netfilter: xtables: stackptr should be percpu

commit f3c5c1bfd4 (netfilter: xtables: make ip_tables reentrant)
introduced a performance regression, because stackptr array is shared by
all cpus, adding cache line ping pongs. (16 cpus share a 64 bytes cache
line)

Fix this using alloc_percpu()

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-By: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>

4 files changed, 6 insertions, 13 deletions

diff --git a/include/linux/netfilter/x_tables.h b/include/linux/netfilter/x_tables.h
index c00cc0c4d0b7..24e5d01d27d0 100644
--- a/include/linux/netfilter/x_tables.h
+++ b/include/linux/netfilter/x_tables.h
@@ -397,7 +397,7 @@ struct xt_table_info {
          * @stacksize jumps (number of user chains) can possibly be made.
          */
         unsigned int stacksize;
-        unsigned int *stackptr;
+        unsigned int __percpu *stackptr;
         void ***jumpstack;
         /* ipt_entry tables: one per CPU */
         /* Note : this field MUST be the last one, see XT_TABLE_INFO_SZ */
diff --git a/net/ipv4/netfilter/ip_tables.c b/net/ipv4/netfilter/ip_tables.c
index 63958f3394a5..4b6c5ca610fc 100644
--- a/net/ipv4/netfilter/ip_tables.c
+++ b/net/ipv4/netfilter/ip_tables.c
@@ -336,7 +336,7 @@ ipt_do_table(struct sk_buff *skb,
         cpu        = smp_processor_id();
         table_base = private->entries[cpu];
         jumpstack  = (struct ipt_entry **)private->jumpstack[cpu];
-        stackptr   = &private->stackptr[cpu];
+        stackptr   = per_cpu_ptr(private->stackptr, cpu);
         origptr    = *stackptr;
 
         e = get_entry(table_base, private->hook_entry[hook]);
diff --git a/net/ipv6/netfilter/ip6_tables.c b/net/ipv6/netfilter/ip6_tables.c
index 6f517bd83692..9d2d68f0e605 100644
--- a/net/ipv6/netfilter/ip6_tables.c
+++ b/net/ipv6/netfilter/ip6_tables.c
@@ -363,7 +363,7 @@ ip6t_do_table(struct sk_buff *skb,
         cpu        = smp_processor_id();
         table_base = private->entries[cpu];
         jumpstack  = (struct ip6t_entry **)private->jumpstack[cpu];
-        stackptr   = &private->stackptr[cpu];
+        stackptr   = per_cpu_ptr(private->stackptr, cpu);
         origptr    = *stackptr;
 
         e = get_entry(table_base, private->hook_entry[hook]);
diff --git a/net/netfilter/x_tables.c b/net/netfilter/x_tables.c
index 47b1e7917a9c..e34622fa0003 100644
--- a/net/netfilter/x_tables.c
+++ b/net/netfilter/x_tables.c
@@ -699,10 +699,8 @@ void xt_free_table_info(struct xt_table_info *info)
                 vfree(info->jumpstack);
         else
                 kfree(info->jumpstack);
-        if (sizeof(unsigned int) * nr_cpu_ids > PAGE_SIZE)
-                vfree(info->stackptr);
-        else
-                kfree(info->stackptr);
+
+        free_percpu(info->stackptr);
 
         kfree(info);
 }
@@ -753,14 +751,9 @@ static int xt_jumpstack_alloc(struct xt_table_info *i)
         unsigned int size;
         int cpu;
 
-        size = sizeof(unsigned int) * nr_cpu_ids;
-        if (size > PAGE_SIZE)
-                i->stackptr = vmalloc(size);
-        else
-                i->stackptr = kmalloc(size, GFP_KERNEL);
+        i->stackptr = alloc_percpu(unsigned int);
         if (i->stackptr == NULL)
                 return -ENOMEM;
-        memset(i->stackptr, 0, size);
 
         size = sizeof(void **) * nr_cpu_ids;
         if (size > PAGE_SIZE)