Message ID | 20190401100920.42508-1-edumazet@google.com |
---|---|
State | Accepted |
Delegated to: | David Miller |
Headers | show |
Series | [net] tcp: fix tcp_inet6_sk() for 32bit kernels | expand |
From: Eric Dumazet <edumazet@google.com> Date: Mon, 1 Apr 2019 03:09:20 -0700 > It turns out that struct ipv6_pinfo is not located as we think. > > inet6_sk_generic() and tcp_inet6_sk() disagree on 32bit kernels by 4-bytes, > because struct tcp_sock has 8-bytes alignment, > but ipv6_pinfo size is not a multiple of 8. > > sizeof(struct ipv6_pinfo): 116 (not padded to 8) > > I actually first coded tcp_inet6_sk() as this patch does, but thought > that "container_of(tcp_sk(sk), struct tcp6_sock, tcp)" was cleaner. > > As Julian told me : Nobody should use tcp6_sock.inet6 > directly, it should be accessed via tcp_inet6_sk() or inet6_sk(). > > This happened when we added the first u64 field in struct tcp_sock. > > Fixes: 93a77c11ae79 ("tcp: add tcp_inet6_sk() helper") > Signed-off-by: Eric Dumazet <edumazet@google.com> > Bisected-by: Julian Anastasov <ja@ssi.bg> This looks like a net-next change, because the Fixes tag commit is only there. So that's where I have applied it. Maybe we should change the name of the tcp6_sock.inet6 member if it should never be accessed directly.
On Mon, Apr 1, 2019 at 10:22 AM David Miller <davem@davemloft.net> wrote: > > > This looks like a net-next change, because the Fixes tag commit is only > there. So that's where I have applied it. Oops, wrong script, yes this is a net-next change. > > Maybe we should change the name of the tcp6_sock.inet6 member if it > should never be accessed directly. Or add an __alignof__(u64) if this does not increase memory needs.
Hello, On Mon, 1 Apr 2019, Eric Dumazet wrote: > It turns out that struct ipv6_pinfo is not located as we think. > > inet6_sk_generic() and tcp_inet6_sk() disagree on 32bit kernels by 4-bytes, > because struct tcp_sock has 8-bytes alignment, > but ipv6_pinfo size is not a multiple of 8. > > sizeof(struct ipv6_pinfo): 116 (not padded to 8) > > I actually first coded tcp_inet6_sk() as this patch does, but thought > that "container_of(tcp_sk(sk), struct tcp6_sock, tcp)" was cleaner. > > As Julian told me : Nobody should use tcp6_sock.inet6 > directly, it should be accessed via tcp_inet6_sk() or inet6_sk(). > > This happened when we added the first u64 field in struct tcp_sock. > > Fixes: 93a77c11ae79 ("tcp: add tcp_inet6_sk() helper") > Signed-off-by: Eric Dumazet <edumazet@google.com> > Bisected-by: Julian Anastasov <ja@ssi.bg> > --- > net/ipv6/tcp_ipv6.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c > index eec814fe53b817106bc1d1eaa89dadcb96c974fa..82018bdce863165eba72e1ccf0c12ee558042ae8 100644 > --- a/net/ipv6/tcp_ipv6.c > +++ b/net/ipv6/tcp_ipv6.c > @@ -93,12 +93,13 @@ static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, > /* Helper returning the inet6 address from a given tcp socket. > * It can be used in TCP stack instead of inet6_sk(sk). > * This avoids a dereference and allow compiler optimizations. > + * It is a specialized version of inet6_sk_generic(). > */ > static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk) > { > - struct tcp6_sock *tcp6 = container_of(tcp_sk(sk), struct tcp6_sock, tcp); > + unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo); > > - return &tcp6->inet6; > + return (struct ipv6_pinfo *)(((u8 *)sk) + offset); > } > > static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) Thanks Eric! It is working here. Regards -- Julian Anastasov <ja@ssi.bg>
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c index eec814fe53b817106bc1d1eaa89dadcb96c974fa..82018bdce863165eba72e1ccf0c12ee558042ae8 100644 --- a/net/ipv6/tcp_ipv6.c +++ b/net/ipv6/tcp_ipv6.c @@ -93,12 +93,13 @@ static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, /* Helper returning the inet6 address from a given tcp socket. * It can be used in TCP stack instead of inet6_sk(sk). * This avoids a dereference and allow compiler optimizations. + * It is a specialized version of inet6_sk_generic(). */ static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk) { - struct tcp6_sock *tcp6 = container_of(tcp_sk(sk), struct tcp6_sock, tcp); + unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo); - return &tcp6->inet6; + return (struct ipv6_pinfo *)(((u8 *)sk) + offset); } static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
It turns out that struct ipv6_pinfo is not located as we think. inet6_sk_generic() and tcp_inet6_sk() disagree on 32bit kernels by 4-bytes, because struct tcp_sock has 8-bytes alignment, but ipv6_pinfo size is not a multiple of 8. sizeof(struct ipv6_pinfo): 116 (not padded to 8) I actually first coded tcp_inet6_sk() as this patch does, but thought that "container_of(tcp_sk(sk), struct tcp6_sock, tcp)" was cleaner. As Julian told me : Nobody should use tcp6_sock.inet6 directly, it should be accessed via tcp_inet6_sk() or inet6_sk(). This happened when we added the first u64 field in struct tcp_sock. Fixes: 93a77c11ae79 ("tcp: add tcp_inet6_sk() helper") Signed-off-by: Eric Dumazet <edumazet@google.com> Bisected-by: Julian Anastasov <ja@ssi.bg> --- net/ipv6/tcp_ipv6.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)