linux-stable.git/kernel, branch v3.18.92

perf/core: Fix concurrent sys_perf_event_open() vs. 'move_group' race

2018-01-17T08:29:29+00:00

commit 321027c1fe77f892f4ea07846aeae08cefbbb290 upstream.

Di Shen reported a race between two concurrent sys_perf_event_open()
calls where both try and move the same pre-existing software group
into a hardware context.

The problem is exactly that described in commit:

  f63a8daa5812 ("perf: Fix event->ctx locking")

... where, while we wait for a ctx->mutex acquisition, the event->ctx
relation can have changed under us.

That very same commit failed to recognise sys_perf_event_context() as an
external access vector to the events and thereby didn't apply the
established locking rules correctly.

So while one sys_perf_event_open() call is stuck waiting on
mutex_lock_double(), the other (which owns said locks) moves the group
about. So by the time the former sys_perf_event_open() acquires the
locks, the context we've acquired is stale (and possibly dead).

Apply the established locking rules as per perf_event_ctx_lock_nested()
to the mutex_lock_double() for the 'move_group' case. This obviously means
we need to validate state after we acquire the locks.

Reported-by: Di Shen (Keen Lab)
Tested-by: John Dias 
Signed-off-by: Peter Zijlstra (Intel) 
Cc: Alexander Shishkin 
Cc: Arnaldo Carvalho de Melo 
Cc: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Cc: Kees Cook 
Cc: Linus Torvalds 
Cc: Min Chong 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Cc: Thomas Gleixner 
Cc: Vince Weaver 
Fixes: f63a8daa5812 ("perf: Fix event->ctx locking")
Link: http://lkml.kernel.org/r/20170106131444.GZ3174@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar 
[bwh: Backported to 3.16:
 - Use ACCESS_ONCE() instead of READ_ONCE()
 - Test perf_event::group_flags instead of group_caps
 - Add the err_locked cleanup block, which we didn't need before
 - Adjust context]
Signed-off-by: Ben Hutchings 
Signed-off-by: Suren Baghdasaryan 
Signed-off-by: Amit Pundir 
Signed-off-by: Greg Kroah-Hartman

kernel/signal.c: remove the no longer needed SIGNAL_UNKILLABLE check in complete_signal()

2018-01-17T08:29:27+00:00

commit 426915796ccaf9c2bd9bb06dc5702225957bc2e5 upstream.

complete_signal() checks SIGNAL_UNKILLABLE before it starts to destroy
the thread group, today this is wrong in many ways.

If nothing else, fatal_signal_pending() should always imply that the
whole thread group (except ->group_exit_task if it is not NULL) is
killed, this check breaks the rule.

After the previous changes we can rely on sig_task_ignored();
sig_fatal(sig) && SIGNAL_UNKILLABLE can only be true if we actually want
to kill this task and sig == SIGKILL OR it is traced and debugger can
intercept the signal.

This should hopefully fix the problem reported by Dmitry.  This
test-case

	static int init(void *arg)
	{
		for (;;)
			pause();
	}

	int main(void)
	{
		char stack[16 * 1024];

		for (;;) {
			int pid = clone(init, stack + sizeof(stack)/2,
					CLONE_NEWPID | SIGCHLD, NULL);
			assert(pid > 0);

			assert(ptrace(PTRACE_ATTACH, pid, 0, 0) == 0);
			assert(waitpid(-1, NULL, WSTOPPED) == pid);

			assert(ptrace(PTRACE_DETACH, pid, 0, SIGSTOP) == 0);
			assert(syscall(__NR_tkill, pid, SIGKILL) == 0);
			assert(pid == wait(NULL));
		}
	}

triggers the WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING)) in
task_participate_group_stop().  do_signal_stop()->signal_group_exit()
checks SIGNAL_GROUP_EXIT and return false, but task_set_jobctl_pending()
checks fatal_signal_pending() and does not set JOBCTL_STOP_PENDING.

And his should fix the minor security problem reported by Kyle,
SECCOMP_RET_TRACE can miss fatal_signal_pending() the same way if the
task is the root of a pid namespace.

Link: http://lkml.kernel.org/r/20171103184246.GD21036@redhat.com
Signed-off-by: Oleg Nesterov 
Reported-by: Dmitry Vyukov 
Reported-by: Kyle Huey 
Reviewed-by: Kees Cook 
Tested-by: Kyle Huey 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

kernel/signal.c: protect the SIGNAL_UNKILLABLE tasks from !sig_kernel_only() signals

2018-01-17T08:29:27+00:00

commit ac25385089f673560867eb5179228a44ade0cfc1 upstream.

Change sig_task_ignored() to drop the SIG_DFL && !sig_kernel_only()
signals even if force == T.  This simplifies the next change and this
matches the same check in get_signal() which will drop these signals
anyway.

Link: http://lkml.kernel.org/r/20171103184227.GC21036@redhat.com
Signed-off-by: Oleg Nesterov 
Tested-by: Kyle Huey 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

kernel/signal.c: protect the traced SIGNAL_UNKILLABLE tasks from SIGKILL

2018-01-17T08:29:27+00:00

commit 628c1bcba204052d19b686b5bac149a644cdb72e upstream.

The comment in sig_ignored() says "Tracers may want to know about even
ignored signals" but SIGKILL can not be reported to debugger and it is
just wrong to return 0 in this case: SIGKILL should only kill the
SIGNAL_UNKILLABLE task if it comes from the parent ns.

Change sig_ignored() to ignore ->ptrace if sig == SIGKILL and rely on
sig_task_ignored().

SISGTOP coming from within the namespace is not really right too but at
least debugger can intercept it, and we can't drop it here because this
will break "gdb -p 1": ptrace_attach() won't work.  Perhaps we will add
another ->ptrace check later, we will see.

Link: http://lkml.kernel.org/r/20171103184206.GB21036@redhat.com
Signed-off-by: Oleg Nesterov 
Tested-by: Kyle Huey 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

kernel/acct.c: fix the acct->needcheck check in check_free_space()

2018-01-17T08:29:26+00:00

commit 4d9570158b6260f449e317a5f9ed030c2504a615 upstream.

As Tsukada explains, the time_is_before_jiffies(acct->needcheck) check
is very wrong, we need time_is_after_jiffies() to make sys_acct() work.

Ignoring the overflows, the code should "goto out" if needcheck >
jiffies, while currently it checks "needcheck < jiffies" and thus in the
likely case check_free_space() does nothing until jiffies overflow.

In particular this means that sys_acct() is simply broken, acct_on()
sets acct->needcheck = jiffies and expects that check_free_space()
should set acct->active = 1 after the free-space check, but this won't
happen if jiffies increments in between.

This was broken by commit 32dc73086015 ("get rid of timer in
kern/acct.c") in 2011, then another (correct) commit 795a2f22a8ea
("acct() should honour the limits from the very beginning") made the
problem more visible.

Link: http://lkml.kernel.org/r/20171213133940.GA6554@redhat.com
Fixes: 32dc73086015 ("get rid of timer in kern/acct.c")
Reported-by: TSUKADA Koutaro 
Suggested-by: TSUKADA Koutaro 
Signed-off-by: Oleg Nesterov 
Cc: Al Viro 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

ring-buffer: Mask out the info bits when returning buffer page length

2018-01-02T19:04:58+00:00

commit 45d8b80c2ac5d21cd1e2954431fb676bc2b1e099 upstream.

Two info bits were added to the "commit" part of the ring buffer data page
when returned to be consumed. This was to inform the user space readers that
events have been missed, and that the count may be stored at the end of the
page.

What wasn't handled, was the splice code that actually called a function to
return the length of the data in order to zero out the rest of the page
before sending it up to user space. These data bits were returned with the
length making the value negative, and that negative value was not checked.
It was compared to PAGE_SIZE, and only used if the size was less than
PAGE_SIZE. Luckily PAGE_SIZE is unsigned long which made the compare an
unsigned compare, meaning the negative size value did not end up causing a
large portion of memory to be randomly zeroed out.

Fixes: 66a8cb95ed040 ("ring-buffer: Add place holder recording of dropped events")
Signed-off-by: Steven Rostedt (VMware) 
Signed-off-by: Greg Kroah-Hartman

tracing: Fix crash when it fails to alloc ring buffer

2018-01-02T19:04:57+00:00

commit 24f2aaf952ee0b59f31c3a18b8b36c9e3d3c2cf5 upstream.

Double free of the ring buffer happens when it fails to alloc new
ring buffer instance for max_buffer if TRACER_MAX_TRACE is configured.
The root cause is that the pointer is not set to NULL after the buffer
is freed in allocate_trace_buffers(), and the freeing of the ring
buffer is invoked again later if the pointer is not equal to Null,
as:

instance_mkdir()
    |-allocate_trace_buffers()
        |-allocate_trace_buffer(tr, &tr->trace_buffer...)
	|-allocate_trace_buffer(tr, &tr->max_buffer...)

