在 2012-7-6，1:42，MORITA Kazutaka <[email protected]> 写道：
> At Fri, 6 Jul 2012 01:33:44 +0800, > Yunkai Zhang wrote: >> >> On Fri, Jul 6, 2012 at 12:57 AM, MORITA Kazutaka >> <[email protected]> wrote: >>> At Thu, 5 Jul 2012 18:34:10 +0800, >>> Yunkai Zhang wrote: >>>> >>>> From: Yunkai Zhang <[email protected]> >>>> >>>> When connection is dead, we should: >>>> 1) do some clear actions about this connection. >>>> 2) unregister it's fd from epoll. >>>> 3) destroy this connection directly, since it's fd has been unregistered, >>>> sheep couldn't receive any new events of this fd. >>>> *Note*: If we use client_decref() instread of destroy_client(), this fd >>>> may not be destroyed when ci->refcnt is larger than 1, than this fd >>>> could >>>> not be closed in the future. >>>> >>>> Now I add a helper named clear_client() to do these works. >>>> >>>> Signed-off-by: Yunkai Zhang <[email protected]> >>>> --- >>>> sheep/sdnet.c | 20 ++++++++++++-------- >>>> 1 file changed, 12 insertions(+), 8 deletions(-) >>>> >>>> diff --git a/sheep/sdnet.c b/sheep/sdnet.c >>>> index c13cdb0..b493428 100644 >>>> --- a/sheep/sdnet.c >>>> +++ b/sheep/sdnet.c >>>> @@ -370,6 +370,7 @@ static void requeue_request(struct request *req) >>>> >>>> static void client_incref(struct client_info *ci); >>>> static void client_decref(struct client_info *ci); >>>> +static void clear_client(struct client_info *ci); >>>> >>>> static struct request *alloc_local_request(void *data, int data_length) >>>> { >>>> @@ -473,11 +474,10 @@ void put_request(struct request *req) >>>> if (conn_tx_on(&ci->conn)) { >>>> dprintf("connection seems to be dead\n"); >>>> free_request(req); >>>> + clear_client(ci); >>> >>> We cannot free the client_info here because there may be other >>> inflight requests which use it, no? >> >> >> Yes, there may be other inflight request, but since conn_tx_on() >> failed and then we can't send back response to the client, why not >> free it? I think it can make program simple. > > If we call put_request against the inflight requests after freeing the > client_info, conn_tx_on(&ci->conn) can lead to segfaults because ci is > no longer a valid value. Ho, you are right! I'll give v2 tomorrow. > > Thanks, > > Kazutaka -- sheepdog mailing list [email protected] http://lists.wpkg.org/mailman/listinfo/sheepdog
