Commit 916f6efa authored by Florian Westphal's avatar Florian Westphal Committed by Pablo Neira Ayuso

netfilter: never get/set skb->tstamp

setting net.netfilter.nf_conntrack_timestamp=1 breaks xmit with fq
scheduler.  skb->tstamp might be "refreshed" using ktime_get_real(),
but fq expects CLOCK_MONOTONIC.

This patch removes all places in netfilter that check/set skb->tstamp:

1. To fix the bogus "start" time seen with conntrack timestamping for
   outgoing packets, never use skb->tstamp and always use current time.
2. In nfqueue and nflog, only use skb->tstamp for incoming packets,
   as determined by current hook (prerouting, input, forward).
3. xt_time has to use system clock as well rather than skb->tstamp.
   We could still use skb->tstamp for prerouting/input/foward, but
   I see no advantage to make this conditional.

Fixes: fb420d5d ("tcp/fq: move back to CLOCK_MONOTONIC")
Cc: Eric Dumazet <edumazet@google.com>
Reported-by: default avatarMichal Soltys <soltys@ziu.info>
Signed-off-by: default avatarFlorian Westphal <fw@strlen.de>
Acked-by: default avatarEric Dumazet <edumazet@google.com>
Signed-off-by: default avatarPablo Neira Ayuso <pablo@netfilter.org>
parent 7caa56f0
......@@ -1017,12 +1017,9 @@ __nf_conntrack_confirm(struct sk_buff *skb)
/* set conntrack timestamp, if enabled. */
tstamp = nf_conn_tstamp_find(ct);
if (tstamp) {
if (skb->tstamp == 0)
__net_timestamp(skb);
if (tstamp)
tstamp->start = ktime_get_real_ns();
tstamp->start = ktime_to_ns(skb->tstamp);
}
/* Since the lookup is lockless, hash insertion must be done after
* starting the timer and setting the CONFIRMED bit. The RCU barriers
* guarantee that no other CPU can find the conntrack before the above
......
......@@ -540,7 +540,7 @@ __build_packet_message(struct nfnl_log_net *log,
goto nla_put_failure;
}
if (skb->tstamp) {
if (hooknum <= NF_INET_FORWARD && skb->tstamp) {
struct nfulnl_msg_packet_timestamp ts;
struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
ts.sec = cpu_to_be64(kts.tv_sec);
......
......@@ -582,7 +582,7 @@ nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
if (nfqnl_put_bridge(entry, skb) < 0)
goto nla_put_failure;
if (entskb->tstamp) {
if (entry->state.hook <= NF_INET_FORWARD && entskb->tstamp) {
struct nfqnl_msg_packet_timestamp ts;
struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
......
......@@ -163,19 +163,24 @@ time_mt(const struct sk_buff *skb, struct xt_action_param *par)
s64 stamp;
/*
* We cannot use get_seconds() instead of __net_timestamp() here.
* We need real time here, but we can neither use skb->tstamp
* nor __net_timestamp().
*
* skb->tstamp and skb->skb_mstamp_ns overlap, however, they
* use different clock types (real vs monotonic).
*
* Suppose you have two rules:
* 1. match before 13:00
* 2. match after 13:00
* 1. match before 13:00
* 2. match after 13:00
*
* If you match against processing time (get_seconds) it
* may happen that the same packet matches both rules if
* it arrived at the right moment before 13:00.
* it arrived at the right moment before 13:00, so it would be
* better to check skb->tstamp and set it via __net_timestamp()
* if needed. This however breaks outgoing packets tx timestamp,
* and causes them to get delayed forever by fq packet scheduler.
*/
if (skb->tstamp == 0)
__net_timestamp((struct sk_buff *)skb);
stamp = ktime_to_ns(skb->tstamp);
stamp = div_s64(stamp, NSEC_PER_SEC);
stamp = get_seconds();
if (info->flags & XT_TIME_LOCAL_TZ)
/* Adjust for local timezone */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment