sch_tbf: handle too small burst

If a too small burst is inadvertently set on TBF, we might trigger
a bug in tbf_segment(), as 'skb' instead of 'segs' was used in a
qdisc_reshape_fail() call.

tc qdisc add dev eth0 root handle 1: tbf latency 50ms burst 1KB rate
50mbit

Fix the bug, and add a warning, as such configuration is not
going to work anyway for non GSO packets.

(For some reason, one has to use a burst >= 1520 to get a working
configuration, even with old kernels. This is a probable iproute2/tc
bug)

Based on a report and initial patch from Yang Yingliang

Fixes: e43ac79a4bc6 ("sch_tbf: segment too big GSO packets")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/sched/sch_tbf.c b/net/sched/sch_tbf.c
index 68f9859..a609005 100644 (file)
--- a/net/sched/sch_tbf.c
+++ b/net/sched/sch_tbf.c
@@ -21,6 +21,7 @@
 #include <net/netlink.h>
 #include <net/sch_generic.h>
 #include <net/pkt_sched.h>
+#include <net/tcp.h>
 
 
 /*     Simple Token Bucket Filter.
@@ -117,6 +118,22 @@ struct tbf_sched_data {
 };
 
 
+/*
+ * Return length of individual segments of a gso packet,
+ * including all headers (MAC, IP, TCP/UDP)
+ */
+static unsigned int skb_gso_seglen(const struct sk_buff *skb)
+{
+       unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
+       const struct skb_shared_info *shinfo = skb_shinfo(skb);
+
+       if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
+               hdr_len += tcp_hdrlen(skb);
+       else
+               hdr_len += sizeof(struct udphdr);
+       return hdr_len + shinfo->gso_size;
+}
+
 /* GSO packet is too big, segment it so that tbf can transmit
  * each segment in time
  */
@@ -136,12 +153,8 @@ static int tbf_segment(struct sk_buff *skb, struct Qdisc *sch)
        while (segs) {
                nskb = segs->next;
                segs->next = NULL;
-               if (likely(segs->len <= q->max_size)) {
-                       qdisc_skb_cb(segs)->pkt_len = segs->len;
-                       ret = qdisc_enqueue(segs, q->qdisc);
-               } else {
-                       ret = qdisc_reshape_fail(skb, sch);
-               }
+               qdisc_skb_cb(segs)->pkt_len = segs->len;
+               ret = qdisc_enqueue(segs, q->qdisc);
                if (ret != NET_XMIT_SUCCESS) {
                        if (net_xmit_drop_count(ret))
                                sch->qstats.drops++;
@@ -163,7 +176,7 @@ static int tbf_enqueue(struct sk_buff *skb, struct Qdisc *sch)
        int ret;
 
        if (qdisc_pkt_len(skb) > q->max_size) {
-               if (skb_is_gso(skb))
+               if (skb_is_gso(skb) && skb_gso_seglen(skb) <= q->max_size)
                        return tbf_segment(skb, sch);
                return qdisc_reshape_fail(skb, sch);
        }
@@ -319,6 +332,11 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
        if (max_size < 0)
                goto done;
 
+       if (max_size < psched_mtu(qdisc_dev(sch)))
+               pr_warn_ratelimited("sch_tbf: burst %u is lower than device %s mtu (%u) !\n",
+                                   max_size, qdisc_dev(sch)->name,
+                                   psched_mtu(qdisc_dev(sch)));
+
        if (q->qdisc != &noop_qdisc) {
                err = fifo_set_limit(q->qdisc, qopt->limit);
                if (err)