Re: [libvirt] [PATCH] systemd: fix build without dbus

On Mon, Jan 19, 2015 at 02:13:12PM +0100, Martin Kletzander wrote: > On Mon, Jan 19, 2015 at 12:34:52PM +0000, Daniel P. Berrange wrote: > >The virDBusMethodCall method has a DBusError as one of its > >parameters. If the caller wants to pass a non-NULL value > >for this, it immediately makes the calling code require > >DBus at build time. This has led to breakage of non-DBus > >builds several times. It is desirable that only the virdbus.c > >file should need WITH_DBUS conditionals, so we must ideally > >remove the DBusError parameter from the method. > > > >We can't simply raise a libvirt error, since the whole point > >of this parameter is to give the callers a way to check if > >the error is one they want to ignore, without having the logs > >polluted with an error message. So, we add a virErrorPtr > >parameter which the caller can then either ignore or raise > >using virSetError. > > > >Signed-off-by: Daniel P. Berrange <berrange(a)redhat.com&gt; > >--- > >src/util/virdbus.c | 31 +++++++++++++++++++------------ > >src/util/virdbus.h | 4 ++-- > >src/util/virfirewall.c | 34 ++++++++++------------------------ > >src/util/virsystemd.c | 15 +++++++-------- > >4 files changed, 38 insertions(+), 46 deletions(-) > > > >This is a build-breaker fix, but I'm not pushing since the > >fix is too complicated to go in without some review. > > > [...] > >diff --git a/src/util/virdbus.h b/src/util/virdbus.h > >index d0c7de2..e2b8d2b 100644 > >--- a/src/util/virdbus.h > >+++ b/src/util/virdbus.h > >@@ -28,7 +28,7 @@ > ># else > ># define DBusConnection void > ># define DBusMessage void > >-# define DBusError void > > Using virError instead of DBusError is fine, but > > >+# define dbus_message_unref(m) do {} while (0) > > ewww, can't we just cleanly separate DBus code from libvirt code > instead of adding more of these? This is required because we removed the WITH_DBUS conditional from virsystemd. To eliminate this would require that we remove DBusMessage from the virdbus.h header files too. That is doable but a bigger job and not really needed to fix this current build problem, so I think this is appropriate fix for this current patch. > > >diff --git a/src/util/virfirewall.c b/src/util/virfirewall.c > >index b536912..a01b703 100644 > >--- a/src/util/virfirewall.c > >+++ b/src/util/virfirewall.c > [...] > >@@ -789,10 +777,13 @@ virFirewallApplyRuleFirewallD(virFirewallRulePtr rule, > > "sa&s", > > ipv, > > (int)rule->argsLen, > >- rule->args) < 0) > >+ rule->args) < 0) { > >+ VIR_ERROR("Here fail"); > > Leftover from debugging? Opps, yes. > > goto cleanup; > >+ } > > > >- if (dbus_error_is_set(&error)) { > >+ VIR_ERROR("Error %d: %s\n", error.level, error.message); > >+ if (error.level == VIR_ERR_ERROR) { > > /* > > * As of firewalld-0.3.9.3-1.fc20.noarch the name and > > * message fields in the error look like > [...] > >@@ -862,12 +850,10 @@ virFirewallApplyRule(virFirewallPtr firewall, > > if (virFirewallApplyRuleDirect(rule, ignoreErrors, &output) < 0) > > return -1; > > break; > >-#if WITH_DBUS > > case VIR_FIREWALL_BACKEND_FIREWALLD: > > if (virFirewallApplyRuleFirewallD(rule, ignoreErrors, &output) < 0) > > Another thing that's pre-existing, but we are adding more and more of > them. I don't like that libvirt is still using systemd stuff even if > configured not to (--without-systemd). The --without-systemd flag is about use of the libsystemd-daemon.so library, which is used for daemon startup notification. Everything else related to systemd is just plain DBus API calls, so it is not appropriate to tie them to the --without-systemd flag. All DBus calls are dealt with by runtime checks for the service, rather than compile time checks for a lbirary. > >diff --git a/src/util/virsystemd.c b/src/util/virsystemd.c > >index 3eea5c2..0b71b26 100644 > >--- a/src/util/virsystemd.c > >+++ b/src/util/virsystemd.c > >@@ -252,8 +252,8 @@ int virSystemdCreateMachine(const char *name, > > > > VIR_DEBUG("Attempting to create machine via systemd"); > > if (virAtomicIntGet(&hasCreateWithNetwork)) { > >- DBusError error; > >- dbus_error_init(&error); > >+ virError error; > >+ memset(&error, 0, sizeof(error)); > > > > if (virDBusCallMethod(conn, > > NULL, > >@@ -280,21 +280,20 @@ int virSystemdCreateMachine(const char *name, > > "Before", "as", 1, "libvirt-guests.service") < 0) > > goto cleanup; > > > >- if (dbus_error_is_set(&error)) { > >+ if (error.level == VIR_ERR_ERROR) { > > if (STREQ_NULLABLE("org.freedesktop.DBus.Error.UnknownMethod", > >- error.name)) { > >+ error.str1)) { > > VIR_INFO("CreateMachineWithNetwork isn't supported, switching " > > "to legacy CreateMachine method for systemd-machined"); > >- dbus_error_free(&error); > >+ virResetError(&error); > > virAtomicIntSet(&hasCreateWithNetwork, 0); > > /* Could re-structure without Using goto, but this > > * avoids another atomic read which would trigger > > * another memory barrier */ > > goto fallback; > > } > >- virReportError(VIR_ERR_DBUS_SERVICE, > >- _("CreateMachineWithNetwork: %s"), > >- error.message ? error.message : _("unknown error")); > >+ virSetError(&error); > >+ virResetError(&error); > > goto cleanup; > > } > > } else { > >-- > > This is pretty hackish IMHO. Since > "org.freedesktop.DBus.Error.UnknownMethod" isn't specific for systemd, > we can make virDBusCall() report whether unknown method was called or > not and then decide what to do with it two layers up in the stack. It > also simplifies virSystemdCreateMachine() a lot. The "org.freedesktop.DBus.Error.UnknownMethod" method is a standard error code that is to be raised by DBus services, when the method requested does not exist. This is exactly the scenario we are dealing with here. We tried to run the CreateMachineWithNetwork method and it doesn't exist, so looking for the UnknownMethod error name is the correct handling for this scenario and how DBus error handling is intended to work.