On Fri, 2017-03-03 at 07:22 -0800, Eric Dumazet wrote:
> On Fri, Mar 3, 2017 at 7:12 AM, Dmitry Vyukov <dvyu...@google.com> wrote:
> > The first bot that picked this up started spewing:
> >
> > BUG: spinlock recursion on CPU#1, syz-executor2/9452
> 
> Yes. The bug is not about locking the listener, but protecting fields
> of struct dccp_request_sock
> 
> I will provide a patch, once I reach the office and after the breakfast ;)

OK here is what I suggest to fix the races.

diff --git a/include/linux/dccp.h b/include/linux/dccp.h
index 
61d042bbbf607253033d9948b291cab2322814ba..68449293c4b6233c1a1d4133b1819376a9310225
 100644
--- a/include/linux/dccp.h
+++ b/include/linux/dccp.h
@@ -163,6 +163,7 @@ struct dccp_request_sock {
        __u64                    dreq_isr;
        __u64                    dreq_gsr;
        __be32                   dreq_service;
+       spinlock_t               dreq_lock;
        struct list_head         dreq_featneg;
        __u32                    dreq_timestamp_echo;
        __u32                    dreq_timestamp_time;
diff --git a/net/dccp/minisocks.c b/net/dccp/minisocks.c
index 
e267e6f4c9a5566b369a03a600a408e5bd41cbad..abd07a443219853b022bef41cb072e90ff8f07f0
 100644
--- a/net/dccp/minisocks.c
+++ b/net/dccp/minisocks.c
@@ -142,6 +142,13 @@ struct sock *dccp_check_req(struct sock *sk, struct 
sk_buff *skb,
        struct dccp_request_sock *dreq = dccp_rsk(req);
        bool own_req;
 
+       /* TCP/DCCP listeners became lockless.
+        * DCCP stores complex state in its request_sock, so we need
+        * a protection for them, now this code runs without being protected
+        * by the parent (listener) lock.
+        */
+       spin_lock_bh(&dreq->dreq_lock);
+
        /* Check for retransmitted REQUEST */
        if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) {
 
@@ -156,7 +163,7 @@ struct sock *dccp_check_req(struct sock *sk, struct sk_buff 
*skb,
                        inet_rtx_syn_ack(sk, req);
                }
                /* Network Duplicate, discard packet */
-               return NULL;
+               goto out;
        }
 
        DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
@@ -182,20 +189,20 @@ struct sock *dccp_check_req(struct sock *sk, struct 
sk_buff *skb,
 
        child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
                                                         req, &own_req);
-       if (!child)
-               goto listen_overflow;
-
-       return inet_csk_complete_hashdance(sk, child, req, own_req);
+       if (child) {
+               child = inet_csk_complete_hashdance(sk, child, req, own_req);
+               goto out;
+       }
 
-listen_overflow:
-       dccp_pr_debug("listen_overflow!\n");
        DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 drop:
        if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET)
                req->rsk_ops->send_reset(sk, skb);
 
        inet_csk_reqsk_queue_drop(sk, req);
-       return NULL;
+out:
+       spin_unlock_bh(&dreq->dreq_lock);
+       return child;
 }
 
 EXPORT_SYMBOL_GPL(dccp_check_req);
@@ -246,6 +253,7 @@ int dccp_reqsk_init(struct request_sock *req,
 {
        struct dccp_request_sock *dreq = dccp_rsk(req);
 
+       spin_lock_init(&dreq->dreq_lock);
        inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport;
        inet_rsk(req)->ir_num      = ntohs(dccp_hdr(skb)->dccph_dport);
        inet_rsk(req)->acked       = 0;


Reply via email to