Re: [PROBLEM] [ADDITIONAL DIAG] seltests: net/forwarding/sch_ets.sh [HANG]
From: Mirsad Todorovac
Date: Fri Jul 21 2023 - 20:55:09 EST
On 7/22/23 02:31, Mirsad Todorovac wrote:
On 7/20/23 18:25, Mirsad Todorovac wrote:
On 7/20/23 18:07, Mirsad Todorovac wrote:
On 7/20/23 11:43, Petr Machata wrote:
Mirsad Todorovac <mirsad.todorovac@xxxxxxxxxxxx> writes:
Using the same config for 6.5-rc2 on Ubuntu 22.04 LTS and 22.10, the execution
stop at the exact same line on both boxes (os I reckon it is more than an
accident):
# selftests: net/forwarding: sch_ets.sh
# TEST: ping vlan 10 [ OK ]
# TEST: ping vlan 11 [ OK ]
# TEST: ping vlan 12 [ OK ]
# Running in priomap mode
# Testing ets bands 3 strict 3, streams 0 1
# TEST: band 0 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 1 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 3 strict 3, streams 1 2
# TEST: band 1 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 2 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 0 1
# TEST: band 0 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 1 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 1 2
# TEST: bands 1:2 [ OK ]
# INFO: Expected ratio 2.00 Measured ratio 1.99
# Testing ets bands 3 quanta 3300 3300 3300, streams 0 1 2
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 1.00 Measured ratio .99
# TEST: bands 0:2 [ OK ]
# INFO: Expected ratio 1.00 Measured ratio 1.00
# Testing ets bands 3 quanta 5000 3500 1500, streams 0 1 2
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 1.42 Measured ratio 1.42
# TEST: bands 0:2 [ OK ]
# INFO: Expected ratio 3.33 Measured ratio 3.33
# Testing ets bands 3 quanta 5000 8000 1500, streams 0 1 2
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 1.60 Measured ratio 1.59
# TEST: bands 0:2 [ OK ]
# INFO: Expected ratio 3.33 Measured ratio 3.33
# Testing ets bands 2 quanta 5000 2500, streams 0 1
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 2.00 Measured ratio 1.99
# Running in classifier mode
# Testing ets bands 3 strict 3, streams 0 1
# TEST: band 0 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 1 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 3 strict 3, streams 1 2
# TEST: band 1 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 2 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 0 1
I tried to run 'set -x' enabled version standalone, but that one finished
correctly (?).
It could be something previous scripts left, but right now I don't have a clue.
I can attempt to rerun all tests with sch_ets.sh bash 'set -x' enabled later today.
If you run it standalone without set -x, does it finish as well?
Added that. Yes, standlone run finishes correctly, with or without 'set -x':
root@defiant:/home/marvin/linux/kernel/linux_torvalds/tools/testing/selftests/net/forwarding# ./sch_ets.sh
TEST: ping vlan 10 [ OK ]
TEST: ping vlan 11 [ OK ]
TEST: ping vlan 12 [ OK ]
Running in priomap mode
Testing ets bands 3 strict 3, streams 0 1
TEST: band 0 [ OK ]
INFO: Expected ratio >95% Measured ratio 100.00
TEST: band 1 [ OK ]
INFO: Expected ratio <5% Measured ratio 0
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 strict 3, streams 1 2
TEST: band 1 [ OK ]
INFO: Expected ratio >95% Measured ratio 100.00
TEST: band 2 [ OK ]
INFO: Expected ratio <5% Measured ratio 0
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 0 1
TEST: band 0 [ OK ]
INFO: Expected ratio >95% Measured ratio 100.00
TEST: band 1 [ OK ]
INFO: Expected ratio <5% Measured ratio 0
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 1 2
TEST: bands 1:2 [ OK ]
INFO: Expected ratio 2.00 Measured ratio 1.99
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 quanta 3300 3300 3300, streams 0 1 2
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 1.00 Measured ratio 1.00
