]> git.baikalelectronics.ru Git - kernel.git/commitdiff
netfilter: nf_tables: unlink table before deleting it
authorFlorian Westphal <fw@strlen.de>
Mon, 13 Sep 2021 12:42:33 +0000 (14:42 +0200)
committerPablo Neira Ayuso <pablo@netfilter.org>
Tue, 21 Sep 2021 01:46:55 +0000 (03:46 +0200)
syzbot reports following UAF:
BUG: KASAN: use-after-free in memcmp+0x18f/0x1c0 lib/string.c:955
 nla_strcmp+0xf2/0x130 lib/nlattr.c:836
 nft_table_lookup.part.0+0x1a2/0x460 net/netfilter/nf_tables_api.c:570
 nft_table_lookup net/netfilter/nf_tables_api.c:4064 [inline]
 nf_tables_getset+0x1b3/0x860 net/netfilter/nf_tables_api.c:4064
 nfnetlink_rcv_msg+0x659/0x13f0 net/netfilter/nfnetlink.c:285
 netlink_rcv_skb+0x153/0x420 net/netlink/af_netlink.c:2504

Problem is that all get operations are lockless, so the commit_mutex
held by nft_rcv_nl_event() isn't enough to stop a parallel GET request
from doing read-accesses to the table object even after synchronize_rcu().

To avoid this, unlink the table first and store the table objects in
on-stack scratch space.

Fixes: 6001a930ce03 ("netfilter: nftables: introduce table ownership")
Reported-and-tested-by: syzbot+f31660cf279b0557160c@syzkaller.appspotmail.com
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
net/netfilter/nf_tables_api.c

index 081437dd75b7eba8bb71ac4b9ee46a6ab5c7fd40..33e771cd847c3279d2552634b0f3026522ce36a1 100644 (file)
@@ -9599,7 +9599,6 @@ static void __nft_release_table(struct net *net, struct nft_table *table)
                table->use--;
                nf_tables_chain_destroy(&ctx);
        }
-       list_del(&table->list);
        nf_tables_table_destroy(&ctx);
 }
 
@@ -9612,6 +9611,8 @@ static void __nft_release_tables(struct net *net)
                if (nft_table_has_owner(table))
                        continue;
 
+               list_del(&table->list);
+
                __nft_release_table(net, table);
        }
 }
@@ -9619,31 +9620,38 @@ static void __nft_release_tables(struct net *net)
 static int nft_rcv_nl_event(struct notifier_block *this, unsigned long event,
                            void *ptr)
 {
+       struct nft_table *table, *to_delete[8];
        struct nftables_pernet *nft_net;
        struct netlink_notify *n = ptr;
-       struct nft_table *table, *nt;
        struct net *net = n->net;
-       bool release = false;
+       unsigned int deleted;
+       bool restart = false;
 
        if (event != NETLINK_URELEASE || n->protocol != NETLINK_NETFILTER)
                return NOTIFY_DONE;
 
        nft_net = nft_pernet(net);
+       deleted = 0;
        mutex_lock(&nft_net->commit_mutex);
+again:
        list_for_each_entry(table, &nft_net->tables, list) {
                if (nft_table_has_owner(table) &&
                    n->portid == table->nlpid) {
                        __nft_release_hook(net, table);
-                       release = true;
+                       list_del_rcu(&table->list);
+                       to_delete[deleted++] = table;
+                       if (deleted >= ARRAY_SIZE(to_delete))
+                               break;
                }
        }
-       if (release) {
+       if (deleted) {
+               restart = deleted >= ARRAY_SIZE(to_delete);
                synchronize_rcu();
-               list_for_each_entry_safe(table, nt, &nft_net->tables, list) {
-                       if (nft_table_has_owner(table) &&
-                           n->portid == table->nlpid)
-                               __nft_release_table(net, table);
-               }
+               while (deleted)
+                       __nft_release_table(net, to_delete[--deleted]);
+
+               if (restart)
+                       goto again;
        }
        mutex_unlock(&nft_net->commit_mutex);