[CRIU] [PATCHv3 14/30] page-xfer: dump compatible iovec

[Pavel Emelyanov]

On 07/05/2016 02:47 PM, Dmitry Safonov wrote:
> On 07/05/2016 02:35 PM, Pavel Emelyanov wrote:
>>
>>>>> ========================== Run zdtm/static/env00 in h ==========================
>>>>> Start test
>>>>> ./env00 --pidfile=env00.pid --outfile=env00.out --envname=ENV_00_TEST
>>>>> Run criu pre-dump
>>>>> Run criu pre-dump
>>>>> Run criu dump
>>>>> Run criu restore
>>>>> =[log]=> dump/zdtm/static/env00/24/3/restore.log
>>>>> ------------------------ grep Error ------------------------
>>>>> (00.035223)     24: Error (mem.c:826): Page entry address 400000 outside of VMA 7fffddbcf000-7fffddbd1000
>>>>> (00.035323) Error (cr-restore.c:1885): Restoring FAILED.
>>>>> ------------------------ ERROR OVER ------------------------
>>>>> ################# Test zdtm/static/env00 FAIL at CRIU restore ##################
>>>>> ##################################### FAIL #####################################
>>>>>
>>>>> https://github.com/xemul/criu/issues/185
>>>>
>>>> The pagemap image for test becomes unsorted.
>>>
>>> Yes, thanks, I have reproduced and also found image unsorted.
>>> Working on fix.
>>
>>
>> diff --git a/criu/page-xfer.c b/criu/page-xfer.c
>> index 8c2c799..967a587 100644
>> --- a/criu/page-xfer.c
>> +++ b/criu/page-xfer.c
>> @@ -370,7 +370,7 @@ int page_xfer_dump_pages(struct page_xfer *xfer, struct page_pipe *pp,
>>                                 struct iovec hole = get_iov(pp->holes, cur_hole,
>>                                                 pp->compat_iov);
>>
>> -                               if (hole.iov_base < iov.iov_base)
>> +                               if (hole.iov_base >= iov.iov_base)
>>                                         break;
>>                                 ret = page_xfer_dump_hole(xfer, &hole, off);
>>                                 if (ret)
>>
>> Is it just like this?
> 
> 
> And while we on it, don't you mind, if I change:
> static struct iovec get_iov(...)
> to this:
> static struct iovec *get_iov(...)

>From my perspective if we want optimization like this we'd better do 

static void get_iov(struct iov *into, ...);

;)

-- Pavel

> And introduce put_iov(struct iovec *, bool compat)
> to eleminate needless copy here.
> On the moment of patch I wanted to return pointer to static
> iovec_compat, but found the need in reentrancy, so left
> coping here, but it's pretty annoing.
> .
>