linux-kernel - Re: [PATCH RFC] time,signal: protect resource use statistics with seqlock

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: <20140815145813.GA15379@redhat.com>
Date:	Fri, 15 Aug 2014 16:58:13 +0200
From:	Oleg Nesterov <oleg@...hat.com>
To:	Rik van Riel <riel@...hat.com>
Cc:	linux-kernel@...r.kernel.org,
	Peter Zijlstra <peterz@...radead.org>,
	Hidetoshi Seto <seto.hidetoshi@...fujitsu.com>,
	Frank Mayhar <fmayhar@...gle.com>,
	Frederic Weisbecker <fweisbec@...hat.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Sanjay Rao <srao@...hat.com>,
	Larry Woodman <lwoodman@...hat.com>
Subject: Re: [PATCH RFC] time,signal: protect resource use statistics with
	seqlock

On 08/14, Rik van Riel wrote:
>
> @@ -288,18 +288,31 @@ void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
>  	struct signal_struct *sig = tsk->signal;
>  	cputime_t utime, stime;
>  	struct task_struct *t;
> -
> -	times->utime = sig->utime;
> -	times->stime = sig->stime;
> -	times->sum_exec_runtime = sig->sum_sched_runtime;
> +	unsigned int seq, nextseq;
>  
>  	rcu_read_lock();
> -	for_each_thread(tsk, t) {
> -		task_cputime(t, &utime, &stime);
> -		times->utime += utime;
> -		times->stime += stime;
> -		times->sum_exec_runtime += task_sched_runtime(t);
> -	}
> +	/* Attempt a lockless read on the first round. */
> +	nextseq = 0;
> +	do {
> +		seq = nextseq;
> +		read_seqbegin_or_lock(&sig->stats_lock, &seq);
> +		times->utime = sig->utime;
> +		times->stime = sig->stime;
> +		times->sum_exec_runtime = sig->sum_sched_runtime;
> +
> +		for_each_thread(tsk, t) {
> +			task_cputime(t, &utime, &stime);
> +			times->utime += utime;
> +			times->stime += stime;
> +			times->sum_exec_runtime += task_sched_runtime(t);
> +		}
> +		/*
> +		 * If a writer is currently active, seq will be odd, and
> +		 * read_seqbegin_or_lock will take the lock.
> +		 */
> +		nextseq = raw_read_seqcount(&sig->stats_lock.seqcount);
> +	} while (need_seqretry(&sig->stats_lock, seq));
> +	done_seqretry(&sig->stats_lock, seq);
>  	rcu_read_unlock();
>  }

I still think this is not right. Let me quote my previous email,

	> @@ -288,18 +288,31 @@ void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
	>  	struct signal_struct *sig = tsk->signal;
	>  	cputime_t utime, stime;
	>  	struct task_struct *t;
	> -
	> -	times->utime = sig->utime;
	> -	times->stime = sig->stime;
	> -	times->sum_exec_runtime = sig->sum_sched_runtime;
	> +	unsigned int seq, nextseq;
	>
	>  	rcu_read_lock();

	Almost cosmetic nit, but afaics this patch expands the rcu critical section
	for no reason. We only need rcu_read_lock/unlock around for_each_thread()
	below.

	> +	nextseq = 0;
	> +	do {
	> +		seq = nextseq;
	> +		read_seqbegin_or_lock(&sig->stats_lock, &seq);
	> +		times->utime = sig->utime;
	> +		times->stime = sig->stime;
	> +		times->sum_exec_runtime = sig->sum_sched_runtime;
	> +
	> +		for_each_thread(tsk, t) {
	> +			task_cputime(t, &utime, &stime);
	> +			times->utime += utime;
	> +			times->stime += stime;
	> +			times->sum_exec_runtime += task_sched_runtime(t);
	> +		}
	> +		/*
	> +		 * If a writer is currently active, seq will be odd, and
	> +		 * read_seqbegin_or_lock will take the lock.
	> +		 */
	> +		nextseq = raw_read_seqcount(&sig->stats_lock.seqcount);
	> +	} while (need_seqretry(&sig->stats_lock, seq));
	> +	done_seqretry(&sig->stats_lock, seq);

	Hmm. It seems that read_seqbegin_or_lock() is not used correctly. I mean,
	this code still can livelock in theory. Just suppose that anoter CPU does
	write_seqlock/write_sequnlock right after read_seqbegin_or_lock(). In this
	case "seq & 1" will be never true and thus "or_lock" will never happen.

	IMO, this should be fixed. Either we should guarantee the forward progress
	or we should not play with read_seqbegin_or_lock() at all. This code assumes
	that sooner or later "nextseq = raw_read_seqcount()" should return the odd
	counter, but in theory this can never happen.

	And if we want to fix this we do not need 2 counters, just we need to set
	"seq = 1" manually after need_seqretry() == T. Say, like __dentry_path() does.
	(but unlike __dentry_path() we do not need to worry about rcu_read_unlock so
	the code will be simpler).

	I am wondering if it makes sense to introduce

		bool read_seqretry_or_lock(const seqlock_t *sl, int *seq)
		{
			if (*seq & 1) {
				read_sequnlock_excl(lock);
				return false;
			}
		
			if (!read_seqretry(lock, *seq))
				return false;
		
			*seq = 1;
			return true;
		}

Or I missed your reply?

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/