On 11/12/2014 08:05 PM, Minfei Huang wrote:
> On 11/11/14 at 04:54pm, Minfei Huang wrote:
>> If the page pfn calculated by read_from_splitblock_table is bigger than
>> pfn_needed, the variable pfn_needed will leak.
>>
>> makedumpfile cannot assign the pfn averagely bacause of condition
>> pfn_needed invalid.
>>
>> Signed-off-by: Minfei Huang<mhuang at redhat.com>
>> ---
>> makedumpfile.c | 5 +++--
>> 1 file changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/makedumpfile.c b/makedumpfile.c
>> index 59c4952..8807a90 100644
>> --- a/makedumpfile.c
>> +++ b/makedumpfile.c
>> @@ -8415,7 +8415,7 @@ calculate_end_pfn_by_splitblock(mdf_pfn_t start_pfn,
>> return info->max_mapnr;
>>
>> mdf_pfn_t end_pfn;
>> - long long pfn_needed, offset;
>> + long long pfn_needed, offset, per_splitblock_pfn;
>> char *splitblock_value_offset;
>>
>> pfn_needed = info->num_dumpable / info->num_dumpfile;
>> @@ -8424,7 +8424,8 @@ calculate_end_pfn_by_splitblock(mdf_pfn_t start_pfn,
>> end_pfn = start_pfn;
>>
>> while (*cur_splitblock_num< splitblock->num&& pfn_needed> 0) {
>> - pfn_needed -= read_from_splitblock_table(splitblock_value_offset);
>> + per_splitblock_pfn = read_from_splitblock_table(splitblock_value_offset);
>> + pfn_needed = pfn_needed< per_splitblock_pfn ? 0 : pfn_needed - per_splitblock_pfn;
> Hi, Wenjiang!
>
> Sorry, my emall client didnot receive the mail by you, so I reply it
> here.
>
> The split->table is an array to record the pfn count which we need dump.
> And the memory is divided by the size of info->splitblock_size, the pfn
> count in each entry will be stored in the entry of split->table.
>
> For the purpose, we want to average allocation to the pfns.
>
> Here is a case:
> There are 5 entries in the split->table, and the value is 4, 6, 4, 5, 5.
> We want to split four pieces to write to four files, and every file will
> be write 6 pfns.
> Using the function calculate_end_pfn_by_splitblock, the first file will
> be assigned all of the pfns(24 pfns).
>
> pfn_needed is 6((4+6+4+5+5) / 4) when entry the function.
> By the end of first loop, the variable pfn_needed = 2(6 - 4).
> Then move on, the variable will leak to become huge number by the end of
> second loop, because the return value of function read_from_splitblock_table
> is 6. The variable pfn_needed is ~4(2 - 6), it is huge.
> And the loop will break util *cur_splitblock_num == splitblock->num.
pfn_needed is (long long). If it is -4, won't the loop break?
>
> For above case, it is no sense to use split function, so we should add
> the condition whether pfn_needed is bigger than the value of
> read_from_splitblock_table.
>
> Thanks
> Minfei
>> splitblock_value_offset += splitblock->entry_size;
>> ++*cur_splitblock_num;
>> }
>> --
>> 1.8.3.1
>>
--
Thanks
Zhou Wenjian
