sock_diag: define destruction multicast groups
authorCraig Gallek <kraig@google.com>
Mon, 15 Jun 2015 15:26:18 +0000 (11:26 -0400)
committerDavid S. Miller <davem@davemloft.net>
Tue, 16 Jun 2015 02:49:22 +0000 (19:49 -0700)
These groups will contain socket-destruction events for
AF_INET/AF_INET6, IPPROTO_TCP/IPPROTO_UDP.

Near the end of socket destruction, a check for listeners is
performed.  In the presence of a listener, rather than completely
cleanup the socket, a unit of work will be added to a private
work queue which will first broadcast information about the socket
and then finish the cleanup operation.

Signed-off-by: Craig Gallek <kraig@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/sock_diag.h
include/net/sock.h
include/uapi/linux/sock_diag.h
net/core/sock.c
net/core/sock_diag.c

index 083ac388098e43413b4eee3238558c180b747847..fddebc6174697a311be8aa06d1e30ed90135d41b 100644 (file)
@@ -1,7 +1,10 @@
 #ifndef __SOCK_DIAG_H__
 #define __SOCK_DIAG_H__
 
+#include <linux/netlink.h>
 #include <linux/user_namespace.h>
+#include <net/net_namespace.h>
+#include <net/sock.h>
 #include <uapi/linux/sock_diag.h>
 
 struct sk_buff;
@@ -11,6 +14,7 @@ struct sock;
 struct sock_diag_handler {
        __u8 family;
        int (*dump)(struct sk_buff *skb, struct nlmsghdr *nlh);
+       int (*get_info)(struct sk_buff *skb, struct sock *sk);
 };
 
 int sock_diag_register(const struct sock_diag_handler *h);
@@ -26,4 +30,42 @@ int sock_diag_put_meminfo(struct sock *sk, struct sk_buff *skb, int attr);
 int sock_diag_put_filterinfo(bool may_report_filterinfo, struct sock *sk,
                             struct sk_buff *skb, int attrtype);
 
+static inline
+enum sknetlink_groups sock_diag_destroy_group(const struct sock *sk)
+{
+       switch (sk->sk_family) {
+       case AF_INET:
+               switch (sk->sk_protocol) {
+               case IPPROTO_TCP:
+                       return SKNLGRP_INET_TCP_DESTROY;
+               case IPPROTO_UDP:
+                       return SKNLGRP_INET_UDP_DESTROY;
+               default:
+                       return SKNLGRP_NONE;
+               }
+       case AF_INET6:
+               switch (sk->sk_protocol) {
+               case IPPROTO_TCP:
+                       return SKNLGRP_INET6_TCP_DESTROY;
+               case IPPROTO_UDP:
+                       return SKNLGRP_INET6_UDP_DESTROY;
+               default:
+                       return SKNLGRP_NONE;
+               }
+       default:
+               return SKNLGRP_NONE;
+       }
+}
+
+static inline
+bool sock_diag_has_destroy_listeners(const struct sock *sk)
+{
+       const struct net *n = sock_net(sk);
+       const enum sknetlink_groups group = sock_diag_destroy_group(sk);
+
+       return group != SKNLGRP_NONE && n->diag_nlsk &&
+               netlink_has_listeners(n->diag_nlsk, group);
+}
+void sock_diag_broadcast_destroy(struct sock *sk);
+
 #endif
index 26c1c3171e004eb6bf7afa5f8a0b8f6672cb590f..3e8258699270bb4d2e476dad4f34992e9644d34e 100644 (file)
@@ -1518,6 +1518,7 @@ static inline void unlock_sock_fast(struct sock *sk, bool slow)
 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
                      struct proto *prot, int kern);
 void sk_free(struct sock *sk);
+void sk_destruct(struct sock *sk);
 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
 
 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
index b00e29efb1619bb1943e7aa7062bb64c10f8b995..49230d36f9ce783011bd23edb19ba8be0fb751f1 100644 (file)
@@ -23,4 +23,14 @@ enum {
        SK_MEMINFO_VARS,
 };
 
+enum sknetlink_groups {
+       SKNLGRP_NONE,
+       SKNLGRP_INET_TCP_DESTROY,
+       SKNLGRP_INET_UDP_DESTROY,
+       SKNLGRP_INET6_TCP_DESTROY,
+       SKNLGRP_INET6_UDP_DESTROY,
+       __SKNLGRP_MAX,
+};
+#define SKNLGRP_MAX    (__SKNLGRP_MAX - 1)
+
 #endif /* _UAPI__SOCK_DIAG_H__ */
index 7063c329c1b60f4b881e38b553ff7da0539a93fd..1e1fe9a68d835983d760d50f9ef6a11309ffcfc1 100644 (file)
 #include <linux/ipsec.h>
 #include <net/cls_cgroup.h>
 #include <net/netprio_cgroup.h>
+#include <linux/sock_diag.h>
 
 #include <linux/filter.h>
 
@@ -1423,7 +1424,7 @@ struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 }
 EXPORT_SYMBOL(sk_alloc);
 
-static void __sk_free(struct sock *sk)
+void sk_destruct(struct sock *sk)
 {
        struct sk_filter *filter;
 
