[v2,1/3] Add the latent_entropy gcc plugin
diff mbox

Message ID 20160531013145.612696c12f2ef744af739803@gmail.com
State New
Headers show

Commit Message

Emese Revfy May 30, 2016, 11:31 p.m. UTC
This plugin mitigates the problem of the kernel having too little entropy during
and after boot for generating crypto keys.

It creates a local variable in every marked function. The value of this variable is
modified by randomly chosen operations (add, xor and rol) and
random values (gcc generates them at compile time and the stack pointer at runtime).
It depends on the control flow (e.g., loops, conditions).

Before the function returns the plugin writes this local variable
into the latent_entropy global variable. The value of this global variable is
added to the kernel entropy pool in do_one_initcall() and _do_fork().

Signed-off-by: Emese Revfy <re.emese@gmail.com>
---
 arch/Kconfig                                |  18 ++
 arch/powerpc/kernel/Makefile                |   8 +-
 include/linux/random.h                      |  10 +
 init/main.c                                 |   1 +
 kernel/fork.c                               |   1 +
 mm/page_alloc.c                             |   5 +
 scripts/Makefile.gcc-plugins                |  10 +-
 scripts/gcc-plugins/Makefile                |   1 +
 scripts/gcc-plugins/latent_entropy_plugin.c | 454 ++++++++++++++++++++++++++++
 9 files changed, 502 insertions(+), 6 deletions(-)
 create mode 100644 scripts/gcc-plugins/latent_entropy_plugin.c

Comments

Andrew Morton June 1, 2016, 7:42 p.m. UTC | #1
On Tue, 31 May 2016 01:31:45 +0200 Emese Revfy <re.emese@gmail.com> wrote:

> This plugin mitigates the problem of the kernel having too little entropy during
> and after boot for generating crypto keys.
> 
> It creates a local variable in every marked function. The value of this variable is
> modified by randomly chosen operations (add, xor and rol) and
> random values (gcc generates them at compile time and the stack pointer at runtime).
> It depends on the control flow (e.g., loops, conditions).
> 
> Before the function returns the plugin writes this local variable
> into the latent_entropy global variable. The value of this global variable is
> added to the kernel entropy pool in do_one_initcall() and _do_fork().

I don't think I'm really understanding.  Won't this produce the same
value on each and every boot?
Emese Revfy June 3, 2016, 5:42 p.m. UTC | #2
On Wed, 1 Jun 2016 12:42:27 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Tue, 31 May 2016 01:31:45 +0200 Emese Revfy <re.emese@gmail.com> wrote:
> 
> > This plugin mitigates the problem of the kernel having too little entropy during
> > and after boot for generating crypto keys.
> > 
> > It creates a local variable in every marked function. The value of this variable is
> > modified by randomly chosen operations (add, xor and rol) and
> > random values (gcc generates them at compile time and the stack pointer at runtime).
> > It depends on the control flow (e.g., loops, conditions).
> > 
> > Before the function returns the plugin writes this local variable
> > into the latent_entropy global variable. The value of this global variable is
> > added to the kernel entropy pool in do_one_initcall() and _do_fork().
> 
> I don't think I'm really understanding.  Won't this produce the same
> value on each and every boot?

No, because of interrupts and intentional data races.
David Brown June 6, 2016, 1:38 p.m. UTC | #3
On Fri, Jun 03, 2016 at 07:42:52PM +0200, Emese Revfy wrote:
>On Wed, 1 Jun 2016 12:42:27 -0700
>Andrew Morton <akpm@linux-foundation.org> wrote:
>
>> On Tue, 31 May 2016 01:31:45 +0200 Emese Revfy <re.emese@gmail.com> wrote:
>>
>> > This plugin mitigates the problem of the kernel having too little entropy during
>> > and after boot for generating crypto keys.
>> >
>> > It creates a local variable in every marked function. The value of this variable is
>> > modified by randomly chosen operations (add, xor and rol) and
>> > random values (gcc generates them at compile time and the stack pointer at runtime).
>> > It depends on the control flow (e.g., loops, conditions).
>> >
>> > Before the function returns the plugin writes this local variable
>> > into the latent_entropy global variable. The value of this global variable is
>> > added to the kernel entropy pool in do_one_initcall() and _do_fork().
>>
>> I don't think I'm really understanding.  Won't this produce the same
>> value on each and every boot?
>
>No, because of interrupts and intentional data races.

