From: Florian Westphal fw@strlen.de
[ Upstream commit 916f6efae62305796e012e7c3a7884a267cbacbf ]
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: fb420d5d91c1 ("tcp/fq: move back to CLOCK_MONOTONIC") Cc: Eric Dumazet edumazet@google.com Reported-by: Michal Soltys soltys@ziu.info Signed-off-by: Florian Westphal fw@strlen.de Acked-by: Eric Dumazet edumazet@google.com Signed-off-by: Pablo Neira Ayuso pablo@netfilter.org Signed-off-by: Sasha Levin sashal@kernel.org --- net/netfilter/nf_conntrack_core.c | 7 ++----- net/netfilter/nfnetlink_log.c | 2 +- net/netfilter/nfnetlink_queue.c | 2 +- net/netfilter/xt_time.c | 23 ++++++++++++++--------- 4 files changed, 18 insertions(+), 16 deletions(-)
diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c index 1982faf21ebb..d7ac2f82bb6d 100644 --- a/net/netfilter/nf_conntrack_core.c +++ b/net/netfilter/nf_conntrack_core.c @@ -983,12 +983,9 @@ __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 diff --git a/net/netfilter/nfnetlink_log.c b/net/netfilter/nfnetlink_log.c index b1f9c5303f02..0b3347570265 100644 --- a/net/netfilter/nfnetlink_log.c +++ b/net/netfilter/nfnetlink_log.c @@ -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); diff --git a/net/netfilter/nfnetlink_queue.c b/net/netfilter/nfnetlink_queue.c index 0dcc3592d053..e057b2961d31 100644 --- a/net/netfilter/nfnetlink_queue.c +++ b/net/netfilter/nfnetlink_queue.c @@ -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);
diff --git a/net/netfilter/xt_time.c b/net/netfilter/xt_time.c index c13bcd0ab491..8dbb4d48f2ed 100644 --- a/net/netfilter/xt_time.c +++ b/net/netfilter/xt_time.c @@ -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 */