Re: [RFC PATCH 1B] credential-cache--daemon: close input before exit-ing

[Ramsay Jones <ramsay@ramsayjones.plus.com>]

On 07/07/2022 14:00, Ævar Arnfjörð Bjarmason wrote:
> 
> On Thu, Jul 07 2022, Ramsay Jones wrote:
> 
>> Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
>> ---
>>  builtin/credential-cache--daemon.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/builtin/credential-cache--daemon.c b/builtin/credential-cache--daemon.c
>> index 4c6c89ab0d..556393498f 100644
>> --- a/builtin/credential-cache--daemon.c
>> +++ b/builtin/credential-cache--daemon.c
>> @@ -138,6 +138,7 @@ static void serve_one_client(FILE *in, FILE *out)
>>  		 * process actually ends, which closes the socket and gives
>>  		 * them EOF.
>>  		 */
>> +		fclose(in);
>>  		exit(0);
>>  	}
>>  	else if (!strcmp(action.buf, "erase"))
> 
> This is called by a function that will also close stdout, shouldn't we
> be closing it here?

Err, do you mean 'out' rather than 'stdout'? Actually, it doesn't
matter either way! :)

Let me quote from [1], which (hopefully) explains the context:

"""
Anyway, I started playing around with flushing/closing of 'FILE *' streams
before the 'exit' call, to change the order, relative to the socket-file
deletion in the 'atexit' function (or the closing of the listen-socket
descriptor, come to that). In particular, I found that if I were to close
the 'in'put stream, then the client would receive an EOF and exit normally
(ie no error return from read_in_full() above).

[fclose(in); fclose(out) also works, but fclose(out) on its own does not.
fflush() in various combinations did not work at all].
"""

So, my intent was to find the _minimum_ change necessary to fix the test.
This is not a 'real' patch - it was only to demonstrate that there appears
to be some order dependency problem induced by the 'atexit' clean-up from
exit()-ing the server.

ATB,
Ramsay Jones

[1] https://lore.kernel.org/git/9dc3e85f-a532-6cff-de11-1dfb2e4bc6b6@ramsayjones.plus.com/