[PATCH v2 net-next 0/6] af_unix: Introduce per-netns socket hash table.

[PATCH v2 net-next 0/6] af_unix: Introduce per-netns socket hash table.

[Kuniyuki Iwashima]

This series replaces unix_socket_table with a per-netns hash table and
reduces lock contention and time on iterating over the list.

Note the 3rd-6th patches can be a single patch, but for ease of review,
they are split into small changes without breakage.


Changes:
  v2:
    3rd:
      * Update changelog
      * Remove holes from per-netns hash table structure
      * Use kvmalloc_array() instead of kmalloc() (Eric Dumazet)
      * Remove unnecessary parts in af_unix_init() (Eric Dumazet)
      * Move `err_sysctl` label into ifdef block (kernel test robot)
      * Remove struct netns_unix from struct net if CONFIG_UNIX is disabled
    4th:
      * Use spin_lock_nested() (kernel test robot)

  v1: https://lore.kernel.org/netdev/20220616234714.4291-1-kuniyu@amazon.com/


Kuniyuki Iwashima (6):
  af_unix: Clean up some sock_net() uses.
  af_unix: Include the whole hash table size in UNIX_HASH_SIZE.
  af_unix: Define a per-netns hash table.
  af_unix: Acquire/Release per-netns hash table's locks.
  af_unix: Put a socket into a per-netns hash table.
  af_unix: Remove unix_table_locks.

 include/net/af_unix.h       |   5 +-
 include/net/net_namespace.h |   2 +
 include/net/netns/unix.h    |   6 +
 net/unix/af_unix.c          | 216 +++++++++++++++++++-----------------
 net/unix/diag.c             |  46 ++++----
 5 files changed, 145 insertions(+), 130 deletions(-)

-- 
2.30.2

[PATCH v2 net-next 1/6] af_unix: Clean up some sock_net() uses.

[Kuniyuki Iwashima]

Some functions define a net pointer only for one-shot use.  Others call
sock_net() redundantly even when a net pointer is available.  Let's fix
these and make the code simpler.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 net/unix/af_unix.c | 33 ++++++++++++++-------------------
 net/unix/diag.c    |  3 +--
 2 files changed, 15 insertions(+), 21 deletions(-)

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 3453e0053f76..990257f02e7c 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -932,7 +932,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock, int kern,
 	memset(&u->scm_stat, 0, sizeof(struct scm_stat));
 	unix_insert_unbound_socket(sk);
 
-	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
+	sock_prot_inuse_add(net, sk->sk_prot, 1);
 
 	return sk;
 
@@ -1293,9 +1293,8 @@ static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 			      int alen, int flags)
 {
-	struct sock *sk = sock->sk;
-	struct net *net = sock_net(sk);
 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
+	struct sock *sk = sock->sk;
 	struct sock *other;
 	int err;
 
@@ -1316,7 +1315,7 @@ static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 		}
 
 restart:
-		other = unix_find_other(net, sunaddr, alen, sock->type);
+		other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
 		if (IS_ERR(other)) {
 			err = PTR_ERR(other);
 			goto out;
@@ -1404,15 +1403,13 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 			       int addr_len, int flags)
 {
 	struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
-	struct sock *sk = sock->sk;
-	struct net *net = sock_net(sk);
+	struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
 	struct unix_sock *u = unix_sk(sk), *newu, *otheru;
-	struct sock *newsk = NULL;
-	struct sock *other = NULL;
+	struct net *net = sock_net(sk);
 	struct sk_buff *skb = NULL;
-	int st;
-	int err;
 	long timeo;
+	int err;
+	int st;
 
 	err = unix_validate_addr(sunaddr, addr_len);
 	if (err)
@@ -1432,7 +1429,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 	 */
 
 	/* create new sock for complete connection */
-	newsk = unix_create1(sock_net(sk), NULL, 0, sock->type);
+	newsk = unix_create1(net, NULL, 0, sock->type);
 	if (IS_ERR(newsk)) {
 		err = PTR_ERR(newsk);
 		newsk = NULL;
@@ -1840,17 +1837,15 @@ static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
 static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 			      size_t len)
 {
-	struct sock *sk = sock->sk;
-	struct net *net = sock_net(sk);
-	struct unix_sock *u = unix_sk(sk);
 	DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
-	struct sock *other = NULL;
-	int err;
-	struct sk_buff *skb;
-	long timeo;
+	struct sock *sk = sock->sk, *other = NULL;
+	struct unix_sock *u = unix_sk(sk);
 	struct scm_cookie scm;
+	struct sk_buff *skb;
 	int data_len = 0;
 	int sk_locked;
+	long timeo;
+	int err;
 
 	wait_for_unix_gc();
 	err = scm_send(sock, msg, &scm, false);
@@ -1917,7 +1912,7 @@ static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
 		if (sunaddr == NULL)
 			goto out_free;
 
-		other = unix_find_other(net, sunaddr, msg->msg_namelen,
+		other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
 					sk->sk_type);
 		if (IS_ERR(other)) {
 			err = PTR_ERR(other);
diff --git a/net/unix/diag.c b/net/unix/diag.c
index bb0b5ea1655f..4e3dc8179fa4 100644
--- a/net/unix/diag.c
+++ b/net/unix/diag.c
@@ -308,7 +308,6 @@ static int unix_diag_get_exact(struct sk_buff *in_skb,
 static int unix_diag_handler_dump(struct sk_buff *skb, struct nlmsghdr *h)
 {
 	int hdrlen = sizeof(struct unix_diag_req);
-	struct net *net = sock_net(skb->sk);
 
 	if (nlmsg_len(h) < hdrlen)
 		return -EINVAL;
@@ -317,7 +316,7 @@ static int unix_diag_handler_dump(struct sk_buff *skb, struct nlmsghdr *h)
 		struct netlink_dump_control c = {
 			.dump = unix_diag_dump,
 		};
-		return netlink_dump_start(net->diag_nlsk, skb, h, &c);
+		return netlink_dump_start(sock_net(skb->sk)->diag_nlsk, skb, h, &c);
 	} else
 		return unix_diag_get_exact(skb, h, nlmsg_data(h));
 }
-- 
2.30.2

[PATCH v2 net-next 2/6] af_unix: Include the whole hash table size in UNIX_HASH_SIZE.

[Kuniyuki Iwashima]

Currently, the size of AF_UNIX hash table is UNIX_HASH_SIZE * 2,
the first half for bind()ed sockets and the second half for unbound
ones.  UNIX_HASH_SIZE * 2 is used to define the table and iterate
over it.

In some places, we use ARRAY_SIZE(unix_socket_table) instead of
UNIX_HASH_SIZE * 2.  However, we cannot use it anymore because we
will allocate the hash table dynamically.  Then, we would have to
add UNIX_HASH_SIZE * 2 in many places, which would be troublesome.

