[PATCH] tracing: move static old_tracer to trace_iterator

[Lai Jiangshan]

static old_tracer is global for all processes.

So there is a potential bug when:
current_trace and static old_tracer are changed by other processes,
current_trace and static old_tracer are match with each other.
but *iter->trace and *current_trace are not match.

This patch move old_tracer to trace_iterator, and make it not global.

Signed-off-by: Lai Jiangshan <la...@cn.fujitsu.com>
---
diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
index 3ca9485..925897e 100644
--- a/include/linux/ftrace_event.h
+++ b/include/linux/ftrace_event.h
@@ -47,6 +47,7 @@ struct trace_entry {
*/
struct trace_iterator {
struct trace_array *tr;
+ struct tracer *old_tracer;
struct tracer *trace;
void *private;
int cpu_file;
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 5314c90..bbb2229 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -1667,7 +1667,6 @@ static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
static void *s_start(struct seq_file *m, loff_t *pos)
{
struct trace_iterator *iter = m->private;
- static struct tracer *old_tracer;
int cpu_file = iter->cpu_file;
void *p = NULL;
loff_t l = 0;
@@ -1675,8 +1674,8 @@ static void *s_start(struct seq_file *m, loff_t *pos)

/* copy the tracer to avoid using a global lock all around */
mutex_lock(&trace_types_lock);
- if (unlikely(old_tracer != current_trace && current_trace)) {
- old_tracer = current_trace;
+ if (unlikely(iter->old_tracer != current_trace && current_trace)) {
+ iter->old_tracer = current_trace;
*iter->trace = *current_trace;
}
mutex_unlock(&trace_types_lock);
@@ -3080,7 +3079,6 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,
size_t cnt, loff_t *ppos)
{
struct trace_iterator *iter = filp->private_data;
- static struct tracer *old_tracer;
ssize_t sret;

/* return any leftover data */
@@ -3092,8 +3090,8 @@ tracing_read_pipe(struct file *filp, char __user *ubuf,

/* copy the tracer to avoid using a global lock all around */
mutex_lock(&trace_types_lock);
- if (unlikely(old_tracer != current_trace && current_trace)) {
- old_tracer = current_trace;
+ if (unlikely(iter->old_tracer != current_trace && current_trace)) {
+ iter->old_tracer = current_trace;
*iter->trace = *current_trace;
}
mutex_unlock(&trace_types_lock);
@@ -3242,15 +3240,14 @@ static ssize_t tracing_splice_read_pipe(struct file *filp,
.ops = &tracing_pipe_buf_ops,
.spd_release = tracing_spd_release_pipe,
};
- static struct tracer *old_tracer;
ssize_t ret;
size_t rem;
unsigned int i;

/* copy the tracer to avoid using a global lock all around */
mutex_lock(&trace_types_lock);
- if (unlikely(old_tracer != current_trace && current_trace)) {
- old_tracer = current_trace;
+ if (unlikely(iter->old_tracer != current_trace && current_trace)) {
+ iter->old_tracer = current_trace;
*iter->trace = *current_trace;
}
mutex_unlock(&trace_types_lock);

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

[Frederic Weisbecker]

On Tue, Jan 26, 2010 at 03:28:13PM +0800, Lai Jiangshan wrote:
>
> static old_tracer is global for all processes.
>
> So there is a potential bug when:
> current_trace and static old_tracer are changed by other processes,
> current_trace and static old_tracer are match with each other.
> but *iter->trace and *current_trace are not match.
>
> This patch move old_tracer to trace_iterator, and make it not global.
>
> Signed-off-by: Lai Jiangshan <la...@cn.fujitsu.com>

We should actually get rid of old_tracer and have iter->trace
to be a pointer so that the check can be summarized with

if (iter->trace != current_trace)

I don't exactly recall why we've made iter->trace a copy
of the current_trace. I remember I did it this uglyness,
but I don't remember exactly why. Probably for bad reasons
that should be fixed another way.

Ok, having a deeper look, I think it was because some tracers
implement callbacks whereas others may not, so the
simple:

if (iter->trace->splice_read)
ret = iter->trace->splice_read(...)

would be racy if we weren't copying current_trace.

To fix this, I would rather try to ensure every tracers
have stub callbacks for those unimplemented. This is going
to be much proper than this dance between current_tracer,
old_tracer and iter->trace.

What do you think?