r2268 - trunk/varnish-cache/bin/varnishd
phk at projects.linpro.no
phk at projects.linpro.no
Tue Nov 20 15:20:36 CET 2007
Author: phk
Date: 2007-11-20 15:20:36 +0100 (Tue, 20 Nov 2007)
New Revision: 2268
Modified:
trunk/varnish-cache/bin/varnishd/cache_acceptor_kqueue.c
Log:
Handle that the sessions may be killed the worker thread before
we get around to delete the kevent.
Modified: trunk/varnish-cache/bin/varnishd/cache_acceptor_kqueue.c
===================================================================
--- trunk/varnish-cache/bin/varnishd/cache_acceptor_kqueue.c 2007-11-20 12:11:04 UTC (rev 2267)
+++ trunk/varnish-cache/bin/varnishd/cache_acceptor_kqueue.c 2007-11-20 14:20:36 UTC (rev 2268)
@@ -125,7 +125,7 @@
static pthread_t vca_kqueue_thread;
static int kq = -1;
-struct vbitmap *vca_kqueue_bits;
+static struct vbitmap *vca_kqueue_bits;
static VTAILQ_HEAD(,sess) sesshead = VTAILQ_HEAD_INITIALIZER(sesshead);
@@ -137,13 +137,25 @@
static void
vca_kq_sess(struct sess *sp, short arm)
{
+ int i;
CHECK_OBJ_NOTNULL(sp, SESS_MAGIC);
if (sp->fd < 0)
return;
EV_SET(&ki[nki], sp->fd, EVFILT_READ, arm, 0, 0, sp);
if (++nki == NKEV) {
- AZ(kevent(kq, ki, nki, NULL, 0, NULL));
+ i = kevent(kq, ki, nki, NULL, 0, NULL);
+ assert(i <= 0);
+ if (i < 0) {
+ /*
+ * We do not push kevents into the kernel before passing the session off
+ * to a worker thread, so by the time we get around to delete the event
+ * the fd may be closed and we get an ENOENT back once we do flush.
+ *
+ * XXX: Can we get EBADF if the client closes during this window ?
+ */
+ assert(errno == ENOENT);
+ }
nki = 0;
}
}
@@ -188,6 +200,7 @@
sp, (unsigned long)kp->data, kp->flags,
(kp->flags & EV_EOF) ? " EOF" : "");
#endif
+ assert(sp->fd == kp->ident);
if (kp->data > 0) {
i = HTC_Rx(sp->htc);
if (i == 0)
@@ -233,8 +246,10 @@
assert(n >= 1 && n <= NKEV);
nki = 0;
for (kp = ke, j = 0; j < n; j++, kp++) {
- if (kp->flags & EV_ERROR)
+ if (kp->flags & EV_ERROR) {
+ /* See comment in vca_kq_sess() */
continue;
+ }
if (kp->filter == EVFILT_TIMER) {
dotimer = 1;
continue;
More information about the varnish-commit
mailing list