diff mbox series

[net-next,v6,1/6] neighbour: Add hlist_node to struct neighbour

Message ID 20241021102102.2560279-2-gnaaman@drivenets.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series neighbour: Improve neigh_flush_dev performance | expand

Checks

Context Check Description
netdev/series_format success Posting correctly formatted
netdev/tree_selection success Clearly marked for net-next, async
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 18 this patch: 18
netdev/build_tools success Errors and warnings before: 157 (+0) this patch: 157 (+0)
netdev/cc_maintainers success CCed 4 of 4 maintainers
netdev/build_clang success Errors and warnings before: 36 this patch: 36
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 2624 this patch: 2624
netdev/checkpatch warning WARNING: line length of 81 exceeds 80 columns WARNING: line length of 82 exceeds 80 columns WARNING: line length of 83 exceeds 80 columns WARNING: line length of 84 exceeds 80 columns
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0
netdev/contest fail net-next-2024-10-21--18-00 (tests: 766)

Commit Message

Gilad Naaman Oct. 21, 2024, 10:20 a.m. UTC 
Add a doubly-linked node to neighbours, so that they
can be deleted without iterating the entire bucket they're in.

Signed-off-by: Gilad Naaman <gnaaman@drivenets.com>
---
 include/net/neighbour.h |  2 ++
 net/core/neighbour.c    | 40 ++++++++++++++++++++++++++++++++++++++--
 2 files changed, 40 insertions(+), 2 deletions(-)

Comments

Kuniyuki Iwashima Oct. 21, 2024, 7:11 p.m. UTC | #1 
From: Gilad Naaman <gnaaman@drivenets.com>
Date: Mon, 21 Oct 2024 10:20:53 +0000
> @@ -530,27 +532,47 @@ static void neigh_get_hash_rnd(u32 *x)
>  
>  static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
>  {
> +	size_t hash_heads_size = (1 << shift) * sizeof(struct hlist_head);
>  	size_t size = (1 << shift) * sizeof(struct neighbour *);
> -	struct neigh_hash_table *ret;
>  	struct neighbour __rcu **buckets;
> +	struct hlist_head *hash_heads;
> +	struct neigh_hash_table *ret;
>  	int i;
>  
> +	hash_heads = NULL;

nit: This init is not needed.


> @@ -564,6 +586,8 @@ static void neigh_hash_free_rcu(struct rcu_head *head)
>  						    rcu);
>  	size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
>  	struct neighbour __rcu **buckets = nht->hash_buckets;
> +	size_t hash_heads_size = (1 << nht->hash_shift) * sizeof(struct hlist_head);
> +	struct hlist_head *hash_heads = nht->hash_heads;

nit: reverse xmas tree order.
Gilad Naaman Oct. 21, 2024, 7:36 p.m. UTC | #2 
> > @@ -530,27 +532,47 @@ static void neigh_get_hash_rnd(u32 *x)
> >  
> >  static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
> >  {
> > +	size_t hash_heads_size = (1 << shift) * sizeof(struct hlist_head);
> >  	size_t size = (1 << shift) * sizeof(struct neighbour *);
> > -	struct neigh_hash_table *ret;
> >  	struct neighbour __rcu **buckets;
> > +	struct hlist_head *hash_heads;
> > +	struct neigh_hash_table *ret;
> >  	int i;
> >  
> > +	hash_heads = NULL;
> 
> nit: This init is not needed.

This is needed in order to prevent unitialized memory access if we failed to
allocate `buckets`.

If possible I'd prefer to leave this as-is, given that this is rewritten later,
in commit 5.

> > @@ -564,6 +586,8 @@ static void neigh_hash_free_rcu(struct rcu_head *head)
> >  						    rcu);
> >  	size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
> >  	struct neighbour __rcu **buckets = nht->hash_buckets;
> > +	size_t hash_heads_size = (1 << nht->hash_shift) * sizeof(struct hlist_head);
> > +	struct hlist_head *hash_heads = nht->hash_heads;
> 
> nit: reverse xmas tree order.

Apologies, I thought I squashed all of these, I'll be more thorough next time.
Kuniyuki Iwashima Oct. 21, 2024, 7:40 p.m. UTC | #3 
From: Gilad Naaman <gnaaman@drivenets.com>
Date: Mon, 21 Oct 2024 19:36:22 +0000
> > > @@ -530,27 +532,47 @@ static void neigh_get_hash_rnd(u32 *x)
> > >  
> > >  static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
> > >  {
> > > +	size_t hash_heads_size = (1 << shift) * sizeof(struct hlist_head);
> > >  	size_t size = (1 << shift) * sizeof(struct neighbour *);
> > > -	struct neigh_hash_table *ret;
> > >  	struct neighbour __rcu **buckets;
> > > +	struct hlist_head *hash_heads;
> > > +	struct neigh_hash_table *ret;
> > >  	int i;
> > >  
> > > +	hash_heads = NULL;
> > 
> > nit: This init is not needed.
> 
> This is needed in order to prevent unitialized memory access if we failed to
> allocate `buckets`.

Ah, you are right.

> 
> If possible I'd prefer to leave this as-is, given that this is rewritten later,
> in commit 5.

Sounds good.  Let's remove that in patch 5.
diff mbox series

Patch

diff --git a/include/net/neighbour.h b/include/net/neighbour.h
index 3887ed9e5026..0402447854c7 100644
--- a/include/net/neighbour.h
+++ b/include/net/neighbour.h
@@ -136,6 +136,7 @@  struct neigh_statistics {
 
 struct neighbour {
 	struct neighbour __rcu	*next;
+	struct hlist_node	hash;
 	struct neigh_table	*tbl;
 	struct neigh_parms	*parms;
 	unsigned long		confirmed;
@@ -191,6 +192,7 @@  struct pneigh_entry {
 
 struct neigh_hash_table {
 	struct neighbour __rcu	**hash_buckets;
+	struct hlist_head	*hash_heads;
 	unsigned int		hash_shift;
 	__u32			hash_rnd[NEIGH_NUM_HASH_RND];
 	struct rcu_head		rcu;
diff --git a/net/core/neighbour.c b/net/core/neighbour.c
index 395ae1626eef..45c8df801dfb 100644
--- a/net/core/neighbour.c
+++ b/net/core/neighbour.c
@@ -217,6 +217,7 @@  static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
 		neigh = rcu_dereference_protected(n->next,
 						  lockdep_is_held(&tbl->lock));
 		rcu_assign_pointer(*np, neigh);
+		hlist_del_rcu(&n->hash);
 		neigh_mark_dead(n);
 		retval = true;
 	}
@@ -403,6 +404,7 @@  static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
 			rcu_assign_pointer(*np,
 				   rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));
+			hlist_del_rcu(&n->hash);
 			write_lock(&n->lock);
 			neigh_del_timer(n);
 			neigh_mark_dead(n);
@@ -530,27 +532,47 @@  static void neigh_get_hash_rnd(u32 *x)
 
 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 {
+	size_t hash_heads_size = (1 << shift) * sizeof(struct hlist_head);
 	size_t size = (1 << shift) * sizeof(struct neighbour *);
-	struct neigh_hash_table *ret;
 	struct neighbour __rcu **buckets;
+	struct hlist_head *hash_heads;
+	struct neigh_hash_table *ret;
 	int i;
 
+	hash_heads = NULL;
+
 	ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 	if (!ret)
 		return NULL;
 	if (size <= PAGE_SIZE) {
 		buckets = kzalloc(size, GFP_ATOMIC);
+
+		if (buckets) {
+			hash_heads = kzalloc(hash_heads_size, GFP_ATOMIC);
+			if (!hash_heads)
+				kfree(buckets);
+		}
 	} else {
 		buckets = (struct neighbour __rcu **)
 			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 					   get_order(size));
 		kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
+
+		if (buckets) {
+			hash_heads = (struct hlist_head *)
+				__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
+						 get_order(hash_heads_size));
+			kmemleak_alloc(hash_heads, hash_heads_size, 1, GFP_ATOMIC);
+			if (!hash_heads)
+				free_pages((unsigned long)buckets, get_order(size));
+		}
 	}
-	if (!buckets) {
+	if (!buckets || !hash_heads) {
 		kfree(ret);
 		return NULL;
 	}
 	ret->hash_buckets = buckets;
+	ret->hash_heads = hash_heads;
 	ret->hash_shift = shift;
 	for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 		neigh_get_hash_rnd(&ret->hash_rnd[i]);
@@ -564,6 +586,8 @@  static void neigh_hash_free_rcu(struct rcu_head *head)
 						    rcu);
 	size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 	struct neighbour __rcu **buckets = nht->hash_buckets;
+	size_t hash_heads_size = (1 << nht->hash_shift) * sizeof(struct hlist_head);
+	struct hlist_head *hash_heads = nht->hash_heads;
 
 	if (size <= PAGE_SIZE) {
 		kfree(buckets);
@@ -571,6 +595,13 @@  static void neigh_hash_free_rcu(struct rcu_head *head)
 		kmemleak_free(buckets);
 		free_pages((unsigned long)buckets, get_order(size));
 	}
+
+	if (hash_heads_size < PAGE_SIZE) {
+		kfree(hash_heads);
+	} else {
+		kmemleak_free(hash_heads);
+		free_pages((unsigned long)hash_heads, get_order(hash_heads_size));
+	}
 	kfree(nht);
 }
 
@@ -607,6 +638,8 @@  static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 						new_nht->hash_buckets[hash],
 						lockdep_is_held(&tbl->lock)));
 			rcu_assign_pointer(new_nht->hash_buckets[hash], n);
+			hlist_del_rcu(&n->hash);
+			hlist_add_head_rcu(&n->hash, &new_nht->hash_heads[hash]);
 		}
 	}
 
@@ -717,6 +750,7 @@  ___neigh_create(struct neigh_table *tbl, const void *pkey,
 			   rcu_dereference_protected(nht->hash_buckets[hash_val],
 						     lockdep_is_held(&tbl->lock)));
 	rcu_assign_pointer(nht->hash_buckets[hash_val], n);
+	hlist_add_head_rcu(&n->hash, &nht->hash_heads[hash_val]);
 	write_unlock_bh(&tbl->lock);
 	neigh_dbg(2, "neigh %p is created\n", n);
 	rc = n;
@@ -1002,6 +1036,7 @@  static void neigh_periodic_work(struct work_struct *work)
 				rcu_assign_pointer(*np,
 					rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));
+				hlist_del_rcu(&n->hash);
 				neigh_mark_dead(n);
 				write_unlock(&n->lock);
 				neigh_cleanup_and_release(n);
@@ -3131,6 +3166,7 @@  void __neigh_for_each_release(struct neigh_table *tbl,
 				rcu_assign_pointer(*np,
 					rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));
+				hlist_del_rcu(&n->hash);
 				neigh_mark_dead(n);
 			} else
 				np = &n->next;