[PATCH] net: sctp: Fix IPv6 ancestor_size calc in sctp_copy_descendant

When calculating ancestor_size with IPv6 enabled, simply using
sizeof(struct ipv6_pinfo) doesn't account for extra bytes needed for
alignment in the struct sctp6_sock. On x86, there aren't any extra
bytes, but on ARM the ipv6_pinfo structure is aligned on an 8-byte
boundary so there were 4 pad bytes that were omitted from the
ancestor_size calculation.  This would lead to corruption of the
pd_lobby pointers, causing an oops when trying to free the sctp
structure on socket close.

Signed-off-by: Henry Ptasinski <redacted>
---
 net/sctp/socket.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 836615f71a7d..a6358c81f087 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c

@@ -9220,12 +9220,14 @@ void sctp_copy_sock(struct sock *newsk, struct sock *sk,
 static inline void sctp_copy_descendant(struct sock *sk_to,
 					const struct sock *sk_from)
 {
-	int ancestor_size = sizeof(struct inet_sock) +
-			    sizeof(struct sctp_sock) -
-			    offsetof(struct sctp_sock, pd_lobby);
+	size_t ancestor_size = sizeof(struct inet_sock);
 
 	if (sk_from->sk_family == PF_INET6)
-		ancestor_size += sizeof(struct ipv6_pinfo);
+		ancestor_size += sizeof(struct sctp6_sock);
+	else
+		ancestor_size += sizeof(struct sctp_sock);
+
+	ancestor_size -= offsetof(struct sctp_sock, pd_lobby);
 
 	__inet_sk_copy_descendant(sk_to, sk_from, ancestor_size);
 }

-- 
2.28.0.681.g6f77f65b4e-goog

On Fri, Sep 18, 2020 at 01:56:10AM +0000, Henry Ptasinski wrote:

When calculating ancestor_size with IPv6 enabled, simply using
sizeof(struct ipv6_pinfo) doesn't account for extra bytes needed for
alignment in the struct sctp6_sock. On x86, there aren't any extra
bytes, but on ARM the ipv6_pinfo structure is aligned on an 8-byte
boundary so there were 4 pad bytes that were omitted from the
ancestor_size calculation.  This would lead to corruption of the
pd_lobby pointers, causing an oops when trying to free the sctp
structure on socket close.

Makes sense.

Signed-off-by: Henry Ptasinski <redacted>

Please add a:
Fixes: 636d25d557d1 ("sctp: not copy sctp_sock pd_lobby in sctp_copy_descendant")

          ^^^^^^  I'll send a patch to fix/remove this.

 					const struct sock *sk_from)
 {
-	int ancestor_size = sizeof(struct inet_sock) +
-			    sizeof(struct sctp_sock) -
-			    offsetof(struct sctp_sock, pd_lobby);

                                                       ^^^^^^^^

Then, as this patch is actually fixing the aforementioned commit,
please also update the comment on sctp_sock definition, as pd_lobby
now is also skipped.

+	size_t ancestor_size = sizeof(struct inet_sock);
 
 	if (sk_from->sk_family == PF_INET6)
-		ancestor_size += sizeof(struct ipv6_pinfo);
+		ancestor_size += sizeof(struct sctp6_sock);

As you probably noticed by the build bot email already, there need to
be some protection to building without IPv6 enabled.

To avoid ifdefs here, something similar to how
inet_sk_copy_descendant() is done is probably welcomed, but please
feel free to be creative. :-)

+	else
+		ancestor_size += sizeof(struct sctp_sock);
+
+	ancestor_size -= offsetof(struct sctp_sock, pd_lobby);
 
 	__inet_sk_copy_descendant(sk_to, sk_from, ancestor_size);
 }
-- 
2.28.0.681.g6f77f65b4e-goog

When calculating ancestor_size with IPv6 enabled, simply using
sizeof(struct ipv6_pinfo) doesn't account for extra bytes needed for
alignment in the struct sctp6_sock. On x86, there aren't any extra
bytes, but on ARM the ipv6_pinfo structure is aligned on an 8-byte
boundary so there were 4 pad bytes that were omitted from the
ancestor_size calculation.  This would lead to corruption of the
pd_lobby pointers, causing an oops when trying to free the sctp
structure on socket close.

Fixes: 636d25d557d1 ("sctp: not copy sctp_sock pd_lobby in sctp_copy_descendant")
Signed-off-by: Henry Ptasinski <redacted>
---
 include/net/sctp/structs.h | 8 +++++---
 net/sctp/socket.c          | 9 +++------
 2 files changed, 8 insertions(+), 9 deletions(-)

diff --git a/include/net/sctp/structs.h b/include/net/sctp/structs.h
index b33f1aefad09..0bdff38eb4bb 100644
--- a/include/net/sctp/structs.h
+++ b/include/net/sctp/structs.h

@@ -226,12 +226,14 @@ struct sctp_sock {
 		data_ready_signalled:1;
 
 	atomic_t pd_mode;
+
+	/* Fields after this point will be skipped on copies, like on accept
+	 * and peeloff operations
+	 */
+
 	/* Receive to here while partial delivery is in effect. */
 	struct sk_buff_head pd_lobby;
 
-	/* These must be the last fields, as they will skipped on copies,
-	 * like on accept and peeloff operations
-	 */
 	struct list_head auto_asconf_list;
 	int do_auto_asconf;
 };

diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 836615f71a7d..53d0a4161df3 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c

@@ -9220,13 +9220,10 @@ void sctp_copy_sock(struct sock *newsk, struct sock *sk,
 static inline void sctp_copy_descendant(struct sock *sk_to,
 					const struct sock *sk_from)
 {
-	int ancestor_size = sizeof(struct inet_sock) +
-			    sizeof(struct sctp_sock) -
-			    offsetof(struct sctp_sock, pd_lobby);
-
-	if (sk_from->sk_family == PF_INET6)
-		ancestor_size += sizeof(struct ipv6_pinfo);
+	size_t ancestor_size = sizeof(struct inet_sock);
 
+	ancestor_size += sk_from->sk_prot->obj_size;
+	ancestor_size -= offsetof(struct sctp_sock, pd_lobby);
 	__inet_sk_copy_descendant(sk_to, sk_from, ancestor_size);
 }
 

-- 
2.28.0.681.g6f77f65b4e-goog

On Sat, Sep 19, 2020 at 12:12:11AM +0000, Henry Ptasinski wrote:

When calculating ancestor_size with IPv6 enabled, simply using
sizeof(struct ipv6_pinfo) doesn't account for extra bytes needed for
alignment in the struct sctp6_sock. On x86, there aren't any extra
bytes, but on ARM the ipv6_pinfo structure is aligned on an 8-byte
boundary so there were 4 pad bytes that were omitted from the
ancestor_size calculation.  This would lead to corruption of the
pd_lobby pointers, causing an oops when trying to free the sctp
structure on socket close.

Fixes: 636d25d557d1 ("sctp: not copy sctp_sock pd_lobby in sctp_copy_descendant")
Signed-off-by: Henry Ptasinski <redacted>

Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>

...

 {
-	int ancestor_size = sizeof(struct inet_sock) +
-			    sizeof(struct sctp_sock) -
-			    offsetof(struct sctp_sock, pd_lobby);
-
-	if (sk_from->sk_family == PF_INET6)
-		ancestor_size += sizeof(struct ipv6_pinfo);
+	size_t ancestor_size = sizeof(struct inet_sock);
 
+	ancestor_size += sk_from->sk_prot->obj_size;

Heh, of course. Nice one.

+	ancestor_size -= offsetof(struct sctp_sock, pd_lobby);
 	__inet_sk_copy_descendant(sk_to, sk_from, ancestor_size);
 }
 
-- 
2.28.0.681.g6f77f65b4e-goog

From: Henry Ptasinski <redacted>
Date: Sat, 19 Sep 2020 00:12:11 +0000

When calculating ancestor_size with IPv6 enabled, simply using
sizeof(struct ipv6_pinfo) doesn't account for extra bytes needed for
alignment in the struct sctp6_sock. On x86, there aren't any extra
bytes, but on ARM the ipv6_pinfo structure is aligned on an 8-byte
boundary so there were 4 pad bytes that were omitted from the
ancestor_size calculation.  This would lead to corruption of the
pd_lobby pointers, causing an oops when trying to free the sctp
structure on socket close.

Fixes: 636d25d557d1 ("sctp: not copy sctp_sock pd_lobby in sctp_copy_descendant")
Signed-off-by: Henry Ptasinski <redacted>

Applied and queued up for -stable, thank you.