A soft lockup issue was observed with following log:

  watchdog: BUG: soft lockup - CPU#1 stuck for 104s! [tc:94]
  CPU: 1 UID: 0 PID: 94 Comm: tc Tainted: G L 6.18.0-rc4-00007-gc9cfc122f037 #425 PREEMPT(voluntary)
  RIP: 0010:qdisc_match_from_root+0x0/0x70
  Call Trace:
   <TASK>
   qdisc_tree_reduce_backlog+0xec/0x110
   fq_change+0x2e0/0x6a0
   qdisc_create+0x138/0x4e0
   tc_modify_qdisc+0x5b9/0x9d0
   rtnetlink_rcv_msg+0x15a/0x400
   netlink_rcv_skb+0x55/0x100
   netlink_unicast+0x257/0x380
   netlink_sendmsg+0x1e2/0x420
   ____sys_sendmsg+0x313/0x340
   ___sys_sendmsg+0x82/0xd0
   __sys_sendmsg+0x6c/0xd0
   </TASK>

The issue can be reproduced by:
  tc qdisc add dev eth0 root noqueue
  tc qdisc add dev eth0 handle ffff:0 ingress
  tc qdisc add dev eth0 handle ffe0:0 parent ffff:a fq

A fq qdisc was created in __tc_modify_qdisc(), when the input parent major
'ffff' is equal to the ingress qdisc handle major and the complete parent
handle is not TC_H_ROOT or TC_H_INGRESS, which leads to a infinite loop in
qdisc_tree_reduce_backlog().

The infinite loop scenario:

  qdisc_tree_reduce_backlog

    // init sch is fq qdisc, parent is ffff000a
    while ((parentid = sch->parent)) {

      // query qdisc by handle ffff0000
      sch = qdisc_lookup_rcu(qdisc_dev(sch), TC_H_MAJ(parentid));

      // return ingress qdisc, sch->parent is fffffff1
      if (sch == NULL) {
      ...
    }

Commit 2e95c4384438 ("net/sched: stop qdisc_tree_reduce_backlog on
TC_H_ROOT") break the loop only when parent TC_H_ROOT is reached. However,
qdisc_lookup_rcu() will return the same qdisc when input an ingress qdisc
with major 'ffff'. If the parent TC_H_INGRESS is reached, also break the
loop.

Fixes: 2e95c4384438 ("net/sched: stop qdisc_tree_reduce_backlog on TC_H_ROOT")
Signed-off-by: Wang Liang <wangliang74@huawei.com>
---
 net/sched/sch_api.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index 1e058b46d3e1..b4d6fe6b6812 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -784,7 +784,7 @@ void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
 	drops = max_t(int, n, 0);
 	rcu_read_lock();
 	while ((parentid = sch->parent)) {
-		if (parentid == TC_H_ROOT)
+		if (parentid == TC_H_ROOT || parentid == TC_H_INGRESS)
 			break;
 
 		if (sch->flags & TCQ_F_NOPARENT)
-- 
2.34.1

On Tue, Nov 4, 2025 at 9:00 PM Wang Liang <wangliang74@huawei.com> wrote:
>
> A soft lockup issue was observed with following log:
>
>   watchdog: BUG: soft lockup - CPU#1 stuck for 104s! [tc:94]
>   CPU: 1 UID: 0 PID: 94 Comm: tc Tainted: G L 6.18.0-rc4-00007-gc9cfc122f037 #425 PREEMPT(voluntary)
>   RIP: 0010:qdisc_match_from_root+0x0/0x70
>   Call Trace:
>    <TASK>
>    qdisc_tree_reduce_backlog+0xec/0x110
>    fq_change+0x2e0/0x6a0
>    qdisc_create+0x138/0x4e0
>    tc_modify_qdisc+0x5b9/0x9d0
>    rtnetlink_rcv_msg+0x15a/0x400
>    netlink_rcv_skb+0x55/0x100
>    netlink_unicast+0x257/0x380
>    netlink_sendmsg+0x1e2/0x420
>    ____sys_sendmsg+0x313/0x340
>    ___sys_sendmsg+0x82/0xd0
>    __sys_sendmsg+0x6c/0xd0
>    </TASK>
>
> The issue can be reproduced by:
>   tc qdisc add dev eth0 root noqueue
>   tc qdisc add dev eth0 handle ffff:0 ingress
>   tc qdisc add dev eth0 handle ffe0:0 parent ffff:a fq
>
> A fq qdisc was created in __tc_modify_qdisc(), when the input parent major
> 'ffff' is equal to the ingress qdisc handle major and the complete parent
> handle is not TC_H_ROOT or TC_H_INGRESS, which leads to a infinite loop in
> qdisc_tree_reduce_backlog().
>

Good test case, but i dont think this is the right solution.
That config is bogus and should never have been accepted by the kernel.
We'll look into it..

cheers,
jamal


> The infinite loop scenario:
>
>   qdisc_tree_reduce_backlog
>
>     // init sch is fq qdisc, parent is ffff000a
>     while ((parentid = sch->parent)) {
>
>       // query qdisc by handle ffff0000
>       sch = qdisc_lookup_rcu(qdisc_dev(sch), TC_H_MAJ(parentid));
>
>       // return ingress qdisc, sch->parent is fffffff1
>       if (sch == NULL) {
>       ...
>     }
>
> Commit 2e95c4384438 ("net/sched: stop qdisc_tree_reduce_backlog on
> TC_H_ROOT") break the loop only when parent TC_H_ROOT is reached. However,
> qdisc_lookup_rcu() will return the same qdisc when input an ingress qdisc
> with major 'ffff'. If the parent TC_H_INGRESS is reached, also break the
> loop.
>
> Fixes: 2e95c4384438 ("net/sched: stop qdisc_tree_reduce_backlog on TC_H_ROOT")
> Signed-off-by: Wang Liang <wangliang74@huawei.com>
> ---
>  net/sched/sch_api.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
> index 1e058b46d3e1..b4d6fe6b6812 100644
> --- a/net/sched/sch_api.c
> +++ b/net/sched/sch_api.c
> @@ -784,7 +784,7 @@ void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
>         drops = max_t(int, n, 0);
>         rcu_read_lock();
>         while ((parentid = sch->parent)) {
> -               if (parentid == TC_H_ROOT)
> +               if (parentid == TC_H_ROOT || parentid == TC_H_INGRESS)
>                         break;
>
>                 if (sch->flags & TCQ_F_NOPARENT)
> --
> 2.34.1
>