On Tue, Aug 20, 2013 at 04:04:18PM -0700, Andrew Morton wrote:
>On Tue, 20 Aug 2013 14:54:53 +0800 Wanpeng Li <liwanp@xxxxxxxxxxxxxxxxxx> wrote:
>
>> preallocate_pmds will continue to preallocate pmds even if failure
>> occurrence, and then free all the preallocate pmds if there is
>> failure, this patch fix it by stop preallocate if failure occurrence
>> and go to free path.
>>
>> ...
>>
>> --- a/arch/x86/mm/pgtable.c
>> +++ b/arch/x86/mm/pgtable.c
>> @@ -196,21 +196,18 @@ static void free_pmds(pmd_t *pmds[])
>> static int preallocate_pmds(pmd_t *pmds[])
>> {
>> int i;
>> - bool failed = false;
>>
>> for(i = 0; i < PREALLOCATED_PMDS; i++) {
>> pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP);
>> if (pmd == NULL)
>> - failed = true;
>> + goto err;
>> pmds[i] = pmd;
>> }
>>
>> - if (failed) {
>> - free_pmds(pmds);
>> - return -ENOMEM;
>> - }
>> -
>> return 0;
>> +err:
>> + free_pmds(pmds);
>> + return -ENOMEM;
>> }
>
Hi Andrew,
>Nope. If the error path is taken, free_pmds() will free uninitialised
>items from pmds[], which is a local in pgd_alloc() and contains random
>stack junk. The kernel will crash.
>
>You could pass an nr_pmds argument to free_pmds(), or zero out the
>remaining items on the error path. However, although the current code
>is a bit kooky, I don't see that it is harmful in any way.
>
There is a check in free_pmds():
if (pmds[i])
free_page((unsigned long)pmds[i]);
which will avoid the issue you mentioned.
In addition, the codes in pgd_alloc will skip free pmds if preallocate pmds
failure which will avoid free pmds twice. Am I miss something? ;-)
Regards,
Wanpeng Li
>> Reviewed-by: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
>
>Ahem.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>