@@ -1451,6 +1452,14 @@ static void __sk_free(struct sock *sk)
        sk_prot_free(sk->sk_prot_creator, sk);
 }
 
+static void __sk_free(struct sock *sk)
+{
+       if (unlikely(sock_diag_has_destroy_listeners(sk)))
+               sock_diag_broadcast_destroy(sk);
+       else
+               sk_destruct(sk);
+}
+
 void sk_free(struct sock *sk)
 {
        /*
index 74dddf84adcdd7fea05ca93d94f97c6558917ae1..d79866c5f8bc51dac4cc4f3445a88f8ac1e6babf 100644 (file)
@@ -5,6 +5,9 @@
 #include <net/net_namespace.h>
 #include <linux/module.h>
 #include <net/sock.h>
+#include <linux/kernel.h>
+#include <linux/tcp.h>
+#include <linux/workqueue.h>
 
 #include <linux/inet_diag.h>
 #include <linux/sock_diag.h>
@@ -12,6 +15,7 @@
 static const struct sock_diag_handler *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;
 
 static u64 sock_gen_cookie(struct sock *sk)
 {
@@ -101,6 +105,62 @@ out:
 }
 EXPORT_SYMBOL(sock_diag_put_filterinfo);
 
+struct broadcast_sk {
+       struct sock *sk;
+       struct work_struct work;
+};
+
+static size_t sock_diag_nlmsg_size(void)
+{
+       return NLMSG_ALIGN(sizeof(struct inet_diag_msg)
+              + nla_total_size(sizeof(u8)) /* INET_DIAG_PROTOCOL */
+              + nla_total_size(sizeof(struct tcp_info))); /* INET_DIAG_INFO */
+}
+
+static void sock_diag_broadcast_destroy_work(struct work_struct *work)
+{
+       struct broadcast_sk *bsk =
+               container_of(work, struct broadcast_sk, work);
+       struct sock *sk = bsk->sk;
+       const struct sock_diag_handler *hndl;
+       struct sk_buff *skb;
+       const enum sknetlink_groups group = sock_diag_destroy_group(sk);
+       int err = -1;
+
+       WARN_ON(group == SKNLGRP_NONE);
+
+       skb = nlmsg_new(sock_diag_nlmsg_size(), GFP_KERNEL);
+       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);
+
+       if (!err)
+               nlmsg_multicast(sock_net(sk)->diag_nlsk, skb, 0, group,
+                               GFP_KERNEL);
+       else
+               kfree_skb(skb);
+out:
+       sk_destruct(sk);
+       kfree(bsk);
+}
+
+void sock_diag_broadcast_destroy(struct sock *sk)
+{
+       /* Note, this function is often called from an interrupt context. */
+       struct broadcast_sk *bsk =
+               kmalloc(sizeof(struct broadcast_sk), GFP_ATOMIC);
+       if (!bsk)
+               return sk_destruct(sk);
+       bsk->sk = sk;
+       INIT_WORK(&bsk->work, sock_diag_broadcast_destroy_work);
+       queue_work(broadcast_wq, &bsk->work);
+}
+
 void sock_diag_register_inet_compat(int (*fn)(struct sk_buff *skb, struct nlmsghdr *nlh))
 {
        mutex_lock(&sock_diag_table_mutex);
@@ -211,10 +271,32 @@ static void sock_diag_rcv(struct sk_buff *skb)
        mutex_unlock(&sock_diag_mutex);
 }
 
+static int sock_diag_bind(struct net *net, int group)
+{
+       switch (group) {
+       case SKNLGRP_INET_TCP_DESTROY:
+       case SKNLGRP_INET_UDP_DESTROY:
+               if (!sock_diag_handlers[AF_INET])
+                       request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
+                                      NETLINK_SOCK_DIAG, AF_INET);
+               break;
+       case SKNLGRP_INET6_TCP_DESTROY:
+       case SKNLGRP_INET6_UDP_DESTROY:
+               if (!sock_diag_handlers[AF_INET6])
+                       request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
+                                      NETLINK_SOCK_DIAG, AF_INET);
+               break;
+       }
+       return 0;
+}
+
 static int __net_init diag_net_init(struct net *net)
 {
        struct netlink_kernel_cfg cfg = {
+               .groups = SKNLGRP_MAX,
                .input  = sock_diag_rcv,
+               .bind   = sock_diag_bind,
+               .flags  = NL_CFG_F_NONROOT_RECV,
        };
 
        net->diag_nlsk = netlink_kernel_create(net, NETLINK_SOCK_DIAG, &cfg);
@@ -234,12 +316,15 @@ static struct pernet_operations diag_net_ops = {
 
 static int __init sock_diag_init(void)
 {
+       broadcast_wq = alloc_workqueue("sock_diag_events", 0, 0);
+       BUG_ON(!broadcast_wq);
        return register_pernet_subsys(&diag_net_ops);
 }
 
 static void __exit sock_diag_exit(void)
 {
        unregister_pernet_subsys(&diag_net_ops);
+       destroy_workqueue(broadcast_wq);
 }
 
 module_init(sock_diag_init);