TEST: bands 0:2 [ OK ]
INFO: Expected ratio 1.00 Measured ratio .99
killing MZ
killed MZ
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 quanta 5000 3500 1500, streams 0 1 2
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 1.42 Measured ratio 1.42
TEST: bands 0:2 [ OK ]
INFO: Expected ratio 3.33 Measured ratio 3.33
killing MZ
killed MZ
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 quanta 5000 8000 1500, streams 0 1 2
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 1.60 Measured ratio 1.59
TEST: bands 0:2 [ OK ]
INFO: Expected ratio 3.33 Measured ratio 3.33
killing MZ
killed MZ
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 2 quanta 5000 2500, streams 0 1
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 2.00 Measured ratio 1.99
killing MZ
killed MZ
killing MZ
killed MZ
Running in classifier mode
Testing ets bands 3 strict 3, streams 0 1
TEST: band 0 [ OK ]
INFO: Expected ratio >95% Measured ratio 100.00
TEST: band 1 [ OK ]
INFO: Expected ratio <5% Measured ratio 0
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 strict 3, streams 1 2
TEST: band 1 [ OK ]
INFO: Expected ratio >95% Measured ratio 100.00
TEST: band 2 [ OK ]
INFO: Expected ratio <5% Measured ratio 0
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 0 1
TEST: band 0 [ OK ]
INFO: Expected ratio >95% Measured ratio 100.00
TEST: band 1 [ OK ]
INFO: Expected ratio <5% Measured ratio 0
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 1 2
TEST: bands 1:2 [ OK ]
INFO: Expected ratio 2.00 Measured ratio 1.99
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 quanta 3300 3300 3300, streams 0 1 2
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 1.00 Measured ratio .99
TEST: bands 0:2 [ OK ]
INFO: Expected ratio 1.00 Measured ratio .99
killing MZ
killed MZ
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 quanta 5000 3500 1500, streams 0 1 2
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 1.42 Measured ratio 1.42
TEST: bands 0:2 [ OK ]
INFO: Expected ratio 3.33 Measured ratio 3.33
killing MZ
killed MZ
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 3 quanta 5000 8000 1500, streams 0 1 2
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 1.60 Measured ratio 1.60
TEST: bands 0:2 [ OK ]
INFO: Expected ratio 3.33 Measured ratio 3.33
killing MZ
killed MZ
killing MZ
killed MZ
killing MZ
killed MZ
Testing ets bands 2 quanta 5000 2500, streams 0 1
TEST: bands 0:1 [ OK ]
INFO: Expected ratio 2.00 Measured ratio 2.00
killing MZ
killed MZ
killing MZ
killed MZ
root@defiant:/home/marvin/linux/kernel/linux_torvalds/tools/testing/selftests/net/forwarding#
That would imply that the reproducer needs to include the previous tests as
well.
This is entirely possible, as timeouts and CTRL+C events do not seem to be caught
and the cleanup is not done ...
sch_ets_core.sh: trap cleanup EXIT
Some tests time out even after settings:timeout=240, so IMHO this should be taken into account.
Best regards,
Mirsad Todorovac
It looks like the test is stuck in ets_test_mixed in classifier_mode.
A way to run just this test would be:
TESTS="classifier_mode ets_test_mixed" ./sch_ets.sh
Looking at the code, the only place that I can see that waits on
anything is the "{ kill %% && wait %%; } 2>/dev/null" line in
stop_traffic() (in lib.sh). Maybe something like this would let
us see if that's the case:
modified tools/testing/selftests/net/forwarding/lib.sh
@@ -1468,8 +1470,10 @@ start_tcp_traffic()
stop_traffic()
{
+ echo killing MZ
# Suppress noise from killing mausezahn.
{ kill %% && wait %%; } 2>/dev/null
+ echo killed MZ
}
FYI, this is the [incomplete] list of the tests that time out even after
being assigned long timeout of 240s instead of default.
