From: Eric Dumazet <edumazet@xxxxxxxxxx>
commit d07418afea8f1d9896aaf9dc5ae47ac4f45b220c upstream.
net/ipv4/fib_semantics.c uses an hash table of 256 slots,
keyed by device ifindexes: fib_info_devhash[DEVINDEX_HASHSIZE]
Problem is that with network namespaces, devices tend
to use the same ifindex.
lo device for instance has a fixed ifindex of one,
for all network namespaces.
This means that hosts with thousands of netns spend
a lot of time looking at some hash buckets with thousands
of elements, notably at netns dismantle.
Simply add a per netns perturbation (net_hash_mix())
to spread elements more uniformely.
Also change fib_devindex_hashfn() to use more entropy.
Fixes: aa79e66eee5d ("net: Make ifindex generation per-net namespace")
Signed-off-by: Eric Dumazet <edumazet@xxxxxxxxxx>
Reviewed-by: David Ahern <dsahern@xxxxxxxxxx>
Signed-off-by: Jakub Kicinski <kuba@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
net/ipv4/fib_semantics.c | 36 +++++++++++++++++-------------------
1 file changed, 17 insertions(+), 19 deletions(-)
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -29,6 +29,7 @@
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/netlink.h>
+#include <linux/hash.h>
#include <net/arp.h>
#include <net/ip.h>
@@ -318,11 +319,15 @@ static inline int nh_comp(struct fib_inf
static inline unsigned int fib_devindex_hashfn(unsigned int val)
{
- unsigned int mask = DEVINDEX_HASHSIZE - 1;
+ return hash_32(val, DEVINDEX_HASHBITS);
+}
+
+static struct hlist_head *
+fib_info_devhash_bucket(const struct net_device *dev)
+{
+ u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
- return (val ^
- (val >> DEVINDEX_HASHBITS) ^
- (val >> (DEVINDEX_HASHBITS * 2))) & mask;
+ return &fib_info_devhash[fib_devindex_hashfn(val)];
}
static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
@@ -432,12 +437,11 @@ int ip_fib_check_default(__be32 gw, stru
{
struct hlist_head *head;
struct fib_nh *nh;
- unsigned int hash;
spin_lock(&fib_info_lock);
- hash = fib_devindex_hashfn(dev->ifindex);
- head = &fib_info_devhash[hash];
+ head = fib_info_devhash_bucket(dev);
+
hlist_for_each_entry(nh, head, nh_hash) {
if (nh->fib_nh_dev == dev &&
nh->fib_nh_gw4 == gw &&
@@ -1594,12 +1598,10 @@ link_it:
} else {
change_nexthops(fi) {
struct hlist_head *head;
- unsigned int hash;
if (!nexthop_nh->fib_nh_dev)
continue;
- hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
- head = &fib_info_devhash[hash];
+ head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
hlist_add_head(&nexthop_nh->nh_hash, head);
} endfor_nexthops(fi)
}
@@ -1940,8 +1942,7 @@ void fib_nhc_update_mtu(struct fib_nh_co
void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
{
- unsigned int hash = fib_devindex_hashfn(dev->ifindex);
- struct hlist_head *head = &fib_info_devhash[hash];
+ struct hlist_head *head = fib_info_devhash_bucket(dev);
struct fib_nh *nh;
hlist_for_each_entry(nh, head, nh_hash) {
@@ -1960,12 +1961,11 @@ void fib_sync_mtu(struct net_device *dev
*/
int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
{
- int ret = 0;
- int scope = RT_SCOPE_NOWHERE;
+ struct hlist_head *head = fib_info_devhash_bucket(dev);
struct fib_info *prev_fi = NULL;
- unsigned int hash = fib_devindex_hashfn(dev->ifindex);
- struct hlist_head *head = &fib_info_devhash[hash];
+ int scope = RT_SCOPE_NOWHERE;
struct fib_nh *nh;
+ int ret = 0;
if (force)
scope = -1;
@@ -2110,7 +2110,6 @@ out:
int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
{
struct fib_info *prev_fi;
- unsigned int hash;
struct hlist_head *head;
struct fib_nh *nh;
int ret;
@@ -2126,8 +2125,7 @@ int fib_sync_up(struct net_device *dev,
}
prev_fi = NULL;
- hash = fib_devindex_hashfn(dev->ifindex);
- head = &fib_info_devhash[hash];
+ head = fib_info_devhash_bucket(dev);
ret = 0;
hlist_for_each_entry(nh, head, nh_hash) {
