[Gluster-devel] catching unitialized structures
jdarcy at redhat.com
Tue Apr 30 14:24:13 UTC 2013
On 04/30/2013 07:31 AM, Vijay Bellur wrote:
> On 04/30/2013 06:10 AM, Jeff Darcy wrote:
>> On 04/29/2013 01:38 PM, Vijay Bellur wrote:
>>> There are a lot of places where we make an
>>> implicit assumption that GF_CALLOC and the likes memset the memory area
>>> to zero.
>> Actually I was a bit disturbed recently when I found that at least one
>> member of that family (don't remember which) *doesn't* do that. I meant
>> to go through and check which code relied on that unmet assumption, but
>> then something else came up and I never got back to it.
> I did a quick scan of mem-pool.c but failed to notice anything obvious. If you
> happen to notice it again, it should be worth a patch.
Mem-pool does zero memory, but none of GF_CALLOC, GF_MALLOC, or GF_REALLOC do.
For example, GF_CALLOC just calls __gf_calloc which just calls calloc.
108 req_size = nmemb * size;
109 tot_size = req_size + GF_MEM_HEADER_SIZE + GF_MEM_TRAILER_SIZE;
111 ptr = calloc (1, tot_size);
The only thing after that is gf_mem_set_acct_info, which doesn't (and
shouldn't) zero memory. Nonetheless, the assumption you refer to is common in
code which calls these functions. Code which has been converted to use
mem-pool is safe, but that leaves a lot of code which is still unsafe.
More information about the Gluster-devel