perf/core: Fix locking for children siblings group read
authorJiri Olsa <jolsa@kernel.org>
Thu, 20 Jul 2017 14:14:55 +0000 (16:14 +0200)
committerIngo Molnar <mingo@kernel.org>
Fri, 21 Jul 2017 07:54:23 +0000 (09:54 +0200)
We're missing ctx lock when iterating children siblings
within the perf_read path for group reading. Following
race and crash can happen:

User space doing read syscall on event group leader:

T1:
  perf_read
    lock event->ctx->mutex
    perf_read_group
      lock leader->child_mutex
      __perf_read_group_add(child)
        list_for_each_entry(sub, &leader->sibling_list, group_entry)

---->   sub might be invalid at this point, because it could
        get removed via perf_event_exit_task_context in T2

Child exiting and cleaning up its events:

T2:
  perf_event_exit_task_context
    lock ctx->mutex
    list_for_each_entry_safe(child_event, next, &child_ctx->event_list,...
      perf_event_exit_event(child)
        lock ctx->lock
        perf_group_detach(child)
        unlock ctx->lock

---->   child is removed from sibling_list without any sync
        with T1 path above

        ...
        free_event(child)

Before the child is removed from the leader's child_list,
(and thus is omitted from perf_read_group processing), we
need to ensure that perf_read_group touches child's
siblings under its ctx->lock.

Peter further notes:

| One additional note; this bug got exposed by commit:
|
|   ba5213ae6b88 ("perf/core: Correct event creation with PERF_FORMAT_GROUP")
|
| which made it possible to actually trigger this code-path.

Tested-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: ba5213ae6b88 ("perf/core: Correct event creation with PERF_FORMAT_GROUP")
Link: http://lkml.kernel.org/r/20170720141455.2106-1-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/core.c

index c9cdbd3967709cfd8456c8e48ae09c2238936d32..c17c0881fd36aef7200f3991b12bdd5a4322b765 100644 (file)
@@ -4372,7 +4372,9 @@ EXPORT_SYMBOL_GPL(perf_event_read_value);
 static int __perf_read_group_add(struct perf_event *leader,
                                        u64 read_format, u64 *values)
 {
+       struct perf_event_context *ctx = leader->ctx;
        struct perf_event *sub;
+       unsigned long flags;
        int n = 1; /* skip @nr */
        int ret;
 
@@ -4402,12 +4404,15 @@ static int __perf_read_group_add(struct perf_event *leader,
        if (read_format & PERF_FORMAT_ID)
                values[n++] = primary_event_id(leader);
 
+       raw_spin_lock_irqsave(&ctx->lock, flags);
+
        list_for_each_entry(sub, &leader->sibling_list, group_entry) {
                values[n++] += perf_event_count(sub);
                if (read_format & PERF_FORMAT_ID)
                        values[n++] = primary_event_id(sub);
        }
 
+       raw_spin_unlock_irqrestore(&ctx->lock, flags);
        return 0;
 }