Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Vineet Gupta <vineet.gupta1@synopsys.com>
To:     Peter Xu <peterx@redhat.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
CC:     Alexey Brodkin <alexey.brodkin@synopsys.com>,
        "Eric W. Biederman" <ebiederm@xmission.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        "Souptick Joarder" <jrdr.linux@gmail.com>,
        Andrea Arcangeli <aarcange@redhat.com>,
        "linux-snps-arc@lists.infradead.org" 
        <linux-snps-arc@lists.infradead.org>
Subject: Re: [PATCH RFC] mm: arc: fix potential double realease of mmap_sem
Thread-Topic: [PATCH RFC] mm: arc: fix potential double realease of mmap_sem
Thread-Index: AQHUcZJWAhL05uLd1Uaujch30fnJhg==
Date:   Tue, 6 Nov 2018 00:48:31 +0000
Message-ID: <C2D7FE5348E1B147BCA15975FBA23075013A590E5E@us01wembx1.internal.synopsys.com>
References: <20181101032354.19351-1-peterx@redhat.com>
Accept-Language: en-US
Content-Language: en-US
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 10/31/18 8:24 PM, Peter Xu wrote:=0A=
> In do_page_fault() of ARC we have:=0A=
>=0A=
>         ...=0A=
> 	fault =3D handle_mm_fault(vma, address, flags);=0A=
>=0A=
>         /* If Pagefault was interrupted by SIGKILL, exit page fault "earl=
y" */=0A=
>         if (unlikely(fatal_signal_pending(current))) {=0A=
>                 if ((fault & VM_FAULT_ERROR) && !(fault & VM_FAULT_RETRY)=
)=0A=
>                         up_read(&mm->mmap_sem);     <---------------- [1]=
=0A=
>                 if (user_mode(regs))=0A=
>                         return;=0A=
>         }=0A=
>         ...=0A=
> 	if (likely(!(fault & VM_FAULT_ERROR))) {=0A=
>                 ...=0A=
> 		return;=0A=
> 	}=0A=
>=0A=
> 	if (fault & VM_FAULT_OOM)=0A=
> 		goto out_of_memory;                <----------------- [2]=0A=
> 	else if (fault & VM_FAULT_SIGSEGV)=0A=
> 		goto bad_area;                     <----------------- [3]=0A=
> 	else if (fault & VM_FAULT_SIGBUS)=0A=
> 		goto do_sigbus;                    <----------------- [4]=0A=
>=0A=
> Logically it's possible that we might try to release the mmap_sem twice=
=0A=
> by having a scenario like:=0A=
>=0A=
>         - task received SIGKILL,=0A=
>         - task handled kernel mode page fault,=0A=
>         - handle_mm_fault() returned with one of VM_FAULT_ERROR,=0A=
>=0A=
> Then we'll go into path [1] to release the mmap_sem, however we won't=0A=
> return immediately since user_mode(regs) check will fail (a kernel page=
=0A=
> fault).  Then we might go into either [2]-[4] and either of them will=0A=
> try to release the mmap_sem again.=0A=
>=0A=
> To fix this, we only release the mmap_sem at [1] when we're sure we'll=0A=
> quit immediately (after we checked with user_mode(regs)).=0A=
=0A=
Hmm, do_page_fault() needs a serious makeover. There's a known problem in t=
he area=0A=
you touched (with test case) where we fail to relinquish the mmap_sem for w=
hich=0A=
Alexey had provided a fix. But I'm going to redo this part now and CC you f=
olks=0A=
for review. OK ?=0A=
=0A=
>=0A=
> CC: Vineet Gupta <vgupta@synopsys.com>=0A=
> CC: "Eric W. Biederman" <ebiederm@xmission.com>=0A=
> CC: Peter Xu <peterx@redhat.com>=0A=
> CC: Andrew Morton <akpm@linux-foundation.org>=0A=
> CC: Souptick Joarder <jrdr.linux@gmail.com>=0A=
> CC: Andrea Arcangeli <aarcange@redhat.com>=0A=
> CC: linux-snps-arc@lists.infradead.org=0A=
> CC: linux-kernel@vger.kernel.org=0A=
> Signed-off-by: Peter Xu <peterx@redhat.com>=0A=
> ---=0A=
>=0A=
> I noticed this only by reading the code.  Neither have I verified the=0A=
> issue, nor have I tested the patch since I even don't know how to (I'm=0A=
> totally unfamiliar with the arc architecture).  However I'm posting this=
=0A=
> out first to see whether there's any quick feedback, and in case it's a=
=0A=
> valid issue that we've ignored.=0A=
> ---=0A=
>  arch/arc/mm/fault.c | 7 +++----=0A=
>  1 file changed, 3 insertions(+), 4 deletions(-)=0A=
>=0A=
> diff --git a/arch/arc/mm/fault.c b/arch/arc/mm/fault.c=0A=
> index c9da6102eb4f..2d28c3dad5c1 100644=0A=
> --- a/arch/arc/mm/fault.c=0A=
> +++ b/arch/arc/mm/fault.c=0A=
> @@ -142,11 +142,10 @@ void do_page_fault(unsigned long address, struct pt=
_regs *regs)=0A=
>  	fault =3D handle_mm_fault(vma, address, flags);=0A=
>  =0A=
>  	/* If Pagefault was interrupted by SIGKILL, exit page fault "early" */=
=0A=
> -	if (unlikely(fatal_signal_pending(current))) {=0A=
> -		if ((fault & VM_FAULT_ERROR) && !(fault & VM_FAULT_RETRY))=0A=
> +	if (unlikely(fatal_signal_pending(current) && user_mode(regs))) {=0A=
> +		if (!(fault & VM_FAULT_RETRY))=0A=
>  			up_read(&mm->mmap_sem);=0A=
> -		if (user_mode(regs))=0A=
> -			return;=0A=
> +		return;=0A=
>  	}=0A=
>  =0A=
>  	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);=0A=
=0A=