perf: Close install vs. exit race
authorPeter Zijlstra <peterz@infradead.org>
Wed, 24 Feb 2016 17:45:40 +0000 (18:45 +0100)
committerIngo Molnar <mingo@kernel.org>
Thu, 25 Feb 2016 07:42:32 +0000 (08:42 +0100)
Consider the following scenario:

  CPU0 CPU1

  ctx = find_get_ctx();
perf_event_exit_task_context()
  mutex_lock(&ctx->mutex);
  perf_install_in_context(ctx, ...);
    /* NO-OP */
  mutex_unlock(&ctx->mutex);

  ...

  perf_release()
    WARN_ON_ONCE(event->state != STATE_EXIT);

Since the event doesn't pass through perf_remove_from_context()
because perf_install_in_context() NO-OPs because the ctx is dead, and
perf_event_exit_task_context() will not observe the event because its
not attached yet, the event->state will not be set.

Solve this by revalidating ctx->task after we acquire ctx->mutex and
failing the event creation as a whole.

Tested-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: 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 <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: dvyukov@google.com
Cc: eranian@google.com
Cc: oleg@redhat.com
Cc: panand@redhat.com
Cc: sasha.levin@oracle.com
Cc: vince@deater.net
Link: http://lkml.kernel.org/r/20160224174947.626853419@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/core.c

index 0d58522103cd7e73d76a7935119077f33d68082b..d7b0316e3465faf742d8add01883c173dc32c67d 100644 (file)
@@ -2158,13 +2158,15 @@ perf_install_in_context(struct perf_event_context *ctx,
         */
        raw_spin_lock_irq(&ctx->lock);
        task = ctx->task;
+
        /*
-        * Worse, we cannot even rely on the ctx actually existing anymore. If
-        * between find_get_context() and perf_install_in_context() the task
-        * went through perf_event_exit_task() its dead and we should not be
-        * adding new events.
+        * If between ctx = find_get_context() and mutex_lock(&ctx->mutex) the
+        * ctx gets destroyed, we must not install an event into it.
+        *
+        * This is normally tested for after we acquire the mutex, so this is
+        * a sanity check.
         */
-       if (task == TASK_TOMBSTONE) {
+       if (WARN_ON_ONCE(task == TASK_TOMBSTONE)) {
                raw_spin_unlock_irq(&ctx->lock);
                return;
        }
@@ -8389,10 +8391,19 @@ SYSCALL_DEFINE5(perf_event_open,
        if (move_group) {
                gctx = group_leader->ctx;
                mutex_lock_double(&gctx->mutex, &ctx->mutex);
+               if (gctx->task == TASK_TOMBSTONE) {
+                       err = -ESRCH;
+                       goto err_locked;
+               }
        } else {
                mutex_lock(&ctx->mutex);
        }
 
+       if (ctx->task == TASK_TOMBSTONE) {
+               err = -ESRCH;
+               goto err_locked;
+       }
+
        if (!perf_event_validate_size(event)) {
                err = -E2BIG;
                goto err_locked;
@@ -8563,12 +8574,14 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
 
        WARN_ON_ONCE(ctx->parent_ctx);
        mutex_lock(&ctx->mutex);
+       if (ctx->task == TASK_TOMBSTONE) {
+               err = -ESRCH;
+               goto err_unlock;
+       }
+
        if (!exclusive_event_installable(event, ctx)) {
-               mutex_unlock(&ctx->mutex);
-               perf_unpin_context(ctx);
-               put_ctx(ctx);
                err = -EBUSY;
-               goto err_free;
+               goto err_unlock;
        }
 
        perf_install_in_context(ctx, event, cpu);
@@ -8577,6 +8590,10 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
 
        return event;
 
+err_unlock:
+       mutex_unlock(&ctx->mutex);
+       perf_unpin_context(ctx);
+       put_ctx(ctx);
 err_free:
        free_event(event);
 err: