Re: [PATCH RFC net-next] net: drop_monitor: Add debugfs support

From
Wang Liang <hidden>
Date
2025-10-16 06:26:44
Also in
lkml
Reply to
Eric Dumazet, 2025-10-15
在 2025/10/15 18:40, Eric Dumazet 写道:
On Wed, Oct 15, 2025 at 2:51 AM Wang Liang [off-list ref] wrote:
quoted
This patch add debugfs interfaces for drop monitor. Similar to kmemleak, we
can use the monitor by below commands:

   echo clear > /sys/kernel/debug/drop_monitor/trace
   echo start > /sys/kernel/debug/drop_monitor/trace
   echo stop  > /sys/kernel/debug/drop_monitor/trace
   cat /sys/kernel/debug/drop_monitor/trace

The trace skb number limit can be set dynamically:

   cat /sys/kernel/debug/drop_monitor/trace_limit
   echo 200 > /sys/kernel/debug/drop_monitor/trace_limit

Compare to original netlink method, the callstack dump is supported. There
is a example for received udp packet with error checksum:

   reason   : UDP_CSUM (11)
   pc       : udp_queue_rcv_one_skb+0x14b/0x350
   len      : 12
   protocol : 0x0800
   stack    :
     sk_skb_reason_drop+0x8f/0x120
     udp_queue_rcv_one_skb+0x14b/0x350
     udp_unicast_rcv_skb+0x71/0x90
     ip_protocol_deliver_rcu+0xa6/0x160
     ip_local_deliver_finish+0x90/0x100
     ip_sublist_rcv_finish+0x65/0x80
     ip_sublist_rcv+0x130/0x1c0
     ip_list_rcv+0xf7/0x130
     __netif_receive_skb_list_core+0x21d/0x240
     netif_receive_skb_list_internal+0x186/0x2b0
     napi_complete_done+0x78/0x190
     e1000_clean+0x27f/0x860
     __napi_poll+0x25/0x1e0
     net_rx_action+0x2ca/0x330
     handle_softirqs+0xbc/0x290
     irq_exit_rcu+0x90/0xb0

It's more friendly to use and not need user application to cooperate.
Furthermore, it is easier to add new feature. We can add reason/ip/port
filter by debugfs parameters, like ftrace, rather than netlink msg.
I do not understand the fascination with net/core/drop_monitor.c,
which looks very old school to me,
and misses all the features,  flexibility, scalability  that 'perf',
eBPF tracing, bpftrace, .... have today.

Adding  /sys/kernel/debug/drop_monitor/* is even more old school.

Not mentioning the maintenance burden.

For me the choice is easy :

# CONFIG_NET_DROP_MONITOR is not set

perf record -ag -e skb:kfree_skb sleep 1

perf script # or perf report

Thank you for taking time to review this patch!

My initially thought was that the drop_monitor may cover more drop
positions (not just kfree_skb), show more skb info, filter skb by ip/port
debugfs parameter (not support now), and not need userspace tools.

Currently perf is indeed a better choice, adding debugfs is not necessary.

Thanks!
Thread (5 messages) 5 messages, 4 authors, last reply 2025-10-16
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help