linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] workqueue/tracing: Copy workqueue name to buffer in trace event
@ 2021-03-18 15:55 Steven Rostedt
  0 siblings, 0 replies; only message in thread
From: Steven Rostedt @ 2021-03-18 15:55 UTC (permalink / raw)
  To: LKML; +Cc: Tejun Heo, Linus Torvalds, Ingo Molnar, Andrew Morton

From: "Steven Rostedt (VMware)" <rostedt@goodmis.org>

The trace event "workqueue_queue_work" references an unsafe string in
dereferencing the name of the workqueue. As the name is allocated, it
could later be freed, and the pointer to that string could stay on the
tracing buffer. If the trace buffer is read after the string is freed, it
will reference an unsafe pointer.

I added a new verifier to make sure that all strings referenced in the
output of the trace buffer is safe to read and this triggered on the
workqueue_queue_work trace event:

workqueue_queue_work: work struct=00000000b2b235c7 function=gc_worker workqueue=(0xffff888100051160:events_power_efficient)[UNSAFE-MEMORY] req_cpu=256 cpu=1
workqueue_queue_work: work struct=00000000c344caec function=flush_to_ldisc workqueue=(0xffff888100054d60:events_unbound)[UNSAFE-MEMORY] req_cpu=256 cpu=4294967295
workqueue_queue_work: work struct=00000000b2b235c7 function=gc_worker workqueue=(0xffff888100051160:events_power_efficient)[UNSAFE-MEMORY] req_cpu=256 cpu=1
workqueue_queue_work: work struct=000000000b238b3f function=vmstat_update workqueue=(0xffff8881000c3760:mm_percpu_wq)[UNSAFE-MEMORY] req_cpu=1 cpu=1

Also, if this event is read via a user space application like perf or
trace-cmd, the name would only be an address and useless information:

workqueue_queue_work: work struct=0xffff953f80b4b918 function=disk_events_workfn workqueue=ffff953f8005d378 req_cpu=8192 cpu=5

Cc: stable@vger.kernel.org
Fixes: 7bf9c4a88e3e3 ("workqueue: tracing the name of the workqueue instead of it's address")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---

[
  This is another reason I need to get my verifier into the kernel,
  because it would have triggered a WARN_ON() and stopped the broken
  patch from getting in, in the first place.
]

 include/trace/events/workqueue.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/include/trace/events/workqueue.h b/include/trace/events/workqueue.h
index 970cc2ea2850..6154a2e72bce 100644
--- a/include/trace/events/workqueue.h
+++ b/include/trace/events/workqueue.h
@@ -30,7 +30,7 @@ TRACE_EVENT(workqueue_queue_work,
 	TP_STRUCT__entry(
 		__field( void *,	work	)
 		__field( void *,	function)
-		__field( const char *,	workqueue)
+		__string( workqueue,	pwq->wq->name)
 		__field( unsigned int,	req_cpu	)
 		__field( unsigned int,	cpu	)
 	),
@@ -38,13 +38,13 @@ TRACE_EVENT(workqueue_queue_work,
 	TP_fast_assign(
 		__entry->work		= work;
 		__entry->function	= work->func;
-		__entry->workqueue	= pwq->wq->name;
+		__assign_str(workqueue, pwq->wq->name);
 		__entry->req_cpu	= req_cpu;
 		__entry->cpu		= pwq->pool->cpu;
 	),
 
 	TP_printk("work struct=%p function=%ps workqueue=%s req_cpu=%u cpu=%u",
-		  __entry->work, __entry->function, __entry->workqueue,
+		  __entry->work, __entry->function, __get_str(workqueue),
 		  __entry->req_cpu, __entry->cpu)
 );
 
-- 
2.29.2


^ permalink raw reply related	[flat|nested] only message in thread

only message in thread, other threads:[~2021-03-18 15:56 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-03-18 15:55 [PATCH] workqueue/tracing: Copy workqueue name to buffer in trace event Steven Rostedt

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).