ipv4: Restore old dst_free() behavior.

commit 404e0a8b6a55 (net: ipv4: fix RCU races on dst refcounts) tried
to solve a race but added a problem at device/fib dismantle time :

We really want to call dst_free() as soon as possible, even if sockets
still have dst in their cache.
dst_release() calls in free_fib_info_rcu() are not welcomed.

Root of the problem was that now we also cache output routes (in
nh_rth_output), we must use call_rcu() instead of call_rcu_bh() in
rt_free(), because output route lookups are done in process context.

Based on feedback and initial patch from David Miller (adding another
call_rcu_bh() call in fib, but it appears it was not the right fix)

I left the inet_sk_rx_dst_set() helper and added __rcu attributes
to nh_rth_output and nh_rth_input to better document what is going on in
this code.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/include/net/dst.h b/include/net/dst.h
index 31a9fd39edb6b10fc283e1a8680dc3f63e3dbfee..baf59789006427bc7c373449b62b271e6e777d15 100644 (file)
--- a/include/net/dst.h
+++ b/include/net/dst.h
@@ -61,7 +61,6 @@ struct dst_entry {
 #define DST_NOPEER             0x0040
 #define DST_FAKE_RTABLE                0x0080
 #define DST_XFRM_TUNNEL                0x0100
-#define DST_RCU_FREE           0x0200
 
        unsigned short          pending_confirm;
 
@@ -383,6 +382,12 @@ static inline void dst_free(struct dst_entry *dst)
        __dst_free(dst);
 }
 
+static inline void dst_rcu_free(struct rcu_head *head)
+{
+       struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
+       dst_free(dst);
+}
+
 static inline void dst_confirm(struct dst_entry *dst)
 {
        dst->pending_confirm = 1;

diff --git a/include/net/inet_sock.h b/include/net/inet_sock.h
index e3fd34c83ac95554280754df6bd34dc54343a8d7..83b567fe194163b74f734cedaf1dffff28e914a6 100644 (file)
--- a/include/net/inet_sock.h
+++ b/include/net/inet_sock.h
@@ -253,13 +253,9 @@ static inline void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb
 {
        struct dst_entry *dst = skb_dst(skb);
 
-       if (atomic_inc_not_zero(&dst->__refcnt)) {
-               if (!(dst->flags & DST_RCU_FREE))
-                       dst->flags |= DST_RCU_FREE;
-
-               sk->sk_rx_dst = dst;
-               inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
-       }
+       dst_hold(dst);
+       sk->sk_rx_dst = dst;
+       inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
 }
 
 #endif /* _INET_SOCK_H */

diff --git a/include/net/ip_fib.h b/include/net/ip_fib.h
index e69c3a47153de4826fdbdd8a42d64a0bb2eed974..e521a03515b170cbff8c94d65e08f8cd69d88627 100644 (file)
--- a/include/net/ip_fib.h
+++ b/include/net/ip_fib.h
@@ -81,8 +81,8 @@ struct fib_nh {
        __be32                  nh_gw;
        __be32                  nh_saddr;
        int                     nh_saddr_genid;
-       struct rtable           *nh_rth_output;
-       struct rtable           *nh_rth_input;
+       struct rtable __rcu     *nh_rth_output;
+       struct rtable __rcu     *nh_rth_input;
        struct fnhe_hash_bucket *nh_exceptions;
 };
 

diff --git a/net/core/dst.c b/net/core/dst.c
index d9e33ebe170fc42d757504e4a1392bf16e5933a1..069d51d29414a3245f9b60b66c534f78ce3b1752 100644 (file)
--- a/net/core/dst.c
+++ b/net/core/dst.c
@@ -258,15 +258,6 @@ again:
 }
 EXPORT_SYMBOL(dst_destroy);
 
-static void dst_rcu_destroy(struct rcu_head *head)
-{
-       struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
-
-       dst = dst_destroy(dst);
-       if (dst)
-               __dst_free(dst);
-}
-
 void dst_release(struct dst_entry *dst)
 {
        if (dst) {
@@ -274,14 +265,10 @@ void dst_release(struct dst_entry *dst)
 
                newrefcnt = atomic_dec_return(&dst->__refcnt);
                WARN_ON(newrefcnt < 0);
-               if (unlikely(dst->flags & (DST_NOCACHE | DST_RCU_FREE)) && !newrefcnt) {
-                       if (dst->flags & DST_RCU_FREE) {
-                               call_rcu_bh(&dst->rcu_head, dst_rcu_destroy);
-                       } else {
-                               dst = dst_destroy(dst);
-                               if (dst)
-                                       __dst_free(dst);
-                       }
+               if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt) {
+                       dst = dst_destroy(dst);
+                       if (dst)
+                               __dst_free(dst);
                }
        }
 }
@@ -333,14 +320,11 @@ EXPORT_SYMBOL(__dst_destroy_metrics_generic);
  */
 void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
 {
-       bool hold;
-
        WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
        /* If dst not in cache, we must take a reference, because
         * dst_release() will destroy dst as soon as its refcount becomes zero
         */
-       hold = (dst->flags & (DST_NOCACHE | DST_RCU_FREE)) == DST_NOCACHE;
-       if (unlikely(hold)) {
+       if (unlikely(dst->flags & DST_NOCACHE)) {
                dst_hold(dst);
                skb_dst_set(skb, dst);
        } else {

diff --git a/net/decnet/dn_route.c b/net/decnet/dn_route.c
index 26719779ad8ee595d013beed4987ce8b6d591e41..85a3604c87c8d2cac6cac4d7049d9ba75877bf66 100644 (file)
--- a/net/decnet/dn_route.c
+++ b/net/decnet/dn_route.c
@@ -184,12 +184,6 @@ static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
        return dn_rt_hash_mask & (unsigned int)tmp;
 }
 