          // allocate fail(-ENOMEM),first free
          // and the buffer pointer is not set to null
        |-ring_buffer_free(tr->trace_buffer.buffer)

       // out_free_tr
    |-free_trace_buffers()
        |-free_trace_buffer(&tr->trace_buffer);

	      //if trace_buffer is not null, free again
	    |-ring_buffer_free(buf->buffer)
                |-rb_free_cpu_buffer(buffer->buffers[cpu])
                    // ring_buffer_per_cpu is null, and
                    // crash in ring_buffer_per_cpu->pages

Link: http://lkml.kernel.org/r/20171226071253.8968-1-chunyan.zhang@spreadtrum.com

Fixes: 737223fbca3b1 ("tracing: Consolidate buffer allocation code")
Signed-off-by: Jing Xia 
Signed-off-by: Chunyan Zhang 
Signed-off-by: Steven Rostedt (VMware) 
Signed-off-by: Greg Kroah-Hartman

tracing: Fix possible double free on failure of allocating trace buffer

2018-01-02T19:04:57+00:00

commit 4397f04575c44e1440ec2e49b6302785c95fd2f8 upstream.

Jing Xia and Chunyan Zhang reported that on failing to allocate part of the
tracing buffer, memory is freed, but the pointers that point to them are not
initialized back to NULL, and later paths may try to free the freed memory
again. Jing and Chunyan fixed one of the locations that does this, but
missed a spot.

Link: http://lkml.kernel.org/r/20171226071253.8968-1-chunyan.zhang@spreadtrum.com

Fixes: 737223fbca3b1 ("tracing: Consolidate buffer allocation code")
Reported-by: Jing Xia 
Reported-by: Chunyan Zhang 
Signed-off-by: Steven Rostedt (VMware) 
Signed-off-by: Greg Kroah-Hartman

tracing: Remove extra zeroing out of the ring buffer page

2018-01-02T19:04:57+00:00

commit 6b7e633fe9c24682df550e5311f47fb524701586 upstream.

The ring_buffer_read_page() takes care of zeroing out any extra data in the
page that it returns. There's no need to zero it out again from the
consumer. It was removed from one consumer of this function, but
read_buffers_splice_read() did not remove it, and worse, it contained a
nasty bug because of it.

Fixes: 2711ca237a084 ("ring-buffer: Move zeroing out excess in page to ring buffer code")
Signed-off-by: Steven Rostedt (VMware) 
Signed-off-by: Greg Kroah-Hartman

sched/deadline: Use deadline instead of period when calculating overflow

2017-12-20T09:01:30+00:00

[ Upstream commit 2317d5f1c34913bac5971d93d69fb6c31bb74670 ]

I was testing Daniel's changes with his test case, and tweaked it a
little. Instead of having the runtime equal to the deadline, I
increased the deadline ten fold.

Daniel's test case had:

	attr.sched_runtime  = 2 * 1000 * 1000;		/* 2 ms */
	attr.sched_deadline = 2 * 1000 * 1000;		/* 2 ms */
	attr.sched_period   = 2 * 1000 * 1000 * 1000;	/* 2 s */

To make it more interesting, I changed it to:

	attr.sched_runtime  =  2 * 1000 * 1000;		/* 2 ms */
	attr.sched_deadline = 20 * 1000 * 1000;		/* 20 ms */
	attr.sched_period   =  2 * 1000 * 1000 * 1000;	/* 2 s */

The results were rather surprising. The behavior that Daniel's patch
was fixing came back. The task started using much more than .1% of the
CPU. More like 20%.

Looking into this I found that it was due to the dl_entity_overflow()
constantly returning true. That's because it uses the relative period
against relative runtime vs the absolute deadline against absolute
runtime.

  runtime / (deadline - t) > dl_runtime / dl_period

There's even a comment mentioning this, and saying that when relative
deadline equals relative period, that the equation is the same as using
deadline instead of period. That comment is backwards! What we really
want is:

  runtime / (deadline - t) > dl_runtime / dl_deadline

We care about if the runtime can make its deadline, not its period. And
then we can say "when the deadline equals the period, the equation is
the same as using dl_period instead of dl_deadline".

After correcting this, now when the task gets enqueued, it can throttle
correctly, and Daniel's fix to the throttling of sleeping deadline
tasks works even when the runtime and deadline are not the same.

Signed-off-by: Steven Rostedt (VMware) 
Signed-off-by: Peter Zijlstra (Intel) 
Reviewed-by: Daniel Bristot de Oliveira 
Cc: Juri Lelli 
Cc: Linus Torvalds 
Cc: Luca Abeni 
Cc: Mike Galbraith 
Cc: Peter Zijlstra 
Cc: Romulo Silva de Oliveira 
Cc: Steven Rostedt 
Cc: Thomas Gleixner 
Cc: Tommaso Cucinotta 
Link: http://lkml.kernel.org/r/02135a27f1ae3fe5fd032568a5a2f370e190e8d7.1488392936.git.bristot@redhat.com
Signed-off-by: Ingo Molnar 
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman