Skip to contentSkip to conversation

: calling dccp_v{4,6}_reqsk_send_ack is a BUG

PATCH In review Arnaldo Carvalho de Melo 1 participant 1 message Nov 10, 2006 Networking

lore
Share & tools
lore
#0 lore raw

PATCH 6/10 : calling dccp_v{4,6}_reqsk_send_ack is a BUG

This patch removes two functions, the send_ack functions of request_sock, which are not called/used by the DCCP code. It is correct that these functions are not called, below is a justification why calling these functions (on a passive socket in the LISTEN/RESPOND state) would mean a DCCP protocol violation.

A) Background: using request_sock in TCP:
-----------------------------------------

	req->rsk_ops->send_ack is called only once, in tcp_check_req() of
        net/ipv4/tcp_minisocks.c This single case deals with a socket in the
        SYN-RECEIVED state; RFC 793 on page 69 describes the case for
        SYN-RECEIVED and other states: "If an incoming segment is not
        acceptable, an acknowledgment should be sent in reply (unless the
        RST bit is set, if so drop the segment and return):

                       <SEQ=SND.NXT><ACK=RCV.NXT><CTL¬K>"

        (SND.NXT = ISS + 1, RCV.NEXT = SEG.SEQ + 1 if SYN-RECEIVED was
         entered from LISTEN)

       This is referenced by the following code segment:

        if (paws_reject || !tcp_in_window(...)) {
                /* Out of window: send ACK and drop. */
                if (!(flg & TCP_FLAG_RST))
                        req->rsk_ops->send_ack(skb, req);
                /* ... */
                return NULL;
        }

        This is the _only_ occurrence where rsk_ops->send_ack() is called
        for TCP.

B) The Problem: protocol constraints of DCCP:
---------------------------------------------
(1) rsk_ops->send_ack is called _nowhere_ throughout the DCCP code (grep confirms) (2) This is correct, due to the following:
(a) RFC 4340, sec. 8.5, step 3: The only acceptable packets in LISTEN state
are DCCP-Request or packets with a valid InitCookie option (InitCookie is an optional, not mandatory, protocol feature). Only if such a packet arrives, RESPOND is entered after sending a DCCP-Response. Then, in step 11, receiving any packet other than DCCP-Request, will trigger the transition to OPEN (`Established') => the request_sock will become a full socket here.
(b) DCCP uses the Sync where TCP would use an ACK - in the case of
sequence-invalid packets (RFC 4340, 7.5.4). This however is already handled by calling dccp_send_sync in dccp_check_seqno(), __dcp_rcv_established(), dccp_rcv_state_process(), dccp_rcv_closereq().

c) Conclusion:
--------------
In DCCP the function pointer send_ack of struct request_sock_ops is unnecessary, and calling it would mean a protocol violation. The DCCP code is correct in this regard in that it does not call the send_ack virtual function pointer, but the existence of two functions - which are never called and are not necessary - is highly confusing - see e.g. the posting on
http://www.mail-archive.com/[email protected]/msg00419.html

Therefore, the attached patch replaces the functions with a verbose message (if debugging is enabled), and otherwise flags a BUG(). This solution preserves the generic structure of the request_sock while making sure that RFC 4340 is not violated by the implementation.

Commiter note: I've demoted the BUG() all to the equivalente of a WARN_ON, i.e. print a message and dump the stack.

------------------------------------------------------------------------------

 dccp.h      |    2 ++
 ipv4.c      |   48 +-----------------------------------------------
 ipv6.c      |   57 +--------------------------------------------------------
 minisocks.c |   10 ++++++++++
 4 files changed, 14 insertions(+), 103 deletions(-)

------------------------------------------------------------------------------

net/dccp/dccp.h +2 -0
index 8964b18..3d4b4a9 100644
@@ -131,6 +131,8 @@ extern int  dccp_retransmit_skb(struct s
 
 extern void dccp_send_ack(struct sock *sk);
 extern void dccp_send_delayed_ack(struct sock *sk);
+extern void dccp_reqsk_send_ack(struct sk_buff *sk, struct request_sock *rsk);
+
 extern void dccp_send_sync(struct sock *sk, const u64 seq,
 			   const enum dccp_pkt_type pkt_type);
 
net/dccp/ipv4.c +1 -47

60 lines not on this page · raw message

net/dccp/ipv6.c +1 -56

75 lines not on this page · raw message

net/dccp/minisocks.c +10 -0
index 9045438..5f3e1a4 100644
@@ -11,6 +11,7 @@
  */
 
 #include <linux/dccp.h>
+#include <linux/kernel.h>
 #include <linux/skbuff.h>
 #include <linux/timer.h>
 
@@ -283,3 +284,12 @@ int dccp_child_process(struct sock *pare
 }
 
 EXPORT_SYMBOL_GPL(dccp_child_process);
+
+void dccp_reqsk_send_ack(struct sk_buff *skb, struct request_sock *rsk)
+{
+	pr_info(KERN_WARNING "DCCP: ACK packets are never sent in "
+			     "LISTEN/RESPOND state\n");
+	dump_stack();
+}
+
+EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack);