Hi Oleg,
On Thu, 31 Oct 2013 19:16:54 +0100, Oleg Nesterov wrote:
> On 10/29, Namhyung Kim wrote:
>>
>> @@ -630,6 +653,19 @@ probe_event_enable(struct trace_uprobe *tu, int flag,
>> filter_func_t filter)
>> if (trace_probe_is_enabled(&tu->p))
>> return -EINTR;
>>
>> + if (atomic_inc_return(&uprobe_buffer_ref) == 1) {
>> + int cpu;
>> +
>> + uprobe_cpu_buffer = __alloc_percpu(PAGE_SIZE, PAGE_SIZE);
>> + if (uprobe_cpu_buffer == NULL) {
>> + atomic_dec(&uprobe_buffer_ref);
>> + return -ENOMEM;
>> + }
>> +
>> + for_each_possible_cpu(cpu)
>> + mutex_init(&per_cpu(uprobe_cpu_mutex, cpu));
>> + }
>> +
>> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>>
>> tu->p.flags |= flag;
>> @@ -646,6 +682,11 @@ static void probe_event_disable(struct trace_uprobe
>> *tu, int flag)
>> if (!trace_probe_is_enabled(&tu->p))
>> return;
>>
>> + if (atomic_dec_and_test(&uprobe_buffer_ref)) {
>> + free_percpu(uprobe_cpu_buffer);
>> + uprobe_cpu_buffer = NULL;
>> + }
>> +
>> WARN_ON(!uprobe_filter_is_empty(&tu->filter));
>
> Do we really need atomic_t? probe_event_enable/disable is called under
> event_mutex and we rely on this fact anyway.
Looking at the code, it seems probe_event_enable/disable() is called
without event_mutex when it called from sys_perf_event_open(). So we
still need to protect refcount from concurrent accesses.
>
> Otherwise this logic looks racy even with atomic_t, another thread could
> use the uninitialized uprobe_cpu_buffer/mutex if it registers another probe
> and the handler runs before we complete the initialization, no?
But yeah, this is indeed a problem. Thanks for pointing it out. I'll
put a mutex to prevent such cases.
Thanks,
Namhyung
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
