Re: [PATCH 6/9] signal: Fold do_group_exit into get_signal fixing io_uring threads

[Eric W. Biederman]

Kees Cook <keescook@xxxxxxxxxxxx> writes:

> On Thu, Jun 24, 2021 at 02:02:16PM -0500, Eric W. Biederman wrote:
>> 
>> Forld do_group_exit into get_signal as it is the last caller.
>> 
>> Move the group_exit logic above the PF_IO_WORKER exit, ensuring
>> that if an PF_IO_WORKER catches SIGKILL every thread in
>> the thread group will exit not just the the PF_IO_WORKER.
>> 
>> Now that the information is easily available only set PF_SIGNALED
>> when it was a signal that caused the exit.
>> 
>> Signed-off-by: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx>
>> ---
>>  include/linux/sched/task.h |  1 -
>>  kernel/exit.c              | 31 -------------------------------
>>  kernel/signal.c            | 35 +++++++++++++++++++++++++----------
>>  3 files changed, 25 insertions(+), 42 deletions(-)
>> 
>> diff --git a/include/linux/sched/task.h b/include/linux/sched/task.h
>> index ef02be869cf2..45525512e3d0 100644
>> --- a/include/linux/sched/task.h
>> +++ b/include/linux/sched/task.h
>> @@ -77,7 +77,6 @@ static inline void exit_thread(struct task_struct *tsk)
>>  {
>>  }
>>  #endif
>> -extern void do_group_exit(int);
>>  
>>  extern void exit_files(struct task_struct *);
>>  extern void exit_itimers(struct signal_struct *);
>> diff --git a/kernel/exit.c b/kernel/exit.c
>> index 921519d80b56..635f434122b7 100644
>> --- a/kernel/exit.c
>> +++ b/kernel/exit.c
>> @@ -892,37 +892,6 @@ SYSCALL_DEFINE1(exit, int, error_code)
>>  	do_exit((error_code&0xff)<<8);
>>  }
>>  
>> -/*
>> - * Take down every thread in the group.  This is called by fatal signals
>> - * as well as by sys_exit_group (below).
>> - */
>> -void
>> -do_group_exit(int exit_code)
>> -{
>> -	struct signal_struct *sig = current->signal;
>> -
>> -	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
>> -
>> -	if (signal_group_exit(sig))
>> -		exit_code = sig->group_exit_code;
>> -	else if (!thread_group_empty(current)) {
>> -		struct sighand_struct *const sighand = current->sighand;
>> -
>> -		spin_lock_irq(&sighand->siglock);
>> -		if (signal_group_exit(sig))
>> -			/* Another thread got here before we took the lock.  */
>> -			exit_code = sig->group_exit_code;
>> -		else {
>> -			sig->group_exit_code = exit_code;
>> -			sig->flags = SIGNAL_GROUP_EXIT;
>> -			zap_other_threads(current);
>
> Oh, now I see it: the "new code" in start_group_exit() is an open-coded
> zap_other_threads()? That wasn't clear to me, but makes sense now.

Pretty much.  I think zap_other_threads has actually muddied the waters
quite a bit by putting reuse in the wrong place.

>> -		}
>> -		spin_unlock_irq(&sighand->siglock);
>> -	}
>> -
>> -	do_exit(exit_code);
>> -	/* NOTREACHED */
>> -}
>>  
>>  /*
>>   * this kills every thread in the thread group. Note that any externally
>> diff --git a/kernel/signal.c b/kernel/signal.c
>> index c79c010ca5f3..95a076af600a 100644
>> --- a/kernel/signal.c
>> +++ b/kernel/signal.c
>> @@ -2646,6 +2646,7 @@ bool get_signal(struct ksignal *ksig)
>>  {
>>  	struct sighand_struct *sighand = current->sighand;
>>  	struct signal_struct *signal = current->signal;
>> +	int exit_code;
>>  	int signr;
>>  
>>  	if (unlikely(current->task_works))
>> @@ -2848,8 +2849,6 @@ bool get_signal(struct ksignal *ksig)
>>  		/*
>>  		 * Anything else is fatal, maybe with a core dump.
>>  		 */
>> -		current->flags |= PF_SIGNALED;
>> -
>>  		if (sig_kernel_coredump(signr)) {
>>  			if (print_fatal_signals)
>>  				print_fatal_signal(ksig->info.si_signo);
>> @@ -2857,14 +2856,33 @@ bool get_signal(struct ksignal *ksig)
>>  			/*
>>  			 * If it was able to dump core, this kills all
>>  			 * other threads in the group and synchronizes with
>> -			 * their demise.  If we lost the race with another
>> -			 * thread getting here, it set group_exit_code
>> -			 * first and our do_group_exit call below will use
>> -			 * that value and ignore the one we pass it.
>> +			 * their demise.  If  another thread makes it
>> +			 * to do_coredump first, it will set group_exit_code
>> +			 * which will be passed to do_exit.
>>  			 */
>>  			do_coredump(&ksig->info);
>>  		}
>>  
>> +		/*
>> +		 * Death signals, no core dump.
>> +		 */
>> +		exit_code = signr;
>> +		if (signal_group_exit(signal)) {
>> +			exit_code = signal->group_exit_code;
>> +		} else {
>> +			spin_lock_irq(&sighand->siglock);
>> +			if (signal_group_exit(signal)) {
>> +				/* Another thread got here before we took the lock.  */
>> +				exit_code = signal->group_exit_code;
>> +			} else {
>> +				start_group_exit_locked(signal, exit_code);
>
> And here's the "if we didn't already do start_group_exit(), do it here".
> And that state is entirely captured via the SIGNAL_GROUP_EXIT flag.
> Cool.

Yes.  At least when the dust clears. 

>> +			}
>> +			spin_unlock_irq(&sighand->siglock);
>> +		}
>> +
>> +		if (exit_code & 0x7f)
>> +			current->flags |= PF_SIGNALED;
>> +
>>  		/*
>>  		 * PF_IO_WORKER threads will catch and exit on fatal signals
>>  		 * themselves. They have cleanup that must be performed, so
>> @@ -2873,10 +2891,7 @@ bool get_signal(struct ksignal *ksig)
>>  		if (current->flags & PF_IO_WORKER)
>>  			goto out;
>>  
>> -		/*
>> -		 * Death signals, no core dump.
>> -		 */
>> -		do_group_exit(ksig->info.si_signo);
>> +		do_exit(exit_code);
>>  		/* NOTREACHED */
>>  	}
>>  	spin_unlock_irq(&sighand->siglock);
>> -- 
>> 2.20.1
>>