We used to only use sk_rmem_schedule when we potentially overflow
the sk_rcvbuf. That allowed us to use sk_rmem[1] space on top of
existing sk_recvbuf space which is way more that needed and thus
incorrectly detecting memory pressure conditions.
We need to call sk_rmem_schedule at all times (whenever sk_rcvbuf is
not locked), letting it correctly catch memory pressure conditions.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@xxxxxx>
---
net/sctp/ulpevent.c | 12 +++++++-----
1 files changed, 7 insertions(+), 5 deletions(-)
diff --git a/net/sctp/ulpevent.c b/net/sctp/ulpevent.c
index a1f654a..71a97c7 100644
--- a/net/sctp/ulpevent.c
+++ b/net/sctp/ulpevent.c
@@ -687,6 +687,7 @@ struct sctp_ulpevent *sctp_ulpevent_make_rcvmsg(struct sctp_association *asoc,
struct sk_buff *skb;
size_t padding, len;
int rx_count;
+ int pressure = 0;
/*
* check to see if we need to make space for this
@@ -698,12 +699,13 @@ struct sctp_ulpevent *sctp_ulpevent_make_rcvmsg(struct sctp_association *asoc,
else
rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
- if (rx_count >= asoc->base.sk->sk_rcvbuf) {
+ if (asoc->base.sk->sk_userlocks & SOCK_RCVBUF_LOCK)
+ pressure = sk_rmem_schedule(asoc->base.sk,
+ chunk->skb->truesize);
+
+ if (rx_count >= asoc->base.sk->sk_rcvbuf || pressure)
+ goto fail;
- if ((asoc->base.sk->sk_userlocks & SOCK_RCVBUF_LOCK) ||
- (!sk_rmem_schedule(asoc->base.sk, chunk->skb->truesize)))
- goto fail;
- }
/* Clone the original skb, sharing the data. */
skb = skb_clone(chunk->skb, gfp);
--
1.5.2.5
--
To unsubscribe from this list: send the line "unsubscribe linux-sctp" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
