On Wed 06-07-16 14:08:28, Nikolay Borisov wrote:
> On 07/06/2016 01:22 PM, Jan Kara wrote:
> > On Mon 04-07-16 11:49:27, Nikolay Borisov wrote:
> >> Hello again Jan,
> >>
> >> On 06/03/2016 12:19 PM, Jan Kara wrote:
> >>> Hi,
> [SNIP]
> >> The crash again points to test_bit in info_idq_free. I followed
> >> your advise to search for the address and here is what I got:
> >>
> >> crash> search -m ffffffff00000000 d9c01f11
> >>
> >> ffff88000181e030: d9c01927d9c01f11
> >> ffff880996894680: ffffffffd9c01f11
> >> ffff881d5019b858: ffffffffd9c01f11
> >> ffff881d5019b998: ffffffffd9c01f11 - <stack frame of crash_kexec>
> >> ffff881d5019bbe8: ffffffffd9c01f11 - <stack frame of page_fault)
> >> ffffffff8181e030: d9c01927d9c01f11
> >>
> >> So two of the values are in the stack frames of function involved,
> >> in the crash so I'd say they are of no interest. What's interesting
> >> is that ffffffff8181e030 seems to be quota_magics:
> >>
> >> readelf -s vmlinux-4.4.10-clouder1 | grep ffffffff8181e030
> >> 15605: ffffffff8181e030 12 OBJECT LOCAL DEFAULT 4 quota_magics.24849
> >>
> >> #define V2_INITQMAGICS {\
> >> 0xd9c01f11, /* USRQUOTA */\
> >> 0xd9c01927, /* GRPQUOTA */\
> >> 0xd9c03f14, /* PRJQUOTA */\
> >> }
> >>
> >> So it seems that somehow the USRQUOTA magic values overwrites
> >> the dquot pointer. Looking at the code I'm not entirely
> >> sure how this can happen though.
> >
> > This is indeed interesting. Can you dump full struct ext4_inode * of the inode
> > for which dquot_free_inode() was crashing? Command
> >
> > kmem -s ffff880996894680
>
> Unfortunately I can't use that command, since my kernel is not compiled
> with SLUB_DEBUG. However, I was able to retrieve the contents of both
> ext4_inode_info and inode and they seem to be in some semi-freed state.
> I've put the contents of the ext4_inode_info and containing inode here:
>
> http://pastie.org/private/c1ptyg8abgzqbphmb9mng
>
> I also verified that those struct indeed look legitimate since the
> inode->i_sb is indeed the superblock pointer of the fs which crashed.
> And the i_dquot member of the ext4_inode_info holds the USRQUOTA magic
> value:
>
> i_dquot = {0xffffffffd9c01f11, 0x0, 0x0},
OK, so somewhat interesting thing is that the inode is actually a socket
inode. The i_mode is somewhat unusual - 0141226 - i.e. it is S_IFSOCK |
S_ISVTX | S_IWUSR | S_IWGRP | S_IWOTH | S_IROTH - but I guess that may be
fine. Other than that nothing really out of ordinary.
The strange thing is that for socket inode we don't do any real quota
operations anyway. We will just initialize pointers in i_dquot array (to
zeros when allocating inode and later in __dquot_initialize()) and that's
it. So I'm not sure how the corrupted value could get to the i_dquot array.
I have one idea - since the value 0xffffffffd9c01f11 is unique, could you
add checks in __dquot_initialize(), dquot_transfer(), dquot_drop() to see
where the value got into i_dquot array?
Honza
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
