[for-next][PATCH 1/2] tracing: Add barrier to trace_printk() buffer nesting modification

[for-next][PATCH 1/2] tracing: Add barrier to trace_printk() buffer nesting modification

[Steven Rostedt]

[-- Attachment #1: 0001-tracing-Add-barrier-to-trace_printk-buffer-nesting-m.patch --]
[-- Type: text/plain, Size: 1953 bytes --]

From: "Steven Rostedt (VMware)" <rostedt@goodmis.org>

trace_printk() uses 4 buffers, one for each context (normal, softirq, irq
and NMI), such that it does not need to worry about one context preempting
the other. There's a nesting counter that gets incremented to figure out
which buffer to use. If the context gets preempted by another context which
calls trace_printk() it will increment the counter and use the next buffer,
and restore the counter when it is finished.

The problem is that gcc may optimize the modification of the buffer nesting
counter and it may not be incremented in memory before the buffer is used.
If this happens, and the context gets interrupted by another context, it
could pick the same buffer and corrupt the one that is being used.

Compiler barriers need to be added after the nesting variable is incremented
and before it is decremented to prevent usage of the context buffers by more
than one context at the same time.

Cc: Andy Lutomirski <luto@kernel.org>
Cc: stable@vger.kernel.org
Fixes: e2ace00117 ("tracing: Choose static tp_printk buffer by explicit nesting count")
Hat-tip-to: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---
 kernel/trace/trace.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 30338a835a51..78842557eea0 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -2802,11 +2802,17 @@ static char *get_trace_buf(void)
 	if (!buffer || buffer->nesting >= 4)
 		return NULL;
 
-	return &buffer->buffer[buffer->nesting++][0];
+	buffer->nesting++;
+
+	/* Interrupts must see nesting incremented before we use the buffer */
+	barrier();
+	return &buffer->buffer[buffer->nesting][0];
 }
 
 static void put_trace_buf(void)
 {
+	/* Don't let the decrement of nesting leak before this */
+	barrier();
 	this_cpu_dec(trace_percpu_buffer->nesting);
 }
 
-- 
2.13.2

[for-next][PATCH 2/2] tracing: Fix clear of RECORDED_TGID flag when disabling trace event

[Steven Rostedt]

[-- Attachment #1: 0002-tracing-Fix-clear-of-RECORDED_TGID-flag-when-disabli.patch --]
[-- Type: text/plain, Size: 1180 bytes --]

From: Chunyu Hu <chuhu@redhat.com>

When disabling one trace event, the RECORDED_TGID flag in the event
file is not correctly cleared. It's clearing RECORDED_CMD flag when
it should clear RECORDED_TGID flag.

Link: http://lkml.kernel.org/r/1504589806-8425-1-git-send-email-chuhu@redhat.com

Cc: Joel Fernandes <joelaf@google.com>
Cc: stable@vger.kernel.org
Fixes: d914ba37d7 ("tracing: Add support for recording tgid of tasks")
Signed-off-by: Chunyu Hu <chuhu@redhat.com>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---
 kernel/trace/trace_events.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
index c93540c5df21..87468398b9ed 100644
--- a/kernel/trace/trace_events.c
+++ b/kernel/trace/trace_events.c
@@ -406,7 +406,7 @@ static int __ftrace_event_enable_disable(struct trace_event_file *file,
 
 			if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 				tracing_stop_tgid_record();
-				clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
+				clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 			}
 
 			call->class->reg(call, TRACE_REG_UNREGISTER, file);
-- 
2.13.2