[PATCH] ring-buffer: Prevent inconsistent operation on cpu_buffer->resize_disabled

[Tze-nan Wu]

Sometimes, write to buffer_size_kb can be permanently failure if we change
the cpu_online_mask between two for_each_online_buffer_cpu loops in
function ring_buffer_reset_online_cpus.

The number of increasing and decreasing on cpu_buffer->resize_disable
may be inconsistent, leading the resize_disabled in some CPUs becoming
none zero after ring_buffer_reset_online_cpus return.

This issue can be reproduced by "echo 0 > trace" and hotplug cpu at the
same time. After reproducing succeess, we can find out the attempt to
write to buffer_size_kb node failure every time.

This patch prevent the inconsistent increasing and decreasing on
cpu_buffer->resize_disabled by copying the cpu_online_mask at the
beginning of the function.

But I wonder if there's any side-effect of this patch,
since the behavior changed, if we turn on a cpu between the two loops,
reset_disabled_cpu_buffer() of that cpu won't be run as before,
meaning the cpu_buffer on that cpu just awake will not be cleaned up.

Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Tze-nan Wu <Tze-nan.Wu@xxxxxxxxxxxx>
---
 kernel/trace/ring_buffer.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 76a2d91eecad..468f46bba71e 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -288,9 +288,6 @@ EXPORT_SYMBOL_GPL(ring_buffer_event_data);
 #define for_each_buffer_cpu(buffer, cpu)		\
 	for_each_cpu(cpu, buffer->cpumask)
 
-#define for_each_online_buffer_cpu(buffer, cpu)		\
-	for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask)
-
 #define TS_SHIFT	27
 #define TS_MASK		((1ULL << TS_SHIFT) - 1)
 #define TS_DELTA_TEST	(~TS_MASK)
@@ -5353,12 +5350,15 @@ EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
 void ring_buffer_reset_online_cpus(struct trace_buffer *buffer)
 {
 	struct ring_buffer_per_cpu *cpu_buffer;
+	cpumask_var_t reset_online_mask;
 	int cpu;
 
 	/* prevent another thread from changing buffer sizes */
 	mutex_lock(&buffer->mutex);
 
-	for_each_online_buffer_cpu(buffer, cpu) {
+	cpumask_copy(reset_online_mask, cpu_online_mask);
+
+	for_each_cpu_and(cpu, buffer->cpumask, reset_online_mask) {
 		cpu_buffer = buffer->buffers[cpu];
 
 		atomic_inc(&cpu_buffer->resize_disabled);
@@ -5368,7 +5368,7 @@ void ring_buffer_reset_online_cpus(struct trace_buffer *buffer)
 	/* Make sure all commits have finished */
 	synchronize_rcu();
 
-	for_each_online_buffer_cpu(buffer, cpu) {
+	for_each_cpu_and(cpu, buffer->cpumask, reset_online_mask) {
 		cpu_buffer = buffer->buffers[cpu];
 
 		reset_disabled_cpu_buffer(cpu_buffer);
-- 
2.18.0