This patch adapts the UNIX_HASH_SIZE definition to include bound
and unbound sockets and defines a new UNIX_HASH_MOD macro to ease
calculations.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 include/net/af_unix.h |  7 ++++---
 net/unix/af_unix.c    | 18 +++++++++---------
 net/unix/diag.c       |  6 ++----
 3 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/include/net/af_unix.h b/include/net/af_unix.h
index a7ef624ed726..acb56e463db1 100644
--- a/include/net/af_unix.h
+++ b/include/net/af_unix.h
@@ -16,12 +16,13 @@ void wait_for_unix_gc(void);
 struct sock *unix_get_socket(struct file *filp);
 struct sock *unix_peer_get(struct sock *sk);
 
-#define UNIX_HASH_SIZE	256
+#define UNIX_HASH_MOD	(256 - 1)
+#define UNIX_HASH_SIZE	(256 * 2)
 #define UNIX_HASH_BITS	8
 
 extern unsigned int unix_tot_inflight;
-extern spinlock_t unix_table_locks[2 * UNIX_HASH_SIZE];
-extern struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
+extern spinlock_t unix_table_locks[UNIX_HASH_SIZE];
+extern struct hlist_head unix_socket_table[UNIX_HASH_SIZE];
 
 struct unix_address {
 	refcount_t	refcnt;
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 990257f02e7c..c0804ae9c96a 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -118,9 +118,9 @@
 
 #include "scm.h"
 
-spinlock_t unix_table_locks[2 * UNIX_HASH_SIZE];
+spinlock_t unix_table_locks[UNIX_HASH_SIZE];
 EXPORT_SYMBOL_GPL(unix_table_locks);
-struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
+struct hlist_head unix_socket_table[UNIX_HASH_SIZE];
 EXPORT_SYMBOL_GPL(unix_socket_table);
 static atomic_long_t unix_nr_socks;
 
@@ -137,12 +137,12 @@ static unsigned int unix_unbound_hash(struct sock *sk)
 	hash ^= hash >> 8;
 	hash ^= sk->sk_type;
 
-	return UNIX_HASH_SIZE + (hash & (UNIX_HASH_SIZE - 1));
+	return UNIX_HASH_MOD + 1 + (hash & UNIX_HASH_MOD);
 }
 
 static unsigned int unix_bsd_hash(struct inode *i)
 {
-	return i->i_ino & (UNIX_HASH_SIZE - 1);
+	return i->i_ino & UNIX_HASH_MOD;
 }
 
 static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
@@ -155,14 +155,14 @@ static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
 	hash ^= hash >> 8;
 	hash ^= type;
 
-	return hash & (UNIX_HASH_SIZE - 1);
+	return hash & UNIX_HASH_MOD;
 }
 
 static void unix_table_double_lock(unsigned int hash1, unsigned int hash2)
 {
 	/* hash1 and hash2 is never the same because
-	 * one is between 0 and UNIX_HASH_SIZE - 1, and
-	 * another is between UNIX_HASH_SIZE and UNIX_HASH_SIZE * 2.
+	 * one is between 0 and UNIX_HASH_MOD, and
+	 * another is between UNIX_HASH_MOD + 1 and UNIX_HASH_SIZE - 1.
 	 */
 	if (hash1 > hash2)
 		swap(hash1, hash2);
@@ -3239,7 +3239,7 @@ static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
 	unsigned long bucket = get_bucket(*pos);
 	struct sock *sk;
 
-	while (bucket < ARRAY_SIZE(unix_socket_table)) {
+	while (bucket < UNIX_HASH_SIZE) {
 		spin_lock(&unix_table_locks[bucket]);
 
 		sk = unix_from_bucket(seq, pos);
@@ -3666,7 +3666,7 @@ static int __init af_unix_init(void)
 
 	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
 
-	for (i = 0; i < 2 * UNIX_HASH_SIZE; i++)
+	for (i = 0; i < UNIX_HASH_SIZE; i++)
 		spin_lock_init(&unix_table_locks[i]);
 
 	rc = proto_register(&unix_dgram_proto, 1);
diff --git a/net/unix/diag.c b/net/unix/diag.c
index 4e3dc8179fa4..c5d1cca72aa5 100644
--- a/net/unix/diag.c
+++ b/net/unix/diag.c
@@ -204,9 +204,7 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 	s_slot = cb->args[0];
 	num = s_num = cb->args[1];
 
-	for (slot = s_slot;
-	     slot < ARRAY_SIZE(unix_socket_table);
-	     s_num = 0, slot++) {
+	for (slot = s_slot; slot < UNIX_HASH_SIZE; s_num = 0, slot++) {
 		struct sock *sk;
 
 		num = 0;
@@ -242,7 +240,7 @@ static struct sock *unix_lookup_by_ino(unsigned int ino)
 	struct sock *sk;
 	int i;
 
-	for (i = 0; i < ARRAY_SIZE(unix_socket_table); i++) {
+	for (i = 0; i < UNIX_HASH_SIZE; i++) {
 		spin_lock(&unix_table_locks[i]);
 		sk_for_each(sk, &unix_socket_table[i])
 			if (ino == sock_i_ino(sk)) {
-- 
2.30.2

[PATCH v2 net-next 3/6] af_unix: Define a per-netns hash table.

[Kuniyuki Iwashima]

This commit adds a per netns hash table for AF_UNIX, which size is fixed
as UNIX_HASH_SIZE for now.

The first implementation defines a per-netns hash table as a single array
of lock and list:

	struct unix_hashbucket {
		spinlock_t		lock;
		struct hlist_head	head;
	};

	struct netns_unix {
		struct unix_hashbucket	*hash;
		...
	};

But, Eric pointed out memory cost that the structure has holes because of
sizeof(spinlock_t), which is 4 (or more if LOCKDEP is enabled). [0]  It
could be expensive on a host with thousands of netns and few AF_UNIX
sockets.  For this reason, a per-netns hash table uses two dense arrays.

	struct unix_table {
		spinlock_t		*locks;
		struct hlist_head	*buckets;
	};

	struct netns_unix {
		struct unix_table	table;
		...
	};

Note the length of the list has a significant impact rather than lock
contention, so having shared locks can be an option.  But, per-netns
locks and lists still perform better than the global locks and per-netns
lists. [1]

Also, this patch adds a change so that struct netns_unix disappears from
struct net if CONFIG_UNIX is disabled.

[0]: https://lore.kernel.org/netdev/CANn89iLVxO5aqx16azNU7p7Z-nz5NrnM5QTqOzueVxEnkVTxyg@mail.gmail.com/
[1]: https://lore.kernel.org/netdev/20220617175215.1769-1-kuniyu@amazon.com/

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 include/net/net_namespace.h |  2 ++
 include/net/netns/unix.h    |  6 ++++++
 net/unix/af_unix.c          | 38 +++++++++++++++++++++++++++++++------
 3 files changed, 40 insertions(+), 6 deletions(-)

diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h
index c4f5601f6e32..20a2992901c2 100644
--- a/include/net/net_namespace.h
+++ b/include/net/net_namespace.h
@@ -120,7 +120,9 @@ struct net {
 	struct netns_core	core;
 	struct netns_mib	mib;
 	struct netns_packet	packet;
+#if IS_ENABLED(CONFIG_UNIX)
 	struct netns_unix	unx;
+#endif
 	struct netns_nexthop	nexthop;
 	struct netns_ipv4	ipv4;
 #if IS_ENABLED(CONFIG_IPV6)
diff --git a/include/net/netns/unix.h b/include/net/netns/unix.h
index 91a3d7e39198..6f1a33df061d 100644
--- a/include/net/netns/unix.h
+++ b/include/net/netns/unix.h
@@ -5,8 +5,14 @@
 #ifndef __NETNS_UNIX_H__
 #define __NETNS_UNIX_H__
 
+struct unix_table {
+	spinlock_t		*locks;
+	struct hlist_head	*buckets;
+};
+
 struct ctl_table_header;
 struct netns_unix {
+	struct unix_table	table;
 	int			sysctl_max_dgram_qlen;
 	struct ctl_table_header	*ctl;
 };
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index c0804ae9c96a..cdd12881a39d 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -3559,7 +3559,7 @@ static const struct net_proto_family unix_family_ops = {
 
 static int __net_init unix_net_init(struct net *net)
 {
-	int error = -ENOMEM;
+	int i;
 
 	net->unx.sysctl_max_dgram_qlen = 10;
 	if (unix_sysctl_register(net))
@@ -3567,18 +3567,44 @@ static int __net_init unix_net_init(struct net *net)
 
 #ifdef CONFIG_PROC_FS
 	if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
-			sizeof(struct seq_net_private))) {
-		unix_sysctl_unregister(net);
-		goto out;
+			     sizeof(struct seq_net_private)))
+		goto err_sysctl;
+#endif
+
+	net->unx.table.locks = kvmalloc_array(UNIX_HASH_SIZE,
+					      sizeof(spinlock_t), GFP_KERNEL);
+	if (!net->unx.table.locks)
+		goto err_proc;
+
+	net->unx.table.buckets = kvmalloc_array(UNIX_HASH_SIZE,
+						sizeof(struct hlist_head),
+						GFP_KERNEL);
+	if (!net->unx.table.buckets)
+		goto free_locks;
+
+	for (i = 0; i < UNIX_HASH_SIZE; i++) {
+		spin_lock_init(&net->unx.table.locks[i]);
+		INIT_HLIST_HEAD(&net->unx.table.buckets[i]);
 	}
+
+	return 0;
+
+free_locks:
+	kvfree(net->unx.table.locks);
+err_proc:
+#ifdef CONFIG_PROC_FS
+	remove_proc_entry("unix", net->proc_net);
+err_sysctl:
 #endif
-	error = 0;
+	unix_sysctl_unregister(net);
 out:
-	return error;
+	return -ENOMEM;
 }
 
 static void __net_exit unix_net_exit(struct net *net)
 {
+	kvfree(net->unx.table.buckets);
+	kvfree(net->unx.table.locks);
 	unix_sysctl_unregister(net);
 	remove_proc_entry("unix", net->proc_net);
 }
-- 
2.30.2

[PATCH v2 net-next 4/6] af_unix: Acquire/Release per-netns hash table's locks.

[Kuniyuki Iwashima]

This commit adds extra spin_lock/spin_unlock() for a per-netns
hash table inside the existing ones for unix_table_locks.

As of this commit, sockets are still linked in the global hash
table.  After putting sockets in a per-netns hash table and
removing the old one in the next patch, we remove the global
locks in the last patch.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 net/unix/af_unix.c | 75 +++++++++++++++++++++++++++++++---------------
 net/unix/diag.c    | 23 +++++++++-----
 2 files changed, 66 insertions(+), 32 deletions(-)

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index cdd12881a39d..79f8fc5cdce8 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -158,7 +158,8 @@ static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
 	return hash & UNIX_HASH_MOD;
 }
 
-static void unix_table_double_lock(unsigned int hash1, unsigned int hash2)
+static void unix_table_double_lock(struct net *net,
+				   unsigned int hash1, unsigned int hash2)
 {
 	/* hash1 and hash2 is never the same because
 	 * one is between 0 and UNIX_HASH_MOD, and
@@ -169,10 +170,17 @@ static void unix_table_double_lock(unsigned int hash1, unsigned int hash2)
 
 	spin_lock(&unix_table_locks[hash1]);
 	spin_lock_nested(&unix_table_locks[hash2], SINGLE_DEPTH_NESTING);
+
+	spin_lock(&net->unx.table.locks[hash1]);
+	spin_lock_nested(&net->unx.table.locks[hash2], SINGLE_DEPTH_NESTING);
 }
 
-static void unix_table_double_unlock(unsigned int hash1, unsigned int hash2)
+static void unix_table_double_unlock(struct net *net,
+				     unsigned int hash1, unsigned int hash2)
 {
+	spin_unlock(&net->unx.table.locks[hash1]);
+	spin_unlock(&net->unx.table.locks[hash2]);
+
 	spin_unlock(&unix_table_locks[hash1]);
 	spin_unlock(&unix_table_locks[hash2]);
 }
@@ -316,17 +324,21 @@ static void __unix_set_addr_hash(struct sock *sk, struct unix_address *addr,
 	__unix_insert_socket(sk);
 }
 
-static void unix_remove_socket(struct sock *sk)
+static void unix_remove_socket(struct net *net, struct sock *sk)
 {
 	spin_lock(&unix_table_locks[sk->sk_hash]);
+	spin_lock(&net->unx.table.locks[sk->sk_hash]);
 	__unix_remove_socket(sk);
+	spin_unlock(&net->unx.table.locks[sk->sk_hash]);
 	spin_unlock(&unix_table_locks[sk->sk_hash]);
 }
 
-static void unix_insert_unbound_socket(struct sock *sk)
+static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
 {
 	spin_lock(&unix_table_locks[sk->sk_hash]);
+	spin_lock(&net->unx.table.locks[sk->sk_hash]);
 	__unix_insert_socket(sk);
+	spin_unlock(&net->unx.table.locks[sk->sk_hash]);
 	spin_unlock(&unix_table_locks[sk->sk_hash]);
 }
 
@@ -356,28 +368,33 @@ static inline struct sock *unix_find_socket_byname(struct net *net,
 	struct sock *s;
 
 	spin_lock(&unix_table_locks[hash]);
+	spin_lock(&net->unx.table.locks[hash]);
 	s = __unix_find_socket_byname(net, sunname, len, hash);
 	if (s)
 		sock_hold(s);
+	spin_unlock(&net->unx.table.locks[hash]);
 	spin_unlock(&unix_table_locks[hash]);
 	return s;
 }
 
-static struct sock *unix_find_socket_byinode(struct inode *i)
+static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 {
 	unsigned int hash = unix_bsd_hash(i);
 	struct sock *s;
 
 	spin_lock(&unix_table_locks[hash]);
+	spin_lock(&net->unx.table.locks[hash]);
 	sk_for_each(s, &unix_socket_table[hash]) {
 		struct dentry *dentry = unix_sk(s)->path.dentry;
 
 		if (dentry && d_backing_inode(dentry) == i) {
 			sock_hold(s);
+			spin_unlock(&net->unx.table.locks[hash]);
 			spin_unlock(&unix_table_locks[hash]);
 			return s;
 		}
 	}
+	spin_unlock(&net->unx.table.locks[hash]);
 	spin_unlock(&unix_table_locks[hash]);
 	return NULL;
 }
@@ -576,12 +593,12 @@ static void unix_sock_destructor(struct sock *sk)
 static void unix_release_sock(struct sock *sk, int embrion)
 {
 	struct unix_sock *u = unix_sk(sk);
-	struct path path;
 	struct sock *skpair;
 	struct sk_buff *skb;
+	struct path path;
 	int state;
 
-	unix_remove_socket(sk);
+	unix_remove_socket(sock_net(sk), sk);
 
