Glibc bug

[Armin]

$ thunderbird/thunderbird &
[2] 22682
null:profile-after-change
added
added
[object ChromeWindow]:mail-startup-done
thunderbird: res_query.c:251: __libc_res_nquery: Assertion `hp != hp2' failed.

It is mentioned here:

​http://sourceware.org/bugzilla/show_bug.cgi?id=13013

And as it says, it just happens under some conditions. This is at least 20th time today I started thunderbird and it just crashed once with that error.

Archlinux has a patch for it:

​https://projects.archlinux.org/svntogit/packages.git/plain/trunk/glibc-2.15-fix-res_query-assert.patch?h=packages/glibc

[Armin]

And now even Firefox crashed with the very same error.

[bdubbs@…]

Armin,

That patch is for 2,15 and doesn't apply. I made a new patch by hand (attached). Can you see if it works for you.

[Armin]

I am unable to reproduce it. It just happened twice last night. Not anymore ...I am not sure what triggers it.

[ken@…]

Apparently, it's been around since 2.14 (if what you now see is indeed the same bug). The arch patch *does* apply to 2.16.0 for me. Back in the old days of glibc maintenance, the testcase (using particular nameservers) apparently didn't trigger it.

[bdubbs@…]

Replying to ken@…:

The arch patch *does* apply to 2.16.0 for me.

The patch from the description that says glibc-2.15-fix-res_query-assert.patch or the attached patch? I attached the updated patch that I did by hand. There were some significant changes in the 2.16 code that prevented the 2.15 patch from applying.

[ken@…]

Replying to bdubbs@…:

Replying to ken@…:

The arch patch *does* apply to 2.16.0 for me.

The patch from the description that says glibc-2.15-fix-res_query-assert.patch or the attached patch? I attached the updated patch that I did by hand. There were some significant changes in the 2.16 code that prevented the 2.15 patch from applying.

The one from arch, glibc-2.15-fix-res_query-assert.patch. Yours is created as -p0, and the only differences I can see are formatting at about line 248 hp2 = hp; (yours is not aligned with the other lines), and you kept in hp2 = hp; after the asset, before adding hp2 = (HEADER *) *answerp2; (the compiler will probably optimise theat first hp2= away).

[bdubbs@…]

Fixed at revision 9965. This was a very tricky bug. The resolver code was not handling problem DNS packets properly. It was very intermittent and depended on the name server and even then didn't always fail.