Wouldn't that result in the value having one of a small number of
values, then?  Even if it was just one of thousands or millions of
values, it would make the search space quite small.

David
Kees Cook June 6, 2016, 3:50 p.m. UTC | #4
On Mon, Jun 6, 2016 at 6:38 AM, David Brown <david.brown@linaro.org> wrote:
> On Fri, Jun 03, 2016 at 07:42:52PM +0200, Emese Revfy wrote:
>>
>> On Wed, 1 Jun 2016 12:42:27 -0700
>> Andrew Morton <akpm@linux-foundation.org> wrote:
>>
>>> On Tue, 31 May 2016 01:31:45 +0200 Emese Revfy <re.emese@gmail.com>
>>> wrote:
>>>
>>> > This plugin mitigates the problem of the kernel having too little
>>> > entropy during
>>> > and after boot for generating crypto keys.
>>> >
>>> > It creates a local variable in every marked function. The value of this
>>> > variable is
>>> > modified by randomly chosen operations (add, xor and rol) and
>>> > random values (gcc generates them at compile time and the stack pointer
>>> > at runtime).
>>> > It depends on the control flow (e.g., loops, conditions).
>>> >
>>> > Before the function returns the plugin writes this local variable
>>> > into the latent_entropy global variable. The value of this global
>>> > variable is
>>> > added to the kernel entropy pool in do_one_initcall() and _do_fork().
>>>
>>> I don't think I'm really understanding.  Won't this produce the same
>>> value on each and every boot?
>>
>>
>> No, because of interrupts and intentional data races.
>
>
> Wouldn't that result in the value having one of a small number of
> values, then?  Even if it was just one of thousands or millions of
> values, it would make the search space quite small.

My understanding is that it's not cryptographically secure, but it
provides a way for different machines and different boots to end up
with different seeds here, which is a big improvement over some of the
embedded devices that all boot with the same entropy every time.

I would, however, like to see the documentation improved to describe
the "How" and "Why". The "What" is pretty well covered. Adding
comments to the plugin for kernel developers (not compiler developers)
would help a lot: assume the reader knows nothing about gcc plugins.
:)

-Kees
pageexec@freemail.hu June 6, 2016, 7:30 p.m. UTC | #5
On 6 Jun 2016 at 7:38, David Brown wrote:

> On Fri, Jun 03, 2016 at 07:42:52PM +0200, Emese Revfy wrote:
> >On Wed, 1 Jun 2016 12:42:27 -0700
> >Andrew Morton <akpm@linux-foundation.org> wrote:
> >
> >> I don't think I'm really understanding.  Won't this produce the same
> >> value on each and every boot?
> >
> >No, because of interrupts and intentional data races.
> 
> Wouldn't that result in the value having one of a small number of
> values, then?  Even if it was just one of thousands or millions of
> values, it would make the search space quite small.

what matters for latent entropy is not the actual values fed into the entropy
pool (they're effectively compile time constants save for runtime data dependent
computations) but the precise sequence of them. interrupts stir this sequence
and thus extract entropy. perhaps as a small example imagine that an uninterrupted
kernel boot sequence feeds these values into the entropy pool:
  A B C

now imagine that a single interrupt can occur around any one of these values:
  I A B C
  A I B C
  A B I C
  A B C I

this way we can obtain 4 different final pool states that translate into up
to 2 bits of latent entropy (depends on how probable each sequence is). note
that this works regardless whether the underlying hardware has a high resolution
timer whose values the interrupt handler would feed into the pool.

the kernel boot process executes many of the above sequences with each sequence
potentially having a different length (the number of __init functions and initcalls
depends on the kernel config, initcalls execute for different lengths of time,
interrupt windows have different lengths, etc). how all this translates into
something measurable is a good question as i said elsewhere in the thread, my
completely unscientific guess would be that the number of interrupts is somehow
proportional to the extracted latent entropy.