 	/* Clear state */
 	unix_state_lock(sk);
@@ -930,7 +947,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock, int kern,
 	init_waitqueue_head(&u->peer_wait);
 	init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
 	memset(&u->scm_stat, 0, sizeof(struct scm_stat));
-	unix_insert_unbound_socket(sk);
+	unix_insert_unbound_socket(net, sk);
 
 	sock_prot_inuse_add(net, sk->sk_prot, 1);
 
@@ -1015,7 +1032,7 @@ static struct sock *unix_find_bsd(struct net *net, struct sockaddr_un *sunaddr,
 	if (!S_ISSOCK(inode->i_mode))
 		goto path_put;
 
-	sk = unix_find_socket_byinode(inode);
+	sk = unix_find_socket_byinode(net, inode);
 	if (!sk)
 		goto path_put;
 
@@ -1074,6 +1091,7 @@ static int unix_autobind(struct sock *sk)
 {
 	unsigned int new_hash, old_hash = sk->sk_hash;
 	struct unix_sock *u = unix_sk(sk);
+	struct net *net = sock_net(sk);
 	struct unix_address *addr;
 	u32 lastnum, ordernum;
 	int err;
@@ -1102,11 +1120,10 @@ static int unix_autobind(struct sock *sk)
 	sprintf(addr->name->sun_path + 1, "%05x", ordernum);
 
 	new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
-	unix_table_double_lock(old_hash, new_hash);
+	unix_table_double_lock(net, old_hash, new_hash);
 
-	if (__unix_find_socket_byname(sock_net(sk), addr->name, addr->len,
-				      new_hash)) {
-		unix_table_double_unlock(old_hash, new_hash);
+	if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
+		unix_table_double_unlock(net, old_hash, new_hash);
 
 		/* __unix_find_socket_byname() may take long time if many names
 		 * are already in use.
@@ -1124,7 +1141,7 @@ static int unix_autobind(struct sock *sk)
 	}
 
 	__unix_set_addr_hash(sk, addr, new_hash);
-	unix_table_double_unlock(old_hash, new_hash);
+	unix_table_double_unlock(net, old_hash, new_hash);
 	err = 0;
 
 out:	mutex_unlock(&u->bindlock);
@@ -1138,6 +1155,7 @@ static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
 	       (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
 	unsigned int new_hash, old_hash = sk->sk_hash;
 	struct unix_sock *u = unix_sk(sk);
+	struct net *net = sock_net(sk);
 	struct user_namespace *ns; // barf...
 	struct unix_address *addr;
 	struct dentry *dentry;
@@ -1178,11 +1196,11 @@ static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
 		goto out_unlock;
 
 	new_hash = unix_bsd_hash(d_backing_inode(dentry));
-	unix_table_double_lock(old_hash, new_hash);
+	unix_table_double_lock(net, old_hash, new_hash);
 	u->path.mnt = mntget(parent.mnt);
 	u->path.dentry = dget(dentry);
 	__unix_set_addr_hash(sk, addr, new_hash);
-	unix_table_double_unlock(old_hash, new_hash);
+	unix_table_double_unlock(net, old_hash, new_hash);
 	mutex_unlock(&u->bindlock);
 	done_path_create(&parent, dentry);
 	return 0;
@@ -1205,6 +1223,7 @@ static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
 {
 	unsigned int new_hash, old_hash = sk->sk_hash;
 	struct unix_sock *u = unix_sk(sk);
+	struct net *net = sock_net(sk);
 	struct unix_address *addr;
 	int err;
 
@@ -1222,19 +1241,18 @@ static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
 	}
 
 	new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
-	unix_table_double_lock(old_hash, new_hash);
+	unix_table_double_lock(net, old_hash, new_hash);
 
-	if (__unix_find_socket_byname(sock_net(sk), addr->name, addr->len,
-				      new_hash))
+	if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
 		goto out_spin;
 
 	__unix_set_addr_hash(sk, addr, new_hash);
-	unix_table_double_unlock(old_hash, new_hash);
+	unix_table_double_unlock(net, old_hash, new_hash);
 	mutex_unlock(&u->bindlock);
 	return 0;
 
 out_spin:
-	unix_table_double_unlock(old_hash, new_hash);
+	unix_table_double_unlock(net, old_hash, new_hash);
 	err = -EADDRINUSE;
 out_mutex:
 	mutex_unlock(&u->bindlock);
@@ -3237,15 +3255,18 @@ static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
 static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
 {
 	unsigned long bucket = get_bucket(*pos);
+	struct net *net = seq_file_net(seq);
 	struct sock *sk;
 
 	while (bucket < UNIX_HASH_SIZE) {
 		spin_lock(&unix_table_locks[bucket]);
+		spin_lock(&net->unx.table.locks[bucket]);
 
 		sk = unix_from_bucket(seq, pos);
 		if (sk)
 			return sk;
 
+		spin_unlock(&net->unx.table.locks[bucket]);
 		spin_unlock(&unix_table_locks[bucket]);
 
 		*pos = set_bucket_offset(++bucket, 1);
@@ -3258,11 +3279,13 @@ static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
 				  loff_t *pos)
 {
 	unsigned long bucket = get_bucket(*pos);
+	struct net *net = seq_file_net(seq);
 
 	for (sk = sk_next(sk); sk; sk = sk_next(sk))
-		if (sock_net(sk) == seq_file_net(seq))
+		if (sock_net(sk) == net)
 			return sk;
 
+	spin_unlock(&net->unx.table.locks[bucket]);
 	spin_unlock(&unix_table_locks[bucket]);
 
 	*pos = set_bucket_offset(++bucket, 1);
@@ -3292,8 +3315,10 @@ static void unix_seq_stop(struct seq_file *seq, void *v)
 {
 	struct sock *sk = v;
 
-	if (sk)
+	if (sk) {
+		spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
 		spin_unlock(&unix_table_locks[sk->sk_hash]);
+	}
 }
 
 static int unix_seq_show(struct seq_file *seq, void *v)
@@ -3381,6 +3406,7 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 
 {
 	struct bpf_unix_iter_state *iter = seq->private;
+	struct net *net = seq_file_net(seq);
 	unsigned int expected = 1;
 	struct sock *sk;
 
@@ -3388,7 +3414,7 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 	iter->batch[iter->end_sk++] = start_sk;
 
 	for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
-		if (sock_net(sk) != seq_file_net(seq))
+		if (sock_net(sk) != net)
 			continue;
 
 		if (iter->end_sk < iter->max_sk) {
@@ -3399,6 +3425,7 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 		expected++;
 	}
 
+	spin_unlock(&net->unx.table.locks[start_sk->sk_hash]);
 	spin_unlock(&unix_table_locks[start_sk->sk_hash]);
 
 	return expected;
diff --git a/net/unix/diag.c b/net/unix/diag.c
index c5d1cca72aa5..7fc377435114 100644
--- a/net/unix/diag.c
+++ b/net/unix/diag.c
@@ -195,9 +195,9 @@ static int sk_diag_dump(struct sock *sk, struct sk_buff *skb, struct unix_diag_r
 
 static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 {
-	struct unix_diag_req *req;
-	int num, s_num, slot, s_slot;
 	struct net *net = sock_net(skb->sk);
+	int num, s_num, slot, s_slot;
+	struct unix_diag_req *req;
 
 	req = nlmsg_data(cb->nlh);
 
@@ -209,6 +209,7 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 
 		num = 0;
 		spin_lock(&unix_table_locks[slot]);
+		spin_lock(&net->unx.table.locks[slot]);
 		sk_for_each(sk, &unix_socket_table[slot]) {
 			if (!net_eq(sock_net(sk), net))
 				continue;
@@ -220,12 +221,14 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 					 NETLINK_CB(cb->skb).portid,
 					 cb->nlh->nlmsg_seq,
 					 NLM_F_MULTI) < 0) {
+				spin_unlock(&net->unx.table.locks[slot]);
 				spin_unlock(&unix_table_locks[slot]);
 				goto done;
 			}
 next:
 			num++;
 		}
+		spin_unlock(&net->unx.table.locks[slot]);
 		spin_unlock(&unix_table_locks[slot]);
 	}
 done:
@@ -235,19 +238,22 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 	return skb->len;
 }
 
-static struct sock *unix_lookup_by_ino(unsigned int ino)
+static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
 {
 	struct sock *sk;
 	int i;
 
 	for (i = 0; i < UNIX_HASH_SIZE; i++) {
 		spin_lock(&unix_table_locks[i]);
+		spin_lock(&net->unx.table.locks[i]);
 		sk_for_each(sk, &unix_socket_table[i])
 			if (ino == sock_i_ino(sk)) {
 				sock_hold(sk);
+				spin_unlock(&net->unx.table.locks[i]);
 				spin_unlock(&unix_table_locks[i]);
 				return sk;
 			}
+		spin_unlock(&net->unx.table.locks[i]);
 		spin_unlock(&unix_table_locks[i]);
 	}
 	return NULL;
@@ -257,16 +263,17 @@ static int unix_diag_get_exact(struct sk_buff *in_skb,
 			       const struct nlmsghdr *nlh,
 			       struct unix_diag_req *req)
 {
-	int err = -EINVAL;
-	struct sock *sk;
-	struct sk_buff *rep;
-	unsigned int extra_len;
 	struct net *net = sock_net(in_skb->sk);
+	unsigned int extra_len;
+	struct sk_buff *rep;
+	struct sock *sk;
+	int err;
 
+	err = -EINVAL;
 	if (req->udiag_ino == 0)
 		goto out_nosk;
 
-	sk = unix_lookup_by_ino(req->udiag_ino);
+	sk = unix_lookup_by_ino(net, req->udiag_ino);
 	err = -ENOENT;
 	if (sk == NULL)
 		goto out_nosk;
-- 
2.30.2

[PATCH v2 net-next 5/6] af_unix: Put a socket into a per-netns hash table.

[Kuniyuki Iwashima]

This commit replaces the global hash table with a per-netns one and removes
the global one.

We now link a socket in each netns's hash table so we can save some netns
comparisons when iterating through a hash bucket.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 include/net/af_unix.h |  1 -
 net/unix/af_unix.c    | 50 +++++++++++++++++--------------------------
 net/unix/diag.c       |  9 +++-----
 3 files changed, 23 insertions(+), 37 deletions(-)

diff --git a/include/net/af_unix.h b/include/net/af_unix.h
index acb56e463db1..b1748c9b6db2 100644
--- a/include/net/af_unix.h
+++ b/include/net/af_unix.h
@@ -22,7 +22,6 @@ struct sock *unix_peer_get(struct sock *sk);
 
 extern unsigned int unix_tot_inflight;
 extern spinlock_t unix_table_locks[UNIX_HASH_SIZE];
-extern struct hlist_head unix_socket_table[UNIX_HASH_SIZE];
 
 struct unix_address {
 	refcount_t	refcnt;
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 79f8fc5cdce8..9d0b07235dbc 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -120,8 +120,6 @@
 
 spinlock_t unix_table_locks[UNIX_HASH_SIZE];
 EXPORT_SYMBOL_GPL(unix_table_locks);
-struct hlist_head unix_socket_table[UNIX_HASH_SIZE];
-EXPORT_SYMBOL_GPL(unix_socket_table);
 static atomic_long_t unix_nr_socks;
 
 /* SMP locking strategy:
@@ -308,20 +306,20 @@ static void __unix_remove_socket(struct sock *sk)
 	sk_del_node_init(sk);
 }
 
-static void __unix_insert_socket(struct sock *sk)
+static void __unix_insert_socket(struct net *net, struct sock *sk)
 {
 	DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
-	sk_add_node(sk, &unix_socket_table[sk->sk_hash]);
+	sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
 }
 
-static void __unix_set_addr_hash(struct sock *sk, struct unix_address *addr,
-				 unsigned int hash)
+static void __unix_set_addr_hash(struct net *net, struct sock *sk,
+				 struct unix_address *addr, unsigned int hash)
 {
 	__unix_remove_socket(sk);
 	smp_store_release(&unix_sk(sk)->addr, addr);
 
 	sk->sk_hash = hash;
-	__unix_insert_socket(sk);
+	__unix_insert_socket(net, sk);
 }
 
 static void unix_remove_socket(struct net *net, struct sock *sk)
@@ -337,7 +335,7 @@ static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
 {
 	spin_lock(&unix_table_locks[sk->sk_hash]);
 	spin_lock(&net->unx.table.locks[sk->sk_hash]);
-	__unix_insert_socket(sk);
+	__unix_insert_socket(net, sk);
 	spin_unlock(&net->unx.table.locks[sk->sk_hash]);
 	spin_unlock(&unix_table_locks[sk->sk_hash]);
 }
@@ -348,12 +346,9 @@ static struct sock *__unix_find_socket_byname(struct net *net,
 {
 	struct sock *s;
 
-	sk_for_each(s, &unix_socket_table[hash]) {
+	sk_for_each(s, &net->unx.table.buckets[hash]) {
 		struct unix_sock *u = unix_sk(s);
 
-		if (!net_eq(sock_net(s), net))
-			continue;
-
 		if (u->addr->len == len &&
 		    !memcmp(u->addr->name, sunname, len))
 			return s;
@@ -384,7 +379,7 @@ static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 
 	spin_lock(&unix_table_locks[hash]);
 	spin_lock(&net->unx.table.locks[hash]);
-	sk_for_each(s, &unix_socket_table[hash]) {
+	sk_for_each(s, &net->unx.table.buckets[hash]) {
 		struct dentry *dentry = unix_sk(s)->path.dentry;
 
 		if (dentry && d_backing_inode(dentry) == i) {
@@ -1140,7 +1135,7 @@ static int unix_autobind(struct sock *sk)
 		goto retry;
 	}
 
-	__unix_set_addr_hash(sk, addr, new_hash);
+	__unix_set_addr_hash(net, sk, addr, new_hash);
 	unix_table_double_unlock(net, old_hash, new_hash);
 	err = 0;
 
@@ -1199,7 +1194,7 @@ static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
 	unix_table_double_lock(net, old_hash, new_hash);
 	u->path.mnt = mntget(parent.mnt);
 	u->path.dentry = dget(dentry);
-	__unix_set_addr_hash(sk, addr, new_hash);
+	__unix_set_addr_hash(net, sk, addr, new_hash);
 	unix_table_double_unlock(net, old_hash, new_hash);
 	mutex_unlock(&u->bindlock);
 	done_path_create(&parent, dentry);
@@ -1246,7 +1241,7 @@ static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
 	if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
 		goto out_spin;
 
-	__unix_set_addr_hash(sk, addr, new_hash);
+	__unix_set_addr_hash(net, sk, addr, new_hash);
 	unix_table_double_unlock(net, old_hash, new_hash);
 	mutex_unlock(&u->bindlock);
 	return 0;
@@ -3239,12 +3234,11 @@ static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
 {
 	unsigned long offset = get_offset(*pos);
 	unsigned long bucket = get_bucket(*pos);
-	struct sock *sk;
 	unsigned long count = 0;
+	struct sock *sk;
 
-	for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
-		if (sock_net(sk) != seq_file_net(seq))
-			continue;
+	for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
+	     sk; sk = sk_next(sk)) {
 		if (++count == offset)
 			break;
 	}
@@ -3279,13 +3273,13 @@ static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
 				  loff_t *pos)
 {
 	unsigned long bucket = get_bucket(*pos);
-	struct net *net = seq_file_net(seq);
 
-	for (sk = sk_next(sk); sk; sk = sk_next(sk))
-		if (sock_net(sk) == net)
-			return sk;
+	sk = sk_next(sk);
+	if (sk)
+		return sk;
+
 
-	spin_unlock(&net->unx.table.locks[bucket]);
+	spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
 	spin_unlock(&unix_table_locks[bucket]);
 
 	*pos = set_bucket_offset(++bucket, 1);
@@ -3406,7 +3400,6 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 
 {
 	struct bpf_unix_iter_state *iter = seq->private;
-	struct net *net = seq_file_net(seq);
 	unsigned int expected = 1;
 	struct sock *sk;
 
@@ -3414,9 +3407,6 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 	iter->batch[iter->end_sk++] = start_sk;
 
 	for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
-		if (sock_net(sk) != net)
-			continue;
-
 		if (iter->end_sk < iter->max_sk) {
 			sock_hold(sk);
 			iter->batch[iter->end_sk++] = sk;
@@ -3425,7 +3415,7 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 		expected++;
 	}
 
-	spin_unlock(&net->unx.table.locks[start_sk->sk_hash]);
+	spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
 	spin_unlock(&unix_table_locks[start_sk->sk_hash]);
 
 	return expected;
diff --git a/net/unix/diag.c b/net/unix/diag.c
index 7fc377435114..4d0f0ca6a1eb 100644
--- a/net/unix/diag.c
+++ b/net/unix/diag.c
@@ -210,9 +210,7 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 		num = 0;
 		spin_lock(&unix_table_locks[slot]);
 		spin_lock(&net->unx.table.locks[slot]);
-		sk_for_each(sk, &unix_socket_table[slot]) {
-			if (!net_eq(sock_net(sk), net))
-				continue;
+		sk_for_each(sk, &net->unx.table.buckets[slot]) {
 			if (num < s_num)
 				goto next;
 			if (!(req->udiag_states & (1 << sk->sk_state)))
@@ -246,13 +244,14 @@ static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
 	for (i = 0; i < UNIX_HASH_SIZE; i++) {
 		spin_lock(&unix_table_locks[i]);
 		spin_lock(&net->unx.table.locks[i]);
-		sk_for_each(sk, &unix_socket_table[i])
+		sk_for_each(sk, &net->unx.table.buckets[i]) {
 			if (ino == sock_i_ino(sk)) {
 				sock_hold(sk);
 				spin_unlock(&net->unx.table.locks[i]);
 				spin_unlock(&unix_table_locks[i]);
 				return sk;
 			}
+		}
 		spin_unlock(&net->unx.table.locks[i]);
 		spin_unlock(&unix_table_locks[i]);
 	}
@@ -277,8 +276,6 @@ static int unix_diag_get_exact(struct sk_buff *in_skb,
 	err = -ENOENT;
 	if (sk == NULL)
 		goto out_nosk;
-	if (!net_eq(sock_net(sk), net))
-		goto out;
 
 	err = sock_diag_check_cookie(sk, req->udiag_cookie);
 	if (err)
-- 
2.30.2

[PATCH v2 net-next 6/6] af_unix: Remove unix_table_locks.

[Kuniyuki Iwashima]

unix_table_locks are to protect the global hash table, unix_socket_table.
The previous commit removed it, so let's clean up the unnecessary locks.

Here is a test result on EC2 c5.9xlarge where 10 processes run concurrently
in different netns and bind 100,000 sockets for each.

  without this series : 1m 38s
  with this series    :    11s

It is ~10x faster because the global hash table is split into 10 netns in
this case.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 include/net/af_unix.h |  1 -
 net/unix/af_unix.c    | 30 ++----------------------------
 net/unix/diag.c       |  5 -----
 3 files changed, 2 insertions(+), 34 deletions(-)

diff --git a/include/net/af_unix.h b/include/net/af_unix.h
index b1748c9b6db2..480fa579787e 100644
--- a/include/net/af_unix.h
+++ b/include/net/af_unix.h
@@ -21,7 +21,6 @@ struct sock *unix_peer_get(struct sock *sk);
 #define UNIX_HASH_BITS	8
 
 extern unsigned int unix_tot_inflight;
-extern spinlock_t unix_table_locks[UNIX_HASH_SIZE];
 
 struct unix_address {
 	refcount_t	refcnt;
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 9d0b07235dbc..eaf5b5b429c1 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -118,8 +118,6 @@
 
 #include "scm.h"
 
-spinlock_t unix_table_locks[UNIX_HASH_SIZE];
-EXPORT_SYMBOL_GPL(unix_table_locks);
 static atomic_long_t unix_nr_socks;
 
 /* SMP locking strategy:
@@ -166,9 +164,6 @@ static void unix_table_double_lock(struct net *net,
 	if (hash1 > hash2)
 		swap(hash1, hash2);
 
-	spin_lock(&unix_table_locks[hash1]);
-	spin_lock_nested(&unix_table_locks[hash2], SINGLE_DEPTH_NESTING);
-
 	spin_lock(&net->unx.table.locks[hash1]);
 	spin_lock_nested(&net->unx.table.locks[hash2], SINGLE_DEPTH_NESTING);
 }
@@ -178,9 +173,6 @@ static void unix_table_double_unlock(struct net *net,
 {
 	spin_unlock(&net->unx.table.locks[hash1]);
 	spin_unlock(&net->unx.table.locks[hash2]);
-
-	spin_unlock(&unix_table_locks[hash1]);
-	spin_unlock(&unix_table_locks[hash2]);
 }
 
 #ifdef CONFIG_SECURITY_NETWORK
@@ -324,20 +316,16 @@ static void __unix_set_addr_hash(struct net *net, struct sock *sk,
 
 static void unix_remove_socket(struct net *net, struct sock *sk)
 {
-	spin_lock(&unix_table_locks[sk->sk_hash]);
 	spin_lock(&net->unx.table.locks[sk->sk_hash]);
 	__unix_remove_socket(sk);
 	spin_unlock(&net->unx.table.locks[sk->sk_hash]);
-	spin_unlock(&unix_table_locks[sk->sk_hash]);
 }
 
 static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
 {
-	spin_lock(&unix_table_locks[sk->sk_hash]);
 	spin_lock(&net->unx.table.locks[sk->sk_hash]);
 	__unix_insert_socket(net, sk);
 	spin_unlock(&net->unx.table.locks[sk->sk_hash]);
-	spin_unlock(&unix_table_locks[sk->sk_hash]);
 }
 
 static struct sock *__unix_find_socket_byname(struct net *net,
@@ -362,13 +350,11 @@ static inline struct sock *unix_find_socket_byname(struct net *net,
 {
 	struct sock *s;
 
