: Let dccp_epoch return
RFC PATCH 1/1 : Let dccp_epoch return
This is an idea-test, not a full patch. But I have found it to work.
The idea is to start the timer when DCCP is loaded. This gives 2^32-1 usecs = circa 49 days before the clock wraps around.
What I have tested is the RTT sampling for the initial SYN RTT and that worked well with the hack below.
Here is a transcript (after the patch)
$ iperf -c 10.0.0.1 -d ------------------------------------------------------------ Client connecting to 10.0.0.1, DCCP port 5001 NOTE: running in bytestream-mode (maximum speed) DCCP datagram buffer size: 104 KByte (default) ------------------------------------------------------------ [ 3] local 10.0.0.176 port 53425 connected with 10.0.0.1 port 5001 Aug 21 15:04:04 gerrit kernel: delta = 171 <Aug 21 15:04:04 gerrit kernel: SYN RTT = 171us <= [ 3] 0.0-10.0 sec 112 MBytes 93.9 Mbits/se
Will try to clean this up for the other cases - but is the idea ok?
--- net/dccp/dccp.h | 1 + net/dccp/input.c | 4 ++-- net/dccp/options.c | 2 +- net/dccp/proto.c | 7 +++++++ 4 files changed, 11 insertions(+), 3 deletions(-)
net/dccp/dccp.h +1 -0
@@ -413,6 +413,7 @@ extern int dccp_insert_options(struct soextern int dccp_insert_option_elapsed_time (struct sock *sk,struct sk_buff *skb,u32 elapsed_time); +extern u32 dccp_timestamp (void );extern int dccp_insert_option_timestamp (struct sock *sk,struct sk_buff *skb); extern int dccp_insert_option(struct sock *sk, struct sk_buff *skb,
net/dccp/options.c +1 -1
@@ -365,7 +365,7 @@ EXPORT_SYMBOL_GPL(dccp_insert_option_elaint dccp_insert_option_timestamp (struct sock *sk,struct sk_buff *skb) { -__be32 now =htonl (((suseconds_t )ktime_to_us (ktime_get_real ())) /10 ); +__be32 now =htonl (dccp_timestamp () /10 );/* yes this will overflow but that is the point as we want a * 10 usec 32 bit timer which mean it wraps every 11.9 hours */
net/dccp/proto.c +7 -0
@@ -52,6 +52,12 @@ struct inet_hashinfo __cacheline_alignedEXPORT_SYMBOL_GPL (dccp_hashinfo); +static ktime_t dccp_timestamp_seed; +u32 dccp_timestamp (void ) +{ +return ktime_us_delta (ktime_get_real (), dccp_timestamp_seed); +} +/* Forward Declarations. */ static void dccp_handle_passive_close (struct sock *sk);
@@ -1134,6 +1140,7 @@ static int __init dccp_init(void) rc =dccp_sysctl_init ();if (rc) goto out_ackvec_exit; + dccp_timestamp_seed =ktime_get_real (); out:return rc; out_ackvec_exit:
net/dccp/input.c +2 -2
@@ -648,8 +648,8 @@ u32 dccp_sample_rtt(struct sock *sk, ktiDCCP_WARN ("packet without timestamp echo\n" );return DCCP_SANE_RTT_MAX ; } -ktime_sub_us (t_recv, or->dccpor_timestamp_echo *10 ); - delta =ktime_to_us (t_recv); + delta =dccp_timestamp () - or->dccpor_timestamp_echo *10 ; +printk ("delta = %lld\n" , (long long )delta); }else delta =ktime_us_delta (t_recv, *t_hist);
| The idea is to start the timer when DCCP is loaded. This gives | 2^32-1 usecs = circa 49 days before the clock wraps around.
Oops, it is not milliseconds but microseconds ... that means the counter wraps around every 71 seconds. If the maximum RTT is 3 seconds, then this means wrap-around every 20 RTTs.
Maybe this is too much?