networking: explain isrv_run() API
No binary (build failed)
—
(log yok)
| @@ -23,7 +23,32 @@ int isrv_register_fd(isrv_state_t *state, int peer, int fd); | |
| void isrv_close_fd(isrv_state_t *state, int fd); | |
| int isrv_register_peer(isrv_state_t *state, void *param); | |
| - | /* driver */ |
| + | /* Driver: |
| + | * |
| + | * Select on listen_fd for <linger_timeout> (or forever if 0). |
| + | * |
| + | * If we time out and we have no peers, exit. |
| + | * If we have peers, call do_timeout(peer_param), |
| + | * if it returns !0, peer is removed. |
| + | * |
| + | * If listen_fd is active, accept new connection ("peer"), |
| + | * call new_peer() on it, and if it returns 1, |
| + | * and add it to fds to select on. |
| + | * Now, select will wait for <timeout>, not <linger_timeout> |
| + | * (as long as we we have more than zero clients). |
| + | * |
| + | * If a peer's fd is active, we call do_rd() on it if read |
| + | * mask bit was set, |
| + | * and then do_wr() if write mask bit was also set. |
| + | * If either returns !0, peer is removed. |
| + | * Reaching this place also resets timeout counter for this peer. |
| + | * |
| + | * Note that peer must indicate that he wants to be selected |
| + | * for read and/or write using isrv_want_rd()/isrv_want_wr() |
| + | * [can be called in new_peer() or in do_rd()/do_wr()]. |
| + | * If it never wants to be selected for write, do_wr() |
| + | * will never be called (can be NULL). |
| + | */ |
| void isrv_run( | |
| int listen_fd, | |
| int (*new_peer)(isrv_state_t *state, int fd), | |