Re: [INTERNALS-WIN] OPcache: auto address mapping fallback desired?

On 17.03.2019 at 20:28, Anatol Belski wrote:

> On Sun, 2019-03-17 at 15:51 +0100, Christoph M. Becker wrote:
>
>> I've noticed that the win32 model of OPcache's shared memory may map
>> it
>> to an automatic address determinded by the OS.  I wonder whether this
>> is
>> actually desired, and if so, why we try to map to hard-coded
>> addresses
>> in the first place.  If the auto adress mapping is not desired, we
>> should apply something like this patch:
>> <https://gist.github.com/cmb69/80e2ecc2d46f5255a315a268fcdd6f94>.
>
> Thanks for the checks. The flow as i read it is currently
> 
> - if no mmap_base specified, map a fixed address
> - if mmap_base specified, map to it
> - if any of above failed, obtain an address from OS
> 
> The point is, that once the mapping is successfull, it's going to be
> saved and read by another process. The doc of MapViewOfFileEx says
> 
> [quote]
> While it is possible to specify an address that is safe now (not used
> by the operating system), there is no guarantee that the address will
> remain safe over time. Therefore, it is better to let the operating
> system choose the address. In this case, you would not store pointers
> in the memory mapped file, you would store offsets from the base of the
> file mapping so that the mapping can be used at any address.
> [/quote]
> 
> Those fixed addresses are considered safe from experience, so they're
> used when no base address has been specified. But also, from the
> experience, none of those methods is ideal.

Thanks, Anatol, this makes sense.  So sticking with the current flow is
fine.

> Removing the possibility to use a system chosen base address will
> effectively mean, that processes would fail to reattach if any of the
> fixed addresses is unusable on the given system. Of course one can say,
> then user should seek for a usable address, but that's probably a bit
> too much of burden for a user :)
> 
> Also you check on ASLR why these failures happen
> 
> https://en.wikipedia.org/wiki/Address_space_layout_randomization
> 
> On a moders system, it doesn't seem possible to disable ASLR at all,
> which is good for security, but causes issues with Opcache. That's why
> we've invented also the file_cache_fallback.

-- 
Christoph M. Becker