Re: xt_hashlimig build error (was Re: [RFC 01/17] x86/asm/64: Remove the restore_c_regs_and_iret label)

From: Vishwanath Pai <vpai@akamai.com>
To: Linus Torvalds <torvalds@linux-foundation.org>,
	Ingo Molnar <mingo@kernel.org>,
	Igor Lubashev <ilubashe@akamai.com>,
	Josh Hunt <johunt@akamai.com>,
	Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Borislav Petkov <bp@alien8.de>, Andy Lutomirski <luto@kernel.org>,
	the arch/x86 maintainers <x86@kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Brian Gerst <brgerst@gmail.com>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Juergen Gross <jgross@suse.com>,
	Boris Ostrovsky <boris.ostrovsky@oracle.com>,
	Kees Cook <keescook@chromium.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	"David S. Miller" <davem@davemloft.net>,
	Pablo Neira Ayuso <pablo@netfilter.org>
Subject: Re: xt_hashlimig build error (was Re: [RFC 01/17] x86/asm/64: Remove the restore_c_regs_and_iret label)
Date: Thu, 7 Sep 2017 14:25:04 -0400	[thread overview]
Message-ID: <69b38985-e094-ddcc-6f7e-d6e5cc2c657e@akamai.com> (raw)
In-Reply-To: <CA+55aFznorYJd8OyV09eHt4zrRrrv2aEq_D=Q8k2w+CmgJ=+cw@mail.gmail.com>

On 09/07/2017 01:51 PM, Linus Torvalds wrote:
> On Thu, Sep 7, 2017 at 3:29 AM, Ingo Molnar <mingo@kernel.org> wrote:
>>
>> not the best of kernels, 32-bit allyesconfig doesn't even appear to build:
>>
>>   net/netfilter/xt_hashlimit.o: In function `hashlimit_mt_common.isra.6':
>>   xt_hashlimit.c:(.text+0x1146): undefined reference to `__udivdi3'
> 
> I think this is due to commit bea74641e378 ("netfilter: xt_hashlimit:
> add rate match mode").
> 
> It adds a 64-bit divide in user2rate_bytes() afaik, and to make things
> worse it seems to be a really stupid one too.
> 
> Although I guess "worse" is not bad when the stupidity of it should
> mean that it's easy to avoid the 64-bit issue.
> 
> Oddly, user2rate() that actually *does* need a 64-bit divide, seems to
> do it right and use "div64_u64()" to do so.
> 
> But user2rate_bytes() could easily avoid any 64-bit issues, since it
> divides the max 32-bit (unsigned) number with a 64-bit unsigned
> number.
> 
> It would be easy to just say
> 
>  - "if high 32 bits are set, result is 0"
> 
>  - else do a 32-bit divide
> 
> or just use "div64_u64()" in that code too.
> 
> But honestly, that math is odd in other ways too (is that "r-1"
> _supposed_ to underflow to -1 for large 'user' counts?), so somebody
> needs to look at that logic.
> 
> And there might be some other 64-bit divide I missed, so please,
> netfilter people, check the 32-bit build.
> 
>                   Linus
> 

Sorry about the build failure, we have already queued up a fix for this:
https://patchwork.ozlabs.org/patch/810772/

I agree, this could've been easily avoided, I happened to overlook this
particular line. There are other places in xt_hashlimit where we use
64bit division and I believe we have already covered those cases using
div64_u64.

- Vishwanath