From 404e0a8b6a55d5e1cd138c6deb1bca9abdf75d8c Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Sun, 29 Jul 2012 23:20:37 +0000 Subject: net: ipv4: fix RCU races on dst refcounts commit c6cffba4ffa2 (ipv4: Fix input route performance regression.) added various fatal races with dst refcounts. crashes happen on tcp workloads if routes are added/deleted at the same time. The dst_free() calls from free_fib_info_rcu() are clearly racy. We need instead regular dst refcounting (dst_release()) and make sure dst_release() is aware of RCU grace periods : Add DST_RCU_FREE flag so that dst_release() respects an RCU grace period before dst destruction for cached dst Introduce a new inet_sk_rx_dst_set() helper, using atomic_inc_not_zero() to make sure we dont increase a zero refcount (On a dst currently waiting an rcu grace period before destruction) rt_cache_route() must take a reference on the new cached route, and release it if was not able to install it. With this patch, my machines survive various benchmarks. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- include/net/dst.h | 7 +------ include/net/inet_sock.h | 13 +++++++++++++ 2 files changed, 14 insertions(+), 6 deletions(-) (limited to 'include') diff --git a/include/net/dst.h b/include/net/dst.h index baf597890064..31a9fd39edb6 100644 --- a/include/net/dst.h +++ b/include/net/dst.h @@ -61,6 +61,7 @@ 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; @@ -382,12 +383,6 @@ 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 613cfa401672..e3fd34c83ac9 100644 --- a/include/net/inet_sock.h +++ b/include/net/inet_sock.h @@ -249,4 +249,17 @@ static inline __u8 inet_sk_flowi_flags(const struct sock *sk) return flags; } +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; + } +} + #endif /* _INET_SOCK_H */ -- cgit v1.2.3 From 0c7462a2351b4cc502f326aad7fedd04909928be Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Mon, 30 Jul 2012 07:14:29 +0000 Subject: ipv4: remove rt_cache_rebuild_count After IP route cache removal, rt_cache_rebuild_count is no longer used. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- Documentation/networking/ip-sysctl.txt | 6 ------ include/net/netns/ipv4.h | 2 -- net/ipv4/sysctl_net_ipv4.c | 11 ----------- 3 files changed, 19 deletions(-) (limited to 'include') diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt index 406a5226220d..ca447b35b833 100644 --- a/Documentation/networking/ip-sysctl.txt +++ b/Documentation/networking/ip-sysctl.txt @@ -48,12 +48,6 @@ min_adv_mss - INTEGER The advertised MSS depends on the first hop route MTU, but will never be lower than this setting. -rt_cache_rebuild_count - INTEGER - The per net-namespace route cache emergency rebuild threshold. - Any net-namespace having its route cache rebuilt due to - a hash bucket chain being too long more than this many times - will have its route caching disabled - IP Fragmentation: ipfrag_high_thresh - INTEGER diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h index 0ffb8e31f3cd..1474dd65c66f 100644 --- a/include/net/netns/ipv4.h +++ b/include/net/netns/ipv4.h @@ -61,8 +61,6 @@ struct netns_ipv4 { int sysctl_icmp_ratelimit; int sysctl_icmp_ratemask; int sysctl_icmp_errors_use_inbound_ifaddr; - int sysctl_rt_cache_rebuild_count; - int current_rt_cache_rebuild_count; unsigned int sysctl_ping_group_range[2]; long sysctl_tcp_mem[3]; diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c index 5840c3255721..4b6487a68279 100644 --- a/net/ipv4/sysctl_net_ipv4.c +++ b/net/ipv4/sysctl_net_ipv4.c @@ -783,13 +783,6 @@ static struct ctl_table ipv4_net_table[] = { .mode = 0644, .proc_handler = proc_dointvec }, - { - .procname = "rt_cache_rebuild_count", - .data = &init_net.ipv4.sysctl_rt_cache_rebuild_count, - .maxlen = sizeof(int), - .mode = 0644, - .proc_handler = proc_dointvec - }, { .procname = "ping_group_range", .data = &init_net.ipv4.sysctl_ping_group_range, @@ -829,8 +822,6 @@ static __net_init int ipv4_sysctl_init_net(struct net *net) table[5].data = &net->ipv4.sysctl_icmp_ratemask; table[6].data = - &net->ipv4.sysctl_rt_cache_rebuild_count; - table[7].data = &net->ipv4.sysctl_ping_group_range; } @@ -842,8 +833,6 @@ static __net_init int ipv4_sysctl_init_net(struct net *net) net->ipv4.sysctl_ping_group_range[0] = 1; net->ipv4.sysctl_ping_group_range[1] = 0; - net->ipv4.sysctl_rt_cache_rebuild_count = 4; - tcp_init_mem(net); net->ipv4.ipv4_hdr = register_net_sysctl(net, "net/ipv4", table); -- cgit v1.2.3 From 54764bb647b2e847c512acf8d443df965da35000 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Tue, 31 Jul 2012 01:08:23 +0000 Subject: 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 Signed-off-by: David S. Miller --- include/net/dst.h | 7 ++++++- include/net/inet_sock.h | 10 +++------- include/net/ip_fib.h | 4 ++-- net/core/dst.c | 26 +++++--------------------- net/decnet/dn_route.c | 6 ------ net/ipv4/fib_semantics.c | 21 +++++++++++++++++---- net/ipv4/route.c | 26 +++++++++++++++++--------- 7 files changed, 50 insertions(+), 50 deletions(-) (limited to 'include') diff --git a/include/net/dst.h b/include/net/dst.h index 31a9fd39edb6..baf597890064 100644 --- 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 e3fd34c83ac9..83b567fe1941 100644 --- 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 e69c3a47153d..e521a03515b1 100644 --- 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 d9e33ebe170f..069d51d29414 100644 --- 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 26719779ad8e..85a3604c87c8 100644 --- 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 e55171f184f9..625cf185c489 100644 --- 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 d6eabcfe8a90..2bd107477469 100644 --- 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; -- cgit v1.2.3 From d26b3a7c4b3b26319f18bb645de93eba8f4bdcd5 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Tue, 31 Jul 2012 05:45:30 +0000 Subject: ipv4: percpu nh_rth_output cache Input path is mostly run under RCU and doesnt touch dst refcnt But output path on forwarding or UDP workloads hits badly dst refcount, and we have lot of false sharing, for example in ipv4_mtu() when reading rt->rt_pmtu Using a percpu cache for nh_rth_output gives a nice performance increase at a small cost. 24 udpflood test on my 24 cpu machine (dummy0 output device) (each process sends 1.000.000 udp frames, 24 processes are started) before : 5.24 s after : 2.06 s For reference, time on linux-3.5 : 6.60 s Signed-off-by: Eric Dumazet Tested-by: Alexander Duyck Signed-off-by: David S. Miller --- include/net/ip_fib.h | 3 ++- net/ipv4/fib_semantics.c | 20 +++++++++++++++++++- net/ipv4/route.c | 18 +++++++++++++----- 3 files changed, 34 insertions(+), 7 deletions(-) (limited to 'include') diff --git a/include/net/ip_fib.h b/include/net/ip_fib.h index e521a03515b1..e331746029b4 100644 --- a/include/net/ip_fib.h +++ b/include/net/ip_fib.h @@ -21,6 +21,7 @@ #include #include #include +#include struct fib_config { u8 fc_dst_len; @@ -81,7 +82,7 @@ struct fib_nh { __be32 nh_gw; __be32 nh_saddr; int nh_saddr_genid; - struct rtable __rcu *nh_rth_output; + struct rtable __rcu * __percpu *nh_pcpu_rth_output; struct rtable __rcu *nh_rth_input; struct fnhe_hash_bucket *nh_exceptions; }; diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c index 625cf185c489..fe2ca02a1979 100644 --- a/net/ipv4/fib_semantics.c +++ b/net/ipv4/fib_semantics.c @@ -176,6 +176,23 @@ static void rt_nexthop_free(struct rtable __rcu **rtp) dst_free(&rt->dst); } +static void rt_nexthop_free_cpus(struct rtable __rcu * __percpu *rtp) +{ + int cpu; + + if (!rtp) + return; + + for_each_possible_cpu(cpu) { + struct rtable *rt; + + rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); + if (rt) + dst_free(&rt->dst); + } + free_percpu(rtp); +} + /* Release a nexthop info record */ static void free_fib_info_rcu(struct rcu_head *head) { @@ -186,7 +203,7 @@ 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); - rt_nexthop_free(&nexthop_nh->nh_rth_output); + rt_nexthop_free_cpus(nexthop_nh->nh_pcpu_rth_output); rt_nexthop_free(&nexthop_nh->nh_rth_input); } endfor_nexthops(fi); @@ -817,6 +834,7 @@ struct fib_info *fib_create_info(struct fib_config *cfg) fi->fib_nhs = nhs; change_nexthops(fi) { nexthop_nh->nh_parent = fi; + nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *); } endfor_nexthops(fi) if (cfg->fc_mx) { diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 2bd107477469..4f6276ce0af3 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -1206,11 +1206,15 @@ static inline void rt_free(struct rtable *rt) static void rt_cache_route(struct fib_nh *nh, struct rtable *rt) { - struct rtable *orig, *prev, **p = (struct rtable **)&nh->nh_rth_output; + struct rtable *orig, *prev, **p; - if (rt_is_input_route(rt)) + if (rt_is_input_route(rt)) { p = (struct rtable **)&nh->nh_rth_input; - + } else { + if (!nh->nh_pcpu_rth_output) + goto nocache; + p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output); + } orig = *p; prev = cmpxchg(p, orig, rt); @@ -1223,6 +1227,7 @@ static void rt_cache_route(struct fib_nh *nh, struct rtable *rt) * unsuccessful at storing this route into the cache * we really need to set it. */ +nocache: rt->dst.flags |= DST_NOCACHE; } } @@ -1749,8 +1754,11 @@ static struct rtable *__mkroute_output(const struct fib_result *res, fnhe = NULL; if (fi) { fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr); - if (!fnhe) { - rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_output); + if (!fnhe && FIB_RES_NH(*res).nh_pcpu_rth_output) { + struct rtable __rcu **prth; + + prth = __this_cpu_ptr(FIB_RES_NH(*res).nh_pcpu_rth_output); + rth = rcu_dereference(*prth); if (rt_cache_valid(rth)) { dst_hold(&rth->dst); return rth; -- cgit v1.2.3 From c5038a8327b980a5b279fa193163c468011de009 Mon Sep 17 00:00:00 2001 From: David S. Miller Date: Tue, 31 Jul 2012 15:02:02 -0700 Subject: ipv4: Cache routes in nexthop exception entries. Signed-off-by: David S. Miller --- include/net/ip_fib.h | 1 + net/ipv4/fib_semantics.c | 39 +++++++++--------- net/ipv4/route.c | 103 ++++++++++++++++++++++++++--------------------- 3 files changed, 79 insertions(+), 64 deletions(-) (limited to 'include') diff --git a/include/net/ip_fib.h b/include/net/ip_fib.h index e331746029b4..926142ed8d7a 100644 --- a/include/net/ip_fib.h +++ b/include/net/ip_fib.h @@ -55,6 +55,7 @@ struct fib_nh_exception { u32 fnhe_pmtu; __be32 fnhe_gw; unsigned long fnhe_expires; + struct rtable __rcu *fnhe_rth; unsigned long fnhe_stamp; }; diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c index fe2ca02a1979..da80dc14cc76 100644 --- a/net/ipv4/fib_semantics.c +++ b/net/ipv4/fib_semantics.c @@ -140,6 +140,21 @@ const struct fib_prop fib_props[RTN_MAX + 1] = { }, }; +static void rt_fibinfo_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); +} + static void free_nh_exceptions(struct fib_nh *nh) { struct fnhe_hash_bucket *hash = nh->nh_exceptions; @@ -153,6 +168,9 @@ static void free_nh_exceptions(struct fib_nh *nh) struct fib_nh_exception *next; next = rcu_dereference_protected(fnhe->fnhe_next, 1); + + rt_fibinfo_free(&fnhe->fnhe_rth); + kfree(fnhe); fnhe = next; @@ -161,22 +179,7 @@ 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); -} - -static void rt_nexthop_free_cpus(struct rtable __rcu * __percpu *rtp) +static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) { int cpu; @@ -203,8 +206,8 @@ 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); - rt_nexthop_free_cpus(nexthop_nh->nh_pcpu_rth_output); - rt_nexthop_free(&nexthop_nh->nh_rth_input); + rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output); + rt_fibinfo_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 4f6276ce0af3..b102eeb16e34 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -587,11 +587,17 @@ static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, build_sk_flow_key(fl4, sk); } -static DEFINE_SEQLOCK(fnhe_seqlock); +static inline void rt_free(struct rtable *rt) +{ + call_rcu(&rt->dst.rcu_head, dst_rcu_free); +} + +static DEFINE_SPINLOCK(fnhe_lock); static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) { struct fib_nh_exception *fnhe, *oldest; + struct rtable *orig; oldest = rcu_dereference(hash->chain); for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; @@ -599,6 +605,11 @@ static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) oldest = fnhe; } + orig = rcu_dereference(oldest->fnhe_rth); + if (orig) { + RCU_INIT_POINTER(oldest->fnhe_rth, NULL); + rt_free(orig); + } return oldest; } @@ -620,7 +631,7 @@ static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw, int depth; u32 hval = fnhe_hashfun(daddr); - write_seqlock_bh(&fnhe_seqlock); + spin_lock_bh(&fnhe_lock); hash = nh->nh_exceptions; if (!hash) { @@ -667,7 +678,7 @@ static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw, fnhe->fnhe_stamp = jiffies; out_unlock: - write_sequnlock_bh(&fnhe_seqlock); + spin_unlock_bh(&fnhe_lock); return; } @@ -1167,41 +1178,40 @@ static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr) static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, __be32 daddr) { - __be32 fnhe_daddr, gw; - unsigned long expires; - unsigned int seq; - u32 pmtu; - -restart: - seq = read_seqbegin(&fnhe_seqlock); - fnhe_daddr = fnhe->fnhe_daddr; - gw = fnhe->fnhe_gw; - pmtu = fnhe->fnhe_pmtu; - expires = fnhe->fnhe_expires; - if (read_seqretry(&fnhe_seqlock, seq)) - goto restart; - - if (daddr != fnhe_daddr) - return; + spin_lock_bh(&fnhe_lock); - if (pmtu) { - unsigned long diff = expires - jiffies; + if (daddr == fnhe->fnhe_daddr) { + struct rtable *orig; - if (time_before(jiffies, expires)) { - rt->rt_pmtu = pmtu; - dst_set_expires(&rt->dst, diff); + if (fnhe->fnhe_pmtu) { + unsigned long expires = fnhe->fnhe_expires; + unsigned long diff = expires - jiffies; + + if (time_before(jiffies, expires)) { + rt->rt_pmtu = fnhe->fnhe_pmtu; + dst_set_expires(&rt->dst, diff); + } + } + if (fnhe->fnhe_gw) { + rt->rt_flags |= RTCF_REDIRECTED; + rt->rt_gateway = fnhe->fnhe_gw; } - } - if (gw) { - rt->rt_flags |= RTCF_REDIRECTED; - rt->rt_gateway = gw; - } - fnhe->fnhe_stamp = jiffies; -} -static inline void rt_free(struct rtable *rt) -{ - call_rcu(&rt->dst.rcu_head, dst_rcu_free); + orig = rcu_dereference(fnhe->fnhe_rth); + rcu_assign_pointer(fnhe->fnhe_rth, rt); + if (orig) + rt_free(orig); + + fnhe->fnhe_stamp = jiffies; + } else { + /* Routes we intend to cache in nexthop exception 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; + } + spin_unlock_bh(&fnhe_lock); } static void rt_cache_route(struct fib_nh *nh, struct rtable *rt) @@ -1249,13 +1259,13 @@ static void rt_set_nexthop(struct rtable *rt, __be32 daddr, if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) rt->rt_gateway = nh->nh_gw; - if (unlikely(fnhe)) - rt_bind_exception(rt, fnhe, daddr); dst_init_metrics(&rt->dst, fi->fib_metrics, true); #ifdef CONFIG_IP_ROUTE_CLASSID rt->dst.tclassid = nh->nh_tclassid; #endif - if (!(rt->dst.flags & DST_NOCACHE)) + if (unlikely(fnhe)) + rt_bind_exception(rt, fnhe, daddr); + else if (!