Luiz Fernando N. Capitulino skrev:
On Sat, 30 Dec 2006 14:26:52 +0300
Andrey Borzenkov <[EMAIL PROTECTED]> wrote:
| -----BEGIN PGP SIGNED MESSAGE-----
| Hash: SHA1
|
| On Saturday 30 December 2006 13:38, Andrey Borzenkov wrote:
| > On Saturday 30 December 2006 13:13, Thomas Backlund wrote:
| > > Andrey Borzenkov skrev:
| > > > -----BEGIN PGP SIGNED MESSAGE-----
| > > > Hash: SHA1
| > > >
| > > > There was recently long discussion on lkml about mysterious file
| > > > corruptions during some activity (mentioned were - large multifiles
| > > > torrents using rtorrent, packages repository corruption using deb
| > > > packages - I forgot the
| > >
| > > And reported here too:
| > > http://qa.mandriva.com/show_bug.cgi?id=27693
| > >
| > > > distribution). It turned out to be - apparently, very old - race
| > > > between file system and VM writeback code. The commit that fixes it is
| > > > http://git.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=
| > > >co mmit;h=7658cc289288b8ae7dd2c2224549a048431222b3
| > > >
| > > > Now, there was at least one report since then that this commit also
| > > > fixed BDB corruption (NNTP
| > > > Message-ID:<[EMAIL PROTECTED]>) Quoting this:
| > > >
| > > > ===================
| > > > With 2.6.20-rc2-git1, which contain this patch, I have no more Berkeley
| > > > DB corruption with Klibido.¹
| > > > I'm afraid a lot of software project switched to Sqlite,² from BDB,³
| > > > because the bug this patch fix (ie.
| > > > http://bogofilter.sourceforge.net/). I've also thought, since years, it
| > > > was an userland problem.
| > > > ===================
| > > >
| > > > I wonder if this accounts for constant reports about corrupted RPM DB.
| > > > If yes, it probably should be backported to at least 2007 ...
| > >
| > > If IRC the oldest reported kernel triggering this bug is 2.6.5, so ...
| > >
| > > +1
| > >
| > > CC'ing kernel-discuss and Luiz
| >
| > Here is running Linus test program on reiserfs with default mount options
| > under 2.6.19 vanilla:
| >
| > {pts/0}% ./test_vm
| > Writing chunk 71819/71820 (99%)
| > Chunk 18265 corrupted
| > Chunk 18410 corrupted
| > Chunk 18412 corrupted
| > Chunk 18447 corrupted
| > Chunk 18449 corrupted
| > ....
| >
| > I'm going to apply patch and see ...
|
| ... and I am unable to trigger corruption using (adapted) patch. Anyone has
| good procedure for testing RPM DB consistency? :)
|
| Happy New Year to everybody, nice new year present :)
I was aware about the bug but wasn't about the fix (I'm not following
LKML as I should).
Yeah, it was a _long_ and interesting discussion and debbugging around
this bug ;-)
I'll apply the fix, but I've submitted a mdv2007 kernel release already
and QA is finishing the tests, then mdv2007 will have to wait for the
next release (which will happen in a month or so) or the next UC
(which may happen in one week).
Is that ok? Is someone hitting that bug?
I think it's quite OK to make it wait for next uc, so that you can get
the current Security Fixed kernel out without any further delay...
Then people can run the uc kernel if/when they have the problem, just to
provide more testing ;-)
I could try to make a new release for mdv2007 faster, but it'll
be difficult, because there're mdv2006/corpo4 and corpo3 kernels
in the queue...
I think this bug + fix need to be checked on mdv2006/corpo4 and maybe
corpo3 too, since the bug was atleast in 2.6.5 series kernels...
PS: Thanks for CC'ing me Thomas.
No problem...
--
Thomas