cheers,
 PaX Team
Theodore Y. Ts'o June 6, 2016, 11:13 p.m. UTC | #6
On Mon, Jun 06, 2016 at 09:30:12PM +0200, PaX Team wrote:
> 
> what matters for latent entropy is not the actual values fed into the entropy
> pool (they're effectively compile time constants save for runtime data dependent
> computations) but the precise sequence of them. interrupts stir this sequence
> and thus extract entropy. perhaps as a small example imagine that an uninterrupted
> kernel boot sequence feeds these values into the entropy pool:
>   A B C
> 
> now imagine that a single interrupt can occur around any one of these values:
>   I A B C
>   A I B C
>   A B I C
>   A B C I
> 
> this way we can obtain 4 different final pool states that translate into up
> to 2 bits of latent entropy (depends on how probable each sequence is). note
> that this works regardless whether the underlying hardware has a high resolution
> timer whose values the interrupt handler would feed into the pool.

Right, but if it's only about interrupts, we're doing this already
inside modern Linux kernels.  On every single interrupt we are mixing
into a per-CPU "fast mix" pool the IP from the interrupt registers.

Since we're not claiming any additional entropy, I suppose it won't
hurt to do it twice, two different ways, but I'm not sure how much it
will actually help, and by doing the instrumentation in every single
basic block, instead of in the interrupt handler, I would think it
would be cheaper to do it in the interrupt handler.

      	 	       	     - Ted
pageexec@freemail.hu June 7, 2016, 12:19 p.m. UTC | #7
On 6 Jun 2016 at 19:13, Theodore Ts'o wrote:

> On Mon, Jun 06, 2016 at 09:30:12PM +0200, PaX Team wrote:
> > 
> > what matters for latent entropy is not the actual values fed into the entropy
> > pool (they're effectively compile time constants save for runtime data dependent
> > computations) but the precise sequence of them. interrupts stir this sequence
> > and thus extract entropy. perhaps as a small example imagine that an uninterrupted
> > kernel boot sequence feeds these values into the entropy pool:
> >   A B C
> > 
> > now imagine that a single interrupt can occur around any one of these values:
> >   I A B C
> >   A I B C
> >   A B I C
> >   A B C I
> > 
> > this way we can obtain 4 different final pool states that translate into up
> > to 2 bits of latent entropy (depends on how probable each sequence is). note
> > that this works regardless whether the underlying hardware has a high resolution
> > timer whose values the interrupt handler would feed into the pool.
> 
> Right, but if it's only about interrupts,

(i believe that) latent entropy is found in more than just interrupt timing, there're
also data dependent computations that can have entropy, either on a single system or
across a population of them.

> we're doing this already inside modern Linux kernels.  On every single
> interrupt we are mixing into a per-CPU "fast mix" pool the IP from the
> interrupt registers. 

i agree that sampling the kernel register state can have entropy (the plugin
already extracts the current stack pointer) but i'm much less sure about
userland (at least i see no dependence on !user_mode(...)) since an attacker
could feed no entropy into the pool but still get it credited.

cheers,
 PaX Team
Theodore Y. Ts'o June 7, 2016, 1:58 p.m. UTC | #8
On Tue, Jun 07, 2016 at 02:19:14PM +0200, PaX Team wrote:
> (i believe that) latent entropy is found in more than just interrupt timing, there're
> also data dependent computations that can have entropy, either on a single system or
> across a population of them.

It's not clear how much data dependent computations you would have in
kernel space that's not introduced by interrupts, but there would
some, I'm sure.

> > we're doing this already inside modern Linux kernels.  On every single
> > interrupt we are mixing into a per-CPU "fast mix" pool the IP from the
> > interrupt registers. 
> 
> i agree that sampling the kernel register state can have entropy (the plugin
> already extracts the current stack pointer) but i'm much less sure about
> userland (at least i see no dependence on !user_mode(...)) since an attacker
> could feed no entropy into the pool but still get it credited.

Well, the attacker can't control when the interrupts happen, but it
could try to burn power by simply having a thread spin in an infinite
loop ("0: jmp 0"), sure.  Of course, this would be rather noticeable,
and if there were any other jobs running, the attacker would be
degrading the amount of entropy that would be gathered, but not
eliminating it.

All of this goes into the question of how much entropy we can assume
can be gathered per interrupt (or in the case of basic block
instrumentation, per basic block).  IIRC, in the latent_entropy
patches, the assumption is that zero entropy should be credited,
correct?

In the case Linux's current get_interrupt_randomness(), there's a
reason I'm using a very conservative 1/64th of a bit per interrupt.
In practice, on most modern CPU where we have a cycle counter, even if
the bad guy was doing a "0: jmp 0" spinning loop, we would still get
entropy via the cycle counter interacting with what is hopefully a
certain amount of entropy from the interrupt timing.

On a crappy $50 Android phone/tablet from China, using an ancient ARM
chip that doesn't have any cycle counting facilities, we're kind of
screwed, but those devices have lousy batteries, so if you have an
attacker that has disabled the wakelocks and is spinning in an
infinite loop, the battery life won't last long, so the problem will
mostly solve itself when the phone dies.  :-)

       	     	    	     	   	  - Ted
pageexec@freemail.hu June 9, 2016, 5:22 p.m. UTC | #9
On 7 Jun 2016 at 9:58, Theodore Ts'o wrote:

> On Tue, Jun 07, 2016 at 02:19:14PM +0200, PaX Team wrote:
> > (i believe that) latent entropy is found in more than just interrupt timing, there're
> > also data dependent computations that can have entropy, either on a single system or
> > across a population of them.
> 
> It's not clear how much data dependent computations you would have in
> kernel space that's not introduced by interrupts, but there would
> some, I'm sure.

there's plenty of such computations both during boot and later as well. starting
with kernel command line options through parsing firmware provided data to hardware
configurations to processing various queues, lists, trees, file systems, network
packets, etc. as for interrupts specifically, latent entropy can be extracted from
polled devices as well (e.g., i think even modern NICs can be turned into polling
mode under sufficient load as processing packets that way is more efficient).

> > i agree that sampling the kernel register state can have entropy (the plugin
> > already extracts the current stack pointer) but i'm much less sure about
> > userland (at least i see no dependence on !user_mode(...)) since an attacker
> > could feed no entropy into the pool but still get it credited.
> 
> Well, the attacker can't control when the interrupts happen, but it
> could try to burn power by simply having a thread spin in an infinite
> loop ("0: jmp 0"), sure.

yes, that's one obvious way to accomplish it but even normal applications can
behave in a similar way, think about spinning event loops, media decoding, etc
whose sampled insn ptrs may provide less entropy than they get credited for.

> All of this goes into the question of how much entropy we can assume
> can be gathered per interrupt (or in the case of basic block
> instrumentation, per basic block).  IIRC, in the latent_entropy
> patches, the assumption is that zero entropy should be credited,
> correct?

yes, no entropy is credited since i don't know how much there is and i tend to err
on the side of safety which means crediting 0 entropy for latent entropy. of course
the expectation is that it's not actually 0 but to prove any specific value or limit
is beyond my skills at least.

> In the case Linux's current get_interrupt_randomness(), there's a
> reason I'm using a very conservative 1/64th of a bit per interrupt.

i think it's not just per 64 interrupts but also after each elapsed second (i.e.,
whichever condition occurs first), so on an idle system (which i believe is more
likely to occur on exactly those small systems that the referenced paper was concerned
about) the credited entropy could be overestimated.

> In practice, on most modern CPU where we have a cycle counter,

a quick check for get_cycles shows that at least these archs seem to return 0:
arc, avr32, cris, frv, m32r, m68k, xtensa. now you may not think of them as modern,
but they're still used in real life devices. i think that latent entropy is still
an option on them.

cheers,
 PaX Team
Theodore Y. Ts'o June 9, 2016, 7:55 p.m. UTC | #10
On Thu, Jun 09, 2016 at 07:22:29PM +0200, PaX Team wrote:
> > Well, the attacker can't control when the interrupts happen, but it
> > could try to burn power by simply having a thread spin in an infinite
> > loop ("0: jmp 0"), sure.
> 
> yes, that's one obvious way to accomplish it but even normal applications can
> behave in a similar way, think about spinning event loops, media decoding, etc
> whose sampled insn ptrs may provide less entropy than they get credited for.

Sure, as long as we're assuming less than one bit of entropy per
interrupt, even for a loop which which is:

1:   cmpl    $1, -8(%rsp)
     jz	     1b

there would still be *some* uncertainty.  And with an event loop there
would be more instructions to sample.  Granted, the number of cycles
spent in each will be different, so there will be some biasing, but
that's one of the reason why we've been using 1/64 bit per interrupt.

