networking: explain isrv_run() API

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
This commit is contained in:
Denys Vlasenko 2014-01-09 19:58:19 +01:00
parent 89deb22f97
commit 3ea93e853e

View File

@ -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),