[Rpm-maint] rpm4.14 makes perl-RPM4 testsuite to segfault
Thierry Vignaud
thierry.vignaud at gmail.com
Thu Oct 5 11:59:48 UTC 2017
On 5 October 2017 at 13:27, Thierry Vignaud <thierry.vignaud at gmail.com> wrote:
>>>>>>>>> Yeah, I'm getting segfaults all the way to rpm 4.11.x, didn't test
>>>>>>>>> earlier because this already shows it's not a regression in 4.14.x
>>>>>>>>> but
>>>>>>>>> something else. A bug in perl-RPM4 perhaps, as compiling it with -Og
>>>>>>>>> makes
>>>>>>>>> the crash go away, other optimization levels make it blow up with
>>>>>>>>> different
>>>>>>>>> levels of spectacular. I dont see anything obvious in there but that
>>>>>>>>> doesn't
>>>>>>>>> mean much, I know diddly about perl and its extensions.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> I ran it with some added debugging on rpm side which I'm more
>>>>>>>> familiar
>>>>>>>> with, and the crash happens because a totally garbage pointer is
>>>>>>>> passed
>>>>>>>> to
>>>>>>>> headerFree(). Well indeed, it was passing the address of the header
>>>>>>>> pointer
>>>>>>>> variable as the header itself into the callback, and when you try do
>>>>>>>> stuff
>>>>>>>> with that, well...
>>>>>>>>
>>>>>>>> This fixes it:
>>>>>>>>
>>>>>>>> diff --git a/src/RPM4.xs b/src/RPM4.xs
>>>>>>>> index 04c65ee..6604477 100644
>>>>>>>> --- a/src/RPM4.xs
>>>>>>>> +++ b/src/RPM4.xs
>>>>>>>> @@ -246,7 +246,7 @@ static void *
>>>>>>>> s_what = "INST_START";
>>>>>>>> if (h) {
>>>>>>>> mXPUSHs(newSVpv("header", 0));
>>>>>>>> - mXPUSHs(sv_setref_pv(newSVpvs(""), bless_header,
>>>>>>>> &h));
>>>>>>>> + mXPUSHs(sv_setref_pv(newSVpvs(""), bless_header,
>>>>>>>> h));
>>>>>>>> #ifdef HDRPMMEM
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Oh and you'll want to fix the debug printf too, even though it's
>>>>>>> obviously
>>>>>>> harmless (but then useless for debugging):
>>>>>>>
>>>>>>>> PRINTF_NEW(bless_header, &h, -1);
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ^^
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> Blech, one of those days...
>>>>>>
>>>>>> The above is closer to mark but still not quite right and will crash
>>>>>> too,
>>>>>> only in a different way because the refcount is wrong. Here's the real
>>>>>> deal:
>>>>>>
>>>>>> diff --git a/src/RPM4.xs b/src/RPM4.xs
>>>>>> index 04c65ee..f7cfd33 100644
>>>>>> --- a/src/RPM4.xs
>>>>>> +++ b/src/RPM4.xs
>>>>>> @@ -246,9 +246,9 @@ static void *
>>>>>> s_what = "INST_START";
>>>>>> if (h) {
>>>>>> mXPUSHs(newSVpv("header", 0));
>>>>>> - mXPUSHs(sv_setref_pv(newSVpvs(""), bless_header, &h));
>>>>>> + mXPUSHs(sv_setref_pv(newSVpvs(""), bless_header,
>>>>>> headerLink(h)));
>>>>>> #ifdef HDRPMMEM
>>>>>> - PRINTF_NEW(bless_header, &h, -1);
>>>>>> + PRINTF_NEW(bless_header, h, -1);
>>>>>> #endif
>>>>>> }
>>>>>> break;
>>>>>>
>>>>>> - Panu -
>>>>>>
>>>>>
>>>>> Thanks
>>>>> Now remains the issue with several builds on the same spec object
>>>>>
>>>>
>>>> Yeah, I was just about to write about that.
