Re: 2.4.18-rc4-aa1 XFS oopses caused by cpio

[Stephen Lord <lord@sgi.com>]

Svetoslav Slavtchev wrote:

>that's what i got
>
>Unable to handle kernel NULL pointer dereference at virtual address
>00000008
>801dba4e
>*pde = 00000000
>Oops: 0000
>CPU:    0
>EIP:    0010:[<801dba4e>]    Not tainted
>Using defaults from ksymoops -t elf32-i386 -a i386
>EFLAGS: 00010046
>eax: 00000120   ebx: 00000000   ecx: 00000282   edx: 00000000
>esi: 8b745a00   edi: 00000008   ebp: 9aa92000   esp: 9ca31904
>ds: 0018   es: 0018   ss: 0018
>Process cpio (pid: 20435, stackpage=9ca31000)
>Stack: 00000282 00000004 8d5f4200 88dc0248 8d65b8c0 801db30d 88dc0248
>00000008
>       00000004 00000001 8d65b980 8d65b8c0 00000004 88dc0248 801db081
>88dc0248
>       8d65b980 8d65b8c0 00000004 8d65ba40 00000000 00000004 00000004
>8b745b20
>Call Trace: [<801db30d>] [<801db081>] [<801db674>] [<8022db0d>]
>[<80209989>]
>   [<8023bc4d>] [<8022dd94>] [<801db3a2>] [<8022e747>] [<80222538>]
>[<8020a412>]
>   [<8021d4b9>] [<80212496>] [<8020f959>] [<802236ac>] [<802299f5>]
>[<80222f71>]
>   [<80234f0e>] [<801d83bd>] [<80220981>] [<8020e103>] [<8022341a>]
>[<8020e103>]
>   [<80227c17>] [<80235098>] [<80235348>] [<80145966>] [<80145a52>]
>[<8010722b>]
>Code: 8b 4a 08 85 c9 74 1b 8d 74 26 00 8d bc 27 00 00 00 00 43 83
>
>>>EIP; 801dba4e <xfs_alloc_mark_busy+3e/d0>   <=====
>>>
>Trace; 801db30c <xfs_alloc_put_freelist+cc/e0>
>Trace; 801db080 <xfs_alloc_fix_freelist+430/480>
>Trace; 801db674 <xfs_alloc_vextent+124/3c0>
>Trace; 8022db0c <_pagebuf_get_object+3c/140>
>Trace; 80209988 <xfs_ialloc_ag_alloc+1d8/810>
>Trace; 8023bc4c <kmem_zone_zalloc+3c/e0>
>Trace; 8022dd94 <_pagebuf_free_object+f4/120>
>Trace; 801db3a2 <xfs_alloc_read_agf+82/230>
>Trace; 8022e746 <pagebuf_rele+66/80>
>Trace; 80222538 <xfs_trans_brelse+d8/e0>
>Trace; 8020a412 <xfs_dialloc+182/900>
>Trace; 8021d4b8 <xfs_mod_incore_sb_batch+48/a0>
>Trace; 80212496 <xfs_inode_item_unlock+76/80>
>Trace; 8020f958 <xfs_ialloc+58/3e0>
>Trace; 802236ac <xfs_dir_ialloc+8c/290>
>Trace; 802299f4 <xfs_mkdir+374/640>
>Trace; 80222f70 <xfs_trans_free_items+30/90>
>Trace; 80234f0e <linvfs_common_cr+1ae/2b0>
>Trace; 801d83bc <xfs_acl_iaccess+2c/90>
>Trace; 80220980 <xfs_trans_reserve+90/160>
>Trace; 8020e102 <xfs_iunlock+32/60>
>Trace; 8022341a <xfs_dir_lookup_int+ba/2c0>
>Trace; 8020e102 <xfs_iunlock+32/60>
>Trace; 80227c16 <xfs_lookup+a6/110>
>Trace; 80235098 <linvfs_lookup+68/c0>
>Trace; 80235348 <linvfs_mkdir+18/20>
>Trace; 80145966 <vfs_mkdir+106/160>
>Trace; 80145a52 <sys_mkdir+92/e0>
>Trace; 8010722a <system_call+32/38>
>Code;  801dba4e <xfs_alloc_mark_busy+3e/d0>
>00000000 <_EIP>:
>Code;  801dba4e <xfs_alloc_mark_busy+3e/d0>   <=====
>   0:   8b 4a 08                  mov    0x8(%edx),%ecx   <=====
>Code;  801dba50 <xfs_alloc_mark_busy+40/d0>
>   3:   85 c9                     test   %ecx,%ecx
>Code;  801dba52 <xfs_alloc_mark_busy+42/d0>
>   5:   74 1b                     je     22 <_EIP+0x22> 801dba70
><xfs_alloc_mark_busy+60/d0>
>Code;  801dba54 <xfs_alloc_mark_busy+44/d0>
>   7:   8d 74 26 00               lea    0x0(%esi,1),%esi
>Code;  801dba58 <xfs_alloc_mark_busy+48/d0>
>   b:   8d bc 27 00 00 00 00      lea    0x0(%edi,1),%edi
>Code;  801dba60 <xfs_alloc_mark_busy+50/d0>
>  12:   43                        inc    %ebx
>Code;  801dba60 <xfs_alloc_mark_busy+50/d0>
>  13:   83 00 00                  addl   $0x0,(%eax)
>

This is this chunk of code:

        for (bsy = mp->m_perag[agno].pagb_list, n = 0;
c01970e2:       31 db                   xor    %ebx,%ebx
c01970e4:       8b b5 e4 01 00 00       mov    0x1e4(%ebp),%esi
c01970ea:       8b 54 06 20             mov    0x20(%esi,%eax,1),%edx
             n < XFS_PAGB_NUM_SLOTS;
             bsy++, n++) {
                if (bsy->busy_tp == NULL) {
c01970ee:       8b 4a 08                mov    0x8(%edx),%ecx

It looks like bsy is NULL, which means either a corruption problem (agno 
too large),
or a memory allocation problem during mount - did you get any error 
messages when
you mounted the filesystem? Had you performed much I/O before this 
happened, or
was the cpio the first thing you did. If this structure is missing then 
chances are you
would not get far before dying. We have already referenced other fields 
in the m_perag[agno]
structure,  so I lean towards pagb_list itself being NULL.

Can you send me the output of mkfs if you still have it, or failing 
that, the output
of xfs_growfs -n /xxx where /xxx is the mounted filesystem.

This chunk of code is new, you may have found a hole in it - although I 
have been running
it for a couple of months here without problem. It could also be 
something specific to
the aa kernel, I have not looked at this merge of XFS.

Steve