Re: [PATCH 1/2] vmalloc: New flag for flush before releasing pages

From: Nadav Amit
Date: Tue Dec 04 2018 - 20:45:11 EST


> On Dec 4, 2018, at 5:09 PM, Edgecombe, Rick P <rick.p.edgecombe@xxxxxxxxx> wrote:
>
> On Tue, 2018-12-04 at 14:48 -0800, Nadav Amit wrote:
>>> On Dec 4, 2018, at 11:48 AM, Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>>>
>>> On Tue, Dec 4, 2018 at 11:45 AM Nadav Amit <nadav.amit@xxxxxxxxx> wrote:
>>>>> On Dec 4, 2018, at 10:56 AM, Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>>>>>
>>>>> On Mon, Dec 3, 2018 at 5:43 PM Nadav Amit <nadav.amit@xxxxxxxxx> wrote:
>>>>>>> On Nov 27, 2018, at 4:07 PM, Rick Edgecombe <
>>>>>>> rick.p.edgecombe@xxxxxxxxx> wrote:
>>>>>>>
>>>>>>> Since vfree will lazily flush the TLB, but not lazily free the
>>>>>>> underlying pages,
>>>>>>> it often leaves stale TLB entries to freed pages that could get re-
>>>>>>> used. This is
>>>>>>> undesirable for cases where the memory being freed has special
>>>>>>> permissions such
>>>>>>> as executable.
>>>>>>
>>>>>> So I am trying to finish my patch-set for preventing transient W+X
>>>>>> mappings
>>>>>> from taking space, by handling kprobes & ftrace that I missed (thanks
>>>>>> again for
>>>>>> pointing it out).
>>>>>>
>>>>>> But all of the sudden, I donât understand why we have the problem that
>>>>>> this
>>>>>> (your) patch-set deals with at all. We already change the mappings to
>>>>>> make
>>>>>> the memory writable before freeing the memory, so why canât we make it
>>>>>> non-executable at the same time? Actually, why do we make the module
>>>>>> memory,
>>>>>> including its data executable before freeing it???
>>>>>
>>>>> All the code you're looking at is IMO a very awkward and possibly
>>>>> incorrect of doing what's actually necessary: putting the direct map
>>>>> the way it wants to be.
>>>>>
>>>>> Can't we shove this entirely mess into vunmap? Have a flag (as part
>>>>> of vmalloc like in Rick's patch or as a flag passed to a vfree variant
>>>>> directly) that makes the vunmap code that frees the underlying pages
>>>>> also reset their permissions?
>>>>>
>>>>> Right now, we muck with set_memory_rw() and set_memory_nx(), which
>>>>> both have very awkward (and inconsistent with each other!) semantics
>>>>> when called on vmalloc memory. And they have their own flushes, which
>>>>> is inefficient. Maybe the right solution is for vunmap to remove the
>>>>> vmap area PTEs, call into a function like set_memory_rw() that resets
>>>>> the direct maps to their default permissions *without* flushing, and
>>>>> then to do a single flush for everything. Or, even better, to cause
>>>>> the change_page_attr code to do the flush and also to flush the vmap
>>>>> area all at once so that very small free operations can flush single
>>>>> pages instead of flushing globally.
>>>>
>>>> Thanks for the explanation. I read it just after I realized that indeed
>>>> the
>>>> whole purpose of this code is to get cpa_process_alias()
>>>> update the corresponding direct mapping.
>>>>
>>>> This thing (pageattr.c) indeed seems over-engineered and very unintuitive.
>>>> Right now I have a list of patch-sets that I owe, so I donât have the time
>>>> to deal with it.
>>>>
>>>> But, I still think that disable_ro_nx() should not call set_memory_x().
>>>> IIUC, this breaks W+X of the direct-mapping which correspond with the
>>>> module
>>>> memory. Does it ever stop being W+X?? Iâll have another look.
>>>
>>> Dunno. I did once chase down a bug where some memory got freed while
>>> it was still read-only, and the results were hilarious and hard to
>>> debug, since the explosion happened long after the buggy code
>>> finished.
>>
>> This piece of code causes me pain and misery.
>>
>> So, it turns out that the direct map is *not* changed if you just change
>> the NX-bit. See change_page_attr_set_clr():
>>
>> /* No alias checking for _NX bit modifications */
>> checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) !=
>> _PAGE_NX;
>>
>> How many levels of abstraction are broken in the way? What would happen
>> if somebody tries to change the NX-bit and some other bit in the PTE?
>> Luckily, I donât think someone doesâ at least for now.
>>
>> So, again, I think the change I proposed makes sense. nios2 does not have
>> set_memory_x() and it will not be affected.
> Hold on...so on architectures that don't have set_memory_ but do have something
> like NX, wont the executable stale TLB continue to live to re-used pages, and so
> it doesn't fix the problem this patch is trying to address generally? I see at
> least a couple archs use vmalloc and have an exec bit, but don't define
> set_memory_*.

Again, this does not come instead of your patch (the one in this thread).
And if you follow Andyâs suggestion, the patch I propose will not be needed.
However, in the meantime - I see no reason to mark data as executable, even
for a brief period of time.