Message-ID: <20251106223900.3893d7d9@pumpkin>
Date: Thu, 6 Nov 2025 22:39:00 +0000
From: David Laight <david.laight.linux@...il.com>
To: Chuck Lever <cel@...nel.org>
Cc: "stable@...r.kernel.org" <stable@...r.kernel.org>, Andrew Morton
 <akpm@...ux-foundation.org>, David Laight <David.Laight@...LAB.COM>, Linux
 NFS Mailing List <linux-nfs@...r.kernel.org>, Linux List Kernel Mailing
 <linux-kernel@...r.kernel.org>, speedcracker@...mail.com
Subject: Re: Compile Error fs/nfsd/nfs4state.o - clamp() low limit slotsize
 greater than high limit total_avail/scale_factor

On Thu, 6 Nov 2025 14:32:34 -0500
Chuck Lever <cel@...nel.org> wrote:

> On 11/6/25 2:22 PM, David Laight wrote:
> > On Thu, 6 Nov 2025 09:33:28 -0500
> > Chuck Lever <cel@...nel.org> wrote:
> >   
> >> FYI
> >>
> >> https://bugzilla.kernel.org/show_bug.cgi?id=220745  
> > 
> > Ugg - that code is horrid.
> > It seems to have got deleted since, but it is:
> > 
> > 	u32 slotsize = slot_bytes(ca);
> > 	u32 num = ca->maxreqs;
> > 	unsigned long avail, total_avail;
> > 	unsigned int scale_factor;
> > 
> > 	spin_lock(&nfsd_drc_lock);
> > 	if (nfsd_drc_max_mem > nfsd_drc_mem_used)
> > 		total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
> > 	else
> > 		/* We have handed out more space than we chose in
> > 		 * set_max_drc() to allow.  That isn't really a
> > 		 * problem as long as that doesn't make us think we
> > 		 * have lots more due to integer overflow.
> > 		 */
> > 		total_avail = 0;
> > 	avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
> > 	/*
> > 	 * Never use more than a fraction of the remaining memory,
> > 	 * unless it's the only way to give this client a slot.
> > 	 * The chosen fraction is either 1/8 or 1/number of threads,
> > 	 * whichever is smaller.  This ensures there are adequate
> > 	 * slots to support multiple clients per thread.
> > 	 * Give the client one slot even if that would require
> > 	 * over-allocation--it is better than failure.
> > 	 */
> > 	scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
> > 
> > 	avail = clamp_t(unsigned long, avail, slotsize,
> > 			total_avail/scale_factor);
> > 	num = min_t(int, num, avail / slotsize);
> > 	num = max_t(int, num, 1);
> > 
> > Lets rework it a bit...
> > 	if (nfsd_drc_max_mem > nfsd_drc_mem_used) {
> > 		total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
> > 		avail = min(NFSD_MAX_MEM_PER_SESSION, total_avail);
> > 		avail = clamp(avail, n + sizeof(xxx), total_avail/8)
> > 	} else {
> > 		total_avail = 0;
> > 		avail = 0;
> > 		avail = clamp(0, n + sizeof(xxx), 0);
> > 	}
> > 
> > Neither of those clamp() are sane at all - should be clamp(val, lo, hi)
> > with 'lo <= hi' otherwise the result is dependant on the order of the
> > comparisons.
> > The compiler sees the second one and rightly bleats.
> > I can't even guess what the code is actually trying to calculate!
> > 
> > Maybe looking at where the code came from, or the current version might help.  
> 
> The current upstream code is part of a new feature that is not
> appropriate to backport to LTS kernels. I consider that code out of
> play.
> 
> The compiler error showed up in 6.1.y with the recent minmax.h
> changes -- there have been no reported problems in any of the LTS
> kernels until now, including with 32-bit builds.
> 
> The usual guidelines about regressions suggest that the most recent
> backports (ie, minmax.h) are the ones that should be removed or reworked
> to address the compile breakage. I don't think we should address this by
> writing special clean-ups to code that wasn't broken before the minmax.h
> changes. Cleaning that code up is more likely to introduce bugs than
> reverting the minmax.h changes.

No, that code needs fixing. It is broken.....
The compiler warning/error is completely valid.
The result of that clamp() has never been well defined.
It is likely that it always generated the wrong result.
It might be that a much older version of the function exists
before someone changed a pair of conditionals to be a call to clamp().
That old version may well be fine.

	David

> 
> 
> > It MIGHT be that the 'lo' of slotsize was an attempt to ensure that
> > the following 'avail / slotsize' was as least one.
> > Some software archaeology might show that the 'num = max(num, 1)' was added
> > because the code above didn't work.
> > In that case the clamp can be clamp(avail, 0, total_avail/scale_factor)
> > which is just min(avail, total_avail/scale_factor).
> > 
> > The person who rewrote it between 6.1 and 6.18 might now more.
> > 
> > 	David
> > 	  
> >>
> >>
> >> -------- Forwarded Message --------
> >> Subject: Re: Compile Error fs/nfsd/nfs4state.o - clamp() low limit
> >> slotsize greater than high limit total_avail/scale_factor
> >> Date: Thu, 06 Nov 2025 07:29:25 -0500
> >> From: Jeff Layton <jlayton@...nel.org>
> >> To: Mike-SPC via Bugspray Bot <bugbot@...nel.org>, cel@...nel.org,
> >> neilb@...mail.net, trondmy@...nel.org, linux-nfs@...r.kernel.org,
> >> anna@...nel.org, neilb@...wn.name
> >>
> >> On Thu, 2025-11-06 at 11:30 +0000, Mike-SPC via Bugspray Bot wrote:  
> >>> Mike-SPC writes via Kernel.org Bugzilla:
> >>>
> >>> (In reply to Bugspray Bot from comment #5)    
> >>>> Chuck Lever <cel@...nel.org> replies to comment #4:
> >>>>
> >>>> On 11/5/25 7:25 AM, Mike-SPC via Bugspray Bot wrote:    
> >>>>> Mike-SPC writes via Kernel.org Bugzilla:
> >>>>>     
> >>>>>> Have you found a 6.1.y kernel for which the build doesn't fail?    
> >>>>>
> >>>>> Yes. Compiling Version 6.1.155 works without problems.
> >>>>> Versions >= 6.1.156 aren't.    
> >>>>
> >>>> My analysis yesterday suggests that, because the nfs4state.c code hasn't
> >>>> changed, it's probably something elsewhere that introduced this problem.
> >>>> As we can't reproduce the issue, can you use "git bisect" between
> >>>> v6.1.155 and v6.1.156 to find the culprit commit?
> >>>>
> >>>> (via https://msgid.link/ab235dbe-7949-4208-a21a-2cdd50347152@kernel.org)    
> >>>
> >>>
> >>> Yes, your analysis is right (thanks for it).
> >>> After some investigation, the issue appears to be caused by changes introduced in
> >>> include/linux/minmax.h.
> >>>
> >>> I verified this by replacing minmax.h in 6.1.156 with the version from 6.1.155,
> >>> and the kernel then compiles successfully.
> >>>
> >>> The relevant section in the 6.1.156 changelog (https://cdn.kernel.org/pub/linux/kernel/v6.x/ChangeLog-6.1.156) shows several modifications to minmax.h (notably around __clamp_once() and the use of
> >>> BUILD_BUG_ON_MSG(statically_true(ulo > uhi), ...)), which seem to trigger a compile-time assertion when building NFSD.
> >>>
> >>> Replacing the updated header with the previous one resolves the issue, so this appears
> >>> to be a regression introduced by the new clamp() logic.
> >>>
> >>> Could you please advise who is the right person or mailing list to report this issue to
> >>> (minmax.h maintainers, kernel core, or stable tree)?
> >>>     
> >>
> >> I'd let all 3 know, and I'd include the author of the patches that you
> >> suspect are the problem. They'll probably want to revise the one that's
> >> a problem.
> >>
> >> Cheers,  
> >   
> 
> 

Powered by blists - more mailing lists