(rt->dst.flags & DST_NOCACHE)) rt_cache_route(nh, rt); } @@ -1753,22 +1763,23 @@ static struct rtable *__mkroute_output(const struct fib_result *res, fnhe = NULL; if (fi) { - fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr); - if (!fnhe && FIB_RES_NH(*res).nh_pcpu_rth_output) { - struct rtable __rcu **prth; + struct rtable __rcu **prth; + fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr); + if (fnhe) + prth = &fnhe->fnhe_rth; + else prth = __this_cpu_ptr(FIB_RES_NH(*res).nh_pcpu_rth_output); - rth = rcu_dereference(*prth); - if (rt_cache_valid(rth)) { - dst_hold(&rth->dst); - return rth; - } + rth = rcu_dereference(*prth); + if (rt_cache_valid(rth)) { + dst_hold(&rth->dst); + return rth; } } rth = rt_dst_alloc(dev_out, IN_DEV_CONF_GET(in_dev, NOPOLICY), IN_DEV_CONF_GET(in_dev, NOXFRM), - fi && !fnhe); + fi); if (!rth) return ERR_PTR(-ENOBUFS); -- cgit v1.2.3 From caacf05e5ad1abf0a2864863da4e33024bc68ec6 Mon Sep 17 00:00:00 2001 From: David S. Miller Date: Tue, 31 Jul 2012 15:06:50 -0700 Subject: ipv4: Properly purge netdev references on uncached routes. When a device is unregistered, we have to purge all of the references to it that may exist in the entire system. If a route is uncached, we currently have no way of accomplishing this. So create a global list that is scanned when a network device goes down. This mirrors the logic in net/core/dst.c's dst_ifdown(). Signed-off-by: David S. Miller --- include/net/route.h | 3 +++ net/ipv4/fib_frontend.c | 1 + net/ipv4/route.c | 68 ++++++++++++++++++++++++++++++++++++++++++++++--- net/ipv4/xfrm4_policy.c | 1 + 4 files changed, 69 insertions(+), 4 deletions(-) (limited to 'include') diff --git a/include/net/route.h b/include/net/route.h index 8c52bc6f1c90..776a27f1ab78 100644 --- a/include/net/route.h +++ b/include/net/route.h @@ -57,6 +57,8 @@ struct rtable { /* Miscellaneous cached information */ u32 rt_pmtu; + + struct list_head rt_uncached; }; static inline bool rt_is_input_route(const struct rtable *rt) @@ -107,6 +109,7 @@ extern struct ip_rt_acct __percpu *ip_rt_acct; struct in_device; extern int ip_rt_init(void); extern void rt_cache_flush(struct net *net, int how); +extern void rt_flush_dev(struct net_device *dev); extern struct rtable *__ip_route_output_key(struct net *, struct flowi4 *flp); extern struct rtable *ip_route_output_flow(struct net *, struct flowi4 *flp, struct sock *sk); diff --git a/net/ipv4/fib_frontend.c b/net/ipv4/fib_frontend.c index 8732cc7920ed..c43ae3fba792 100644 --- a/net/ipv4/fib_frontend.c +++ b/net/ipv4/fib_frontend.c @@ -1046,6 +1046,7 @@ static int fib_netdev_event(struct notifier_block *this, unsigned long event, vo if (event == NETDEV_UNREGISTER) { fib_disable_ip(dev, 2, -1); + rt_flush_dev(dev); return NOTIFY_DONE; } diff --git a/net/ipv4/route.c b/net/ipv4/route.c index b102eeb16e34..c035251beb07 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -147,6 +147,7 @@ static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb, u32 mtu); static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb); +static void ipv4_dst_destroy(struct dst_entry *dst); static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how) @@ -170,6 +171,7 @@ static struct dst_ops ipv4_dst_ops = { .default_advmss = ipv4_default_advmss, .mtu = ipv4_mtu, .cow_metrics = ipv4_cow_metrics, + .destroy = ipv4_dst_destroy, .ifdown = ipv4_dst_ifdown, .negative_advice = ipv4_negative_advice, .link_failure = ipv4_link_failure, @@ -1175,9 +1177,11 @@ static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr) return NULL; } -static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, +static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, __be32 daddr) { + bool ret = false; + spin_lock_bh(&fnhe_lock); if (daddr == fnhe->fnhe_daddr) { @@ -1203,6 +1207,7 @@ static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, rt_free(orig); fnhe->fnhe_stamp = jiffies; + ret = true; } else { /* Routes we intend to cache in nexthop exception have * the DST_NOCACHE bit clear. However, if we are @@ -1212,11 +1217,14 @@ static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, rt->dst.flags |= DST_NOCACHE; } spin_unlock_bh(&fnhe_lock); + + return ret; } -static void rt_cache_route(struct fib_nh *nh, struct rtable *rt) +static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt) { struct rtable *orig, *prev, **p; + bool ret = true; if (rt_is_input_route(rt)) { p = (struct rtable **)&nh->nh_rth_input; @@ -1239,6 +1247,48 @@ static void rt_cache_route(struct fib_nh *nh, struct rtable *rt) */ nocache: rt->dst.flags |= DST_NOCACHE; + ret = false; + } + + return ret; +} + +static DEFINE_SPINLOCK(rt_uncached_lock); +static LIST_HEAD(rt_uncached_list); + +static void rt_add_uncached_list(struct rtable *rt) +{ + spin_lock_bh(&rt_uncached_lock); + list_add_tail(&rt->rt_uncached, &rt_uncached_list); + spin_unlock_bh(&rt_uncached_lock); +} + +static void ipv4_dst_destroy(struct dst_entry *dst) +{ + struct rtable *rt = (struct rtable *) dst; + + if (dst->flags & DST_NOCACHE) { + spin_lock_bh(&rt_uncached_lock); + list_del(&rt->rt_uncached); + spin_unlock_bh(&rt_uncached_lock); + } +} + +void rt_flush_dev(struct net_device *dev) +{ + if (!list_empty(&rt_uncached_list)) { + struct net *net = dev_net(dev); + struct rtable *rt; + + spin_lock_bh(&rt_uncached_lock); + list_for_each_entry(rt, &rt_uncached_list, rt_uncached) { + if (rt->dst.dev != dev) + continue; + rt->dst.dev = net->loopback_dev; + dev_hold(rt->dst.dev); + dev_put(dev); + } + spin_unlock_bh(&rt_uncached_lock); } } @@ -1254,6 +1304,8 @@ static void rt_set_nexthop(struct rtable *rt, __be32 daddr, struct fib_nh_exception *fnhe, struct fib_info *fi, u16 type, u32 itag) { + bool cached = false; + if (fi) { struct fib_nh *nh = &FIB_RES_NH(*res); @@ -1264,10 +1316,12 @@ static void rt_set_nexthop(struct rtable *rt, __be32 daddr, rt->dst.tclassid = nh->nh_tclassid; #endif if (unlikely(fnhe)) - rt_bind_exception(rt, fnhe, daddr); + cached = rt_bind_exception(rt, fnhe, daddr); else if (!(rt->dst.flags & DST_NOCACHE)) - rt_cache_route(nh, rt); + cached = rt_cache_route(nh, rt); } + if (unlikely(!cached)) + rt_add_uncached_list(rt); #ifdef CONFIG_IP_ROUTE_CLASSID #ifdef CONFIG_IP_MULTIPLE_TABLES @@ -1334,6 +1388,7 @@ static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, rth->rt_iif = 0; rth->rt_pmtu = 0; rth->rt_gateway = 0; + INIT_LIST_HEAD(&rth->rt_uncached); if (our) { rth->dst.input= ip_local_deliver; rth->rt_flags |= RTCF_LOCAL; @@ -1459,6 +1514,7 @@ static int __mkroute_input(struct sk_buff *skb, rth->rt_iif = 0; rth->rt_pmtu = 0; rth->rt_gateway = 0; + INIT_LIST_HEAD(&rth->rt_uncached); rth->dst.input = ip_forward; rth->dst.output = ip_output; @@ -1625,6 +1681,7 @@ local_input: rth->rt_iif = 0; rth->rt_pmtu = 0; rth->rt_gateway = 0; + INIT_LIST_HEAD(&rth->rt_uncached); if (res.type == RTN_UNREACHABLE) { rth->dst.input= ip_error; rth->dst.error= -err; @@ -1792,6 +1849,7 @@ static struct rtable *__mkroute_output(const struct fib_result *res, rth->rt_iif = orig_oif ? : 0; rth->rt_pmtu = 0; rth->rt_gateway = 0; + INIT_LIST_HEAD(&rth->rt_uncached); RT_CACHE_STAT_INC(out_slow_tot); @@ -2071,6 +2129,8 @@ struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_or rt->rt_type = ort->rt_type; rt->rt_gateway = ort->rt_gateway; + INIT_LIST_HEAD(&rt->rt_uncached); + dst_free(new); } diff --git a/net/ipv4/xfrm4_policy.c b/net/ipv4/xfrm4_policy.c index c6281847f16a..681ea2f413e2 100644 --- a/net/ipv4/xfrm4_policy.c +++ b/net/ipv4/xfrm4_policy.c @@ -92,6 +92,7 @@ static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, xdst->u.rt.rt_type = rt->rt_type; xdst->u.rt.rt_gateway = rt->rt_gateway; xdst->u.rt.rt_pmtu = rt->rt_pmtu; + INIT_LIST_HEAD(&xdst->u.rt.rt_uncached); return 0; } -- cgit v1.2.3