linux-kernel - Re: [PATCH v2] kernel/fork: Increase minimum number of allowed threads

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: <20250721192830.6f8f57d9@pumpkin>
Date: Mon, 21 Jul 2025 19:28:30 +0100
From: David Laight <david.laight.linux@...il.com>
To: Hauke Mehrtens <hauke@...ke-m.de>
Cc: Jiri Slaby <jirislaby@...nel.org>, sashal@...nel.org,
 linux-kernel@...r.kernel.org, frederic@...nel.org, david@...hat.com,
 viro@...iv.linux.org.uk, paulmck@...nel.org, stable@...r.kernel.org, Tejun
 Heo <tj@...nel.org>, Lai Jiangshan <jiangshanlai@...il.com>, "Luis R .
 Rodriguez" <mcgrof@...nel.org>
Subject: Re: [PATCH v2] kernel/fork: Increase minimum number of allowed
 threads

On Sun, 20 Jul 2025 17:28:20 +0200
Hauke Mehrtens <hauke@...ke-m.de> wrote:

> Hi,
> 
> I am not exactly sure how I should limit the number of parallel user 
> mode helper calls.
> The user mode helper is calling wait_for_initramfs() so it could be that 
> some calls are getting queued at the early bootup. This is probably the 
> problem I am hitting.
> 
> I do not want to block the device creation till the user mode helper 
> finished. This could also result in a deadlock and would probably slow 
> down bootup.
> 
> When I limit the number of user mode helper calls to 1 and let the 
> others wait in a system queue, I might block other unrelated tasks in 
> the system queue.
> 
> I would create an own queue and let the async user mode helper wait in 
> this queue to only execute one at a time. When one of them needs a long 
> time in user space it would block the others. This workqueue would also 
> be active all the time. After the bootup it would probably not do much 
> work any more.
> 
> I do not like any of these solutions. Do you have better ideas?

Could you put the requests onto a private queue but use a system work queue
function the clear the queue - only starting the function if there isn't
a copy running?

	David

> 
> Hauke
> 
> On 7/18/25 00:52, Hauke Mehrtens wrote:
> > On 7/17/25 23:34, David Laight wrote:  
> >> On Thu, 17 Jul 2025 07:26:59 +0200
> >> Jiri Slaby <jirislaby@...nel.org> wrote:
> >>  
> >>> Cc wqueue & umode helper folks
> >>>
> >>> On 12. 07. 25, 1:08, Hauke Mehrtens wrote:  
> >>>> A modern Linux system creates much more than 20 threads at bootup.
> >>>> When I booted up OpenWrt in qemu the system sometimes failed to boot up
> >>>> when it wanted to create the 419th thread. The VM had 128MB RAM and the
> >>>> calculation in set_max_threads() calculated that max_threads should be
> >>>> set to 419. When the system booted up it tried to notify the user space
> >>>> about every device it created because CONFIG_UEVENT_HELPER was set and
> >>>> used. I counted 1299 calls to call_usermodehelper_setup(), all of
> >>>> them try to create a new thread and call the userspace hotplug 
> >>>> script in
> >>>> it.
> >>>>
> >>>> This fixes bootup of Linux on systems with low memory.
> >>>>
> >>>> I saw the problem with qemu 10.0.2 using these commands:
> >>>> qemu-system-aarch64 -machine virt -cpu cortex-a57 -nographic
> >>>>
> >>>> Cc: stable@...r.kernel.org
> >>>> Signed-off-by: Hauke Mehrtens <hauke@...ke-m.de>
> >>>> ---
> >>>>    kernel/fork.c | 2 +-
> >>>>    1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/kernel/fork.c b/kernel/fork.c
> >>>> index 7966c9a1c163..388299525f3c 100644
> >>>> --- a/kernel/fork.c
> >>>> +++ b/kernel/fork.c
> >>>> @@ -115,7 +115,7 @@
> >>>>    /*
> >>>>     * Minimum number of threads to boot the kernel
> >>>>     */
> >>>> -#define MIN_THREADS 20
> >>>> +#define MIN_THREADS 600  
> >>>
> >>> As David noted, this is not the proper fix. It appears that usermode
> >>> helper should use limited thread pool. I.e. instead of
> >>> system_unbound_wq, alloc_workqueue("", WQ_UNBOUND, max_active) with
> >>> max_active set to max_threads divided by some arbitrary constant (3? 
> >>> 4?)?  
> >>
> >> Or maybe just 1 ?
> >> I'd guess all the threads either block in the same place or just block
> >> each other??  
> > 
> > I will reduce the number of threads. Maybe to max 5 or maybe just one.
> > 
> > I think we should still increase the minimum number of threads, but 
> > something like 60 to 100 should be fine. It is calculated based RAM size 
> > 128MB RAM resulted already in 419 max threads.
> > 
> > Hauke  
>