>>>>
>>>> I totally misunderstood the case initially (that's what happens if you
>>>> try
>>>> to understand the world by peeking through a keyhole), I thought your
>>>> testcase was looking at the number of files by itself and finding one
>>>> more
>>>> than expected. When the error actually comes from rpm itself.
>>>
>>>
>>> Yeah we had a misunderstanding there :-)
>>>
>>>> That file entries get messed when reusing the same spec is probably an
>>>> ages
>>>> old issue, it's just that until now rpm didn't sanity check the file
>>>> data.
>>>>
>>>> The right thing to do is to reload the spec, I doubt it ever *really*
>>>> worked
>>>> correctly.
>>>
>>>
>>> OK I'll commit my fix in the VC then.
>>> There's still one last testsuite failure with RPM4 when verifying a
>>> rpmdb after rebuilding it :
>>>
>>> $ LC_ALL=C perl -Iblib/lib -Iblib/arch t/05transaction.t 2>&1|grep -A2
>>> -B36 "not ok"
>>> error: rpmdb: Enhancename: No such file or directory
>>> error: db5 error(2) from db->verify: No such file or directory
>>> (...)
>>> error: rpmdb: Name: No such file or directory
>>> error: db5 error(2) from db->verify: No such file or directory
>>> not ok 4 - Verify empty
>>> # Failed test 'Verify empty'
>>> # at t/05transaction.t line 24.
>>>
>>> The initdb creates the proper files, but after rebuilddb, there's only
>>> "Packages" (+ the __db.000 env)
>>> This used to work.
>>> Any idea?
>>>
>>
>> Yeah, see
>> https://github.com/rpm-software-management/rpm/commit/b62d85d78b07d6d43ca26fd7cda489c636a5796b
>> :)
>>
>> (and yes that'll be in 4.14.0 too)
>>
>> - Panu -
>
> Thanks.
For the record
BTW, the latest issue I've with rpm-4.14 is that its testsuite has
several debugsources related failures when building it with rpm-4.14
tpoo
http://pkgsubmit.mageia.org/uploads/done/cauldron/core/release/20171005114000.tv.duvel.20120/rpm-4.14.0-0.rc2.4.mga7/build.0.20171005114003.log
Have you tried building rpm-4.14 with rpm-4.14?
ERROR: All 384 tests were run,
19 failed (2 expected failures).
Here's the summary of the failures :
## ------------------------ ##
## Summary of the failures. ##
## ------------------------ ##
Failed tests:
rpm 4.14.0-rc2 test suite test groups:
NUM: FILE-NAME:LINE TEST-GROUP-NAME
KEYWORDS
107: rpmbuild.at:23 rpmbuild -ba *.spec
build
108: rpmbuild.at:41 rpmbuild --rebuild
build
109: rpmbuild.at:54 rpmbuild --short-circuit -bl
build
111: rpmbuild.at:89 rpmbuild -tb
build
120: rpmbuild.at:326 Dependency generation
build
130: rpmbuild.at:761 rpmbuild debuginfo no gdb index included
build debuginfo gdb
131: rpmbuild.at:789 rpmbuild debuginfo unique debug src dir
build debuginfo
132: rpmbuild.at:828 rpmbuild debuginfo no unique debug src dir
build debuginfo
142: rpmbuildid.at:57 rpmbuild buildid alldebug
build debuginfo buildid
143: rpmbuildid.at:149 rpmbuild buildid alldebug unique debug names
build debuginfo buildid
144: rpmbuildid.at:240 rpmbuild buildid separate
build debuginfo buildid
145: rpmbuildid.at:331 rpmbuild buildid separate unique debug names
build debuginfo buildid
146: rpmbuildid.at:421 rpmbuild buildid compat
build debuginfo buildid
147: rpmbuildid.at:525 rpmbuild buildid compat unique debug names
build debuginfo buildid
155: rpmbuildid.at:1153 rpmbuild buildid unique r1 r2
build debuginfo buildid
156: rpmbuildid.at:1201 rpmbuild buildid non-unique r1 r2
build debuginfo buildid
369: rpmpython.at:70 spec parse
python
More information about the Rpm-maint
mailing list