Please ignore my foolish question. I overlooked the __set_current_state(TASK_RUNNING); line in tcp_rcv_established. Sorry. :P On Fri, 14 Jan 2005 17:03:53 -0500, Wen Xu <crocoxu@xxxxxxxxx> wrote: > Hello, > > I am reading 2.4.22 kernel code on TCP/IP networking. And I don't > understand why the following scenario will not occur: > > 1. user thread calls read which in turn calls tcp_recvmsg() on an > established socket > 2. tcp_recvmsg() sees that no data is available on receive_queue, > installs prequeue > 3. tcp_recvmsg() continues and prepare to sleep because it has not > seen any data, but it has not entered tcp_data_wait() yet. > 4. suddenly at this moment a packet comes in > 5. the bottom half calls bh_lock_sock, see that sock->lock.users == 1, > adds the packet to backlog queue > 6. the bottom half exits > 7. tcp_recvmsg calls tcp_data_wait() to sleep > 8. in release_sock() of tcp_data_wait, backlog queue is processed by > tcp_v4_do_rcv() which calls tcp_rcv_established() > 9. in tcp_rcv_established(), as prequeue is installed, packets are > directly copied to user context and nothing gets queued to > sk->receive_queue, so eaten == 1 > 10. at the end of tcp_rcv_established(), because eaten == 1, > sk->data_ready(sk, 0); is *not* called > 11. backlog queue processing finished, return to release_sock in step 8. > 12. continue tcp_data_wait() in step 7, see that receive_queue is > NULL, goes to sleep > > If the above situation occurs, the user thread waiting for data will > go to sleep even though incoming packets have been processed and data > have been copied. Obviously it was not the case. I wonder at which > step did I miss something. > > Thanks a lot. > Wen > -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
