[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200121113235.GA1847@kadam>
Date: Tue, 21 Jan 2020 14:32:36 +0300
From: Dan Carpenter <dan.carpenter@...cle.com>
To: syzbot <syzbot+29125d208b3dae9a7019@...kaller.appspotmail.com>,
pablo@...filter.org
Cc: coreteam@...filter.org, davem@...emloft.net, fw@...len.de,
kadlec@...filter.org, linux-kernel@...r.kernel.org,
netdev@...r.kernel.org, netfilter-devel@...r.kernel.org,
syzkaller-bugs@...glegroups.com
Subject: Re: KASAN: use-after-free Read in __nf_tables_abort
I think I see the problem, but I'm not sure how you want to fix it...
net/netfilter/nf_tables_api.c
942 static int nf_tables_newtable(struct net *net, struct sock *nlsk,
943 struct sk_buff *skb, const struct nlmsghdr *nlh,
944 const struct nlattr * const nla[],
945 struct netlink_ext_ack *extack)
946 {
947 const struct nfgenmsg *nfmsg = nlmsg_data(nlh);
948 u8 genmask = nft_genmask_next(net);
949 int family = nfmsg->nfgen_family;
950 const struct nlattr *attr;
951 struct nft_table *table;
952 u32 flags = 0;
953 struct nft_ctx ctx;
954 int err;
955
956 lockdep_assert_held(&net->nft.commit_mutex);
957 attr = nla[NFTA_TABLE_NAME];
958 table = nft_table_lookup(net, attr, family, genmask);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This is looking up table in net->nft.tables
959 if (IS_ERR(table)) {
960 if (PTR_ERR(table) != -ENOENT)
961 return PTR_ERR(table);
962 } else {
963 if (nlh->nlmsg_flags & NLM_F_EXCL) {
964 NL_SET_BAD_ATTR(extack, attr);
965 return -EEXIST;
966 }
967 if (nlh->nlmsg_flags & NLM_F_REPLACE)
968 return -EOPNOTSUPP;
969
970 nft_ctx_init(&ctx, net, skb, nlh, family, table, NULL, nla);
971 return nf_tables_updtable(&ctx);
^^^^^^^^^^^^^^^^^^^^^^^
Then it adds it to &ctx->net->nft.commit_list
972 }
973
974 if (nla[NFTA_TABLE_FLAGS]) {
975 flags = ntohl(nla_get_be32(nla[NFTA_TABLE_FLAGS]));
976 if (flags & ~NFT_TABLE_F_DORMANT)
977 return -EINVAL;
978 }
979
980 err = -ENOMEM;
981 table = kzalloc(sizeof(*table), GFP_KERNEL);
982 if (table == NULL)
983 goto err_kzalloc;
984
985 table->name = nla_strdup(attr, GFP_KERNEL);
986 if (table->name == NULL)
987 goto err_strdup;
988
989 err = rhltable_init(&table->chains_ht, &nft_chain_ht_params);
990 if (err)
991 goto err_chain_ht;
992
993 INIT_LIST_HEAD(&table->chains);
994 INIT_LIST_HEAD(&table->sets);
995 INIT_LIST_HEAD(&table->objects);
996 INIT_LIST_HEAD(&table->flowtables);
997 table->family = family;
998 table->flags = flags;
999 table->handle = ++table_handle;
1000
1001 nft_ctx_init(&ctx, net, skb, nlh, family, table, NULL, nla);
1002 err = nft_trans_table_add(&ctx, NFT_MSG_NEWTABLE);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Added to ctx->net->nft.commit_list
1003 if (err < 0)
1004 goto err_trans;
1005
1006 list_add_tail_rcu(&table->list, &net->nft.tables);
^^^^^^^^^^^^^^^^
Added to net->nft.tables
1007 return 0;
1008 err_trans:
1009 rhltable_destroy(&table->chains_ht);
1010 err_chain_ht:
1011 kfree(table->name);
1012 err_strdup:
1013 kfree(table);
net/netfilter/nf_tables_api.c
6995 static void nf_tables_commit_release(struct net *net)
6996 {
6997 struct nft_trans *trans;
6998
6999 /* all side effects have to be made visible.
7000 * For example, if a chain named 'foo' has been deleted, a
7001 * new transaction must not find it anymore.
7002 *
7003 * Memory reclaim happens asynchronously from work queue
7004 * to prevent expensive synchronize_rcu() in commit phase.
7005 */
7006 if (list_empty(&net->nft.commit_list)) {
7007 mutex_unlock(&net->nft.commit_mutex);
7008 return;
7009 }
7010
7011 trans = list_last_entry(&net->nft.commit_list,
7012 struct nft_trans, list);
7013 get_net(trans->ctx.net);
7014 WARN_ON_ONCE(trans->put_net);
7015
7016 trans->put_net = true;
7017 spin_lock(&nf_tables_destroy_list_lock);
7018 list_splice_tail_init(&net->nft.commit_list, &nf_tables_destroy_list);
^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^
This starts the process of freeing everything from net->nft.commit_list,
but we need to delete it from the net->nft.tables list as well.
7019 spin_unlock(&nf_tables_destroy_list_lock);
7020
7021 mutex_unlock(&net->nft.commit_mutex);
7022
7023 schedule_work(&trans_destroy_work);
7024 }
regards,
dan carpenter
Powered by blists - more mailing lists