[Devel] [PATCH] fs/fuse kio: skip iostat count for unknown req types

[Pavel Butsykin]

On 31.05.2019 19:06, Kirill Tkhai wrote:
> On 31.05.2019 12:23, Pavel Butsykin wrote:
>> There are still some number of response req types:
>> PCS_CS_WRITE_ZERO_RESP
>> PCS_CS_WRITE_HOLE_RESP
>> PCS_CS_REPLICATEX_RESP
>> PCS_CS_FIEMAP_RESP
>> PCS_CS_MAP_PROP_RESP
>>
>> It makes no sense to count statistics for such requests, so for them the stat
>> count will simply be skipped.
>>
>> Signed-off-by: Pavel Butsykin <pbutsykin at virtuozzo.com>
>> ---
>>   fs/fuse/kio/pcs/fuse_stat.c | 7 ++++---
>>   1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/fuse/kio/pcs/fuse_stat.c b/fs/fuse/kio/pcs/fuse_stat.c
>> index 06be3c75240c..31ff73d6c3dc 100644
>> --- a/fs/fuse/kio/pcs/fuse_stat.c
>> +++ b/fs/fuse/kio/pcs/fuse_stat.c
>> @@ -654,7 +654,6 @@ struct fuse_val_stat *req_stat_entry(struct pcs_fuse_io_stat *io, u32 type)
>>   		default:
>>   			break;
>>   	}
>> -	WARN_ON_ONCE(1);
>>   	return NULL;
>>   }
>>   
>> @@ -663,9 +662,11 @@ static void fuse_iostat_count(struct pcs_fuse_io_stat_sync *iostat,
>>   {
>>   	struct fuse_val_stat *se;
>>   
>> -	write_seqlock(&iostat->seqlock);
>>   	se = req_stat_entry(iostat->CURR(iostat), type);
> 
> Shouldn't iostat->CURR(iostat) be obtained under seqlock?
> 

Yep, iostat->CURR(iostat) should be under seqlock, thanks for fixup!