---- William Liu will@willsroot.io 在 Wed, 2025-12-10 11:31:02 写到：---
Hi zyc, The netdev maintainers and I are aware of this issue. This change was made to prevent a local DOS bug in netem that was extremely trivial to trigger, and some maintainers considered your type of configuration as an extremely rare use case (but imo still a very valid one). I provided a summary of this fix as well as other proposed fixes in both the commit log and [1]. There is currently a patch proposed in [2] to resolve this which technically obeys man page semantics but was previously rejected. Basically, the DOS can be easily resolved by enqueuing duplicated packets to the same netem qdisc, but other maintainers presented valid concerns about changing user visible behavior from the past 2 decades and the original commit message mentioned enqueuing from root was necessary to "avoid problems with qlen accounting with nested qdisc." tc_skb_extensions sounds like a reasonable solution to me that changes no existing behavior and adds no additional restrictions [3], but it has not been further explored or checked for soundness. Hypothetically, it would fix both the DOS bug and allow packets to retain the behavior of enqueuing from root. You can follow the status of the fix in [2] - I do not plan to further volunteer myself and my free time in this specific fix process due to the consistent pattern of unprofessional and patronizing emails from Cong Wang. I am confident the other netdev maintainers should reach a resolution soon. Best, Will [1] https://lore.kernel.org/netdev/PKMd5btHYmJcKSiIJdtxQvZBEfuS4RQkBnE4M-TZkjUq_... [2] https://lore.kernel.org/netdev/20251126195244.88124-4-xiyou.wangcong@gmail.c... [3] https://lore.kernel.org/netdev/CAM0EoMmBdZBzfUAms5-0hH5qF5ODvxWfgqrbHaGT6p3-...
Hi Will,
Thanks for the details. We will read the links you provide.
Zhang Yang Chao