-	spin_lock(&unix_table_locks[hash]);
 	spin_lock(&net->unx.table.locks[hash]);
 	s = __unix_find_socket_byname(net, sunname, len, hash);
 	if (s)
 		sock_hold(s);
 	spin_unlock(&net->unx.table.locks[hash]);
-	spin_unlock(&unix_table_locks[hash]);
 	return s;
 }
 
@@ -377,7 +363,6 @@ static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 	unsigned int hash = unix_bsd_hash(i);
 	struct sock *s;
 
-	spin_lock(&unix_table_locks[hash]);
 	spin_lock(&net->unx.table.locks[hash]);
 	sk_for_each(s, &net->unx.table.buckets[hash]) {
 		struct dentry *dentry = unix_sk(s)->path.dentry;
@@ -385,12 +370,10 @@ static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 		if (dentry && d_backing_inode(dentry) == i) {
 			sock_hold(s);
 			spin_unlock(&net->unx.table.locks[hash]);
-			spin_unlock(&unix_table_locks[hash]);
 			return s;
 		}
 	}
 	spin_unlock(&net->unx.table.locks[hash]);
-	spin_unlock(&unix_table_locks[hash]);
 	return NULL;
 }
 
@@ -3253,7 +3236,6 @@ static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
 	struct sock *sk;
 
 	while (bucket < UNIX_HASH_SIZE) {
-		spin_lock(&unix_table_locks[bucket]);
 		spin_lock(&net->unx.table.locks[bucket]);
 
 		sk = unix_from_bucket(seq, pos);
@@ -3261,7 +3243,6 @@ static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
 			return sk;
 
 		spin_unlock(&net->unx.table.locks[bucket]);
-		spin_unlock(&unix_table_locks[bucket]);
 
 		*pos = set_bucket_offset(++bucket, 1);
 	}
@@ -3280,7 +3261,6 @@ static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
 
 
 	spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
-	spin_unlock(&unix_table_locks[bucket]);
 
 	*pos = set_bucket_offset(++bucket, 1);
 
@@ -3309,10 +3289,8 @@ static void unix_seq_stop(struct seq_file *seq, void *v)
 {
 	struct sock *sk = v;
 
-	if (sk) {
+	if (sk)
 		spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
-		spin_unlock(&unix_table_locks[sk->sk_hash]);
-	}
 }
 
 static int unix_seq_show(struct seq_file *seq, void *v)
@@ -3416,7 +3394,6 @@ static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
 	}
 
 	spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
-	spin_unlock(&unix_table_locks[start_sk->sk_hash]);
 
 	return expected;
 }
@@ -3705,13 +3682,10 @@ static void __init bpf_iter_register(void)
 
 static int __init af_unix_init(void)
 {
-	int i, rc = -1;
+	int rc = -1;
 
 	BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
 
-	for (i = 0; i < UNIX_HASH_SIZE; i++)
-		spin_lock_init(&unix_table_locks[i]);
-
 	rc = proto_register(&unix_dgram_proto, 1);
 	if (rc != 0) {
 		pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
diff --git a/net/unix/diag.c b/net/unix/diag.c
index 4d0f0ca6a1eb..e7f5636daa68 100644
--- a/net/unix/diag.c
+++ b/net/unix/diag.c
@@ -208,7 +208,6 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 		struct sock *sk;
 
 		num = 0;
-		spin_lock(&unix_table_locks[slot]);
 		spin_lock(&net->unx.table.locks[slot]);
 		sk_for_each(sk, &net->unx.table.buckets[slot]) {
 			if (num < s_num)
@@ -220,14 +219,12 @@ static int unix_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 					 cb->nlh->nlmsg_seq,
 					 NLM_F_MULTI) < 0) {
 				spin_unlock(&net->unx.table.locks[slot]);
-				spin_unlock(&unix_table_locks[slot]);
 				goto done;
 			}
 next:
 			num++;
 		}
 		spin_unlock(&net->unx.table.locks[slot]);
-		spin_unlock(&unix_table_locks[slot]);
 	}
 done:
 	cb->args[0] = slot;
@@ -242,7 +239,6 @@ static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
 	int i;
 
 	for (i = 0; i < UNIX_HASH_SIZE; i++) {
-		spin_lock(&unix_table_locks[i]);
 		spin_lock(&net->unx.table.locks[i]);
 		sk_for_each(sk, &net->unx.table.buckets[i]) {
 			if (ino == sock_i_ino(sk)) {
@@ -253,7 +249,6 @@ static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
 			}
 		}
 		spin_unlock(&net->unx.table.locks[i]);
-		spin_unlock(&unix_table_locks[i]);
 	}
 	return NULL;
 }
-- 
2.30.2

Re: [PATCH v2 net-next 6/6] af_unix: Remove unix_table_locks.

[kernel test robot]

Hi Kuniyuki,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on net-next/master]

url:    https://github.com/intel-lab-lkp/linux/commits/Kuniyuki-Iwashima/af_unix-Introduce-per-netns-socket-hash-table/20220621-025503
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git dbca1596bbb08318f5e3b3b99f8ca0a0d3830a65
config: arc-defconfig (https://download.01.org/0day-ci/archive/20220621/202206210532.q32y1W52-lkp@intel.com/config)
compiler: arc-elf-gcc (GCC) 11.3.0
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/intel-lab-lkp/linux/commit/f2cd3aee6929543114a0728717969d9bb2f6fa90
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Kuniyuki-Iwashima/af_unix-Introduce-per-netns-socket-hash-table/20220621-025503
        git checkout f2cd3aee6929543114a0728717969d9bb2f6fa90
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.3.0 make.cross W=1 O=build_dir ARCH=arc SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
Reported-by: kernel test robot <lkp@intel.com>

All errors (new ones prefixed by >>):

   net/unix/diag.c: In function 'unix_lookup_by_ino':
>> net/unix/diag.c:247:46: error: 'unix_table_locks' undeclared (first use in this function); did you mean 'unix_state_lock'?
     247 |                                 spin_unlock(&unix_table_locks[i]);
         |                                              ^~~~~~~~~~~~~~~~
         |                                              unix_state_lock
   net/unix/diag.c:247:46: note: each undeclared identifier is reported only once for each function it appears in


vim +247 net/unix/diag.c

22931d3b906cd0 Pavel Emelyanov   2011-12-15  235  
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  236  static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  237  {
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  238  	struct sock *sk;
afd20b9290e184 Kuniyuki Iwashima 2021-11-24  239  	int i;
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  240  
614d83f20d005b Kuniyuki Iwashima 2022-06-20  241  	for (i = 0; i < UNIX_HASH_SIZE; i++) {
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  242  		spin_lock(&net->unx.table.locks[i]);
868a38bbd9ad2a Kuniyuki Iwashima 2022-06-20  243  		sk_for_each(sk, &net->unx.table.buckets[i]) {
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  244  			if (ino == sock_i_ino(sk)) {
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  245  				sock_hold(sk);
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  246  				spin_unlock(&net->unx.table.locks[i]);
afd20b9290e184 Kuniyuki Iwashima 2021-11-24 @247  				spin_unlock(&unix_table_locks[i]);
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  248  				return sk;
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  249  			}
868a38bbd9ad2a Kuniyuki Iwashima 2022-06-20  250  		}
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  251  		spin_unlock(&net->unx.table.locks[i]);
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  252  	}
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  253  	return NULL;
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  254  }
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  255  

-- 
0-DAY CI Kernel Test Service
https://01.org/lkp

Re: [PATCH v2 net-next 6/6] af_unix: Remove unix_table_locks.

