linux-kernel - Re: [PATCH] staging: android/lowmemorykiller: Don't grab tasklist

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:	Thu, 2 Feb 2012 21:16:59 +0400
From:	Anton Vorontsov <anton.vorontsov@...aro.org>
To:	Oleg Nesterov <oleg@...hat.com>, Greg KH <gregkh@...e.de>
Cc:	Arve Hjønnevåg <arve@...roid.com>,
	KOSAKI Motohiro <kosaki.motohiro@...il.com>,
	San Mehat <san@...gle.com>, Colin Cross <ccross@...roid.com>,
	"Eric W. Biederman" <ebiederm@...ssion.com>,
	linux-kernel@...r.kernel.org, kernel-team@...roid.com,
	linaro-kernel@...ts.linaro.org,
	"Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
Subject: Re: [PATCH] staging: android/lowmemorykiller: Don't grab
 tasklist_lock

On Thu, Feb 02, 2012 at 01:54:41PM +0100, Oleg Nesterov wrote:
> On 02/01, Anton Vorontsov wrote:
> >
> > @@ -132,7 +133,7 @@ static int lowmem_shrink(struct shrinker *s, struct shrink_control *sc)
> >  	}
> >  	selected_oom_adj = min_adj;
> >
> > -	read_lock(&tasklist_lock);
> > +	rcu_read_lock();
> 
> This has the same problem, force_sig() becomes unsafe.

Ouch, I think I finally got it. So, lock_task_sighand() is trying to
gracefully grab the lock, checking if the sighand is not NULL (which means,
per __exit_signal(), that the task is halfway into the grave).

Well, it seems that such a behaviour of force_sig() is not quite obvious,
and there are other offenders out there. E.g. in sysrq code I don't see
anything that prevent the same race.

static void send_sig_all(int sig)
{      
        struct task_struct *p;

        for_each_process(p) {
                if (p->mm && !is_global_init(p))
                        /* Not swapper, init nor kernel thread */
                        force_sig(sig, p);
        }
}

Would the following fix work for the sysrq?

- - - -
From: Anton Vorontsov <anton.vorontsov@...aro.org>
Subject: [PATCH] sysrq: Fix unsafe operations on tasks

sysrq should grab the tasklist lock, otherwise calling force_sig() is
not safe, as it might race with exiting task, which ->sighand might be
set to NULL already.

Signed-off-by: Anton Vorontsov <anton.vorontsov@...aro.org>
---
 drivers/tty/sysrq.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/tty/sysrq.c b/drivers/tty/sysrq.c
index 7867b7c..a1bcad7 100644
--- a/drivers/tty/sysrq.c
+++ b/drivers/tty/sysrq.c
@@ -322,11 +322,13 @@ static void send_sig_all(int sig)
 {
 	struct task_struct *p;
 
+	read_lock(&tasklist_lock);
 	for_each_process(p) {
 		if (p->mm && !is_global_init(p))
 			/* Not swapper, init nor kernel thread */
 			force_sig(sig, p);
 	}
+	read_unlock(&tasklist_lock);
 }
 
 static void sysrq_handle_term(int key)
- - - -

But for LMK I will use send_signal(), as LMK is special.

Plus, while I'm at it, might want to review a bit closer other offenders,
and fix them as well.

> Why do you need force_? Do you really want to kill /sbin/init (or sub-namespace
> init) ?

Nope.

> We could change force_sig_info() to use lock_task_sighand(), but I'd like to
> avoid this. Imho, this interface should be cleanuped, and it should be used
> for synchronous signals only.

OK. Then we should fix the users?

> With or without this patch, sig == NULL is not possible but !mm is not right,
> there could be other other threads with mm != NULL.

I'm not sure I completely follow. In the current LMK code, we check for !mm
because otherwise the potential victim is useless for us (i.e. killing it
will not free much memory anyway).

Thanks!

-- 
Anton Vorontsov
Email: cbouatmailru@...il.com
--
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/