| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20210318073455.17281-2-kurt@linutronix.de>
Date: Thu, 18 Mar 2021 08:34:55 +0100
From: Kurt Kanzenbach <kurt@...utronix.de>
To: Vinicius Costa Gomes <vinicius.gomes@...el.com>
Cc: Vladimir Oltean <olteanv@...il.com>,
Jamal Hadi Salim <jhs@...atatu.com>,
Cong Wang <xiyou.wangcong@...il.com>,
Jiri Pirko <jiri@...nulli.us>,
"David S. Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>,
Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
netdev@...r.kernel.org, Kurt Kanzenbach <kurt@...utronix.de>
Subject: [PATCH net-next 1/1] taprio: Handle short intervals and large packets
When using short intervals e.g. below one millisecond, large packets won't be
transmitted at all. The software implementations checks whether the packet can
be fit into the remaining interval. Therefore, it takes the packet length and
the transmission speed into account. That is correct.
However, for large packets it may be that the transmission time exceeds the
interval resulting in no packet transmission. The same situation works fine with
hardware offloading applied.
The problem has been observed with the following schedule and iperf3:
|tc qdisc replace dev lan1 parent root handle 100 taprio \
| num_tc 8 \
| map 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 \
| queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
| base-time $base \
| sched-entry S 0x40 500000 \
| sched-entry S 0xbf 500000 \
| clockid CLOCK_TAI \
| flags 0x00
[...]
|root@tsn:~# iperf3 -c 192.168.2.105
|Connecting to host 192.168.2.105, port 5201
|[ 5] local 192.168.2.121 port 52610 connected to 192.168.2.105 port 5201
|[ ID] Interval Transfer Bitrate Retr Cwnd
|[ 5] 0.00-1.00 sec 45.2 KBytes 370 Kbits/sec 0 1.41 KBytes
|[ 5] 1.00-2.00 sec 0.00 Bytes 0.00 bits/sec 0 1.41 KBytes
After debugging, it seems that the packet length stored in the SKB is about
7000-8000 bytes. Using a 100 Mbit/s link the transmission time is about 600us
which larger than the interval of 500us.
Therefore, segment the SKB into smaller chunks if the packet is too big. This
yields similar results than the hardware offload:
|root@tsn:~# iperf3 -c 192.168.2.105
|Connecting to host 192.168.2.105, port 5201
|- - - - - - - - - - - - - - - - - - - - - - - - -
|[ ID] Interval Transfer Bitrate Retr
|[ 5] 0.00-10.00 sec 48.9 MBytes 41.0 Mbits/sec 0 sender
|[ 5] 0.00-10.02 sec 48.7 MBytes 40.7 Mbits/sec receiver
Furthermore, the segmentation can be skipped for the full offload case, as the
driver or the hardware is expected to handle this.
Signed-off-by: Kurt Kanzenbach <kurt@...utronix.de>
---
net/sched/sch_taprio.c | 64 +++++++++++++++++++++++++++++++++++-------
1 file changed, 54 insertions(+), 10 deletions(-)
diff --git a/net/sched/sch_taprio.c b/net/sched/sch_taprio.c
index 8287894541e3..922ed6b91abb 100644
--- a/net/sched/sch_taprio.c
+++ b/net/sched/sch_taprio.c
@@ -411,18 +411,10 @@ static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
return txtime;
}
-static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
- struct sk_buff **to_free)
+static int taprio_enqueue_one(struct sk_buff *skb, struct Qdisc *sch,
+ struct Qdisc *child, struct sk_buff **to_free)
{
struct taprio_sched *q = qdisc_priv(sch);
- struct Qdisc *child;
- int queue;
-
- queue = skb_get_queue_mapping(skb);
-
- child = q->qdiscs[queue];
- if (unlikely(!child))
- return qdisc_drop(skb, sch, to_free);
if (skb->sk && sock_flag(skb->sk, SOCK_TXTIME)) {
if (!is_valid_interval(skb, sch))
@@ -439,6 +431,58 @@ static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
return qdisc_enqueue(skb, child, to_free);
}
+static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
+ struct sk_buff **to_free)
+{
+ struct taprio_sched *q = qdisc_priv(sch);
+ struct Qdisc *child;
+ int queue;
+
+ queue = skb_get_queue_mapping(skb);
+
+ child = q->qdiscs[queue];
+ if (unlikely(!child))
+ return qdisc_drop(skb, sch, to_free);
+
+ /* Large packets might not be transmitted when the transmission duration
+ * exceeds any configured interval. Therefore, segment the skb into
+ * smaller chunks. Skip it for the full offload case, as the driver
+ * and/or the hardware is expected to handle this.
+ */
+ if (skb_is_gso(skb) && !FULL_OFFLOAD_IS_ENABLED(q->flags)) {
+ unsigned int slen = 0, numsegs = 0, len = qdisc_pkt_len(skb);
+ netdev_features_t features = netif_skb_features(skb);
+ struct sk_buff *segs, *nskb;
+ int ret;
+
+ segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
+ if (IS_ERR_OR_NULL(segs))
+ return qdisc_drop(skb, sch, to_free);
+
+ skb_list_walk_safe(segs, segs, nskb) {
+ skb_mark_not_on_list(segs);
+ qdisc_skb_cb(segs)->pkt_len = segs->len;
+ slen += segs->len;
+
+ ret = taprio_enqueue_one(segs, sch, child, to_free);
+ if (ret != NET_XMIT_SUCCESS) {
+ if (net_xmit_drop_count(ret))
+ qdisc_qstats_drop(sch);
+ } else {
+ numsegs++;
+ }
+ }
+
+ if (numsegs > 1)
+ qdisc_tree_reduce_backlog(sch, 1 - numsegs, len - slen);
+ consume_skb(skb);
+
+ return numsegs > 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
+ }
+
+ return taprio_enqueue_one(skb, sch, child, to_free);
+}
+
static struct sk_buff *taprio_peek_soft(struct Qdisc *sch)
{
struct taprio_sched *q = qdisc_priv(sch);
--
2.20.1
Powered by blists - more mailing lists