Skip to contentSkip to conversation

sctp: drop a backlogged chunk if its transport was removed

PATCH Superseded Hyunwoo Kim 2 participants 4 messages Aug 14 – Aug 19

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sctp: drop a backlogged chunk if its transport was removed

sctp_rcv() resolves the transport once per packet and leaves it in chunk->transport. If the socket is owned by userspace the packet goes to the socket backlog, and sctp_add_backlog() takes a reference on that transport.

An authenticated ASCONF DEL-IP in an earlier backlogged packet can remove it. sctp_assoc_rm_peer() takes the transport out of the association and calls sctp_transport_free(), which tags it dead and drops the reference the association held. The backlogged packet still holds a reference, so the transport stays around.

The DATA chunk in that packet puts the removed transport back into asoc->peer.last_data_from. Once the packet is done that reference goes away and the transport is freed by RCU, so the next delayed SACK carries the pointer into the SACK chunk and sctp_outq_select_transport() reads the freed transport's state.

Drop the chunk in sctp_backlog_rcv(), next to the existing rcvr->dead check. The peer retransmits it. Guarding the last_data_from assignment is not enough, sctp_assoc_rm_peer() clears more than that one pointer and letting the packet run fills them in again. sctp_wait_for_sndbuf() already uses the dead flag this way on the send side.

Fixes df132eff4638 ("sctp: clear the transport of some out_chunk_list chunks in sctp_assoc_rm_peer")
---
 net/sctp/input.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
net/sctp/input.c +3 -2
index 864741fae4187e..83b8361d149251 100644
@@ -285,9 +285,10 @@ int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 
 	/* If the rcvr is dead then the association or endpoint
 	 * has been deleted and we can safely drop the chunk
-	 * and refs that we are holding.
+	 * and refs that we are holding.  Same if the transport
+	 * we looked up has been removed in the meantime.
 	 */
-	if (rcvr->dead) {
+	if (rcvr->dead || (t && t->dead)) {
 		sctp_chunk_free(chunk);
 		goto done;
 	}
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 49 lines · Hyunwoo Kim

sctp_rcv() resolves the transport once per packet and leaves it in chunk->transport. If the socket is owned by userspace the packet goes to the socket backlog, and sctp_add_backlog() takes a reference on that transport.

An authenticated ASCONF DEL-IP in an earlier backlogged packet can remove it. sctp_assoc_rm_peer() takes the transport out of the association and calls sctp_transport_free(), which tags it dead and drops the reference the association held. The backlogged packet still holds a reference, so the transport stays around.

The DATA chunk in that packet puts the removed transport back into asoc->peer.last_data_from. Once the packet is done that reference goes away and the transport is freed by RCU, so the next delayed SACK carries the pointer into the SACK chunk and sctp_outq_select_transport() reads the freed transport's state.

Drop the chunk in sctp_backlog_rcv(), next to the existing rcvr->dead check. The peer retransmits it. Guarding the last_data_from assignment is not enough, sctp_assoc_rm_peer() clears more than that one pointer and letting the packet run fills them in again. sctp_wait_for_sndbuf() already uses the dead flag this way on the send side.

Fixes df132eff4638 ("sctp: clear the transport of some out_chunk_list chunks in sctp_assoc_rm_peer")
---
 net/sctp/input.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
net/sctp/input.c +3 -3
index 864741fae4187e..83b8361d149251 100644
@@ -285,9 +285,10 @@ int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 
        /* If the rcvr is dead then the association or endpoint
         * has been deleted and we can safely drop the chunk
-        * and refs that we are holding.
+        * and refs that we are holding.  Same if the transport
+        * we looked up has been removed in the meantime.
         */
-       if (rcvr->dead) {
+       if (rcvr->dead || (t && t->dead)) {
                sctp_chunk_free(chunk);
                goto done;
        }
--
Quoted 2 lines

2.43.0

Can you try to move the check to sctp_inq_push() to also cover the case Sashik reported?

https://netdev-ai.bots.linux.dev/sashiko/#/patchset/an-oGfEatacPTSX-%40v4bel

Thanks.

maintainer
#2 lore raw
Quoted 20 lines · Xin Long
Quoted 18 lines · Hyunwoo Kim

sctp_rcv() resolves the transport once per packet and leaves it in chunk->transport. If the socket is owned by userspace the packet goes to the socket backlog, and sctp_add_backlog() takes a reference on that transport.

