leandrof/linux
1
0
Fork 0
forked from Minki/linux
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ipv4: avoid quadratic behavior in netns dismantle

Browse Source 
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: aa79e66eee ("net: Make ifindex generation per-net namespace")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit is contained in:
Eric Dumazet 2022-01-19 02:04:12 -08:00 committed by Jakub Kicinski
parent 8eb896a777
commit d07418afea
1 changed files with 17 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
net/ipv4/fib_semantics.c
View File

@ -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>
@ -319,11 +320,15 @@ static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
static inline unsigned int fib_devindex_hashfn(unsigned int val)
{
unsigned int mask = DEVINDEX_HASHSIZE - 1;
return hash_32(val, DEVINDEX_HASHBITS);
}
return (val ^
(val >> DEVINDEX_HASHBITS) ^
(val >> (DEVINDEX_HASHBITS * 2))) & mask;
static struct hlist_head *
fib_info_devhash_bucket(const struct net_device *dev)
{
u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
return &fib_info_devhash[fib_devindex_hashfn(val)];
}
static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
@ -433,12 +438,11 @@ int ip_fib_check_default(__be32 gw, struct net_device *dev)
{
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 &&
@ -1609,12 +1613,10 @@ 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)
}
@ -1966,8 +1968,7 @@ void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
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) {
@ -1986,12 +1987,11 @@ void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
*/
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;
@ -2136,7 +2136,6 @@ 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;
@ -2152,8 +2151,7 @@ int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
}
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) {