On 21/2/2023 20:46, Eric Dumazet wrote:
On Tue, Feb 21, 2023 at 10:22 AM Hangyu Hua <hbh25y@xxxxxxxxx> wrote:
A useless record can be insert into av_records when dccp_insert_options()
fails after dccp_insert_option_ackvec(). Repeated triggering may cause
av_records to have a lot of useless record with the same avr_ack_seqno.
Fixes: 8b7b6c75c638 ("dccp: Integrate feature-negotiation insertion code")
Signed-off-by: Hangyu Hua <hbh25y@xxxxxxxxx>
---
net/dccp/options.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/net/dccp/options.c b/net/dccp/options.c
index d24cad05001e..8aa4abeb15ea 100644
--- a/net/dccp/options.c
+++ b/net/dccp/options.c
@@ -549,6 +549,8 @@ static void dccp_insert_option_padding(struct sk_buff *skb)
int dccp_insert_options(struct sock *sk, struct sk_buff *skb)
{
struct dccp_sock *dp = dccp_sk(sk);
+ struct dccp_ackvec *av = dp->dccps_hc_rx_ackvec;
+ struct dccp_ackvec_record *avr;
DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
@@ -577,16 +579,22 @@ int dccp_insert_options(struct sock *sk, struct sk_buff *skb)
if (dp->dccps_hc_rx_insert_options) {
if (ccid_hc_rx_insert_options(dp->dccps_hc_rx_ccid, sk, skb))
- return -1;
+ goto delete_ackvec;
dp->dccps_hc_rx_insert_options = 0;
}
if (dp->dccps_timestamp_echo != 0 &&
dccp_insert_option_timestamp_echo(dp, NULL, skb))
- return -1;
+ goto delete_ackvec;
dccp_insert_option_padding(skb);
return 0;
+
+delete_ackvec:
+ avr = dccp_ackvec_lookup(&av->av_records, DCCP_SKB_CB(skb)->dccpd_seq);
Why avr would be not NULL ?
+ list_del(&avr->avr_node);
+ kmem_cache_free(dccp_ackvec_record_slab, avr);
+ return -1;
}
Are you really using DCCP and/or how have you tested this patch ?
net/dccp/ackvec.c:15:static struct kmem_cache *dccp_ackvec_record_slab;
I doubt this patch has been compiled.
I would rather mark DCCP deprecated and stop trying to fix it.
My bad. I will fix these problems.
Thanks,
Hangyu
