perf: Fix race in perf_event_exit_task_context()

There is a race between perf_event_exit_task_context() and
orphans_remove_work() which results in a use-after-free.

We mark ctx->task with TASK_TOMBSTONE to indicate a context is
'dead', under ctx->lock. After which point event_function_call()
on any event of that context will NOP

A concurrent orphans_remove_work() will only hold ctx->mutex for
the list iteration and not serialize against this. Therefore its
possible that orphans_remove_work()'s perf_remove_from_context()
call will fail, but we'll continue to free the event, with the
result of free'd memory still being on lists and everything.

Once perf_event_exit_task_context() gets around to acquiring
ctx->mutex it too will iterate the event list, encounter the
already free'd event and proceed to free it _again_. This fails
with the WARN in free_event().

Plug the race by having perf_event_exit_task_context() hold
ctx::mutex over the whole tear-down, thereby 'naturally'
serializing against all other sites, including the orphan work.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
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: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: alexander.shishkin@linux.intel.com
Cc: dsahern@gmail.com
Cc: namhyung@kernel.org
Link: http://lkml.kernel.org/r/20160125130954.GY6357@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>

1 files changed, 29 insertions, 21 deletions

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 6759f2a332d7..1d243fadfd12 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -8748,14 +8748,40 @@ static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
 {
         struct perf_event_context *child_ctx, *clone_ctx = NULL;
         struct perf_event *child_event, *next;
-        unsigned long flags;
 
         WARN_ON_ONCE(child != current);
 
-        child_ctx = perf_lock_task_context(child, ctxn, &flags);
+        child_ctx = perf_pin_task_context(child, ctxn);
         if (!child_ctx)
                 return;
 
+        /*
+         * In order to reduce the amount of tricky in ctx tear-down, we hold
+         * ctx::mutex over the entire thing. This serializes against almost
+         * everything that wants to access the ctx.
+         *
+         * The exception is sys_perf_event_open() /
+         * perf_event_create_kernel_count() which does find_get_context()
+         * without ctx::mutex (it cannot because of the move_group double mutex
+         * lock thing). See the comments in perf_install_in_context().
+         *
+         * We can recurse on the same lock type through:
+         *
+         *   __perf_event_exit_task()
+         *     sync_child_event()
+         *       put_event()
+         *         mutex_lock(&ctx->mutex)
+         *
+         * But since its the parent context it won't be the same instance.
+         */
+        mutex_lock(&child_ctx->mutex);
+
+        /*
+         * In a single ctx::lock section, de-schedule the events and detach the
+         * context from the task such that we cannot ever get it scheduled back
+         * in.
+         */
+        raw_spin_lock_irq(&child_ctx->lock);
         task_ctx_sched_out(__get_cpu_context(child_ctx), child_ctx);
 
         /*
@@ -8767,14 +8793,8 @@ static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
         WRITE_ONCE(child_ctx->task, TASK_TOMBSTONE);
         put_task_struct(current); /* cannot be last */
 
-        /*
-         * If this context is a clone; unclone it so it can't get
-         * swapped to another process while we're removing all
-         * the events from it.
-         */
         clone_ctx = unclone_ctx(child_ctx);
-        update_context_time(child_ctx);
-        raw_spin_unlock_irqrestore(&child_ctx->lock, flags);
+        raw_spin_unlock_irq(&child_ctx->lock);
 
         if (clone_ctx)
                 put_ctx(clone_ctx);
@@ -8786,18 +8806,6 @@ static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
          */
         perf_event_task(child, child_ctx, 0);
 
-        /*
-         * We can recurse on the same lock type through:
-         *
-         *   __perf_event_exit_task()
-         *     sync_child_event()
-         *       put_event()
-         *         mutex_lock(&ctx->mutex)
-         *
-         * But since its the parent context it won't be the same instance.
-         */
-        mutex_lock(&child_ctx->mutex);
-
         list_for_each_entry_safe(child_event, next, &child_ctx->event_list, event_entry)
                 __perf_event_exit_task(child_event, child_ctx, child);