linux-kernel - Re: [PATCH 1/1] perf/core: fix dangling cgroup pointer in cpuctx

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20250603144414.GC38114@noisy.programming.kicks-ass.net>
Date: Tue, 3 Jun 2025 16:44:14 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Leo Yan <leo.yan@....com>
Cc: Yeoreum Yun <yeoreum.yun@....com>, mingo@...hat.com, mingo@...nel.org,
	acme@...nel.org, namhyung@...nel.org, mark.rutland@....com,
	alexander.shishkin@...ux.intel.com, jolsa@...nel.org,
	irogers@...gle.com, adrian.hunter@...el.com,
	kan.liang@...ux.intel.com, linux-perf-users@...r.kernel.org,
	linux-kernel@...r.kernel.org, David Wang <00107082@....com>
Subject: Re: [PATCH 1/1] perf/core: fix dangling cgroup pointer in cpuctx

On Tue, Jun 03, 2025 at 03:00:40PM +0100, Leo Yan wrote:

> > +	if (event->state > PERF_EVENT_STATE_OFF)
> > +		perf_cgroup_event_disable(event, ctx);
> > +
> 
> As we discussed, seems to me, the issue is caused by an ambigous state
> machine transition:
> 
> When a PMU event state is PERF_EVENT_STATE_EXIT, the current code does
> not transite the state to PERF_EVENT_STATE_OFF. As a result, the
> list_del_event() function skips to clean up cgroup pointer for non OFF
> states. This is different from the code prior to the commit a3c3c6667,
> which transits states EXIT -> INACTIVE -> OFF.

Right.

> My suggestion is not reliable. Roughly read code, except for the
> PERF_EVENT_STATE_EXIT case, I think other error cases should also clean
> up the cgroup pointer.  The reason is I don't see other places to
> clean up the cgroup pointer for these error cases:
> 
>   PERF_EVENT_STATE_REVOKED
>   PERF_EVENT_STATE_DEAD

Those should be done here; on the first transition into these states.

> Only in the PERF_EVENT_STATE_ERROR state, we don't need to cleanup
> cgroup as this has already been handled in merge_sched_in().
> 
> So a correct condition would be:
> 
>     if (event->state > PERF_EVENT_STATE_OFF ||
>         event->state <= PERF_EVENT_STATE_EXIT)
>         perf_cgroup_event_disable(event, ctx);

I'm too tired to get my head straight. I'll look tomorrow.

> And we need to remove the perf_cgroup_event_disable() from
> list_del_event() to avoid duplicate code.
> 
> Perhaps a better approach for code consolidation would be to modify
> the conditions in list_del_event() to ensure the cgroup pointer is
> cleaned up in error cases. However, I'm not confident that this is the
> correct direction, so I would wait for suggestions from the maintainers.

Probably easier to keep here in __perf_remove_from_context() where we
have prev and next state available.

Anyway, I currently have the below, but I'll update once I've had sleep.

---
Subject: perf: Fix dangling cgroup pointer in cpuctx
From: Yeoreum Yun <yeoreum.yun@....com>
Date: Mon, 2 Jun 2025 19:40:49 +0100

From: Yeoreum Yun <yeoreum.yun@....com>

Commit a3c3c6667("perf/core: Fix child_total_time_enabled accounting
bug at task exit") moves the event->state update to before
list_del_event(). This makes the event->state test in list_del_event()
always false; never calling perf_cgroup_event_disable().

As a result, cpuctx->cgrp won't be cleared properly; causing havoc.

Fixes: a3c3c6667("perf/core: Fix child_total_time_enabled accounting bug at task exit")
Signed-off-by: Yeoreum Yun <yeoreum.yun@....com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
Tested-by: David Wang <00107082@....com>
Link: https://lore.kernel.org/all/aD2TspKH%2F7yvfYoO@e129823.arm.com/ [0]
---
 kernel/events/core.c |   21 ++++++---------------
 1 file changed, 6 insertions(+), 15 deletions(-)

--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2120,18 +2120,6 @@ list_del_event(struct perf_event *event,
 	if (event->group_leader == event)
 		del_event_from_groups(event, ctx);
 
-	/*
-	 * If event was in error state, then keep it
-	 * that way, otherwise bogus counts will be
-	 * returned on read(). The only way to get out
-	 * of error state is by explicit re-enabling
-	 * of the event
-	 */
-	if (event->state > PERF_EVENT_STATE_OFF) {
-		perf_cgroup_event_disable(event, ctx);
-		perf_event_set_state(event, PERF_EVENT_STATE_OFF);
-	}
-
 	ctx->generation++;
 	event->pmu_ctx->nr_events--;
 }
@@ -2493,11 +2481,14 @@ __perf_remove_from_context(struct perf_e
 		state = PERF_EVENT_STATE_EXIT;
 	if (flags & DETACH_REVOKE)
 		state = PERF_EVENT_STATE_REVOKED;
-	if (flags & DETACH_DEAD) {
-		event->pending_disable = 1;
+	if (flags & DETACH_DEAD)
 		state = PERF_EVENT_STATE_DEAD;
-	}
+
 	event_sched_out(event, ctx);
+
+	if (event->state > PERF_EVENT_STATE_OFF)
+		perf_cgroup_event_disable(event, ctx);
+
 	perf_event_set_state(event, min(event->state, state));
 
 	if (flags & DETACH_GROUP)