Обсуждение: Avoid resource leak (contrib/postgres_fdw/connection.c)

On 16/03/26 08:45, Ranier Vilela wrote:
> Hi.
> 
> Per Coverity.
> 
> CID 1645716: (#1 of 1): Resource leak (RESOURCE_LEAK)
> 8. leaked_storage: Variable str going out of scope leaks the storage
> str.data points to.
> 
> The function *postgres_fdw_connection* leaks the contents of
> var str.data
> Once that function *cstring_to_text* palloc the contents
> must be necessary to free the var str.data.
> 
> patch attached.
> 

Hi,

Thanks for the patch, it looks correct to me. I've searched for this 
pattern `PG_RETURN_TEXT_P(cstring_to_text(.*.data` in other places and 
I've just found on postgres_fdw/connection.c

I've also search for other cases of `return cstring_to_text(...)` 
usages and I didn't found anything that seems suspicious.

Tests are also passing.

--
Matheus Alcantara
EDB: https://www.enterprisedb.com

On Mon, Mar 16, 2026 at 8:46 PM Ranier Vilela <ranier.vf@gmail.com> wrote:
>
> Hi.
>
> Per Coverity.
>
> CID 1645716: (#1 of 1): Resource leak (RESOURCE_LEAK)
> 8. leaked_storage: Variable str going out of scope leaks the storage str.data points to.
>
> The function *postgres_fdw_connection* leaks the contents of
> var str.data
> Once that function *cstring_to_text* palloc the contents
> must be necessary to free the var str.data.

It seems that postgres_fdw_connection() is expected to be called by
ForeignServerConnectionString() via OidFunctionCall3(),
and to allocate memory (including str.data) in the FDWConnectionContext
memory context created by ForeignServerConnectionString().

After calling postgres_fdw_connection(), ForeignServerConnectionString()
deletes FDWConnectionContext. So it seems to me that any memory
allocated in that context, including str.data, would not leak. No?

Regards,

--
Fujii Masao

On 16/03/26 20:07, Fujii Masao wrote:
> On Mon, Mar 16, 2026 at 8:46 PM Ranier Vilela <ranier.vf@gmail.com> wrote:
>>
>> Hi.
>>
>> Per Coverity.
>>
>> CID 1645716: (#1 of 1): Resource leak (RESOURCE_LEAK)
>> 8. leaked_storage: Variable str going out of scope leaks the storage str.data points to.
>>
>> The function *postgres_fdw_connection* leaks the contents of
>> var str.data
>> Once that function *cstring_to_text* palloc the contents
>> must be necessary to free the var str.data.
> 
> It seems that postgres_fdw_connection() is expected to be called by
> ForeignServerConnectionString() via OidFunctionCall3(),
> and to allocate memory (including str.data) in the FDWConnectionContext
> memory context created by ForeignServerConnectionString().
> 
> After calling postgres_fdw_connection(), ForeignServerConnectionString()
> deletes FDWConnectionContext. So it seems to me that any memory
> allocated in that context, including str.data, would not leak. No?
> 

Yeah, this sounds correct, but I'm wondering if the current pfree 
calls on postgres_fdw_connection() is needed?

My previous review was assuming that any allocated memory on 
postgres_fdw_connection() should be free at the end because of these 
calls, but ForeignServerConnectionString() already assume that this 
function leak memory:
    /*
     * GetForeignServer, GetForeignDataWrapper, and the connection function
     * itself all leak memory into CurrentMemoryContext. Switch to a temporary
     * context for easy cleanup.
     */
    tempContext = AllocSetContextCreate(CurrentMemoryContext,
                                        "FDWConnectionContext",
                                        ALLOCSET_SMALL_SIZES);

--
Matheus Alcantara
EDB: https://www.enterprisedb.com