Re: Wrong atime on recent kernels

["Petr Titěra" <petr@titera.eu>]

Petr Titěra napsal(a):
> john stultz napsal(a):
>> On Thu, 2009-12-17 at 12:04 +0100, Petr Tit��ra wrote:
>>> john stultz napsal(a):
>>>> On Wed, 2009-12-16 at 21:55 +0100, Petr Tit��ra wrote:
>>>>> john stultz napsal(a):
>>>>>> 2009/12/14 Petr Tit��ra <petr@titera.eu>:
>>>>>>> Hello,
>>>>>>>
>>>>>>> I see some strange file modification times recently. It seems to me
>>>>>>> that in some situations, kernel allows to set nanoseconds part 
>>>>>>> of file
>>>>>>> access, modification or change time to 100000000 ns. Problem 
>>>>>>> seems to be in
>>>>>>> some generic part of kernel because I see it on several different
>>>>>>> filesysytems (ext4 and nilf2). These is I've got during my 
>>>>>>> testing on kernel
>>>>>>> 2.6.32-tip-08309-gad8e75a.
>>>>>>>
>>>>>>> File: `./Documentation/dvb/contributors.txt'
>>>>>>> Size: 3035 Blocks: 8 IO Block: 4096 regular file
>>>>>>> Device: fe04h/65028d Inode: 818 Links: 1
>>>>>>> Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
>>>>>>> Access: 2009-12-14 10:29:04.1000000000 +0100
>>>>>>> Modify: 2009-12-14 10:29:04.1000000000 +0100
>>>>>>> Change: 2009-12-14 10:29:04.1000000000 +0100
>>>>>>>
>>>>>>> See that all times of that file ends with 1e6 nanoseconds.
>>>>> I did not test reverting this patch yet, because I did not find
>>>>> reliable way how to reproduce these strange modify times. But as I
>>>>> read your description. Would it be possible that if there would be 
>>>>> bug
>>>>> in your patch i would be observer on mostly quiet system? I'm asking
>>>>> because full day of testing of the system under load did not produce
>>>>> any result, but then when I tried to run "find / | xargs stat" on 
>>>>> idle
>>>>> system I've got several new instances of wrong access time 
>>>>> (filesystem
>>>>> is mounted without noatime)
>>>> Another quick question:
>>>>
>>>> What is the normal behavior you see when this issue is not cropping 
>>>> up?
>>>>
>>>> Do you normally see all 0's in the ns field? Or do you expect to 
>>>> see an
>>>> actual ns value?
>>>>
>>> Sorry to reply again. Previous message did not get to list:
>>>
>>> I see values which seems to be ns times there. My root filesystem is 
>>> ext4 too (recently I do not remeber if I formated it from scratch 
>>> when I reinstalled that system) but I see this happen on other 
>>> filesystems too
>>>
>>> Root filesystem (ext4 may be converted from ext3)
>>>
>>> File: `/etc/sysconfig'
>>> Size: 4096 Blocks: 8 IO Block: 4096 directory
>>> Device: fe00h/65024d Inode: 65282 Links: 7
>>> Access: (0755/drwxr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root)
>>> Access: 2009-12-16 21:14:00.172000000 +0100
>>> Modify: 2009-12-12 11:01:48.1000000000 +0100
>>> Change: 2009-12-12 11:01:48.1000000000 +0100
>>> File: `/etc/sysconfig/prelink'
>>> Size: 1459 Blocks: 8 IO Block: 4096 regular file
>>> Device: fe00h/65024d Inode: 22706 Links: 1
>>> Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
>>> Access: 2009-12-14 10:27:46.912000002 +0100
>>> Modify: 2004-11-23 11:43:08.000000000 +0100
>>> Change: 2009-12-08 22:57:24.656000002 +0100
>>> File: `/etc/sysconfig/i18n'
>>> Size: 47 Blocks: 8 IO Block: 4096 regular file
>>> Device: fe00h/65024d Inode: 48962 Links: 1
>>> Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
>>> Access: 2010-08-27 18:07:21.500013018 +0200
>>> Modify: 2009-06-22 23:33:43.113581313 +0200
>>> Change: 2009-06-22 23:58:39.936318201 +0200
>>
>> So I'm not reproducing this with 2.6.33-rc1 on a fresh ext4 partition on
>> x68_64.
>>
>> File: `virt'
>> Size: 4096 Blocks: 8 IO Block: 4096 directory
>> Device: 804h/2052d Inode: 1868440 Links: 3
>> Access: (0755/drwxr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root)
>> Access: 2009-12-17 21:22:44.692710730 -0500
>> Modify: 2009-12-17 20:14:40.000000000 -0500
>> Change: 2009-12-17 21:20:21.001915208 -0500
>> File: `vmlinux'
>> Size: 21122497 Blocks: 24136 IO Block: 4096 regular file
>> Device: 804h/2052d Inode: 1874435 Links: 1
>> Access: (0755/-rwxr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root)
>> Access: 2009-12-17 21:22:05.381691121 -0500
>> Modify: 2009-12-17 21:22:05.376691754 -0500
>> Change: 2009-12-17 21:22:05.376691754 -0500
>> File: `vmlinux.o'
>> Size: 16701780 Blocks: 32624 IO Block: 4096 regular file
>> Device: 804h/2052d Inode: 1874418 Links: 1
>> Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
>> Access: 2009-12-17 21:22:01.138228732 -0500
>> Modify: 2009-12-17 21:22:01.131229619 -0500
>> Change: 2009-12-17 21:22:01.131229619 -0500
>>
>>
>> Let me know if you find anything that helps narrow this down.
>>
> Hello,
>
> I know its far fetched, but is there something what is preventing 
> xtime.tv_nsec to be exactly 999999999 near the end of update_wall_time 
> in kernel/time/timekeeping.c?
>
> Petr
>

Just to follow up. I'm asking because I see a lot of files with access 
and/or modify times near the top of thousanth of second (see 
`/etc/sysconfig/prelink' in my example) and I thing that addition of 1 
to xtime.tv_nsec ath the end of update_wall_time can 'owerflow' to whole 
second.

Petr
>
>> thanks
>> -john
>>
>>
>>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe 
> linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/