[kernel test robot]

Hi Kuniyuki,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on net-next/master]

url:    https://github.com/intel-lab-lkp/linux/commits/Kuniyuki-Iwashima/af_unix-Introduce-per-netns-socket-hash-table/20220621-025503
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git dbca1596bbb08318f5e3b3b99f8ca0a0d3830a65
config: i386-randconfig-a012-20220620 (https://download.01.org/0day-ci/archive/20220621/202206210554.YYgdQz8e-lkp@intel.com/config)
compiler: clang version 15.0.0 (https://github.com/llvm/llvm-project af6d2a0b6825e71965f3e2701a63c239fa0ad70f)
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/intel-lab-lkp/linux/commit/f2cd3aee6929543114a0728717969d9bb2f6fa90
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Kuniyuki-Iwashima/af_unix-Introduce-per-netns-socket-hash-table/20220621-025503
        git checkout f2cd3aee6929543114a0728717969d9bb2f6fa90
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=i386 SHELL=/bin/bash net/

If you fix the issue, kindly add following tag where applicable
Reported-by: kernel test robot <lkp@intel.com>

All errors (new ones prefixed by >>):

>> net/unix/diag.c:247:18: error: use of undeclared identifier 'unix_table_locks'
                                   spin_unlock(&unix_table_locks[i]);
                                                ^
   1 error generated.


vim +/unix_table_locks +247 net/unix/diag.c

22931d3b906cd0 Pavel Emelyanov   2011-12-15  235  
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  236  static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  237  {
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  238  	struct sock *sk;
afd20b9290e184 Kuniyuki Iwashima 2021-11-24  239  	int i;
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  240  
614d83f20d005b Kuniyuki Iwashima 2022-06-20  241  	for (i = 0; i < UNIX_HASH_SIZE; i++) {
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  242  		spin_lock(&net->unx.table.locks[i]);
868a38bbd9ad2a Kuniyuki Iwashima 2022-06-20  243  		sk_for_each(sk, &net->unx.table.buckets[i]) {
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  244  			if (ino == sock_i_ino(sk)) {
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  245  				sock_hold(sk);
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  246  				spin_unlock(&net->unx.table.locks[i]);
afd20b9290e184 Kuniyuki Iwashima 2021-11-24 @247  				spin_unlock(&unix_table_locks[i]);
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  248  				return sk;
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  249  			}
868a38bbd9ad2a Kuniyuki Iwashima 2022-06-20  250  		}
a50feb1de03567 Kuniyuki Iwashima 2022-06-20  251  		spin_unlock(&net->unx.table.locks[i]);
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  252  	}
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  253  	return NULL;
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  254  }
5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  255  

-- 
0-DAY CI Kernel Test Service
https://01.org/lkp

Re: [PATCH v2 net-next 6/6] af_unix: Remove unix_table_locks.

[Kuniyuki Iwashima]

From:   kernel test robot <lkp@intel.com>
Date:   Tue, 21 Jun 2022 05:39:37 +0800
> Hi Kuniyuki,
> 
> Thank you for the patch! Yet something to improve:
> 
> [auto build test ERROR on net-next/master]
> 
> url:    https://github.com/intel-lab-lkp/linux/commits/Kuniyuki-Iwashima/af_unix-Introduce-per-netns-socket-hash-table/20220621-025503
> base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git dbca1596bbb08318f5e3b3b99f8ca0a0d3830a65
> config: arc-defconfig (https://download.01.org/0day-ci/archive/20220621/202206210532.q32y1W52-lkp@intel.com/config)
> compiler: arc-elf-gcc (GCC) 11.3.0
> reproduce (this is a W=1 build):
>         wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
>         chmod +x ~/bin/make.cross
>         # https://github.com/intel-lab-lkp/linux/commit/f2cd3aee6929543114a0728717969d9bb2f6fa90
>         git remote add linux-review https://github.com/intel-lab-lkp/linux
>         git fetch --no-tags linux-review Kuniyuki-Iwashima/af_unix-Introduce-per-netns-socket-hash-table/20220621-025503
>         git checkout f2cd3aee6929543114a0728717969d9bb2f6fa90
>         # save the config file
>         mkdir build_dir && cp config build_dir/.config
>         COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.3.0 make.cross W=1 O=build_dir ARCH=arc SHELL=/bin/bash
> 
> If you fix the issue, kindly add following tag where applicable
> Reported-by: kernel test robot <lkp@intel.com>
> 
> All errors (new ones prefixed by >>):
> 
>    net/unix/diag.c: In function 'unix_lookup_by_ino':
> >> net/unix/diag.c:247:46: error: 'unix_table_locks' undeclared (first use in this function); did you mean 'unix_state_lock'?
>      247 |                                 spin_unlock(&unix_table_locks[i]);
>          |                                              ^~~~~~~~~~~~~~~~
>          |                                              unix_state_lock
>    net/unix/diag.c:247:46: note: each undeclared identifier is reported only once for each function it appears in

I'm sorry, CONFIG_UNIX_DIAG was not set in my .config.
I'll respin v3.

Best regards,
Kuniyuki


> vim +247 net/unix/diag.c
> 
> 22931d3b906cd0 Pavel Emelyanov   2011-12-15  235  
> a50feb1de03567 Kuniyuki Iwashima 2022-06-20  236  static struct sock *unix_lookup_by_ino(struct net *net, unsigned int ino)
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  237  {
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  238  	struct sock *sk;
> afd20b9290e184 Kuniyuki Iwashima 2021-11-24  239  	int i;
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  240  
> 614d83f20d005b Kuniyuki Iwashima 2022-06-20  241  	for (i = 0; i < UNIX_HASH_SIZE; i++) {
> a50feb1de03567 Kuniyuki Iwashima 2022-06-20  242  		spin_lock(&net->unx.table.locks[i]);
> 868a38bbd9ad2a Kuniyuki Iwashima 2022-06-20  243  		sk_for_each(sk, &net->unx.table.buckets[i]) {
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  244  			if (ino == sock_i_ino(sk)) {
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  245  				sock_hold(sk);
> a50feb1de03567 Kuniyuki Iwashima 2022-06-20  246  				spin_unlock(&net->unx.table.locks[i]);
> afd20b9290e184 Kuniyuki Iwashima 2021-11-24 @247  				spin_unlock(&unix_table_locks[i]);
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  248  				return sk;
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  249  			}
> 868a38bbd9ad2a Kuniyuki Iwashima 2022-06-20  250  		}
> a50feb1de03567 Kuniyuki Iwashima 2022-06-20  251  		spin_unlock(&net->unx.table.locks[i]);
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  252  	}
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  253  	return NULL;
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  254  }
> 5d3cae8bc39dd3 Pavel Emelyanov   2011-12-15  255  
> 
> -- 
> 0-DAY CI Kernel Test Service
> https://01.org/lkp