Re: Urgent Bug report: PPPoE ioctl(PPPIOCCONNECT): Transport endpoint is not connected
From: Guillaume Nault <hidden>
Date: 2021-09-14 08:07:48
On Tue, Sep 14, 2021 at 09:16:55AM +0300, Martin Zaharinov wrote:
Hi Nault
See this stats :
Linux 5.14.2 (testb) 09/14/21 _x86_64_ (12 CPU)
11:33:44 CPU %usr %nice %sys %iowait %irq %soft %steal %guest %gnice %idle
11:33:45 all 1.75 0.00 18.85 0.00 0.00 5.00 0.00 0.00 0.00 74.40
11:33:46 all 1.74 0.00 17.88 0.00 0.00 4.72 0.00 0.00 0.00 75.66
11:33:47 all 2.23 0.00 17.62 0.00 0.00 5.05 0.00 0.00 0.00 75.10
11:33:48 all 1.82 0.00 13.64 0.00 0.00 5.70 0.00 0.00 0.00 78.84
11:33:49 all 1.50 0.00 13.46 0.00 0.00 5.15 0.00 0.00 0.00 79.90
11:33:50 all 3.06 0.00 13.96 0.00 0.00 4.79 0.00 0.00 0.00 78.20
11:33:51 all 1.40 0.00 16.53 0.00 0.00 5.21 0.00 0.00 0.00 76.86
11:33:52 all 4.43 0.00 19.44 0.00 0.00 6.56 0.00 0.00 0.00 69.57
11:33:53 all 1.51 0.00 16.40 0.00 0.00 4.77 0.00 0.00 0.00 77.32
11:33:54 all 1.51 0.00 16.55 0.00 0.00 4.71 0.00 0.00 0.00 77.23
11:33:55 all 1.00 0.00 13.21 0.00 0.00 5.90 0.00 0.00 0.00 79.90
Average: all 2.00 0.00 16.14 0.00 0.00 5.23 0.00 0.00 0.00 76.63
PerfTop: 28046 irqs/sec kernel:96.3% exact: 100.0% lost: 0/0 drop: 0/0 [4000Hz cycles], (all, 12 CPUs)
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
23.37% [nf_conntrack] [k] nf_ct_iterate_cleanup
17.76% [kernel] [k] mutex_spin_on_owner
9.47% [pppoe] [k] pppoe_rcv
7.71% [kernel] [k] osq_lock
2.77% [nf_nat] [k] inet_cmp
2.59% [nf_nat] [k] device_cmp
2.55% [kernel] [k] __local_bh_enable_ip
2.04% [kernel] [k] _raw_spin_lock
1.23% [kernel] [k] __cond_resched
1.16% [kernel] [k] rcu_all_qs
1.13% libfrr.so.0.0.0 [.] 0x00000000000ce970
0.79% [nf_conntrack] [k] nf_conntrack_lock
0.75% libfrr.so.0.0.0 [.] 0x00000000000ce94e
0.53% [kernel] [k] __netif_receive_skb_core.constprop.0
0.46% [kernel] [k] fib_table_lookup
0.46% [ip_tables] [k] ipt_do_table
0.45% [ixgbe] [k] ixgbe_clean_rx_irq
0.37% [kernel] [k] __dev_queue_xmit
0.34% [nf_conntrack] [k] __nf_conntrack_find_get.isra.0
0.33% [ixgbe] [k] ixgbe_clean_tx_irq
0.30% [kernel] [k] menu_select
0.25% [kernel] [k] vlan_do_receive
0.21% [kernel] [k] ip_finish_output2
0.21% [ixgbe] [k] ixgbe_poll
0.20% [kernel] [k] _raw_spin_lock_irqsave
0.19% [kernel] [k] get_rps_cpu
0.19% libc.so.6 [.] 0x0000000000186afa
0.19% [kernel] [k] queued_read_lock_slowpath
0.19% [kernel] [k] do_poll.constprop.0
0.19% [kernel] [k] cpuidle_enter_state
0.18% [kernel] [k] dev_hard_start_xmit
0.18% [kernel] [k] ___slab_alloc.constprop.0
0.17% zebra [.] 0x00000000000b9271
0.16% [kernel] [k] csum_partial_copy_generic
0.16% zebra [.] 0x00000000000b91f1
0.16% [kernel] [k] page_frag_free
0.16% [kernel] [k] kmem_cache_alloc
0.15% [kernel] [k] __skb_flow_dissect
0.15% [kernel] [k] sched_clock
0.15% libc.so.6 [.] 0x00000000000965a2
0.15% [kernel] [k] kmem_cache_free_bulk.part.0
0.15% [pppoe] [k] pppoe_flush_dev
0.15% [ixgbe] [k] ixgbe_tx_map
0.14% [kernel] [k] _raw_spin_lock_bh
