Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v2 1/1] fs/splice: add missing callback for inaccessible
 pages
To:     Dave Hansen <dave.hansen@intel.com>,
        Claudio Imbrenda <imbrenda@linux.ibm.com>,
        viro@zeniv.linux.org.uk
Cc:     david@redhat.com, akpm@linux-foundation.org, aarcange@redhat.com,
        linux-mm@kvack.org, frankja@linux.ibm.com, sfr@canb.auug.org.au,
        jhubbard@nvidia.com, linux-kernel@vger.kernel.org,
        linux-s390@vger.kernel.org, jack@suse.cz, kirill@shutemov.name,
        peterz@infradead.org, sean.j.christopherson@intel.com,
        Ulrich.Weigand@de.ibm.com
References: <20200430143825.3534128-1-imbrenda@linux.ibm.com>
 <1a3f5107-9847-73d4-5059-c6ef9d293551@de.ibm.com>
 <e3e95a35-b0e3-b733-92f4-98bcccbe7ca5@intel.com>
From:   Christian Borntraeger <borntraeger@de.ibm.com>
Autocrypt: addr=borntraeger@de.ibm.com; prefer-encrypt=mutual; keydata=
 xsFNBE6cPPgBEAC2VpALY0UJjGmgAmavkL/iAdqul2/F9ONz42K6NrwmT+SI9CylKHIX+fdf
 J34pLNJDmDVEdeb+brtpwC9JEZOLVE0nb+SR83CsAINJYKG3V1b3Kfs0hydseYKsBYqJTN2j
 CmUXDYq9J7uOyQQ7TNVoQejmpp5ifR4EzwIFfmYDekxRVZDJygD0wL/EzUr8Je3/j548NLyL
 4Uhv6CIPf3TY3/aLVKXdxz/ntbLgMcfZsDoHgDk3lY3r1iwbWwEM2+eYRdSZaR4VD+JRD7p8
 0FBadNwWnBce1fmQp3EklodGi5y7TNZ/CKdJ+jRPAAnw7SINhSd7PhJMruDAJaUlbYaIm23A
 +82g+IGe4z9tRGQ9TAflezVMhT5J3ccu6cpIjjvwDlbxucSmtVi5VtPAMTLmfjYp7VY2Tgr+
 T92v7+V96jAfE3Zy2nq52e8RDdUo/F6faxcumdl+aLhhKLXgrozpoe2nL0Nyc2uqFjkjwXXI
 OBQiaqGeWtxeKJP+O8MIpjyGuHUGzvjNx5S/592TQO3phpT5IFWfMgbu4OreZ9yekDhf7Cvn
 /fkYsiLDz9W6Clihd/xlpm79+jlhm4E3xBPiQOPCZowmHjx57mXVAypOP2Eu+i2nyQrkapaY
 IdisDQfWPdNeHNOiPnPS3+GhVlPcqSJAIWnuO7Ofw1ZVOyg/jwARAQABzUNDaHJpc3RpYW4g
 Qm9ybnRyYWVnZXIgKDJuZCBJQk0gYWRkcmVzcykgPGJvcm50cmFlZ2VyQGxpbnV4LmlibS5j
 b20+wsF5BBMBAgAjBQJdP/hMAhsDBwsJCAcDAgEGFQgCCQoLBBYCAwECHgECF4AACgkQEXu8
 gLWmHHy/pA/+JHjpEnd01A0CCyfVnb5fmcOlQ0LdmoKWLWPvU840q65HycCBFTt6V62cDljB
 kXFFxMNA4y/2wqU0H5/CiL963y3gWIiJsZa4ent+KrHl5GK1nIgbbesfJyA7JqlB0w/E/SuY
 NRQwIWOo/uEvOgXnk/7+rtvBzNaPGoGiiV1LZzeaxBVWrqLtmdi1iulW/0X/AlQPuF9dD1Px
 hx+0mPjZ8ClLpdSp5d0yfpwgHtM1B7KMuQPQZGFKMXXTUd3ceBUGGczsgIMipZWJukqMJiJj
 QIMH0IN7XYErEnhf0GCxJ3xAn/J7iFpPFv8sFZTvukntJXSUssONnwiKuld6ttUaFhSuSoQg
 OFYR5v7pOfinM0FcScPKTkrRsB5iUvpdthLq5qgwdQjmyINt3cb+5aSvBX2nNN135oGOtlb5
 tf4dh00kUR8XFHRrFxXx4Dbaw4PKgV3QLIHKEENlqnthH5t0tahDygQPnSucuXbVQEcDZaL9
 WgJqlRAAj0pG8M6JNU5+2ftTFXoTcoIUbb0KTOibaO9zHVeGegwAvPLLNlKHiHXcgLX1tkjC
 DrvE2Z0e2/4q7wgZgn1kbvz7ZHQZB76OM2mjkFu7QNHlRJ2VXJA8tMXyTgBX6kq1cYMmd/Hl
 OhFrAU3QO1SjCsXA2CDk9MM1471mYB3CTXQuKzXckJnxHkHOwU0ETpw8+AEQAJjyNXvMQdJN
 t07BIPDtbAQk15FfB0hKuyZVs+0lsjPKBZCamAAexNRk11eVGXK/YrqwjChkk60rt3q5i42u
 PpNMO9aS8cLPOfVft89Y654Qd3Rs1WRFIQq9xLjdLfHh0i0jMq5Ty+aiddSXpZ7oU6E+ud+X
 Czs3k5RAnOdW6eV3+v10sUjEGiFNZwzN9Udd6PfKET0J70qjnpY3NuWn5Sp1ZEn6lkq2Zm+G
 9G3FlBRVClT30OWeiRHCYB6e6j1x1u/rSU4JiNYjPwSJA8EPKnt1s/Eeq37qXXvk+9DYiHdT
 PcOa3aNCSbIygD3jyjkg6EV9ZLHibE2R/PMMid9FrqhKh/cwcYn9FrT0FE48/2IBW5mfDpAd
 YvpawQlRz3XJr2rYZJwMUm1y+49+1ZmDclaF3s9dcz2JvuywNq78z/VsUfGz4Sbxy4ShpNpG
 REojRcz/xOK+FqNuBk+HoWKw6OxgRzfNleDvScVmbY6cQQZfGx/T7xlgZjl5Mu/2z+ofeoxb
 vWWM1YCJAT91GFvj29Wvm8OAPN/+SJj8LQazd9uGzVMTz6lFjVtH7YkeW/NZrP6znAwv5P1a
 DdQfiB5F63AX++NlTiyA+GD/ggfRl68LheSskOcxDwgI5TqmaKtX1/8RkrLpnzO3evzkfJb1
 D5qh3wM1t7PZ+JWTluSX8W25ABEBAAHCwV8EGAECAAkFAk6cPPgCGwwACgkQEXu8gLWmHHz8
 2w//VjRlX+tKF3szc0lQi4X0t+pf88uIsvR/a1GRZpppQbn1jgE44hgF559K6/yYemcvTR7r
 6Xt7cjWGS4wfaR0+pkWV+2dbw8Xi4DI07/fN00NoVEpYUUnOnupBgychtVpxkGqsplJZQpng
 v6fauZtyEcUK3dLJH3TdVQDLbUcL4qZpzHbsuUnTWsmNmG4Vi0NsEt1xyd/Wuw+0kM/oFEH1
 4BN6X9xZcG8GYUbVUd8+bmio8ao8m0tzo4pseDZFo4ncDmlFWU6hHnAVfkAs4tqA6/fl7RLN
 JuWBiOL/mP5B6HDQT9JsnaRdzqF73FnU2+WrZPjinHPLeE74istVgjbowvsgUqtzjPIG5pOj
 cAsKoR0M1womzJVRfYauWhYiW/KeECklci4TPBDNx7YhahSUlexfoftltJA8swRshNA/M90/
 i9zDo9ySSZHwsGxG06ZOH5/MzG6HpLja7g8NTgA0TD5YaFm/oOnsQVsf2DeAGPS2xNirmknD
 jaqYefx7yQ7FJXXETd2uVURiDeNEFhVZWb5CiBJM5c6qQMhmkS4VyT7/+raaEGgkEKEgHOWf
 ZDP8BHfXtszHqI3Fo1F4IKFo/AP8GOFFxMRgbvlAs8z/+rEEaQYjxYJqj08raw6P4LFBqozr
 nS4h0HDFPrrp1C2EMVYIQrMokWvlFZbCpsdYbBI=
Message-ID: <3d379d9e-241c-ef3b-dcef-20fdd3b8740d@de.ibm.com>
Date:   Fri, 1 May 2020 09:18:34 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.6.0
MIME-Version: 1.0
In-Reply-To: <e3e95a35-b0e3-b733-92f4-98bcccbe7ca5@intel.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 01.05.20 00:06, Dave Hansen wrote:
> I was also wondering if Claudio was right about the debug patch having
> races.  I went to go look how the s390 code avoids races when pages go
> from accessible->inaccessible.
> 
> Because, if if all of the traps are in place to transform pages from
> inaccessible->accessible, the code *after* those traps is still
> vulnerable.  What *keeps* pages accessible?
> 
> The race avoidance is this, basically:
> 
> 	down_read(&gmap->mm->mmap_sem);
> 	lock_page(page);
>         ptep = get_locked_pte(gmap->mm, uaddr, &ptelock);
> ...
>>         expected = expected_page_refs(page);
>>         if (!page_ref_freeze(page, expected))
>>                 return -EBUSY;
>>         set_bit(PG_arch_1, &page->flags);
>>         rc = uv_call(0, (u64)uvcb);
>>         page_ref_unfreeze(page, expected);
> 
> ... up_read(mmap_sem) / unlock_page() / unlock pte
> 
> I'm assuming that after the uv_call(), the page is inaccessible and I/O
> devices will go boom if they touch the page.
> 
> The page_ref_freeze() ensures that references come between the
> freeze/unfreeze are noticed, but it doesn't actually *stop* new ones for
> users that hold references already.  For the page cache, especially,
> someone could do:
> 
> 	page = find_get_page();
> 	arch_make_page_accessible();
> 					lock_page();
> 	...				make_secure_pte();

Not sure if I got your point here, but this make_secure_pte should bail
out because we actually do check for a calculated refcount value and return
-EBUSY. The find_get_page should have raised this refcount to a value that
would go beyond the expected value, No? 


> 					unlock_page();
> 	get_page();
> 	// ^ OK because I have a ref
> 	// do DMA on inaccessible page
> 
> Because the make_secure_pte() code isn't looking for a *specific*
> 'expected' value, it has no way of noticing that the extra ref snuck in
> there.

I think the expected calcution is actually doing that,giving back the minimum
value when no one else has any references that are valid for I/O.

But I might not have understood what you are trying to tell me?

> 
> I _think_ expected actually needs to be checked for having a specific
> (low) value so that if there's a *possibility* of a reference holder
> acquiring additional references, the page is known to be off-limits.
> mm/migrate.c has a few examples of this, but I'm not quite sure how
> bulletproof they are.  Some of it appears to just be optimizations.
> 
> 
> 
b