sock_diag: allow concurrent operations
authorEric Dumazet <edumazet@google.com>
Mon, 22 Jan 2024 11:26:00 +0000 (11:26 +0000)
committerPaolo Abeni <pabeni@redhat.com>
Tue, 23 Jan 2024 14:13:55 +0000 (15:13 +0100)
sock_diag_broadcast_destroy_work() and __sock_diag_cmd()
are currently using sock_diag_table_mutex to protect
against concurrent sock_diag_handlers[] changes.

This makes inet_diag dump serialized, thus less scalable
than legacy /proc files.

It is time to switch to full RCU protection.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net/core/sock_diag.c

index c53b731f2d6728d113b90732f4df5b011a438038..72009e1f4380dfdcbf43ed08791e5039e74f5c54 100644 (file)
@@ -16,7 +16,7 @@
 #include <linux/inet_diag.h>
 #include <linux/sock_diag.h>
 
-static const struct sock_diag_handler *sock_diag_handlers[AF_MAX];
+static const struct sock_diag_handler __rcu *sock_diag_handlers[AF_MAX];
 static int (*inet_rcv_compat)(struct sk_buff *skb, struct nlmsghdr *nlh);
 static DEFINE_MUTEX(sock_diag_table_mutex);
 static struct workqueue_struct *broadcast_wq;
@@ -122,6 +122,24 @@ static size_t sock_diag_nlmsg_size(void)
               + nla_total_size_64bit(sizeof(struct tcp_info))); /* INET_DIAG_INFO */
 }
 
+static const struct sock_diag_handler *sock_diag_lock_handler(int family)
+{
+       const struct sock_diag_handler *handler;
+
+       rcu_read_lock();
+       handler = rcu_dereference(sock_diag_handlers[family]);
+       if (handler && !try_module_get(handler->owner))
+               handler = NULL;
+       rcu_read_unlock();
+
+       return handler;
+}
+
+static void sock_diag_unlock_handler(const struct sock_diag_handler *handler)
+{
+       module_put(handler->owner);
+}
+
 static void sock_diag_broadcast_destroy_work(struct work_struct *work)
 {
        struct broadcast_sk *bsk =
@@ -138,12 +156,12 @@ static void sock_diag_broadcast_destroy_work(struct work_struct *work)
        if (!skb)
                goto out;
 
-       mutex_lock(&sock_diag_table_mutex);
-       hndl = sock_diag_handlers[sk->sk_family];
-       if (hndl && hndl->get_info)
-               err = hndl->get_info(skb, sk);
-       mutex_unlock(&sock_diag_table_mutex);
-
+       hndl = sock_diag_lock_handler(sk->sk_family);
+       if (hndl) {
+               if (hndl->get_info)
+                       err = hndl->get_info(skb, sk);
+               sock_diag_unlock_handler(hndl);
+       }
        if (!err)
                nlmsg_multicast(sock_net(sk)->diag_nlsk, skb, 0, group,
                                GFP_KERNEL);
@@ -184,33 +202,26 @@ EXPORT_SYMBOL_GPL(sock_diag_unregister_inet_compat);
 
 int sock_diag_register(const struct sock_diag_handler *hndl)
 {
-       int err = 0;
+       int family = hndl->family;
 
-       if (hndl->family >= AF_MAX)
+       if (family >= AF_MAX)
                return -EINVAL;
 
-       mutex_lock(&sock_diag_table_mutex);
-       if (sock_diag_handlers[hndl->family])
-               err = -EBUSY;
-       else
-               WRITE_ONCE(sock_diag_handlers[hndl->family], hndl);
-       mutex_unlock(&sock_diag_table_mutex);
-
-       return err;
+       return !cmpxchg((const struct sock_diag_handler **)
+                               &sock_diag_handlers[family],
+                       NULL, hndl) ? 0 : -EBUSY;
 }
 EXPORT_SYMBOL_GPL(sock_diag_register);
 
-void sock_diag_unregister(const struct sock_diag_handler *hnld)
+void sock_diag_unregister(const struct sock_diag_handler *hndl)
 {
-       int family = hnld->family;
+       int family = hndl->family;
 
        if (family >= AF_MAX)
                return;
 
-       mutex_lock(&sock_diag_table_mutex);
-       BUG_ON(sock_diag_handlers[family] != hnld);
-       WRITE_ONCE(sock_diag_handlers[family], NULL);
-       mutex_unlock(&sock_diag_table_mutex);
+       xchg((const struct sock_diag_handler **)&sock_diag_handlers[family],
+            NULL);
 }
 EXPORT_SYMBOL_GPL(sock_diag_unregister);
 
@@ -227,20 +238,20 @@ static int __sock_diag_cmd(struct sk_buff *skb, struct nlmsghdr *nlh)
                return -EINVAL;
        req->sdiag_family = array_index_nospec(req->sdiag_family, AF_MAX);
 
-       if (READ_ONCE(sock_diag_handlers[req->sdiag_family]) == NULL)
+       if (!rcu_access_pointer(sock_diag_handlers[req->sdiag_family]))
                sock_load_diag_module(req->sdiag_family, 0);
 
-       mutex_lock(&sock_diag_table_mutex);
-       hndl = sock_diag_handlers[req->sdiag_family];
+       hndl = sock_diag_lock_handler(req->sdiag_family);
        if (hndl == NULL)
-               err = -ENOENT;
-       else if (nlh->nlmsg_type == SOCK_DIAG_BY_FAMILY)
+               return -ENOENT;
+
+       if (nlh->nlmsg_type == SOCK_DIAG_BY_FAMILY)
                err = hndl->dump(skb, nlh);
        else if (nlh->nlmsg_type == SOCK_DESTROY && hndl->destroy)
                err = hndl->destroy(skb, nlh);
        else
                err = -EOPNOTSUPP;
-       mutex_unlock(&sock_diag_table_mutex);
+       sock_diag_unlock_handler(hndl);
 
        return err;
 }
@@ -286,12 +297,12 @@ static int sock_diag_bind(struct net *net, int group)
        switch (group) {
        case SKNLGRP_INET_TCP_DESTROY:
        case SKNLGRP_INET_UDP_DESTROY:
-               if (!READ_ONCE(sock_diag_handlers[AF_INET]))
+               if (!rcu_access_pointer(sock_diag_handlers[AF_INET]))
                        sock_load_diag_module(AF_INET, 0);
                break;
        case SKNLGRP_INET6_TCP_DESTROY:
        case SKNLGRP_INET6_UDP_DESTROY:
-               if (!READ_ONCE(sock_diag_handlers[AF_INET6]))
+               if (!rcu_access_pointer(sock_diag_handlers[AF_INET6]))
                        sock_load_diag_module(AF_INET6, 0);
                break;
        }