libvirt/SOURCES/libvirt-RHEL-Remove-glib-2.64.0-workaround-for-GSource-race.patch
2022-05-10 07:15:48 +00:00

90 lines
2.8 KiB
Diff

From b550b671bace6d979a0763a837ce97ddf72e8fc8 Mon Sep 17 00:00:00 2001
Message-Id: <b550b671bace6d979a0763a837ce97ddf72e8fc8@dist-git>
From: Martin Kletzander <mkletzan@redhat.com>
Date: Fri, 28 Jan 2022 20:33:12 +0100
Subject: [PATCH] RHEL: Remove <glib-2.64.0 workaround for GSource race
https://bugzilla.redhat.com/show_bug.cgi?id=2045879
This is to accommodate for RHEL 8 downstream glib being fixed with a backport in
version 2.56.4-12 (Bug 1948988). Another reason for that is that our workaround
does not play nice with a fixed glib, leading to libvirt not properly
dereferencing the monitor socket, eventually leading to EMFILE: Too many open
files.
RHEL-only
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
---
src/util/glibcompat.c | 46 ++++++++-----------------------------------
1 file changed, 8 insertions(+), 38 deletions(-)
diff --git a/src/util/glibcompat.c b/src/util/glibcompat.c
index eb6dcc0111..bf890f10d2 100644
--- a/src/util/glibcompat.c
+++ b/src/util/glibcompat.c
@@ -212,51 +212,21 @@ vir_g_strdup_vprintf(const char *msg, va_list args)
return ret;
}
-
/*
- * If the last reference to a GSource is released in a non-main
- * thread we're exposed to a race condition that causes a
- * crash:
- *
- * https://gitlab.gnome.org/GNOME/glib/-/merge_requests/1358
+ * This is a leftover of a hack that works around glib older than 2.64.0, but
+ * the fix in glib as backported in RHEL-8 to glib2-2.56.4-12 in BZ 1948988:
*
- * Thus we're using an idle func to release our ref...
+ * https://bugzilla.redhat.com/show_bug.cgi?id=1948988
*
- * ...but this imposes a significant performance penalty on
- * I/O intensive workloads which are sensitive to the iterations
- * of the event loop, so avoid the workaround if we know we have
- * new enough glib.
+ * and our workaround coupled with that glib fix started causing leaks, which
+ * surfaced with us not unreferencing the qemu monitor socket in BZ 2045879:
*
- * The function below is used from a header file definition.
+ * https://bugzilla.redhat.com/show_bug.cgi?id=2045879
*
- * Drop when min glib >= 2.64.0
+ * Keeping this wrapper makes it easier to follow with other backports without
+ * conflicts in callers due to the function name change.
*/
-#if GLIB_CHECK_VERSION(2, 64, 0)
void vir_g_source_unref(GSource *src, GMainContext *ctx G_GNUC_UNUSED)
{
g_source_unref(src);
}
-#else
-
-static gboolean
-virEventGLibSourceUnrefIdle(gpointer data)
-{
- GSource *src = data;
-
- g_source_unref(src);
-
- return FALSE;
-}
-
-void vir_g_source_unref(GSource *src, GMainContext *ctx)
-{
- GSource *idle = g_idle_source_new();
-
- g_source_set_callback(idle, virEventGLibSourceUnrefIdle, src, NULL);
-
- g_source_attach(idle, ctx);
-
- g_source_unref(idle);
-}
-
-#endif
--
2.35.1