On 08/15/2011 11:24 AM, Ken'ichi Ohmichi wrote:
>
> Hi Mahesh,
>
> On Fri, 15 Jul 2011 14:46:02 +0530
> Mahesh J Salgaonkar <mahesh at linux.vnet.ibm.com> wrote:
>> @@ -4647,6 +4650,15 @@ write_cache_bufsz(struct cache_data *cd)
>> }
>>
>> int
>> +write_cache_zero(struct cache_data *cd, size_t size)
>> +{
>> + memset(cd->buf + cd->buf_size, 0, size);
>> + cd->buf_size += size;
>> +
>> + return write_cache_bufsz(cd);
>> +}
>
> if cd->buf_size is a little smaller than cd->cache_size and
> the argument "size" is bigger than info->page_size, the above
> a SIGSEGV happens at the above memset().
>
> Previous write_cache_bufsze() call will avoid that :
> ---
> int
> write_cache_zero(struct cache_data *cd, size_t size)
> {
> + if (!write_cache_bufsz(cd))
> + return FALSE;
>
> memset(cd->buf + cd->buf_size, 0, size);
> cd->buf_size += size;
>
> return write_cache_bufsz(cd);
> }
> ---
Nice catch. Agree.
>
>> /*
>> + * ELF note section for erase information
>> + *
>> + * According to elf.h the unused values are 0x15(21) through 0xff. The value
>> + * range 0x1XX, 0x2XX and 0x3XX is been used for PPC, i386 and s390
>> + * respectively.
>> + *
>> + * Using 0xff to be on safer side so that any new Elf Note addition in elf.h
>> + * after 0x15 value would not clash.
>> + */
>> +#ifndef NT_ERASE_INFO
>> +#define NT_ERASE_INFO (0xff) /* Contains erased information. */
>> +#endif
>> +#define ERASEINFO_NOTE_NAME "ERASEINFO"
>> +#define ERASEINFO_NOTE_NAME_BYTES (sizeof(ERASEINFO_NOTE_NAME))
>
> I feel NT_ERASE_INFO(0xff) is not safe and we can use ERASEINFO_NOTE_NAME
> instead.
Do you mean having n_type set to '0' and just using
'ERASEINFO_NOTE_NAME' to identify the eraseinfo note? if Yes, then I
think I am ok with that.
Thanks,
-Mahesh.