An authenticated ASCONF DEL-IP in an earlier backlogged packet can remove it. sctp_assoc_rm_peer() takes the transport out of the association and calls sctp_transport_free(), which tags it dead and drops the reference the association held. The backlogged packet still holds a reference, so the transport stays around.

The DATA chunk in that packet puts the removed transport back into asoc->peer.last_data_from. Once the packet is done that reference goes away and the transport is freed by RCU, so the next delayed SACK carries the pointer into the SACK chunk and sctp_outq_select_transport() reads the freed transport's state.

Quoted 7 lines
Quoted 7 lines

Drop the chunk in sctp_backlog_rcv(), next to the existing rcvr->dead check. The peer retransmits it. Guarding the last_data_from assignment is not enough, sctp_assoc_rm_peer() clears more than that one pointer and letting the packet run fills them in again. sctp_wait_for_sndbuf() already uses the dead flag this way on the send side.

Fixes df132eff4638 ("sctp: clear the transport of some out_chunk_list chunks in sctp_assoc_rm_peer")

Please also double check the 'Fixes' tag, commit df132eff4638 doesn't seem to cause the issue.

Thanks.

Quoted 32 lines
Quoted 24 lines
---
 net/sctp/input.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
net/sctp/input.c +3 -3
index 864741fae4187e..83b8361d149251 100644
@@ -285,9 +285,10 @@ int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 
        /* If the rcvr is dead then the association or endpoint
         * has been deleted and we can safely drop the chunk
-        * and refs that we are holding.
+        * and refs that we are holding.  Same if the transport
+        * we looked up has been removed in the meantime.
         */
-       if (rcvr->dead) {
+       if (rcvr->dead || (t && t->dead)) {
                sctp_chunk_free(chunk);
                goto done;
        }
--
Quoted 2 lines

2.43.0

Can you try to move the check to sctp_inq_push() to also cover the case Sashik reported?

https://netdev-ai.bots.linux.dev/sashiko/#/patchset/an-oGfEatacPTSX-%40v4bel

Thanks.

#3 lore raw
Quoted 27 lines · Xin Long
Quoted 26 lines · Xin Long
Quoted 24 lines · Hyunwoo Kim

sctp_rcv() resolves the transport once per packet and leaves it in chunk->transport. If the socket is owned by userspace the packet goes to the socket backlog, and sctp_add_backlog() takes a reference on that transport.

An authenticated ASCONF DEL-IP in an earlier backlogged packet can remove it. sctp_assoc_rm_peer() takes the transport out of the association and calls sctp_transport_free(), which tags it dead and drops the reference the association held. The backlogged packet still holds a reference, so the transport stays around.

The DATA chunk in that packet puts the removed transport back into asoc->peer.last_data_from. Once the packet is done that reference goes away and the transport is freed by RCU, so the next delayed SACK carries the pointer into the SACK chunk and sctp_outq_select_transport() reads the freed transport's state.

Drop the chunk in sctp_backlog_rcv(), next to the existing rcvr->dead check. The peer retransmits it. Guarding the last_data_from assignment is not enough, sctp_assoc_rm_peer() clears more than that one pointer and letting the packet run fills them in again. sctp_wait_for_sndbuf() already uses the dead flag this way on the send side.

Quoted 3 lines
Quoted 1 line
Quoted 1 line
Fixes df132eff4638 ("sctp: clear the transport of some out_chunk_list chunks in sctp_assoc_rm_peer")

Please also double check the 'Fixes' tag, commit df132eff4638 doesn't seem to cause the issue.

Understood. I'll run a few more tests and then submit v2.

Best regards,
Hyunwoo Kim