Commit f6c12993 authored by Tollef Fog Heen's avatar Tollef Fog Heen

Merge r5227: Update the Solaris "port" based waiter.

Fixes #629


git-svn-id: http://www.varnish-cache.org/svn/branches/2.1@5367 d4fa192b-c00b-0410-8231-f00ffab90ce4
parent a00e8887
......@@ -3,6 +3,7 @@
* Copyright (c) 2006 Linpro AS
* Copyright (c) 2007 OmniTI Computer Consulting, Inc.
* Copyright (c) 2007 Theo Schlossnagle
* Copyright (c) 2010 UPLEX, Nils Goroll
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
......@@ -26,9 +27,6 @@
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* XXX: We need to pass sessions back into the event engine when they are
* reused. Not sure what the most efficient way is for that. For now
* write the session pointer to a pipe which the event engine monitors.
*/
#include "config.h"
......@@ -42,12 +40,10 @@ SVNID("$Id$")
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <math.h>
#include <port.h>
#ifndef HAVE_CLOCK_GETTIME
#include "compat/clock_gettime.h"
#endif
#include <sys/time.h>
#include "shmlog.h"
#include "cache.h"
......@@ -59,47 +55,68 @@ int solaris_dport = -1;
static VTAILQ_HEAD(,sess) sesshead = VTAILQ_HEAD_INITIALIZER(sesshead);
static void
static inline void
vca_add(int fd, void *data)
{
AZ(port_associate(solaris_dport, PORT_SOURCE_FD, fd,
POLLIN | POLLERR | POLLPRI, data));
/*
* POLLIN should be all we need here
*
*/
AZ(port_associate(solaris_dport, PORT_SOURCE_FD, fd, POLLIN, data));
}
static void
static inline void
vca_del(int fd)
{
port_dissociate(solaris_dport, PORT_SOURCE_FD, fd);
}
static void
static inline void
vca_port_ev(port_event_t *ev) {
struct sess *sp;
if(ev->portev_source == PORT_SOURCE_USER) {
sp = ev->portev_user;
CHECK_OBJ_NOTNULL(sp, SESS_MAGIC);
CAST_OBJ_NOTNULL(sp, ev->portev_user, SESS_MAGIC);
assert(sp->fd >= 0);
AZ(sp->obj);
VTAILQ_INSERT_TAIL(&sesshead, sp, list);
vca_add(sp->fd, sp);
} else {
int i;
assert(ev->portev_source == PORT_SOURCE_FD);
CAST_OBJ_NOTNULL(sp, ev->portev_user, SESS_MAGIC);
assert(sp->fd >= 0);
if(ev->portev_events & POLLERR) {
VTAILQ_REMOVE(&sesshead, sp, list);
vca_del(sp->fd);
VTAILQ_REMOVE(&sesshead, sp, list);
vca_close_session(sp, "EOF");
SES_Delete(sp);
return;
}
i = HTC_Rx(sp->htc);
if (i == 0)
if (i == 0) {
/* incomplete header, wait for more data */
vca_add(sp->fd, sp);
return;
if (i > 0) {
VTAILQ_REMOVE(&sesshead, sp, list);
if (sp->fd != -1)
vca_del(sp->fd);
vca_handover(sp, i);
}
/*
* note: the original man page for port_associate(3C) states:
*
* When an event for a PORT_SOURCE_FD object is retrieved,
* the object no longer has an association with the port.
*
* This can be read along the lines of sparing the
* port_dissociate after port_getn(), but in fact,
* port_dissociate should be used
*
* Ref: http://opensolaris.org/jive/thread.jspa?threadID=129476&tstart=0
*/
vca_del(sp->fd);
VTAILQ_REMOVE(&sesshead, sp, list);
/* vca_handover will also handle errors */
vca_handover(sp, i);
}
return;
}
......@@ -109,40 +126,125 @@ vca_main(void *arg)
{
struct sess *sp;
/*
* timeouts:
*
* min_ts : Minimum timeout for port_getn
* min_t : ^ equivalent in floating point representation
*
* max_ts : Maximum timeout for port_getn
* max_t : ^ equivalent in floating point representation
*
* with (nevents == 1), we should always choose the correct port_getn
* timeout to check session timeouts, so max is just a safety measure
* (if this implementation is correct, it could be set to an "infinte"
* value)
*
* with (nevents > 1), min and max define the acceptable range for
* - additional latency of keep-alive connections and
* - additional tolerance for handling session timeouts
*
*/
static struct timespec min_ts = {0L, 100L /*ms*/ * 1000L /*us*/ * 1000L /*ns*/};
static double min_t = 0.1; /* 100 ms*/
static struct timespec max_ts = {1L, 0L}; /* 1 second */
static double max_t = 1.0; /* 1 second */
struct timespec ts;
struct timespec *timeout;
(void)arg;
solaris_dport = port_create();
assert(solaris_dport >= 0);
timeout = &max_ts;
while (1) {
port_event_t ev[MAX_EVENTS];
int nevents, ei;
double deadline;
struct timespec ts;
ts.tv_sec = 0L;
ts.tv_nsec = 50L /*ms*/ * 1000L /*us*/ * 1000L /*ns*/;
int nevents, ei, ret;
double now, deadline;
/*
* XXX Do we want to scale this up dynamically to increase
* efficiency in high throughput situations? - would need to
* start with one to keep latency low at any rate
*
* Note: when increasing nevents, we must lower min_ts
* and max_ts
*/
nevents = 1;
if (port_getn(solaris_dport, ev, MAX_EVENTS, &nevents, &ts)
== 0) {
for (ei=0; ei<nevents; ei++) {
vca_port_ev(ev + ei);
}
/*
* see disucssion in
* - https://issues.apache.org/bugzilla/show_bug.cgi?id=47645
* - http://mail.opensolaris.org/pipermail/networking-discuss/2009-August/011979.html
*
* comment from apr/poll/unix/port.c :
*
* This confusing API can return an event at the same time
* that it reports EINTR or ETIME.
*
*/
ret = port_getn(solaris_dport, ev, MAX_EVENTS, &nevents, timeout);
if (ret < 0)
assert((errno == EINTR) || (errno == ETIME));
for (ei=0; ei<nevents; ei++) {
vca_port_ev(ev + ei);
}
/* check for timeouts */
deadline = TIM_real() - params->sess_timeout;
now = TIM_real();
deadline = now - params->sess_timeout;
/*
* This loop assumes that the oldest sessions are always at the
* beginning of the list (which is the case if we guarantee to
* enqueue at the tail only
*
*/
for (;;) {
sp = VTAILQ_FIRST(&sesshead);
if (sp == NULL)
break;
if (sp->t_open > deadline)
if (sp->t_open > deadline) {
break;
}
VTAILQ_REMOVE(&sesshead, sp, list);
if(sp->fd != -1)
if(sp->fd != -1) {
vca_del(sp->fd);
// XXX: not yet TCP_linger(sp->fd, 0);
}
vca_close_session(sp, "timeout");
SES_Delete(sp);
}
/*
* Calculate the timeout for the next get_portn
*/
if (sp) {
double tmo = (sp->t_open + params->sess_timeout) - now;
/* we should have removed all sps whose timeout has passed */
assert(tmo > 0.0);
if (tmo < min_t) {
timeout = &min_ts;
} else if (tmo > max_t) {
timeout = &max_ts;
} else {
/* TIM_t2ts() ? see #630 */
ts.tv_sec = (int)floor(tmo);
ts.tv_nsec = 1e9 * (tmo - ts.tv_sec);
timeout = &ts;
}
} else {
timeout = &max_ts;
}
}
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment