Hi Vyacheslav,
>On Mon, 2013-01-14 at 16:54 +0100, Sven Eckelmann wrote:
>> The filesystem can end up in a state were the filesystem is full and the
>> returned ss_nongc_ctime is smaller than sui_lastmod of all reclaimable
>> segments. The garbage collector will not clean anything and therefore no new
>> room for new files will be available and ss_nongc_ctime/sui_lastmod will not be
>> updated without using special tools. This makes the filesystem unusable without
>> manual recovery.
>>
>> Signed-off-by: Sven Eckelmann <sven@xxxxxxxxxxxxx>
>> --
>> This problem appeared on a current 3.2 stable kernel (Debian Wheezy build). I
>> am not an FS developer and have therefore not much background knowledge about
>> the NILFS codebase. Nevertheless, this problem hit me quite hard after creating
>> some files on a nilfs partition until it was full and deleting them again.
>>
>> $ for i in `seq 0 150`; do dd if=/dev/zero of=foo$i count=22528; done
>> $ rm foo*
>>
>> Looking at the output debugging output using
>>
>> $ watch -n .5 'df -h;tail /var/log/syslog;'
>>
>> clearly showed that it was not finding any segments to delete. The only problem
>> I could find was the threshold. After "removing" this threshold, I was able to
>> get some clear segments again. I personally cannot explain why the check is
>> there at all. Maybe there is a good reason but the comment above it didn't help
>> much.
>>
>> So, here for completeness the threshold: 1358164666 (aka: Mon Jan 14 12:57:46
>> CET 2013)
>>
>> And here are the output of lssu and lscp:
>>
>> $ lssu --all
>> SEGNUM DATE TIME STAT NBLOCKS
>> 0 2013-01-14 12:58:23 -d- 2047
>> 1 2013-01-14 12:58:23 -d- 2048
>
>[snip]
>
>>
>> Signed-off-by: Sven Eckelmann <sven@xxxxxxxxxxxxx>
>> ---
>> sbin/cleanerd/cleanerd.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/sbin/cleanerd/cleanerd.c b/sbin/cleanerd/cleanerd.c
>> index bfcd893..12ed975 100644
>> --- a/sbin/cleanerd/cleanerd.c
>> +++ b/sbin/cleanerd/cleanerd.c
>> @@ -592,7 +592,7 @@ nilfs_cleanerd_select_segments(struct nilfs_cleanerd *cleanerd,
>> * selected. */
>> thr = (config->cf_selection_policy.p_threshold != 0) ?
>> config->cf_selection_policy.p_threshold :
>> - sustat->ss_nongc_ctime;
>> + ~0ULL;
>>
>
>As I understand the code of nilfs_cleanerd, this code is correct without
>your changing. The ss_nongc_ctime is the creation time of the last
>segment not for GC. When thr is set then it compared with sui_lastmod.
>The sui_lastmod is the timestamp of last modification. So, the
>nilfs_cleanerd works right.
>
>I think that this is a bug on the kernel side. My current vision is that
>in some environment the ns_nongc_ctime can be not updated correctly. So,
>you have such threshold that prevent from segments clearing.
I guess the problem is caused because ss_nongc_ctime is not updated by
the kernel module if dirty blocks created by file operations are written
out through nilfs_clean_segments function. Actually,
nilfs_clean_segments() can roll together normal dirty blocks.
So, if GC is working without interruption, ss_nongc_ctime may not be
updated long time.
If this is the cause of the problem, nilfs2 log writer should be changed
so that nilfs->ns_nongc_ctime is updated with sci->sc_seg_ctime if
there is a file system change (file/directory change or an ifile change).
NILFS_SC_HAVE_DELTA flag may be available for this purpose.
The ss_nongc_ctime check is needed to make GC stoppable.
Without this check, GC can continue endlessly even if the file system
doesn't have any change.
With regards,
Ryusuke Konishi
>Thank you for the issue report.
>
>With the best regards,
>Vyacheslav Dubeyko.
>
>> for (segnum = 0; segnum < sustat->ss_nsegs; segnum += n) {
>> count = (sustat->ss_nsegs - segnum < NILFS_CLEANERD_NSUINFO) ?
>
>
>--
>To unsubscribe from this list: send the line "unsubscribe linux-nilfs" in
>the body of a message to majordomo@xxxxxxxxxxxxxxx
>More majordomo info at http://vger.kernel.org/majordomo-info.html
>
--
To unsubscribe from this list: send the line "unsubscribe linux-nilfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
