 	}
 
 	if (same_flow)
@@ -5355,7 +5360,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
 	if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
 		gro_flush_oldest(gro_head);
 	} else {
-		napi->gro_count++;
+		napi->gro_hash_mask |= (1 << hash);
 		napi->gro_hash[hash].count++;
 	}
 	NAPI_GRO_CB(skb)->count = 1;
@@ -5768,7 +5773,7 @@ bool napi_complete_done(struct napi_struct *n, int work_done)
 				 NAPIF_STATE_IN_BUSY_POLL)))
 		return false;
 
-	if (n->gro_count) {
+	if (n->gro_hash_mask) {
 		unsigned long timeout = 0;
 
 		if (work_done)
@@ -5977,7 +5982,7 @@ static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
 	 */
-	if (napi->gro_count && !napi_disable_pending(napi) &&
+	if (napi->gro_hash_mask && !napi_disable_pending(napi) &&
 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
 		__napi_schedule_irqoff(napi);
 
@@ -5992,7 +5997,7 @@ void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 	INIT_LIST_HEAD(&napi->poll_list);
 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
 	napi->timer.function = napi_watchdog;
-	napi->gro_count = 0;
+	napi->gro_hash_mask = 0;
 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
 		napi->gro_hash[i].count = 0;
@@ -6052,7 +6057,7 @@ void netif_napi_del(struct napi_struct *napi)
 	napi_free_frags(napi);
 
 	flush_gro_hash(napi);
-	napi->gro_count = 0;
+	napi->gro_hash_mask = 0;
 }
 EXPORT_SYMBOL(netif_napi_del);
 
@@ -6094,7 +6099,7 @@ static int napi_poll(struct napi_struct *n, struct list_head *repoll)
 		goto out_unlock;
 	}
 
-	if (n->gro_count) {
+	if (n->gro_hash_mask) {
 		/* flush too old packets
 		 * If HZ < 1000, flush all packets.
 		 */