Skip to contentSkip to conversation

audit: wait_for_auditd() should use TASK_UNINTERRUPTIBLE

PATCH In review Oleg Nesterov 3 participants 5 messages May 24, 2013 – May 31, 2013 Audit

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 audit: wait_for_auditd() should use TASK_UNINTERRUPTIBLE

audit_log_start() does wait_for_auditd() in a loop until audit_backlog_wait_time passes or audit_skb_queue has a room.

If signal_pending() is true this becomes a busy-wait loop, schedule() in TASK_INTERRUPTIBLE won't block.

kernel/audit.c +1 -1
@@ -1056,7 +1056,7 @@ static inline void audit_get_stamp(struc
 static void wait_for_auditd(unsigned long sleep_time)
 {
 	DECLARE_WAITQUEUE(wait, current);
-	set_current_state(TASK_INTERRUPTIBLE);
+	set_current_state(TASK_UNINTERRUPTIBLE);
 	add_wait_queue(&audit_backlog_wait, &wait);
 
 	if (audit_backlog_limit &&
#1 lore raw
Quoted 6 lines · Oleg Nesterov

audit_log_start() does wait_for_auditd() in a loop until audit_backlog_wait_time passes or audit_skb_queue has a room.

If signal_pending() is true this becomes a busy-wait loop, schedule() in TASK_INTERRUPTIBLE won't block.

And the code looks strange imho. I think it should be cleanuped, something like below. But I do not know how can I test this change.

Oleg.

kernel/audit.c +12 -29
@@ -1053,18 +1053,19 @@ static inline void audit_get_stamp(struc
 /*
  * Wait for auditd to drain the queue a little
  */
-static void wait_for_auditd(unsigned long sleep_time)
+static bool wait_for_auditd(gfp_t gfp_mask)
 {
-	DECLARE_WAITQUEUE(wait, current);
-	set_current_state(TASK_INTERRUPTIBLE);
-	add_wait_queue(&audit_backlog_wait, &wait);
-
-	if (audit_backlog_limit &&
-	    skb_queue_len(&audit_skb_queue) > audit_backlog_limit)
-		schedule_timeout(sleep_time);
+	bool can_wait = gfp_mask & __GFP_WAIT;
+	int reserve = can_wait ? 0 : 5;
 
-	__set_current_state(TASK_RUNNING);
-	remove_wait_queue(&audit_backlog_wait, &wait);
+	if (!audit_backlog_limit ||
+	    skb_queue_len(&audit_skb_queue) <= audit_backlog_limit + reserve)
+		return true;
+
+	return can_wait && wait_event_timeout(audit_backlog_wait,
+				!audit_backlog_limit ||
+					skb_queue_len(&audit_skb_queue) <= audit_backlog_limit,
+				audit_backlog_wait_time);
 }
 
 /* Obtain an audit buffer.  This routine does locking to obtain the
@@ -1095,8 +1096,6 @@ struct audit_buffer *audit_log_start(str
 	struct audit_buffer	*ab	= NULL;
 	struct timespec		t;
 	unsigned int		uninitialized_var(serial);
-	int reserve;
-	unsigned long timeout_start = jiffies;
 
 	if (audit_initialized != AUDIT_INITIALIZED)
 		return NULL;
@@ -1104,23 +1103,7 @@ struct audit_buffer *audit_log_start(str
 	if (unlikely(audit_filter_type(type)))
 		return NULL;
 
-	if (gfp_mask & __GFP_WAIT)
-		reserve = 0;
-	else
-		reserve = 5; /* Allow atomic callers to go up to five
-				entries over the normal backlog limit */
-
-	while (audit_backlog_limit
-	       && skb_queue_len(&audit_skb_queue) > audit_backlog_limit + reserve) {
-		if (gfp_mask & __GFP_WAIT && audit_backlog_wait_time) {
-			unsigned long sleep_time;
-
-			sleep_time = timeout_start + audit_backlog_wait_time -
-					jiffies;
-			if ((long)sleep_time > 0)
-				wait_for_auditd(sleep_time);
-			continue;
-		}
+	if (!wait_for_auditd(gfp_mask)) {
 		if (audit_rate_check() && printk_ratelimit())
 			printk(KERN_WARNING
 			       "audit: audit_backlog=%d > "
#2 lore raw
In reply to Oleg Nesterov
Quoted 5 lines · On Fri, 24 May 2013 19:39:25 +0200 Oleg Nesterov <[email protected]> wrote:

audit_log_start() does wait_for_auditd() in a loop until audit_backlog_wait_time passes or audit_skb_queue has a room.

If signal_pending() is true this becomes a busy-wait loop, schedule() in TASK_INTERRUPTIBLE won't block.

And that's game over for a uniprocessor non-preempt machine, yes?

Quoted 1 line

And what did Guy report? "that looks screwy"? "my machine locked up"?

Quoted 9 lines
kernel/audit.c +1 -1
@@ -1056,7 +1056,7 @@ static inline void audit_get_stamp(struc
 static void wait_for_auditd(unsigned long sleep_time)
 {
 	DECLARE_WAITQUEUE(wait, current);
-	set_current_state(TASK_INTERRUPTIBLE);
+	set_current_state(TASK_UNINTERRUPTIBLE);
 	add_wait_queue(&audit_backlog_wait, &wait);
 
 	if (audit_backlog_limit &&
#3 lore raw
Quoted 5 lines · Andrew Morton
Quoted 3 lines · On Fri, 24 May 2013 19:39:25 +0200 Oleg Nesterov <[email protected]> wrote:

audit_log_start() does wait_for_auditd() in a loop until audit_backlog_wait_time passes or audit_skb_queue has a room.

Quoted 9 lines
Quoted 2 lines

If signal_pending() is true this becomes a busy-wait loop, schedule() in TASK_INTERRUPTIBLE won't block.

And that's game over for a uniprocessor non-preempt machine, yes?

Quoted 1 line

And what did Guy report? "that looks screwy"? "my machine locked up"?

Our customer was in fact running a uniprocessor machine, and they reported a system hang.
--Guy

#4 lore raw
In reply to Andrew Morton
Quoted 10 lines · Andrew Morton
Quoted 5 lines · On Fri, 24 May 2013 19:39:25 +0200 Oleg Nesterov <[email protected]> wrote:

audit_log_start() does wait_for_auditd() in a loop until audit_backlog_wait_time passes or audit_skb_queue has a room.

If signal_pending() is true this becomes a busy-wait loop, schedule() in TASK_INTERRUPTIBLE won't block.

And that's game over for a uniprocessor non-preempt machine, yes?

If this task is rt, yes. Otherwise schedule() still does pick_next_task() but this is obviously bad anyway. So I fully agree with "Cc: stable" you added.

Quoted 3 lines
Quoted 1 line

And what did Guy report? "that looks screwy"? "my machine locked up"?

He also investigated the problem and provided the detailed explanation ;)

Oleg.