[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1383932286-25080-4-git-send-email-konrad.wilk@oracle.com>
Date: Fri, 8 Nov 2013 12:38:05 -0500
From: Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>
To: Ian.Campbell@...rix.com, xen-devel@...ts.xenproject.org,
linux-kernel@...r.kernel.org, JBeulich@...e.com,
david.vrabel@...rix.com, boris.ostrovsky@...cle.com
Cc: Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>
Subject: [PATCH 3/4] xen/manage: Guard against user-space initiated poweroff and XenBus.
There is a race case where the user does 'poweroff'
and at the same time the system admin does 'xl shutdown'.
Depending on the race, the system_state will be SYSTEM_RUNNING or
SYSTEM_POWER_OFF. If SYSTEM_RUNNING we just end up making
a duplicate call to 'poweroff' (while it is running).
That will fail or execute (And if executed then it will be
stuck in the reboot_mutex mutex). But nobody will care b/c the
machine is in poweroff sequence.
If the system_state is SYSTEM_POWER_OFF then we end up making
a duplicate call to kernel_power_off. There is no locking
there so we walk in the same steps as what 'poweroff'
has been doing.
The code in kernel/reboot.c has a mutex guarding against multiple
'poweroff' operations. But not against the kernel 'orderly_poweroff'.
As such, lets detect this so that we don't invoke orderly_poweroff
if the user had initiated a poweroff.
This is code by changing the 'shutting_down' to an atomic and
having a reboot notifier. If the 'shutting_down' is set to anything
but SHUTDOWN_INVALID the XenBus handler will not run.
That is exactly what we do in the reboot notifier - we set the
'shutting_down' to SHUTDOWN_POWEROFF.
The reason we change the 'shutting_down' to an atomic is that
the changes to said variable were normally guarded by the XenBus
mutex - "xenwatch_mutex" - guarantting only one caller changing
shutting_down. Since we have now the reboot notifier we have
another user of this variable. Surfacing the 'xenwatch_mutex'
out of XenBus is not a nice way of doing it. Having the
variable however be atomic solves the problem easily.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>
[v2: Don't expose xenwatch_mutex, add comments]
---
drivers/xen/manage.c | 51 ++++++++++++++++++++++++++++++++++++++++++---------
1 file changed, 42 insertions(+), 9 deletions(-)
diff --git a/drivers/xen/manage.c b/drivers/xen/manage.c
index 3f8496c..323703a 100644
--- a/drivers/xen/manage.c
+++ b/drivers/xen/manage.c
@@ -36,8 +36,16 @@ enum shutdown_state {
SHUTDOWN_HALT = 4,
};
-/* Ignore multiple shutdown requests. */
-static enum shutdown_state shutting_down = SHUTDOWN_INVALID;
+/* Ignore multiple shutdown requests. There are two potential race conditions:
+ * - Multiple XenStore 'shutdown' requests. We don't want to run any off
+ * the callbacks in parallel.
+ * - In progress 'poweroff' (initiated inside the guest) and a XenStore
+ * 'shutdown' request. If the poweroff has transitioned 'system_state' to
+ * SYSTEM_POWER_OFF we do not want to call orderly_poweroff. 'system_state'
+ * is not SMP safe so we depend on reboot notifiers to set 'shutting_down'
+ * so that we will ignore XenBus shutdown requests.
+ */
+static atomic_t shutting_down = ATOMIC_INIT(SHUTDOWN_INVALID);
struct suspend_info {
int cancelled;
@@ -109,7 +117,7 @@ static void do_suspend(void)
int err;
struct suspend_info si;
- shutting_down = SHUTDOWN_SUSPEND;
+ atomic_set(&shutting_down, SHUTDOWN_SUSPEND);
#ifdef CONFIG_PREEMPT
/* If the kernel is preemptible, we need to freeze all the processes
@@ -173,7 +181,7 @@ out_thaw:
thaw_processes();
out:
#endif
- shutting_down = SHUTDOWN_INVALID;
+ atomic_set(&shutting_down, SHUTDOWN_INVALID);
}
#endif /* CONFIG_HIBERNATE_CALLBACKS */
@@ -184,7 +192,7 @@ struct shutdown_handler {
static void do_poweroff(void)
{
- shutting_down = SHUTDOWN_POWEROFF;
+ atomic_set(&shutting_down, SHUTDOWN_POWEROFF);
switch (system_state) {
case SYSTEM_BOOTING:
orderly_poweroff(true);
@@ -201,7 +209,7 @@ static void do_poweroff(void)
static void do_reboot(void)
{
- shutting_down = SHUTDOWN_POWEROFF; /* ? */
+ atomic_set(&shutting_down, SHUTDOWN_POWEROFF); /* ? */
ctrl_alt_del();
}
@@ -222,7 +230,7 @@ static void shutdown_handler(struct xenbus_watch *watch,
};
static struct shutdown_handler *handler;
- if (shutting_down != SHUTDOWN_INVALID)
+ if (atomic_read(&shutting_down) != SHUTDOWN_INVALID)
return;
again:
@@ -256,12 +264,29 @@ static void shutdown_handler(struct xenbus_watch *watch,
handler->cb();
} else {
pr_info("Ignoring shutdown request: %s\n", str);
- shutting_down = SHUTDOWN_INVALID;
+ atomic_set(&shutting_down, SHUTDOWN_INVALID);
}
kfree(str);
}
+/*
+ * This function is called when the system is being rebooted.
+ */
+static int
+xen_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
+{
+ switch (event) {
+ case SYS_RESTART:
+ case SYS_HALT:
+ case SYS_POWER_OFF:
+ atomic_set(&shutting_down, SHUTDOWN_POWEROFF);
+ break;
+ default:
+ break;
+ }
+ return NOTIFY_DONE;
+}
#ifdef CONFIG_MAGIC_SYSRQ
static void sysrq_handler(struct xenbus_watch *watch, const char **vec,
unsigned int len)
@@ -302,6 +327,10 @@ static struct xenbus_watch shutdown_watch = {
.callback = shutdown_handler
};
+static struct notifier_block xen_shutdown_notifier = {
+ .notifier_call = xen_system_reboot,
+};
+
static int setup_shutdown_watcher(void)
{
int err;
@@ -319,7 +348,11 @@ static int setup_shutdown_watcher(void)
return err;
}
#endif
-
+ err = register_reboot_notifier(&xen_shutdown_notifier);
+ if (err) {
+ pr_warn("Failed to register shutdown notifier\n");
+ return err;
+ }
return 0;
}
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists