xfrm: Add a state resolution packet queue
authorSteffen Klassert <steffen.klassert@secunet.com>
Tue, 5 Feb 2013 11:52:55 +0000 (12:52 +0100)
committerSteffen Klassert <steffen.klassert@secunet.com>
Wed, 6 Feb 2013 07:31:10 +0000 (08:31 +0100)
As the default, we blackhole packets until the key manager resolves
the states. This patch implements a packet queue where IPsec packets
are queued until the states are resolved. We generate a dummy xfrm
bundle, the output routine of the returned route enqueues the packet
to a per policy queue and arms a timer that checks for state resolution
when dst_output() is called. Once the states are resolved, the packets
are sent out of the queue. If the states are not resolved after some
time, the queue is flushed.

This patch keeps the defaut behaviour to blackhole packets as long
as we have no states. To enable the packet queue the sysctl
xfrm_larval_drop must be switched off.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
include/net/dst.h
include/net/xfrm.h
net/xfrm/xfrm_policy.c

index 9a7881066fb316b02fdd7ed52aaf84d038ded1f6..3da47e0a4a1ffdfce7369980dca71c58cf1606a5 100644 (file)
@@ -61,6 +61,7 @@ struct dst_entry {
 #define DST_NOPEER             0x0040
 #define DST_FAKE_RTABLE                0x0080
 #define DST_XFRM_TUNNEL                0x0100
+#define DST_XFRM_QUEUE         0x0200
 
        unsigned short          pending_confirm;
 
index 814a1baa175c06f134e3b9ee1aefb98c9c1cc539..30f3e5b362ee55a527775156dc6ea3dfedfaea20 100644 (file)
@@ -501,6 +501,12 @@ struct xfrm_policy_walk {
        u32 seq;
 };
 
+struct xfrm_policy_queue {
+       struct sk_buff_head     hold_queue;
+       struct timer_list       hold_timer;
+       unsigned long           timeout;
+};
+
 struct xfrm_policy {
 #ifdef CONFIG_NET_NS
        struct net              *xp_net;
@@ -522,6 +528,7 @@ struct xfrm_policy {
        struct xfrm_lifetime_cfg lft;
        struct xfrm_lifetime_cur curlft;
        struct xfrm_policy_walk_entry walk;
+       struct xfrm_policy_queue polq;
        u8                      type;
        u8                      action;
        u8                      flags;
index 41eabc46f110d9cb607cb24f8af68054b91bcccd..456b11b0f049341d7ba731b66c3ab0f26b2c468a 100644 (file)
 
 #include "xfrm_hash.h"
 
+#define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
+#define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
+#define XFRM_MAX_QUEUE_LEN     100
+
 DEFINE_MUTEX(xfrm_cfg_mutex);
 EXPORT_SYMBOL(xfrm_cfg_mutex);
 
@@ -51,7 +55,7 @@ static struct kmem_cache *xfrm_dst_cache __read_mostly;
 static void xfrm_init_pmtu(struct dst_entry *dst);
 static int stale_bundle(struct dst_entry *dst);
 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
-
+static void xfrm_policy_queue_process(unsigned long arg);
 
 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
                                                int dir);
@@ -287,8 +291,11 @@ struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
                INIT_HLIST_NODE(&policy->byidx);
                rwlock_init(&policy->lock);
                atomic_set(&policy->refcnt, 1);
+               skb_queue_head_init(&policy->polq.hold_queue);
                setup_timer(&policy->timer, xfrm_policy_timer,
                                (unsigned long)policy);
+               setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
+                           (unsigned long)policy);
                policy->flo.ops = &xfrm_policy_fc_ops;
        }
        return policy;
@@ -309,6 +316,16 @@ void xfrm_policy_destroy(struct xfrm_policy *policy)
 }
 EXPORT_SYMBOL(xfrm_policy_destroy);
 
+static void xfrm_queue_purge(struct sk_buff_head *list)
+{
+       struct sk_buff *skb;
+
+       while ((skb = skb_dequeue(list)) != NULL) {
+               dev_put(skb->dev);
+               kfree_skb(skb);
+       }
+}
+
 /* Rule must be locked. Release descentant resources, announce
  * entry dead. The rule must be unlinked from lists to the moment.
  */
@@ -319,6 +336,9 @@ static void xfrm_policy_kill(struct xfrm_policy *policy)
 
        atomic_inc(&policy->genid);
 