-static inline void dst_rcu_free(struct rcu_head *head)
-{
-       struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
-       dst_free(dst);
-}
-
 static inline void dnrt_free(struct dn_route *rt)
 {
        call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);

diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index e55171f184f9731274c568345b9b74c14b2e63cc..625cf185c489b4ad871f152f802bbb0027652c5a 100644 (file)
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -161,6 +161,21 @@ static void free_nh_exceptions(struct fib_nh *nh)
        kfree(hash);
 }
 
+static void rt_nexthop_free(struct rtable __rcu **rtp)
+{
+       struct rtable *rt = rcu_dereference_protected(*rtp, 1);
+
+       if (!rt)
+               return;
+
+       /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
+        * because we waited an RCU grace period before calling
+        * free_fib_info_rcu()
+        */
+
+       dst_free(&rt->dst);
+}
+
 /* Release a nexthop info record */
 static void free_fib_info_rcu(struct rcu_head *head)
 {
@@ -171,10 +186,8 @@ static void free_fib_info_rcu(struct rcu_head *head)
                        dev_put(nexthop_nh->nh_dev);
                if (nexthop_nh->nh_exceptions)
                        free_nh_exceptions(nexthop_nh);
-               if (nexthop_nh->nh_rth_output)
-                       dst_release(&nexthop_nh->nh_rth_output->dst);
-               if (nexthop_nh->nh_rth_input)
-                       dst_release(&nexthop_nh->nh_rth_input->dst);
+               rt_nexthop_free(&nexthop_nh->nh_rth_output);
+               rt_nexthop_free(&nexthop_nh->nh_rth_input);
        } endfor_nexthops(fi);
 
        release_net(fi->fib_net);

diff --git a/net/ipv4/route.c b/net/ipv4/route.c
index d6eabcfe8a9089c315cb6b812d1d327e4fdf1406..2bd1074774692e670e6a07c1e3da46f8c2d4af48 100644 (file)
--- a/net/ipv4/route.c
+++ b/net/ipv4/route.c
@@ -1199,23 +1199,31 @@ restart:
        fnhe->fnhe_stamp = jiffies;
 }
 
+static inline void rt_free(struct rtable *rt)
+{
+       call_rcu(&rt->dst.rcu_head, dst_rcu_free);
+}
+
 static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
 {
-       struct rtable *orig, *prev, **p = &nh->nh_rth_output;
+       struct rtable *orig, *prev, **p = (struct rtable **)&nh->nh_rth_output;
 
        if (rt_is_input_route(rt))
-               p = &nh->nh_rth_input;
+               p = (struct rtable **)&nh->nh_rth_input;
 
        orig = *p;
 
-       rt->dst.flags |= DST_RCU_FREE;
-       dst_hold(&rt->dst);
        prev = cmpxchg(p, orig, rt);
        if (prev == orig) {
                if (orig)
-                       dst_release(&orig->dst);
+                       rt_free(orig);
        } else {
-               dst_release(&rt->dst);
+               /* Routes we intend to cache in the FIB nexthop have
+                * the DST_NOCACHE bit clear.  However, if we are
+                * unsuccessful at storing this route into the cache
+                * we really need to set it.
+                */
+               rt->dst.flags |= DST_NOCACHE;
        }
 }
 
@@ -1412,7 +1420,7 @@ static int __mkroute_input(struct sk_buff *skb,
        do_cache = false;
        if (res->fi) {
                if (!itag) {
-                       rth = FIB_RES_NH(*res).nh_rth_input;
+                       rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
                        if (rt_cache_valid(rth)) {
                                skb_dst_set_noref(skb, &rth->dst);
                                goto out;
@@ -1574,7 +1582,7 @@ local_input:
        do_cache = false;
        if (res.fi) {
                if (!itag) {
-                       rth = FIB_RES_NH(res).nh_rth_input;
+                       rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
                        if (rt_cache_valid(rth)) {
                                skb_dst_set_noref(skb, &rth->dst);
                                err = 0;
@@ -1742,7 +1750,7 @@ static struct rtable *__mkroute_output(const struct fib_result *res,
        if (fi) {
                fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
                if (!fnhe) {
-                       rth = FIB_RES_NH(*res).nh_rth_output;
+                       rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_output);
                        if (rt_cache_valid(rth)) {
                                dst_hold(&rth->dst);
                                return rth;