Test that PASSes but should not

Test that PASSes but should not

[Kris Van Hees]

This is not a good thing:

test/unittest/ip/tst.ipv4localicmp.sh: Running timeout --signal=TERM 41 test/unittest/ip/tst.ipv4localicmp.sh /home/kvanhees/dtrace-bpf-user/build/dtrace
PASS.
-- @@stderr --
BPF: func#0 @0
BPF: func#1 @154
BPF: func#2 @1121
BPF: func#3 @1136
BPF: func#4 @1350
BPF: func#5 @1355
BPF: func#6 @1438
BPF: func#7 @1453
BPF: func#8 @1667
BPF: func#9 @1706
BPF: func#10 @1746
BPF: func#11 @1926
BPF: func#12 @1995
BPF: func#13 @2064
BPF: func#14 @2074
BPF: Tracing programs must provide btf_id
BPF: verification time 33 usec
BPF: stack depth 0+0+0+0+0+0+0+0+0+0+0+0+0+0+0
BPF: processed 0 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
dtrace: could not enable tracing: BPF program load for 'fbt:ipv6:ip6_finish_output:entry' failed: Invalid argument

That failure definitely should cause the test to fail.

Re: Test that PASSes but should not

[Eugene Loh]

And yet it has a .r file?

On 7/24/25 13:25, Kris Van Hees wrote:
> This is not a good thing:
>
> test/unittest/ip/tst.ipv4localicmp.sh: Running timeout --signal=TERM 41 test/unittest/ip/tst.ipv4localicmp.sh /home/kvanhees/dtrace-bpf-user/build/dtrace
> PASS.
> -- @@stderr --
> BPF: func#0 @0
> BPF: func#1 @154
> BPF: func#2 @1121
> BPF: func#3 @1136
> BPF: func#4 @1350
> BPF: func#5 @1355
> BPF: func#6 @1438
> BPF: func#7 @1453
> BPF: func#8 @1667
> BPF: func#9 @1706
> BPF: func#10 @1746
> BPF: func#11 @1926
> BPF: func#12 @1995
> BPF: func#13 @2064
> BPF: func#14 @2074
> BPF: Tracing programs must provide btf_id
> BPF: verification time 33 usec
> BPF: stack depth 0+0+0+0+0+0+0+0+0+0+0+0+0+0+0
> BPF: processed 0 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
> dtrace: could not enable tracing: BPF program load for 'fbt:ipv6:ip6_finish_output:entry' failed: Invalid argument
>
> That failure definitely should cause the test to fail.
>

Re: Test that PASSes but should not

[Kris Van Hees]

On Thu, Jul 24, 2025 at 02:21:36PM -0400, Eugene Loh wrote:
> And yet it has a .r file?

Yes, the problem is less with the test and more with something going wrong with
BTF handling causing the probe to not have a BTF ID and yet it is trying to
load a program for it.  Attachment should not even be tried in that case.

Working out what is wrong and fixing.

> On 7/24/25 13:25, Kris Van Hees wrote:
> > This is not a good thing:
> > 
> > test/unittest/ip/tst.ipv4localicmp.sh: Running timeout --signal=TERM 41 test/unittest/ip/tst.ipv4localicmp.sh /home/kvanhees/dtrace-bpf-user/build/dtrace
> > PASS.
> > -- @@stderr --
> > BPF: func#0 @0
> > BPF: func#1 @154
> > BPF: func#2 @1121
> > BPF: func#3 @1136
> > BPF: func#4 @1350
> > BPF: func#5 @1355
> > BPF: func#6 @1438
> > BPF: func#7 @1453
> > BPF: func#8 @1667
> > BPF: func#9 @1706
> > BPF: func#10 @1746
> > BPF: func#11 @1926
> > BPF: func#12 @1995
> > BPF: func#13 @2064
> > BPF: func#14 @2074
> > BPF: Tracing programs must provide btf_id
> > BPF: verification time 33 usec
> > BPF: stack depth 0+0+0+0+0+0+0+0+0+0+0+0+0+0+0
> > BPF: processed 0 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
> > dtrace: could not enable tracing: BPF program load for 'fbt:ipv6:ip6_finish_output:entry' failed: Invalid argument
> > 
> > That failure definitely should cause the test to fail.
> > 

Re: Test that PASSes but should not

[Kris Van Hees]

On Thu, Jul 24, 2025 at 02:34:00PM -0400, Kris Van Hees wrote:
> On Thu, Jul 24, 2025 at 02:21:36PM -0400, Eugene Loh wrote:
> > And yet it has a .r file?
> 
> Yes, the problem is less with the test and more with something going wrong with
> BTF handling causing the probe to not have a BTF ID and yet it is trying to
> load a program for it.  Attachment should not even be tried in that case.
> 
> Working out what is wrong and fixing.

Patch posted.  Was an issue with the BTF ID lookup code handling modules
whose BTF data didn't get initialized yet.

Re: Test that PASSes but should not

[Eugene Loh]

On 7/24/25 14:57, Kris Van Hees wrote:

> On Thu, Jul 24, 2025 at 02:34:00PM -0400, Kris Van Hees wrote:
>> On Thu, Jul 24, 2025 at 02:21:36PM -0400, Eugene Loh wrote:
>>> And yet it has a .r file?
>> Yes, the problem is less with the test and more with something going wrong with
>> BTF handling causing the probe to not have a BTF ID and yet it is trying to
>> load a program for it.  Attachment should not even be tried in that case.
>>
>> Working out what is wrong and fixing.
> Patch posted.  Was an issue with the BTF ID lookup code handling modules
> whose BTF data didn't get initialized yet.

But how could it dump all that stuff and pass matching against a .r file?

Re: Test that PASSes but should not

[Kris Van Hees]

On Thu, Jul 24, 2025 at 04:54:19PM -0400, Eugene Loh wrote:
> On 7/24/25 14:57, Kris Van Hees wrote:
> 
> > On Thu, Jul 24, 2025 at 02:34:00PM -0400, Kris Van Hees wrote:
> > > On Thu, Jul 24, 2025 at 02:21:36PM -0400, Eugene Loh wrote:
> > > > And yet it has a .r file?
> > > Yes, the problem is less with the test and more with something going wrong with
> > > BTF handling causing the probe to not have a BTF ID and yet it is trying to
> > > load a program for it.  Attachment should not even be tried in that case.
> > > 
> > > Working out what is wrong and fixing.
> > Patch posted.  Was an issue with the BTF ID lookup code handling modules
> > whose BTF data didn't get initialized yet.
> 
> But how could it dump all that stuff and pass matching against a .r file?

No idea - my priority right now was fixing the issue that caused the BPF
verifier error report.  We can look at why it was deemed to PASS even with
that BPF verifer complaint - that is secondary.