+       del_timer(&policy->polq.hold_timer);
+       xfrm_queue_purge(&policy->polq.hold_queue);
+
        if (del_timer(&policy->timer))
                xfrm_pol_put(policy);
 
@@ -562,6 +582,31 @@ static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s
        return 0;
 }
 
+static void xfrm_policy_requeue(struct xfrm_policy *old,
+                               struct xfrm_policy *new)
+{
+       struct xfrm_policy_queue *pq = &old->polq;
+       struct sk_buff_head list;
+
+       __skb_queue_head_init(&list);
+
+       spin_lock_bh(&pq->hold_queue.lock);
+       skb_queue_splice_init(&pq->hold_queue, &list);
+       del_timer(&pq->hold_timer);
+       spin_unlock_bh(&pq->hold_queue.lock);
+
+       if (skb_queue_empty(&list))
+               return;
+
+       pq = &new->polq;
+
+       spin_lock_bh(&pq->hold_queue.lock);
+       skb_queue_splice(&list, &pq->hold_queue);
+       pq->timeout = XFRM_QUEUE_TMO_MIN;
+       mod_timer(&pq->hold_timer, jiffies);
+       spin_unlock_bh(&pq->hold_queue.lock);
+}
+
 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 {
        struct net *net = xp_net(policy);
@@ -603,8 +648,10 @@ int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
        net->xfrm.policy_count[dir]++;
        atomic_inc(&flow_cache_genid);
        rt_genid_bump(net);
-       if (delpol)
+       if (delpol) {
+               xfrm_policy_requeue(delpol, policy);
                __xfrm_policy_unlink(delpol, dir);
+       }
        policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
        hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
        policy->curlft.add_time = get_seconds();
@@ -1115,11 +1162,15 @@ int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
                pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
                __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
        }
-       if (old_pol)
+       if (old_pol) {
+               if (pol)
+                       xfrm_policy_requeue(old_pol, pol);
+
                /* Unlinking succeeds always. This is the only function
                 * allowed to delete or replace socket policy.
                 */
                __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
+       }
        write_unlock_bh(&xfrm_policy_lock);
 
        if (old_pol) {
@@ -1310,6 +1361,8 @@ static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *f
                 * It means we need to try again resolving. */
                if (xdst->num_xfrms > 0)
                        return NULL;
+       } else if (dst->flags & DST_XFRM_QUEUE) {
+               return NULL;
        } else {
                /* Real bundle */
                if (stale_bundle(dst))
@@ -1673,6 +1726,171 @@ xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
        return xdst;
 }
 
+static void xfrm_policy_queue_process(unsigned long arg)
+{
+       int err = 0;
+       struct sk_buff *skb;
+       struct sock *sk;
+       struct dst_entry *dst;
+       struct net_device *dev;
+       struct xfrm_policy *pol = (struct xfrm_policy *)arg;
+       struct xfrm_policy_queue *pq = &pol->polq;
+       struct flowi fl;
+       struct sk_buff_head list;
+
+       spin_lock(&pq->hold_queue.lock);
+       skb = skb_peek(&pq->hold_queue);
+       dst = skb_dst(skb);
+       sk = skb->sk;
+       xfrm_decode_session(skb, &fl, dst->ops->family);
+       spin_unlock(&pq->hold_queue.lock);
+
+       dst_hold(dst->path);
+       dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
+                         sk, 0);
+       if (IS_ERR(dst))
+               goto purge_queue;
+
+       if (dst->flags & DST_XFRM_QUEUE) {
+               dst_release(dst);
+
+               if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
+                       goto purge_queue;
+
+               pq->timeout = pq->timeout << 1;
+               mod_timer(&pq->hold_timer, jiffies + pq->timeout);
+               return;
+       }
+
+       dst_release(dst);
+
+       __skb_queue_head_init(&list);
+
+       spin_lock(&pq->hold_queue.lock);
+       pq->timeout = 0;
+       skb_queue_splice_init(&pq->hold_queue, &list);
+       spin_unlock(&pq->hold_queue.lock);
+
+       while (!skb_queue_empty(&list)) {
+               skb = __skb_dequeue(&list);
+
+               xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
+               dst_hold(skb_dst(skb)->path);
+               dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
+                                 &fl, skb->sk, 0);
+               if (IS_ERR(dst)) {
+                       dev_put(skb->dev);
+                       kfree_skb(skb);
+                       continue;
+               }
+
+               nf_reset(skb);
+               skb_dst_drop(skb);
+               skb_dst_set(skb, dst);
+
+               dev = skb->dev;
+               err = dst_output(skb);
+               dev_put(dev);
+       }
+
+       return;
+
+purge_queue:
+       pq->timeout = 0;
+       xfrm_queue_purge(&pq->hold_queue);
+}
+
+static int xdst_queue_output(struct sk_buff *skb)
+{
+       unsigned long sched_next;
+       struct dst_entry *dst = skb_dst(skb);
+       struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
+       struct xfrm_policy_queue *pq = &xdst->pols[0]->polq;
+
+       if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
+               kfree_skb(skb);
+               return -EAGAIN;
+       }
+
+       skb_dst_force(skb);
+       dev_hold(skb->dev);
+
+       spin_lock_bh(&pq->hold_queue.lock);
+
+       if (!pq->timeout)
+               pq->timeout = XFRM_QUEUE_TMO_MIN;
+
+       sched_next = jiffies + pq->timeout;
+
+       if (del_timer(&pq->hold_timer)) {
+               if (time_before(pq->hold_timer.expires, sched_next))
+                       sched_next = pq->hold_timer.expires;
+       }
+
+       __skb_queue_tail(&pq->hold_queue, skb);
+       mod_timer(&pq->hold_timer, sched_next);
+
+       spin_unlock_bh(&pq->hold_queue.lock);
+
+       return 0;
+}
+
+static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
+                                                struct dst_entry *dst,
+                                                const struct flowi *fl,
+                                                int num_xfrms,
+                                                u16 family)
+{
+       int err;
+       struct net_device *dev;
+       struct dst_entry *dst1;
+       struct xfrm_dst *xdst;
+
+       xdst = xfrm_alloc_dst(net, family);
+       if (IS_ERR(xdst))
+               return xdst;
+
+       if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
+           (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
+               return xdst;
+
+       dst1 = &xdst->u.dst;
+       dst_hold(dst);
+       xdst->route = dst;
+
+       dst_copy_metrics(dst1, dst);
+
+       dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
+       dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
+       dst1->lastuse = jiffies;
+
+       dst1->input = dst_discard;
+       dst1->output = xdst_queue_output;
+
+       dst_hold(dst);
+       dst1->child = dst;
+       dst1->path = dst;
+
+       xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
+
+       err = -ENODEV;
+       dev = dst->dev;
+       if (!dev)
+               goto free_dst;
+
+       err = xfrm_fill_dst(xdst, dev, fl);
+       if (err)
+               goto free_dst;
+
+out:
+       return xdst;
+
+free_dst:
+       dst_release(dst1);
+       xdst = ERR_PTR(err);
+       goto out;
+}
+
 static struct flow_cache_object *
 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
                   struct flow_cache_object *oldflo, void *ctx)
@@ -1751,7 +1969,7 @@ make_dummy_bundle:
        /* We found policies, but there's no bundles to instantiate:
         * either because the policy blocks, has no transformations or
         * we could not build template (no xfrm_states).*/
-       xdst = xfrm_alloc_dst(net, family);
+       xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
        if (IS_ERR(xdst)) {
                xfrm_pols_put(pols, num_pols);
                return ERR_CAST(xdst);
@@ -2359,6 +2577,9 @@ static int xfrm_bundle_ok(struct xfrm_dst *first)
            (dst->dev && !netif_running(dst->dev)))
                return 0;
 
+       if (dst->flags & DST_XFRM_QUEUE)
+               return 1;
+
        last = NULL;
 
        do {