Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab breaks Smack TCP connections

Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab breaks Smack TCP connections

[Casey Schaufler]

Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab 'net: correct
sk_acceptq_is_full()' breaks a system with the Smack LSM.
Reverting this change results in a return to correct behavior.

The Smack testsuite can be found at:
	https://github.com/smack-team/smack-testsuite.git

The failing test is ipv4-tcp-local-peersec.sh, but it seems
that most TCP connections hang with SYN_SENT. Oddly, ssh
to 127.0.0.1 works, but other TCP connections timeout.

Re:Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab breaks Smack TCP connections

[刘亚灿]

Hi Casev:

A quote from the listen(2) man page on my Ubuntu system:
The backlog argument defines the maximum length to which 
the queue of pending connections for sockfd may grow.
I think this implies that the 'backlog' must be greater than zero.
In the test source file (tools/smack-ipv4-tcp-peersec.c) Line 60
I found the following code:
if (listen(firstsock, 0) < 0) {
	    printf("%s-listen\n", argv[0]);
	    exit(1);
}
That means that sock will not accept any requests, 
so client TCP connections hang with SYN_SENT.
In openssh case, it use SSH_LISTEN_BACKLOG as 128.

At 2021-03-30 23:42:04, "Casey Schaufler" <casey@schaufler-ca.com> wrote:
>Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab 'net: correct
>sk_acceptq_is_full()' breaks a system with the Smack LSM.
>Reverting this change results in a return to correct behavior.
>
>The Smack testsuite can be found at:
>	https://github.com/smack-team/smack-testsuite.git
>
>The failing test is ipv4-tcp-local-peersec.sh, but it seems
>that most TCP connections hang with SYN_SENT. Oddly, ssh
>to 127.0.0.1 works, but other TCP connections timeout.
>
>
>
>



 


 

Re: Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab breaks Smack TCP connections

[Casey Schaufler]

On 3/30/2021 7:44 PM, 刘亚灿 wrote:
> Hi Casev:
>
> A quote from the listen(2) man page on my Ubuntu system:
> The backlog argument defines the maximum length to which 
> the queue of pending connections for sockfd may grow.
> I think this implies that the 'backlog' must be greater than zero.
> In the test source file (tools/smack-ipv4-tcp-peersec.c) Line 60
> I found the following code:
> if (listen(firstsock, 0) < 0) {
> 	    printf("%s-listen\n", argv[0]);
> 	    exit(1);
> }
> That means that sock will not accept any requests, 
> so client TCP connections hang with SYN_SENT.

Interesting. Prior to this change the code above was
accepting connections. I also tried code that uses a
backlog of 0 on a system without an LSM and discovered
the same behavior. That is, it accepted connections
with a 0 backlog before the change, and hangs after.

Is this a bug fix?

> In openssh case, it use SSH_LISTEN_BACKLOG as 128.
>
> At 2021-03-30 23:42:04, "Casey Schaufler" <casey@schaufler-ca.com> wrote:
>> Commit f211ac154577ec9ccf07c15f18a6abf0d9bdb4ab 'net: correct
>> sk_acceptq_is_full()' breaks a system with the Smack LSM.
>> Reverting this change results in a return to correct behavior.
>>
>> The Smack testsuite can be found at:
>> 	https://github.com/smack-team/smack-testsuite.git
>>
>> The failing test is ipv4-tcp-local-peersec.sh, but it seems
>> that most TCP connections hang with SYN_SENT. Oddly, ssh
>> to 127.0.0.1 works, but other TCP connections timeout.
>>
>>
>>
>>
>
>
>  
>
>
>