0.14% [kernel] [k] fib_table_flush
0.14% [kernel] [k] native_irq_return_iret
0.14% [kernel] [k] __dev_xmit_skb
0.13% [kernel] [k] nf_hook_slow
0.13% [kernel] [k] fib_lookup_good_nhc
0.12% [kernel] [k] __fget_files
0.12% [kernel] [k] process_backlog
0.12% [xt_dtvqos] [k] 0x00000000000008d1
0.12% [kernel] [k] __list_del_entry_valid
0.12% [kernel] [k] skb_release_data
0.12% [kernel] [k] ip_route_input_slow
0.11% [kernel] [k] netif_skb_features
0.11% [kernel] [k] sock_poll
0.11% [kernel] [k] __schedule
0.11% [kernel] [k] __softirqentry_text_start
And on time of problem when try to write : ip a
to list interface wait 15-20 sec i finaly have options to simulate but users is angry when down internet.Probably some contention on the rtnl lock.
In case need to know why system is overloaded when deconfig ppp interface.
Does it help if you disable conntrack?
Best regards, Martinquoted
On 11 Sep 2021, at 9:26, Martin Zaharinov [off-list ref] wrote: Hi Guillaume Main problem is overload of service because have many finishing ppp (customer) last two day down from 40-50 to 100-200 users and make problem when is happen if try to type : ip a wait 10-20 sec to start list interface . But how to find where is a problem any locking or other. And is there options to make fast remove ppp interface from kernel to reduce this load. Martinquoted
On 7 Sep 2021, at 9:42, Martin Zaharinov [off-list ref] wrote: Perf top from text PerfTop: 28391 irqs/sec kernel:98.0% exact: 100.0% lost: 0/0 drop: 0/0 [4000Hz cycles], (all, 12 CPUs) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 17.01% [nf_conntrack] [k] nf_ct_iterate_cleanup 9.73% [kernel] [k] mutex_spin_on_owner 9.07% [pppoe] [k] pppoe_rcv 2.77% [nf_nat] [k] device_cmp 1.66% [kernel] [k] osq_lock 1.65% [kernel] [k] _raw_spin_lock 1.61% [kernel] [k] __local_bh_enable_ip 1.35% [nf_nat] [k] inet_cmp 1.30% [kernel] [k] __netif_receive_skb_core.constprop.0 1.16% [kernel] [k] menu_select 0.99% [kernel] [k] cpuidle_enter_state 0.96% [ixgbe] [k] ixgbe_clean_rx_irq 0.86% [kernel] [k] __dev_queue_xmit 0.70% [kernel] [k] __cond_resched 0.69% [sch_cake] [k] cake_dequeue 0.67% [nf_tables] [k] nft_do_chain 0.63% [kernel] [k] rcu_all_qs 0.61% [kernel] [k] fib_table_lookup 0.57% [kernel] [k] __schedule 0.57% [kernel] [k] skb_release_data 0.54% [kernel] [k] sched_clock 0.54% [kernel] [k] __copy_skb_header 0.53% [kernel] [k] dev_queue_xmit_nit 0.53% [kernel] [k] _raw_spin_lock_irqsave 0.50% [kernel] [k] kmem_cache_free 0.48% libfrr.so.0.0.0 [.] 0x00000000000ce970 0.47% [ixgbe] [k] ixgbe_clean_tx_irq 0.45% [kernel] [k] timerqueue_add 0.45% [kernel] [k] lapic_next_deadline 0.45% [kernel] [k] csum_partial_copy_generic 0.44% [nf_flow_table] [k] nf_flow_offload_ip_hook 0.44% [kernel] [k] kmem_cache_alloc 0.44% [nf_conntrack] [k] nf_conntrack_lockquoted
