@@ -269,6 +269,12 @@ static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
}
}
+enum {
+ NFNL_BATCH_FAILURE = (1 << 0),
+ NFNL_BATCH_DONE = (1 << 1),
+ NFNL_BATCH_REPLAY = (1 << 2),
+};
+
static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
u_int16_t subsys_id)
{
@@ -276,13 +282,15 @@ static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
struct net *net = sock_net(skb->sk);
const struct nfnetlink_subsystem *ss;
const struct nfnl_callback *nc;
- bool success = true, done = false;
static LIST_HEAD(err_list);
+ u32 status;
int err;
if (subsys_id >= NFNL_SUBSYS_COUNT)
return netlink_ack(skb, nlh, -EINVAL);
replay:
+ status = 0;
+
skb = netlink_skb_clone(oskb, GFP_KERNEL);
if (!skb)
return netlink_ack(oskb, nlh, -ENOMEM);
@@ -336,10 +344,10 @@ replay:
if (type == NFNL_MSG_BATCH_BEGIN) {
/* Malformed: Batch begin twice */
nfnl_err_reset(&err_list);
- success = false;
+ status |= NFNL_BATCH_FAILURE;
goto done;
} else if (type == NFNL_MSG_BATCH_END) {
- done = true;
+ status |= NFNL_BATCH_DONE;
goto done;
} else if (type < NLMSG_MIN_TYPE) {
err = -EINVAL;
@@ -382,11 +390,8 @@ replay:
* original skb.
*/
if (err == -EAGAIN) {
- nfnl_err_reset(&err_list);
- ss->abort(oskb);
- nfnl_unlock(subsys_id);
- kfree_skb(skb);
- goto replay;
+ status |= NFNL_BATCH_REPLAY;
+ goto next;
}
}
ack:
@@ -402,7 +407,7 @@ ack:
*/
nfnl_err_reset(&err_list);
netlink_ack(skb, nlmsg_hdr(oskb), -ENOMEM);
- success = false;
+ status |= NFNL_BATCH_FAILURE;
goto done;
}
/* We don't stop processing the batch on errors, thus,
@@ -410,19 +415,26 @@ ack:
* triggers.
*/
if (err)
- success = false;
+ status |= NFNL_BATCH_FAILURE;
}
-
+next:
msglen = NLMSG_ALIGN(nlh->nlmsg_len);
if (msglen > skb->len)
msglen = skb->len;
skb_pull(skb, msglen);
}
done:
- if (success && done)
+ if (status & NFNL_BATCH_REPLAY) {
+ ss->abort(oskb);
+ nfnl_err_reset(&err_list);
+ nfnl_unlock(subsys_id);
+ kfree_skb(skb);
+ goto replay;
+ } else if (status == NFNL_BATCH_DONE) {
ss->commit(oskb);
- else
+ } else {
ss->abort(oskb);
+ }
nfnl_err_deliver(&err_list, oskb);
nfnl_unlock(subsys_id);
After a fresh boot with no modules in place at all and a large rulesets, the existing nfnetlink_rcv_batch() funcion can take long time to commit the ruleset due to the many abort path. This is specifically a problem for the existing client of this code, ie. nf_tables, since it results in several synchronize_rcu() call in a row. This patch changes the policy to keep full batch processing on missing modules errors so we abort only once. Reported-by: Eric Leblond <eric@regit.org> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> --- net/netfilter/nfnetlink.c | 38 +++++++++++++++++++++++++------------- 1 file changed, 25 insertions(+), 13 deletions(-)