net: sctp: use sk_copy_sanitize for accept sockets
Subsystems (2)
RFC PATCH 3/3 net: sctp: use sk_copy_sanitize for accept sockets
Wang reported an issue that lksctp's test_getname_v6 seems to fail.
The issue is that we do not copy sk_v6_rcv_saddr over to the new socket, although the comment above says so regarding rcv_saddr.
Commit 914e1c8b6980 ("sctp: Inherit all socket options from parent correctly.") originally moved that over to sctp_copy_sock(), but after commit efe4208f47f9 ("ipv6: make lookups simpler and faster") this no longer holds and the actual value of sk_v6_rcv_saddr was no longer being migrated.
The issue seems to be deeper than that though. We always used to do a partial copy of socket members in accept/peeloff case. This means that also newly added members to the kernel's socket representation weren't inherited to spawned sockets, for example, Eric points out that settings of SO_MAX_PACING_RATE would be broken on SCTP as well.
So lets get rid of this restriction by using sk_copy_sanitize(). With this patch, the lksctp test suite also passes again for IPv6.
--- include/net/sctp/sctp.h | 4 +- net/sctp/ipv6.c | 11 +++--- net/sctp/protocol.c | 13 +++--- net/sctp/socket.c | 102 +++++++++++++++++++++++++++++------------------- 4 files changed, 75 insertions(+), 55 deletions(-)
include/net/sctp/sctp.h +2 -2
@@ -105,8 +105,8 @@ void sctp_data_ready(struct sock *sk, int len);unsigned int sctp_poll (struct file *file,struct socket *sock,poll_table *wait);void sctp_sock_rfree (struct sk_buff *skb); -void sctp_copy_sock (struct sock *newsk,struct sock *sk, -struct sctp_association *asoc); +struct sock *sctp_clone_lock (const struct sock *sk,const gfp_t priority, +struct sctp_association *asoc);extern struct percpu_counter sctp_sockets_allocated;int sctp_asconf_mgmt (struct sctp_sock *,struct sctp_sockaddr_entry *);
net/sctp/ipv6.c +5 -6
@@ -639,13 +639,10 @@ static struct sock *sctp_v6_create_accept_sk(struct sock *sk,struct ipv6_pinfo *newnp, *np =inet6_sk (sk);struct sctp6_sock *newsctp6sk; - newsk =sk_alloc (sock_net (sk),PF_INET6 ,GFP_KERNEL , sk->sk_prot); + newsk =sctp_clone_lock (sk,GFP_KERNEL , asoc);if (!newsk) goto out; -sock_init_data (NULL , newsk); - -sctp_copy_sock (newsk, sk, asoc);sock_reset_flag (sk,SOCK_ZAPPED ); newsctp6sk = (struct sctp6_sock *)newsk;
@@ -654,7 +651,6 @@ static struct sock *sctp_v6_create_accept_sk(struct sock *sk,sctp_sk (newsk)->v4mapped =sctp_sk (sk)->v4mapped; newnp =inet6_sk (newsk); -memcpy (newnp, np,sizeof (struct ipv6_pinfo ));/* Initialize sk's sport, dport, rcv_saddr and daddr for getsockname()
@@ -668,8 +664,11 @@ static struct sock *sctp_v6_create_accept_sk(struct sock *sk,sk_common_release (newsk); newsk =NULL ; } - out: +if (newsk) { +bh_unlock_sock (newsk); +sock_put (newsk); + }return newsk; }
net/sctp/protocol.c +6 -7
@@ -544,20 +544,16 @@ static int sctp_v4_is_ce(const struct sk_buff *skb)static struct sock *sctp_v4_create_accept_sk (struct sock *sk,struct sctp_association *asoc) { -struct sock *newsk =sk_alloc (sock_net (sk),PF_INET ,GFP_KERNEL , - sk->sk_prot); +struct sock *newsk;struct inet_sock *newinet; + newsk =sctp_clone_lock (sk,GFP_KERNEL , asoc);if (!newsk) goto out; -sock_init_data (NULL , newsk); - -sctp_copy_sock (newsk, sk, asoc);sock_reset_flag (newsk,SOCK_ZAPPED ); newinet =inet_sk (newsk); - newinet->inet_daddr = asoc->peer.primary_addr.v4.sin_addr.s_addr;sk_refcnt_debug_inc (newsk);
@@ -566,8 +562,11 @@ static struct sock *sctp_v4_create_accept_sk(struct sock *sk,sk_common_release (newsk); newsk =NULL ; } - out: +if (newsk) { +bh_unlock_sock (newsk); +sock_put (newsk); + }return newsk; }
net/sctp/socket.c +62 -40
132 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Quoted 18 lines · Daniel Borkmann
Wang reported an issue that lksctp's
test_getname_v6seems to fail.The issue is that we do not copy
sk_v6_rcv_saddrover to the new socket, although the comment above says so regardingrcv_saddr.Commit
914e1c8b6980("sctp: Inherit all socket options from parent correctly.") originally moved that over tosctp_copy_sock(), but after commitefe4208f47f9("ipv6: make lookups simpler and faster") this no longer holds and the actual value ofsk_v6_rcv_saddrwas no longer being migrated.The issue seems to be deeper than that though. We always used to do a partial copy of socket members in accept/peeloff case. This means that also newly added members to the kernel's socket representation weren't inherited to spawned sockets, for example, Eric points out that settings of
SO_MAX_PACING_RATEwould be broken on SCTP as well.
Quoted 8 lines
So lets get rid of this restriction by using
sk_copy_sanitize(). With this patch, the lksctp test suite also passes again for IPv6.Fixes efe4208f47f9 ("ipv6: make lookups simpler and faster")Reported-by Wang Weidong <[email protected]>Suggested-by Eric Dumazet <[email protected]>Not-yet-signed-off-by Daniel Borkmann <[email protected]>---
One more thing for the non-RFC later on:
net/sctp/ipv6.c static struct sock *sctp_v6_create_accept_sk(struct sock *sk, · line 670
Quoted 4 lines
out:
net/sctp/ipv6.c +3
-0+if (newsk) { +bh_unlock_sock (newsk); +sock_put (newsk);
This also needs to be done in error path.
Quoted 3 lines
net/sctp/ipv6.c +1
-0+ }return newsk; }