sk_wait_data

[Grzegorz Dwornicki]

Hello

Proces have a established tcp socket. It calls sys_revcmsg on that
socket. That function goes all the way to tcp_recvmsg function.

Function tcp_recvmsg.c (defined in net/ipv4/tcp.c) blocks it self on
function sk_wait_data if socket recv queue is empty. I wanted to know
how socked is checked in time. Soo i Looked up this sk_wait_data
function (http://lxr.free-electrons.com/source/net/core/sock.c#L1933):

1933 int sk_wait_data(struct sock *sk, long *timeo)
1934 {
1935         int rc;
1936         DEFINE_WAIT(wait);
1937
1938         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1939         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1940         rc = sk_wait_event(sk, timeo,
!skb_queue_empty(&sk->sk_receive_queue));
1941         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1942         finish_wait(sk_sleep(sk), &wait);
1943         return rc;
1944 }

This function blocks it self on the sk_wait_event macro. Here is it
definition (http://lxr.free-electrons.com/source/include/net/sock.h#L893):

893 #define sk_wait_event(__sk, __timeo, __condition)                       \
894         ({      int __rc;                                               \
895                 release_sock(__sk);                                     \
896                 __rc = __condition;                                     \
897                 if (!__rc) {                                            \
898                         *(__timeo) = schedule_timeout(*(__timeo));      \
899                 }                                                       \
900                 lock_sock(__sk);                                        \
901                 __rc = __condition;                                     \
902                 __rc;                                                   \
903         })

This macro is blocked in the schedule_timeout all the time. I know
this because I added printk functions with some marks easly to target
the line blocking the socket. It hangs on this schedule_timeout and it
magicaly released when some data apear in the revc queue.

How does the kernel know when to "unblock" the process? Function name:
sk_wait_data and macro name: sk_wait_event tells me that there should
be some queue checning from time to time...

Can someone explain this to me?