This is a note to let you know that I've just added the patch titled ipv4: avoid quadratic behavior in netns dismantle to the 5.15-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary The filename of the patch is: ipv4-avoid-quadratic-behavior-in-netns-dismantle.patch and it can be found in the queue-5.15 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable@xxxxxxxxxxxxxxx> know about it. >From d07418afea8f1d9896aaf9dc5ae47ac4f45b220c Mon Sep 17 00:00:00 2001 From: Eric Dumazet <edumazet@xxxxxxxxxx> Date: Wed, 19 Jan 2022 02:04:12 -0800 Subject: ipv4: avoid quadratic behavior in netns dismantle 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> @@ -319,11 +320,15 @@ 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, @@ -433,12 +438,11 @@ 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 && @@ -1607,12 +1611,10 @@ 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) } @@ -1964,8 +1966,7 @@ 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) { @@ -1984,12 +1985,11 @@ 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; @@ -2134,7 +2134,6 @@ 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; @@ -2150,8 +2149,7 @@ 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) { Patches currently in stable-queue which might be from edumazet@xxxxxxxxxx are queue-5.15/ipv4-avoid-quadratic-behavior-in-netns-dismantle.patch queue-5.15/af_unix-annote-lockless-accesses-to-unix_tot_inflight-gc_in_progress.patch queue-5.15/ipv4-update-fib_info_cnt-under-spinlock-protection.patch queue-5.15/ppp-ensure-minimum-packet-size-in-ppp_write.patch queue-5.15/xfrm-fix-a-small-bug-in-xfrm_sa_len.patch