marvin@defiant:~/linux/kernel/linux_torvalds$ grep TIMEOUT ../kselftest-6.5-rc2-net-forwarding-8.log
not ok 13 selftests: net/forwarding: custom_multipath_hash.sh # TIMEOUT 240 seconds
not ok 18 selftests: net/forwarding: gre_custom_multipath_hash.sh # TIMEOUT 240 seconds
not ok 19 selftests: net/forwarding: gre_inner_v4_multipath.sh # TIMEOUT 240 seconds
not ok 21 selftests: net/forwarding: gre_multipath_nh_res.sh # TIMEOUT 240 seconds
not ok 22 selftests: net/forwarding: gre_multipath_nh.sh # TIMEOUT 240 seconds
not ok 27 selftests: net/forwarding: ip6gre_custom_multipath_hash.sh # TIMEOUT 240 seconds
not ok 34 selftests: net/forwarding: ip6gre_inner_v4_multipath.sh # TIMEOUT 240 seconds
not ok 58 selftests: net/forwarding: no_forwarding.sh # TIMEOUT 240 seconds
not ok 67 selftests: net/forwarding: router_mpath_nh_res.sh # TIMEOUT 240 seconds
not ok 68 selftests: net/forwarding: router_mpath_nh.sh # TIMEOUT 240 seconds
not ok 70 selftests: net/forwarding: router_multipath.sh # TIMEOUT 240 seconds
marvin@defiant:~/linux/kernel/linux_torvalds$
The test seem to be stuck despite adding SIGTERM cleanup to all tests
that timed out:
# selftests: net/forwarding: sch_ets.sh
# TEST: ping vlan 10 [ OK ]
# TEST: ping vlan 11 [ OK ]
# TEST: ping vlan 12 [ OK ]
# Running in priomap mode
# Testing ets bands 3 strict 3, streams 0 1
# TEST: band 0 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 1 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 3 strict 3, streams 1 2
# TEST: band 1 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 2 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 0 1
# TEST: band 0 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 1 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 1 2
# TEST: bands 1:2 [ OK ]
# INFO: Expected ratio 2.00 Measured ratio 2.00
# Testing ets bands 3 quanta 3300 3300 3300, streams 0 1 2
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 1.00 Measured ratio .99
# TEST: bands 0:2 [ OK ]
# INFO: Expected ratio 1.00 Measured ratio .99
# Testing ets bands 3 quanta 5000 3500 1500, streams 0 1 2
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 1.42 Measured ratio 1.42
# TEST: bands 0:2 [ OK ]
# INFO: Expected ratio 3.33 Measured ratio 3.33
# Testing ets bands 3 quanta 5000 8000 1500, streams 0 1 2
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 1.60 Measured ratio 1.59
# TEST: bands 0:2 [ OK ]
# INFO: Expected ratio 3.33 Measured ratio 3.33
# Testing ets bands 2 quanta 5000 2500, streams 0 1
# TEST: bands 0:1 [ OK ]
# INFO: Expected ratio 2.00 Measured ratio 1.99
# Running in classifier mode
# Testing ets bands 3 strict 3, streams 0 1
# TEST: band 0 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 1 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 3 strict 3, streams 1 2
# TEST: band 1 [ OK ]
# INFO: Expected ratio >95% Measured ratio 100.00
# TEST: band 2 [ OK ]
# INFO: Expected ratio <5% Measured ratio 0
# Testing ets bands 4 strict 1 quanta 5000 2500 1500, streams 0 1
What is happening is a stuck mousezahn loop:
root 558266 2732 0 01:17 pts/2 00:00:23 mausezahn veth0.10 -p 8000 -A 192.0.2.1 -B 192.0.2.2 -c 0 -a own -b ba:33:37:81:dc:5
root 558273 2732 0 01:17 pts/2 00:00:23 mausezahn veth0.11 -p 8000 -A 192.0.2.17 -B 192.0.2.18 -c 0 -a own -b ba:33:37:81:dc
$ sudo strace -p 558266
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
ioctl(5, SIOCGIFINDEX, {ifr_name="veth0.10"}) = -1 ENODEV (No such device)
I am not sure if this will ever complete, or is it a runaway loop?
Probably it is stuck, for nothing was written to the test log for 71 minutes.
I hope we are coming closer to the cause.
Previously I didn't figure out what hanged.
FWIW, I will post the patches for the scripts that timed out for your review.
Hi,
More homework:
I can trace the infinite mausezahn mode to here:
lib.sh
1420 __start_traffic()
1421 {
1422 local pktsize=$1; shift
1423 local proto=$1; shift
1424 local h_in=$1; shift # Where the traffic egresses the host
1425 local sip=$1; shift
1426 local dip=$1; shift
1427 local dmac=$1; shift
1428
1429 $MZ $h_in -p $pktsize -A $sip -B $dip -c 0 \
1430 -a own -b $dmac -t "$proto" -q "$@" &
1431 sleep 1
1432 }
but I don't know why in this test it runs away for "veth0.10" in an infinite loop
and is obviously nowhere killed for hours.
I seem to have no new ideas now so I guess I should take a break.
Kind regards,
Mirsad