Date: Sun, 5 Aug 2018 12:05:59 -0500 (CDT) From: "Valeri Galtsev" <galtsev@kicp.uchicago.edu> To: "Polytropon" <freebsd@edvax.de> Cc: "JD" <jd1008@gmail.com>, freebsd-questions@freebsd.org Subject: Re: Erase memory on shutdown Message-ID: <64112.108.68.162.197.1533488759.squirrel@cosmo.uchicago.edu> In-Reply-To: <20180805180212.1cbfd1dc.freebsd@edvax.de> References: <acbb3213-e79e-dfde-038f-b1476925cd4a@irk.ru> <20180805172503.e2479108.freebsd@edvax.de> <5B671AF5.7080701@gmail.com> <20180805180212.1cbfd1dc.freebsd@edvax.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, August 5, 2018 11:02 am, Polytropon wrote:
> On Sun, 05 Aug 2018 09:42:45 -0600, JD wrote:
>>
>>
>> On 08/05/2018 09:25 AM, Polytropon wrote:
>> > On Sun, 5 Aug 2018 22:24:16 +0800, thor wrote:
>> >> Just one paranoid question: How to cause FreeBSD to zero all RAM
>> during
>> >> shutdown?
>> > This would imply that the kernel would finally have to
>> > overwrite itself. How can control over zeroing memory
>> > be maintained when the control program itself has been
>> > overwritten? That would be the result of the "all" in
>> > your requirement.
>> >
>> > Sure, you could add some code to the final shutdown
>> > routines to zero the RAM, which is possible, but not
>> > trivial: You need a non-optimized call to memset()
>> > using a custom function pointer.
>> >
>> > static void *(* const volatile memset_ptr)(void *, int, size_t) =
>> memset;
>> > static void secure_memzero(void *p, size_t len)
>> > {
>> > (memset_ptr)(p, 0, len);
>> > }
>> >
>> > void *ram = 0x0; /* RAM start address */
>> > size_t amount = 17179869184UL; /* 16 GB RAM */
>> >
>> > secure_menzero(ram, amount); /* ouch */
>> >
>> > If you add something like this to the kernel, and make
>> > sure your compiler isn't too clever (as to optimize it
>> > into a NOP), you're going to crash the whole system
>> > without actually being sure that at least a part of
>> > the RAM has been zeroed. And even then it might not
>> > work as intended.
>> >
>> > See:
>> >
>> > http://www.daemonology.net/blog/2014-09-04-how-to-zero-a-buffer.html
>> >
>> > http://www.daemonology.net/blog/2014-09-05-erratum.html
>> >
>> > Keep in mind: You're declaring war on intended security
>> > mechanisms if you try to do this. :-)
>> >
>> > However, this is not guaranteed (!) to work, so you
>> > cannot be safe. And you must do it from the kernel.
>> > You cannot use (a privileged) program like dd to
>> > flush /dev/mem and /dev/kmem with /dev/zero output.
>> >
>> > Your best bet is to assume that RAM will be zeroed as soon
>> > as the power-off routine as been completed - no refresh,
>> > no content. Not perfectly secure, though... :-)
>> >
>> > RAM usually isn't zeroed, but marked as "not in use" so
>> > it can be overwritten. Address randomization makes it
>> > hard to protect where something will appear in RAM, and
>> > access to RAM requires certain privileges on a system.
>> >
>> >
>> Just for the heck of it:
>> allocate from contiguous freemem for the program in question,
>> load the program,
>> transfer control to the program which should passed
>> it starting addy, and size, and total mem size.
>> Such a program would work, assuming it is written to watch
>> out for it's addy space, and not overwrite itself.
>
> It's not about the program itself. Assume it is run as
> a user process (by root, for example). It should not
> overwrite itself, it also should not overwrite the
> kernel and other system facilities in memory whose
> removal could cause a kernel crash before the zeroing
> is complete.
>
> That's why I suggested it _could_ be introduced as
> new code into the kernel, to be performed at the
> latest stage of system shutdown.
>
>
>
>> My question is: currently, if the machine is shut down,
>> (i.e. powered off), does ram keep it's content?
>
> Theory: No.
>
> Practice: Yes.
>
>
>
>> If yes,
>>
>> how?
>
> Capacitive effects?
>
>
>
>> for how long?
>
> Until capacity is drained out.
>
>
>
>> Is it static ram, or does the battery provide the power
>> for the ram to remain refreshed?
>
> Usually not.
>
> During normal operations, RAM is refreshed periodically.
> This is the 'D' in DRAM - dynamic.
>
> The most prominent case of static RAM is core memory.
> It keeps it content (after magnetization) until the
> memory is read again. So if you found a core memory matrix
> with a program in it from 1970, the content could still
> be in there... :-)
>
> So you see, even static RAM needs a refresh (after a
> read), usually with the values read.
There is one more incarnation of static RAM that has emerged after ferrite
core (magnetic) RAM your have described. That static RAM is based on CMOS
flip-flops, one flip-flop per one bit of information. Much cheaper dynamic
RAM is based on a single (usually PN-junction based N-channel) FET, and
the charge of capacitance on its gate is what needs to be refreshed in
case of dynamic RAM as it is discharged gradually. Static (CMOS based) RAM
keeps its state forever while the power is present, and it almost does not
consume power, except when you change the bit/information. Static RAM is
[much] more expensive, less dense, that is why in regular computers and
servers it is dynamic RAM that is usually used. To the contrary to
magnetic RAM mentioned above static (CMOS flip-flop based) RAM is still
volatile, however, as it almost doesn't consume power, one would expect
(depending on circuit board design), that after the power shuts off, the
charge on power lead capacitors is drained really slow, and hence the
information in static RAM is preserved for much longer time than that for
dynamic RAM. One only needs to power static RAM on before voltage dropped
below some low value (could be around 1 volt or less depending on chips).
Just my $0.02
Valeri
>
> There are "suspend to RAM" mechanisms where the RAM
> is buffered from the PSU (or in case of laptops: from
> battery) until the power runs out. This means that the
> computer technically isn't off - it's just that the CPU
> isn't running, only the controllers for the memory
> control are performing their normal work.
>
> There are also buffer batteries for "critical RAM"
> that can keep the RAM intact during power outages.
> But I don't know if they also buffer CPU registers
> which would be neccessary to resume operations when
> the power comes back.
>
>
>
> Also see:
>
> https://en.wikipedia.org/wiki/Cold_boot_attack
>
>
>
> --
> Polytropon
> Magdeburg, Germany
> Happy FreeBSD user since 4.0
> Andra moi ennepe, Mousa, ...
> _______________________________________________
> freebsd-questions@freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-questions
> To unsubscribe, send any mail to
> "freebsd-questions-unsubscribe@freebsd.org"
>
++++++++++++++++++++++++++++++++++++++++
Valeri Galtsev
Sr System Administrator
Department of Astronomy and Astrophysics
Kavli Institute for Cosmological Physics
University of Chicago
Phone: 773-702-4247
++++++++++++++++++++++++++++++++++++++++
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?64112.108.68.162.197.1533488759.squirrel>