tcp: Don't allocate tcp_death_row outside of struct netns_ipv4.
authorKuniyuki Iwashima <kuniyu@amazon.com>
Thu, 8 Sep 2022 01:10:18 +0000 (18:10 -0700)
committerJakub Kicinski <kuba@kernel.org>
Tue, 20 Sep 2022 17:21:49 +0000 (10:21 -0700)
commite9bd0cca09d13ac2f08d25e195203e42d4ad1ce8
treec098713403c2ed7e70d310348766a1acf9d5e374
parent08eaef90403110e51861d93e8008a355af467bbe
tcp: Don't allocate tcp_death_row outside of struct netns_ipv4.

We will soon introduce an optional per-netns ehash and access hash
tables via net->ipv4.tcp_death_row->hashinfo instead of &tcp_hashinfo
in most places.

It could harm the fast path because dereferences of two fields in net
and tcp_death_row might incur two extra cache line misses.  To save one
dereference, let's place tcp_death_row back in netns_ipv4 and fetch
hashinfo via net->ipv4.tcp_death_row"."hashinfo.

Note tcp_death_row was initially placed in netns_ipv4, and commit
fbb8295248e1 ("tcp: allocate tcp_death_row outside of struct netns_ipv4")
changed it to a pointer so that we can fire TIME_WAIT timers after freeing
net.  However, we don't do so after commit 04c494e68a13 ("Revert "tcp/dccp:
get rid of inet_twsk_purge()""), so we need not define tcp_death_row as a
pointer.

Also, we move refcount_dec_and_test(&tw_refcount) from tcp_sk_exit() to
tcp_sk_exit_batch() as a debug check.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
include/net/netns/ipv4.h
net/ipv4/inet_timewait_sock.c
net/ipv4/proc.c
net/ipv4/sysctl_net_ipv4.c
net/ipv4/tcp_ipv4.c
net/ipv4/tcp_minisocks.c
net/ipv6/tcp_ipv6.c