Re: [PATCH v3] perf/core: Fix missing wakeup when waiting for context reference

[Mark Rutland <mark.rutland@arm.com>]

On Thu, Apr 18, 2024 at 11:42:09AM +0000, Haifeng Xu wrote:
> In our production environment, we found many hung tasks which are
> blocked for more than 18 hours. Their call traces are like this:
> 
> [346278.191038] __schedule+0x2d8/0x890
> [346278.191046] schedule+0x4e/0xb0
> [346278.191049] perf_event_free_task+0x220/0x270
> [346278.191056] ? init_wait_var_entry+0x50/0x50
> [346278.191060] copy_process+0x663/0x18d0
> [346278.191068] kernel_clone+0x9d/0x3d0
> [346278.191072] __do_sys_clone+0x5d/0x80
> [346278.191076] __x64_sys_clone+0x25/0x30
> [346278.191079] do_syscall_64+0x5c/0xc0
> [346278.191083] ? syscall_exit_to_user_mode+0x27/0x50
> [346278.191086] ? do_syscall_64+0x69/0xc0
> [346278.191088] ? irqentry_exit_to_user_mode+0x9/0x20
> [346278.191092] ? irqentry_exit+0x19/0x30
> [346278.191095] ? exc_page_fault+0x89/0x160
> [346278.191097] ? asm_exc_page_fault+0x8/0x30
> [346278.191102] entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> The task was waiting for the refcount become to 1, but from the vmcore,
> we found the refcount has already been 1. It seems that the task didn't
> get woken up by perf_event_release_kernel() and got stuck forever. The
> below scenario may cause the problem.
> 
> Thread A					Thread B
> ...						...
> perf_event_free_task				perf_event_release_kernel
> 						   ...
> 						   acquire event->child_mutex
> 						   ...
> 						   get_ctx
>    ...						   release event->child_mutex
>    acquire ctx->mutex
>    ...
>    perf_free_event (acquire/release event->child_mutex)
>    ...
>    release ctx->mutex
>    wait_var_event
> 						   acquire ctx->mutex
> 						   acquire event->child_mutex
> 						   # move existing events to free_list
> 						   release event->child_mutex
> 						   release ctx->mutex
> 						   put_ctx
> ...						...
> 
> In this case, all events of the ctx have been freed, so we couldn't
> find the ctx in free_list and Thread A will miss the wakeup. It's thus
> necessary to add a wakeup after dropping the reference.
> 
> Fixes: 1cf8dfe8a661 ("perf/core: Fix race between close() and fork()")
> Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
> Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

FWIW, this looks good to me, but I haven't yet been able to write a test to
exercise this: perf_event_free_task() is only called if
perf_event_init_context() fails or of copy_process() fails partway through, and
while it should be possible to make the latter fail consistently by messing
with cgroups, I haven't had the time to work all that out.

So I think there's a reliable DoS here, but I haven't had the time to go write
that myself. It would be nice if we actually had a test for this.

I reckon that in addition to the Fixes tag, this needs:

Cc: stable@vger.kernel.org

> ---
> Changes since v1:
> - Add the fixed tag.
> - Simplify v1's patch. (Frederic)
> 
> Changes since v2:
> - Use Reviewed-by tag instead of Signed-off-by tag.
> ---
>  kernel/events/core.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 4f0c45ab8d7d..15c35070db6a 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -5340,6 +5340,7 @@ int perf_event_release_kernel(struct perf_event *event)
>  again:
>  	mutex_lock(&event->child_mutex);
>  	list_for_each_entry(child, &event->child_list, child_list) {
> +		void *var = NULL;
>  
>  		/*
>  		 * Cannot change, child events are not migrated, see the
> @@ -5380,11 +5381,23 @@ int perf_event_release_kernel(struct perf_event *event)
>  			 * this can't be the last reference.
>  			 */
>  			put_event(event);
> +		} else {
> +			var = &ctx->refcount;
>  		}
>  
>  		mutex_unlock(&event->child_mutex);
>  		mutex_unlock(&ctx->mutex);
>  		put_ctx(ctx);
> +
> +		if (var) {
> +			/*
> +			 * If perf_event_free_task() has deleted all events from the
> +			 * ctx while the child_mutex got released above, make sure to
> +			 * notify about the preceding put_ctx().
> +			 */
> +			smp_mb(); /* pairs with wait_var_event() */
> +			wake_up_var(var);
> +		}
>  		goto again;
>  	}
>  	mutex_unlock(&event->child_mutex);

I was a bit worrited that we're doing the wakeup with the event->child_mutex
held; AFAICT that looks to be safe, but I'm not a scheduler expert.

FWIW:

Acked-by: Mark Rutland <mark.rutland@arm.com>

Mark.

> -- 
> 2.25.1
>