linux-kernel - Re: [PATCH] mm/kmemleak: use READ_ONCE() for accessing jiffies_scan

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]

Date:   Fri, 11 Jun 2021 09:59:13 +0100
From:   Catalin Marinas <catalin.marinas@....com>
To:     Yanfei Xu <yanfei.xu@...driver.com>
Cc:     akpm@...ux-foundation.org, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mm/kmemleak: use READ_ONCE() for accessing
 jiffies_scan_wait

On Wed, Jun 09, 2021 at 11:56:57PM +0800, Yanfei Xu wrote:
> The stop_scan_thread() and start_scan_thread() cannot really solve
> the problem of concurrent accessing the global jiffies_scan_wait.
> 
> kmemleak_write              kmemleak_scan_thread
>                               while (!kthread_should_stop())
>   stop_scan_thread
>   jiffies_scan_wait = xxx       timeout = jiffies_scan_wait
>   start_scan_thread
> 
> We could replace these with a READ_ONCE() when reading
> jiffies_scan_wait. It also can prevent compiler from reordering the
> jiffies_scan_wait which is in while loop.

I'm ok with READ_ONCE but your patch introduces functional changes.

> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index 92a2d4885808..5ccf3969b7fe 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -1567,7 +1567,7 @@ static int kmemleak_scan_thread(void *arg)
>  	}
>  
>  	while (!kthread_should_stop()) {
> -		signed long timeout = jiffies_scan_wait;
> +		signed long timeout = READ_ONCE(jiffies_scan_wait);
>  
>  		mutex_lock(&scan_mutex);
>  		kmemleak_scan();
> @@ -1812,11 +1812,8 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
>  		ret = kstrtoul(buf + 5, 0, &secs);
>  		if (ret < 0)
>  			goto out;
> -		stop_scan_thread();
> -		if (secs) {
> +		if (secs)
>  			jiffies_scan_wait = msecs_to_jiffies(secs * 1000);

For symmetry, I'd add a WRITE_ONCE here as well.

> -			start_scan_thread();
> -		}

The reason for stop/start_scan_thread() wasn't to protect against
jiffies_scan_wait access but rather to force a new delay. Let's say you
start by default with a 10min delay between scans (default) but you want
to lower it to 1min. With the above removal of stop/start, you'd still
have to wait for 10min until the scanning thread will notice the change.
Also, with secs=0, the expectations is that the thread won't be
restarted but this is removed by your patch.

-- 
Catalin