netlink: Use rhashtable walk interface in diag dump
authorHerbert Xu <herbert@gondor.apana.org.au>
Fri, 19 Aug 2016 08:21:37 +0000 (16:21 +0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 19 Aug 2016 21:40:25 +0000 (14:40 -0700)
This patch converts the diag dumping code to use the rhashtable
walk code instead of going through rhashtable by hand.  The lock
nl_table_lock is now only taken while we process the multicast
list as it's not needed for the rhashtable walk.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/netlink/diag.c

index 8dd836a8dd60ce67c3b2b910f18f80f28cc05b7c..3e3e2534478ace7ff041a6f8787fd7fe9c7fddda 100644 (file)
@@ -63,43 +63,75 @@ out_nlmsg_trim:
 static int __netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
                                int protocol, int s_num)
 {
+       struct rhashtable_iter *hti = (void *)cb->args[2];
        struct netlink_table *tbl = &nl_table[protocol];
-       struct rhashtable *ht = &tbl->hash;
-       const struct bucket_table *htbl = rht_dereference_rcu(ht->tbl, ht);
        struct net *net = sock_net(skb->sk);
        struct netlink_diag_req *req;
        struct netlink_sock *nlsk;
        struct sock *sk;
-       int ret = 0, num = 0, i;
+       int num = 2;
+       int ret = 0;
 
        req = nlmsg_data(cb->nlh);
 
-       for (i = 0; i < htbl->size; i++) {
-               struct rhash_head *pos;
+       if (s_num > 1)
+               goto mc_list;
 
-               rht_for_each_entry_rcu(nlsk, pos, htbl, i, node) {
-                       sk = (struct sock *)nlsk;
+       num--;
 
-                       if (!net_eq(sock_net(sk), net))
-                               continue;
-                       if (num < s_num) {
-                               num++;
+       if (!hti) {
+               hti = kmalloc(sizeof(*hti), GFP_KERNEL);
+               if (!hti)
+                       return -ENOMEM;
+
+               cb->args[2] = (long)hti;
+       }
+
+       if (!s_num)
+               rhashtable_walk_enter(&tbl->hash, hti);
+
+       ret = rhashtable_walk_start(hti);
+       if (ret == -EAGAIN)
+               ret = 0;
+       if (ret)
+               goto stop;
+
+       while ((nlsk = rhashtable_walk_next(hti))) {
+               if (IS_ERR(nlsk)) {
+                       ret = PTR_ERR(nlsk);
+                       if (ret == -EAGAIN) {
+                               ret = 0;
                                continue;
                        }
+                       break;
+               }
 
-                       if (sk_diag_fill(sk, skb, req,
-                                        NETLINK_CB(cb->skb).portid,
-                                        cb->nlh->nlmsg_seq,
-                                        NLM_F_MULTI,
-                                        sock_i_ino(sk)) < 0) {
-                               ret = 1;
-                               goto done;
-                       }
+               sk = (struct sock *)nlsk;
 
-                       num++;
+               if (!net_eq(sock_net(sk), net))
+                       continue;
+
+               if (sk_diag_fill(sk, skb, req,
+                                NETLINK_CB(cb->skb).portid,
+                                cb->nlh->nlmsg_seq,
+                                NLM_F_MULTI,
+                                sock_i_ino(sk)) < 0) {
+                       ret = 1;
+                       break;
                }
        }
 
+stop:
+       rhashtable_walk_stop(hti);
+       if (ret)
+               goto done;
+
+       rhashtable_walk_exit(hti);
+       cb->args[2] = 0;
+       num++;
+
+mc_list:
+       read_lock(&nl_table_lock);
        sk_for_each_bound(sk, &tbl->mc_list) {
                if (sk_hashed(sk))
                        continue;
@@ -116,13 +148,14 @@ static int __netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
                                 NLM_F_MULTI,
                                 sock_i_ino(sk)) < 0) {
                        ret = 1;
-                       goto done;
+                       break;
                }
                num++;
        }
+       read_unlock(&nl_table_lock);
+
 done:
        cb->args[0] = num;
-       cb->args[1] = protocol;
 
        return ret;
 }
@@ -131,20 +164,20 @@ static int netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
 {
        struct netlink_diag_req *req;
        int s_num = cb->args[0];
+       int err = 0;
 
        req = nlmsg_data(cb->nlh);
 
-       rcu_read_lock();
-       read_lock(&nl_table_lock);
-
        if (req->sdiag_protocol == NDIAG_PROTO_ALL) {
                int i;
 
                for (i = cb->args[1]; i < MAX_LINKS; i++) {
-                       if (__netlink_diag_dump(skb, cb, i, s_num))
+                       err = __netlink_diag_dump(skb, cb, i, s_num);
+                       if (err)
                                break;
                        s_num = 0;
                }
+               cb->args[1] = i;
        } else {
                if (req->sdiag_protocol >= MAX_LINKS) {
                        read_unlock(&nl_table_lock);
@@ -152,13 +185,22 @@ static int netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
                        return -ENOENT;
                }
 
-               __netlink_diag_dump(skb, cb, req->sdiag_protocol, s_num);
+               err = __netlink_diag_dump(skb, cb, req->sdiag_protocol, s_num);
        }
 
-       read_unlock(&nl_table_lock);
-       rcu_read_unlock();
+       return err < 0 ? err : skb->len;
+}
+
+static int netlink_diag_dump_done(struct netlink_callback *cb)
+{
+       struct rhashtable_iter *hti = (void *)cb->args[2];
+
+       if (cb->args[0] == 1)
+               rhashtable_walk_exit(hti);
 
-       return skb->len;
+       kfree(hti);
+
+       return 0;
 }
 
 static int netlink_diag_handler_dump(struct sk_buff *skb, struct nlmsghdr *h)
@@ -172,6 +214,7 @@ static int netlink_diag_handler_dump(struct sk_buff *skb, struct nlmsghdr *h)
        if (h->nlmsg_flags & NLM_F_DUMP) {
                struct netlink_dump_control c = {
                        .dump = netlink_diag_dump,
+                       .done = netlink_diag_dump_done,
                };
                return netlink_dump_start(net->diag_nlsk, skb, h, &c);
        } else