VladPolskiy/apache-http-server
1
0
Fork 0
mirror of https://github.com/apache/httpd.git synced 2025-08-20 16:09:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

mpm_event: follow up to r1818804 and r1818951.

Browse Source 
Be more correct in comment about CONN_STATE_WRITE_COMPLETION.
We currently have/need no state to simply wait for readability on a socket,
so the previous comment was misleading. Write completion can't be used for
a simple "wait for read event and come back to process_connection hooks".



git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@1819027 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Yann Ylavic
2017-12-22 10:52:22 +00:00
parent 2ec371c1f9
commit 74e662639d
1 changed files with 20 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
server/mpm/event/event.c
View File

@ -1064,11 +1064,14 @@ read_request:
* The process_connection hooks above should set the connection state
* appropriately upon return, for event MPM to either:
* - do lingering close (CONN_STATE_LINGER),
* - wait for (readability of) the next request according to the keepalive
* - wait for readability of the next request with respect to the keepalive
* timeout (CONN_STATE_CHECK_REQUEST_LINE_READABLE),
* - wait for read/write-ability of the underlying socket according to its
* timeout (CONN_STATE_WRITE_COMPLETION, a legacy name which can also be
* used for readability by setting the sense to CONN_SENSE_WANT_READ),
* - keep flushing the output filters stack in nonblocking mode, and then
* if required wait for read/write-ability of the underlying socket with
* respect to its own timeout (CONN_STATE_WRITE_COMPLETION); since write
* completion at some point may require reads (e.g. SSL_ERROR_WANT_READ),
* an output filter can set the sense to CONN_SENSE_WANT_READ at any time
* for event MPM to do the right thing,
* - suspend the connection (SUSPENDED) such that it now interracts with
* the MPM through suspend/resume_connection() hooks, and/or registered
* poll callbacks (PT_USER), and/or registered timed callbacks triggered
@ -1107,10 +1110,20 @@ read_request:
*/
cs->queue_timestamp = apr_time_now();
notify_suspend(cs);
cs->pfd.reqevents = (
cs->pub.sense == CONN_SENSE_WANT_READ ? APR_POLLIN :
APR_POLLOUT) | APR_POLLHUP | APR_POLLERR;
if (cs->pub.sense == CONN_SENSE_WANT_READ) {
cs->pfd.reqevents = APR_POLLIN;
}
else {
cs->pfd.reqevents = APR_POLLOUT;
}
/* POLLHUP/ERR are usually returned event only (ignored here), but
* some pollset backends may require them in reqevents to do the
* right thing, so it shouldn't hurt.
*/
cs->pfd.reqevents |= APR_POLLHUP | APR_POLLERR;
cs->pub.sense = CONN_SENSE_DEFAULT;
apr_thread_mutex_lock(timeout_mutex);
TO_QUEUE_APPEND(cs->sc->wc_q, cs);
rc = apr_pollset_add(event_pollset, &cs->pfd);