On 7 Sep 2021, at 9:16, Martin Zaharinov [off-list ref] wrote: Hi Sorry for delay but not easy to catch moment . See this is mpstatl 1 : Linux 5.14.1 (demobng) 09/07/21 _x86_64_ (12 CPU) 11:12:16 CPU %usr %nice %sys %iowait %irq %soft %steal %guest %gnice %idle 11:12:17 all 0.17 0.00 6.66 0.00 0.00 4.13 0.00 0.00 0.00 89.05 11:12:18 all 0.25 0.00 8.36 0.00 0.00 4.88 0.00 0.00 0.00 86.51 11:12:19 all 0.26 0.00 9.62 0.00 0.00 3.91 0.00 0.00 0.00 86.21 11:12:20 all 0.85 0.00 6.00 0.00 0.00 4.31 0.00 0.00 0.00 88.84 11:12:21 all 0.08 0.00 4.45 0.00 0.00 4.79 0.00 0.00 0.00 90.67 11:12:22 all 0.17 0.00 9.50 0.00 0.00 4.58 0.00 0.00 0.00 85.75 11:12:23 all 0.00 0.00 6.92 0.00 0.00 2.48 0.00 0.00 0.00 90.61 11:12:24 all 0.17 0.00 5.45 0.00 0.00 4.27 0.00 0.00 0.00 90.11 11:12:25 all 0.25 0.00 5.38 0.00 0.00 4.79 0.00 0.00 0.00 89.58 11:12:26 all 0.60 0.00 1.45 0.00 0.00 2.65 0.00 0.00 0.00 95.30 11:12:27 all 0.42 0.00 6.91 0.00 0.00 4.47 0.00 0.00 0.00 88.20 11:12:28 all 0.00 0.00 6.75 0.00 0.00 4.18 0.00 0.00 0.00 89.07 11:12:29 all 0.17 0.00 3.52 0.00 0.00 5.11 0.00 0.00 0.00 91.20 11:12:30 all 1.45 0.00 10.14 0.00 0.00 3.49 0.00 0.00 0.00 84.92 11:12:31 all 0.09 0.00 5.11 0.00 0.00 4.77 0.00 0.00 0.00 90.03 11:12:32 all 0.25 0.00 3.11 0.00 0.00 4.46 0.00 0.00 0.00 92.17 Average: all 0.32 0.00 6.21 0.00 0.00 4.21 0.00 0.00 0.00 89.26 I attache and one screenshot from perf top (Screenshot is send on preview mail) And I see in lsmod pppoe 20480 8198 pppox 16384 1 pppoe ppp_generic 45056 16364 pppox,pppoe slhc 16384 1 ppp_generic To slow remove pppoe session . And from log : [2021-09-07 11:01:11.129] vlan3020: ebdd1c5d8b5900f6: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:01:53.621] vlan643: ebdd1c5d8b59014e: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:02:00.359] vlan1616: ebdd1c5d8b590195: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:02:05.859] vlan3020: ebdd1c5d8b5900d8: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:02:08.258] vlan3005: ebdd1c5d8b590190: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:02:13.820] vlan643: ebdd1c5d8b590152: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:02:15.839] vlan727: ebdd1c5d8b590144: ioctl(PPPIOCCONNECT): Transport endpoint is not connected [2021-09-07 11:02:20.139] vlan1693: ebdd1c5d8b59019f: ioctl(PPPIOCCONNECT): Transport endpoint is not connectedquoted
On 11 Aug 2021, at 19:48, Guillaume Nault [off-list ref] wrote: On Wed, Aug 11, 2021 at 02:10:32PM +0300, Martin Zaharinov wrote:quoted
And one more that see. Problem is come when accel start finishing sessions, Now in server have 2k users and restart on one of vlans 3 Olt with 400 users and affect other vlans , And problem is start when start destroying dead sessions from vlan with 3 Olt and this affect all other vlans. May be kernel destroy old session slow and entrained other users by locking other sessions. is there a way to speed up the closing of stopped/dead sessions.What are the CPU stats when that happen? Is it users space or kernel space that keeps it busy? One easy way to check is to run "mpstat 1" for a few seconds when the problem occurs.