> yes, no entropy is credited since i don't know how much there is and i tend to err
> on the side of safety which means crediting 0 entropy for latent entropy. of course
> the expectation is that it's not actually 0 but to prove any specific value or limit
> is beyond my skills at least.

Sure, that's fair.

> i think it's not just per 64 interrupts but also after each elapsed second (i.e.,
> whichever condition occurs first), so on an idle system (which i believe is more
> likely to occur on exactly those small systems that the referenced paper was concerned
> about) the credited entropy could be overestimated.

That's a fair concern.  It might be that we should enforce some
minimum (at least 8 interrupts in all cases), but this is where it's
all about hueristics, especially on those systems that don't have random_get_entropy().

> > In practice, on most modern CPU where we have a cycle counter,
> 
> a quick check for get_cycles shows that at least these archs seem to return 0:
> arc, avr32, cris, frv, m32r, m68k, xtensa. now you may not think of them as modern,
> but they're still used in real life devices. i think that latent entropy is still
> an option on them.

It's possible for a system not to have a cycle counter, but to have
something that can be used instead for random_get_entropy.  That's
only being used for the m68k/amiga and mips/R6000[A] cases, but I keep
hoping that the archiecture maintainers for osme of these other
oddball platform (is that better than "non-modern"? :-) will come up
with something, but yes, it is those platforms where I've always been
the most worried.  On the one hand, if the hardware is crap, there's
very little you can do.  Unfortnuately, very often these crap
architectures have a very low BOM cost, so they are most likely to be
used in IOT devices.   :-(

One could try to claim that these IOT devics won't have upgradeable
firmware and, so they'll probably be security disasters even without a
good random number generators, but oddly, that doesn't give me much
solace...

And in the end, that may be the strongest argment for the
latent_entropy plugin.  Even if it doesn't provide a lot of extra
entropy, on those platforms we're going to be so starved of real
entropy that almost anything will be better than what we have today.

	     	    	     	     	    	 - Ted
Kees Cook June 9, 2016, 8:08 p.m. UTC | #11
On Thu, Jun 9, 2016 at 12:55 PM, Theodore Ts'o <tytso@mit.edu> wrote:
> On Thu, Jun 09, 2016 at 07:22:29PM +0200, PaX Team wrote:
>> > Well, the attacker can't control when the interrupts happen, but it
>> > could try to burn power by simply having a thread spin in an infinite
>> > loop ("0: jmp 0"), sure.
>>
>> yes, that's one obvious way to accomplish it but even normal applications can
>> behave in a similar way, think about spinning event loops, media decoding, etc
>> whose sampled insn ptrs may provide less entropy than they get credited for.
>
> Sure, as long as we're assuming less than one bit of entropy per
> interrupt, even for a loop which which is:
>
> 1:   cmpl    $1, -8(%rsp)
>      jz      1b
>
> there would still be *some* uncertainty.  And with an event loop there
> would be more instructions to sample.  Granted, the number of cycles
> spent in each will be different, so there will be some biasing, but
> that's one of the reason why we've been using 1/64 bit per interrupt.
>
>> yes, no entropy is credited since i don't know how much there is and i tend to err
>> on the side of safety which means crediting 0 entropy for latent entropy. of course
>> the expectation is that it's not actually 0 but to prove any specific value or limit
>> is beyond my skills at least.
>
> Sure, that's fair.
>
>> i think it's not just per 64 interrupts but also after each elapsed second (i.e.,
>> whichever condition occurs first), so on an idle system (which i believe is more
>> likely to occur on exactly those small systems that the referenced paper was concerned
>> about) the credited entropy could be overestimated.
>
> That's a fair concern.  It might be that we should enforce some
> minimum (at least 8 interrupts in all cases), but this is where it's
> all about hueristics, especially on those systems that don't have random_get_entropy().
>
>> > In practice, on most modern CPU where we have a cycle counter,
>>
>> a quick check for get_cycles shows that at least these archs seem to return 0:
>> arc, avr32, cris, frv, m32r, m68k, xtensa. now you may not think of them as modern,
>> but they're still used in real life devices. i think that latent entropy is still
>> an option on them.
>
> It's possible for a system not to have a cycle counter, but to have
> something that can be used instead for random_get_entropy.  That's
> only being used for the m68k/amiga and mips/R6000[A] cases, but I keep
> hoping that the archiecture maintainers for osme of these other
> oddball platform (is that better than "non-modern"? :-) will come up
> with something, but yes, it is those platforms where I've always been
> the most worried.  On the one hand, if the hardware is crap, there's
> very little you can do.  Unfortnuately, very often these crap
> architectures have a very low BOM cost, so they are most likely to be
> used in IOT devices.   :-(
>
> One could try to claim that these IOT devics won't have upgradeable
> firmware and, so they'll probably be security disasters even without a
> good random number generators, but oddly, that doesn't give me much
> solace...
>
> And in the end, that may be the strongest argment for the
> latent_entropy plugin.  Even if it doesn't provide a lot of extra
> entropy, on those platforms we're going to be so starved of real
> entropy that almost anything will be better than what we have today.

Yeah, that's been my thinking around this. And on more sane systems,
using latent_entropy doesn't make things worse. :)

-Kees
Kees Cook June 9, 2016, 9:51 p.m. UTC | #12
On Mon, May 30, 2016 at 4:31 PM, Emese Revfy <re.emese@gmail.com> wrote:
> This plugin mitigates the problem of the kernel having too little entropy during
> and after boot for generating crypto keys.
>
> It creates a local variable in every marked function. The value of this variable is
> modified by randomly chosen operations (add, xor and rol) and
> random values (gcc generates them at compile time and the stack pointer at runtime).
> It depends on the control flow (e.g., loops, conditions).
>
> Before the function returns the plugin writes this local variable
> into the latent_entropy global variable. The value of this global variable is
> added to the kernel entropy pool in do_one_initcall() and _do_fork().
>
> Signed-off-by: Emese Revfy <re.emese@gmail.com>
> [...]
> diff --git a/scripts/Makefile.gcc-plugins b/scripts/Makefile.gcc-plugins
> index 5e22b60..cd7902c 100644
> --- a/scripts/Makefile.gcc-plugins
> +++ b/scripts/Makefile.gcc-plugins
> @@ -6,6 +6,12 @@ ifdef CONFIG_GCC_PLUGINS
>
>    gcc-plugin-$(CONFIG_GCC_PLUGIN_CYC_COMPLEXITY)       += cyc_complexity_plugin.so
>
> +  gcc-plugin-$(CONFIG_GCC_PLUGIN_LATENT_ENTROPY)       += latent_entropy_plugin.so
> +  gcc-plugin-cflags-$(CONFIG_GCC_PLUGIN_LATENT_ENTROPY)        += -DLATENT_ENTROPY_PLUGIN
> +  ifdef CONFIG_PAX_LATENT_ENTROPY
> +    DISABLE_LATENT_ENTROPY_PLUGIN                      += -fplugin-arg-latent_entropy_plugin-disable
> +  endif
> +
>    ifdef CONFIG_GCC_PLUGIN_SANCOV
>      ifeq ($(CFLAGS_KCOV),)
>        # It is needed because of the gcc-plugin.sh and gcc version checks.
> @@ -19,9 +25,9 @@ ifdef CONFIG_GCC_PLUGINS
;>      endif
>    endif
>
> -  GCC_PLUGINS_CFLAGS := $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y))
> +  GCC_PLUGINS_CFLAGS := $(strip $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y)) $(gcc-plugin-cflags-y))

Is this change part of latent_entropy, or a general fix to the gcc
plugin infrastructure?

>
> -  export PLUGINCC GCC_PLUGINS_CFLAGS GCC_PLUGIN SANCOV_PLUGIN
> +  export PLUGINCC GCC_PLUGINS_CFLAGS GCC_PLUGIN SANCOV_PLUGIN DISABLE_LATENT_ENTROPY_PLUGIN
>
>    ifeq ($(PLUGINCC),)
>      ifneq ($(GCC_PLUGINS_CFLAGS),)
> [...]
> diff --git a/scripts/gcc-plugins/latent_entropy_plugin.c b/scripts/gcc-plugins/latent_entropy_plugin.c
> new file mode 100644
> index 0000000..f606caa
> --- /dev/null
> +++ b/scripts/gcc-plugins/latent_entropy_plugin.c
> @@ -0,0 +1,454 @@
> +/*
> + * Copyright 2012-2016 by the PaX Team <pageexec@freemail.hu>
> + * Copyright 2016 by Emese Revfy <re.emese@gmail.com>
> + * Licensed under the GPL v2
> + *
> + * Note: the choice of the license means that the compilation process is
> + *       NOT 'eligible' as defined by gcc's library exception to the GPL v3,
> + *       but for the kernel it doesn't matter since it doesn't link against
> + *       any of the gcc libraries
> + *
> + * gcc plugin to help generate a little bit of entropy from program state,
> + * used throughout the uptime of the kernel

I think this comment needs a lot of expanding. What are all the ways
that this plugin makes changes to code? Things I think I see are:
pre-filling data variables with randomness, creating a local_entropy
variable (local to what?), mixing stack pointer (into what?), updating
latent_entropy global.

> + *
> + * TODO:
> + * - add ipa pass to identify not explicitly marked candidate functions
> + * - mix in more program state (function arguments/return values, loop variables, etc)
> + * - more instrumentation control via attribute parameters
> + *
> + * BUGS:
> + * - none known
> + *
> + * Options:
> + * -fplugin-arg-latent_entropy_plugin-disable
> + *
> + * Attribute: __attribute__((latent_entropy))
> + *  The latent_entropy gcc attribute can be only on functions and variables.
> + *  If it is on a function then the plugin will instrument it. If the attribute
> + *  is on a variable then the plugin will initialize it with a random value.
> + *  The variable must be an integer, an integer array type or a structure with integer fields.
> + */
> +
> +#include "gcc-common.h"
> +
> +int plugin_is_GPL_compatible;
> +
> +static GTY(()) tree latent_entropy_decl;
> +
> +static struct plugin_info latent_entropy_plugin_info = {
> +       .version        = "201605292100",
> +       .help           = "disable\tturn off latent entropy instrumentation\n",
> +};
> +
> +static unsigned HOST_WIDE_INT seed;
> +static unsigned HOST_WIDE_INT get_random_const(void)
> +{
> +       unsigned int i;
> +       unsigned HOST_WIDE_INT ret = 0;
> +
> +       for (i = 0; i < 8 * sizeof ret; i++) {
> +               ret = (ret << 1) | (seed & 1);
> +               seed >>= 1;
> +               if (ret & 1)
> +                       seed ^= 0xD800000000000000ULL;
> +       }
> +
> +       return ret;
> +}

Please add some comments above this function about why the seed is
chosen this way, how it is expected to change over the lifetime of the
plugin, etc.

> +static tree handle_latent_entropy_attribute(tree *node, tree name, tree args __unused, int flags __unused, bool *no_add_attrs)

Can you add comments to each section below describing what's being
checked for? Or describe above the function what specific situations
are valid for using the attribute? (The latter patch says "functions",
but also marks other kinds of things.)

> +{
> +       tree type;
> +       unsigned long long mask;
> +#if BUILDING_GCC_VERSION <= 4007
> +       VEC(constructor_elt, gc) *vals;
> +#else
> +       vec<constructor_elt, va_gc> *vals;
> +#endif
> +
> +       switch (TREE_CODE(*node)) {
> +       default:
> +               *no_add_attrs = true;
> +               error("%qE attribute only applies to functions and variables", name);
> +               break;
> +
> +       case VAR_DECL:
> +               if (DECL_INITIAL(*node)) {
> +                       *no_add_attrs = true;
> +                       error("variable %qD with %qE attribute must not be initialized", *node, name);
> +                       break;
> +               }
> +
> +               if (!TREE_STATIC(*node)) {
> +                       *no_add_attrs = true;
> +                       error("variable %qD with %qE attribute must not be local", *node, name);
> +                       break;
> +               }
> +
> +               type = TREE_TYPE(*node);
> +               switch (TREE_CODE(type)) {
> +               default:
> +                       *no_add_attrs = true;
> +                       error("variable %qD with %qE attribute must be an integer or a fixed length integer array type"
> +                               "or a fixed sized structure with integer fields", *node, name);
> +                       break;
> +
> +               case RECORD_TYPE: {
> +                       tree field;
> +                       unsigned int nelt = 0;
> +
> +                       for (field = TYPE_FIELDS(type); field; nelt++, field = TREE_CHAIN(field)) {
> +                               tree fieldtype;
> +
> +                               fieldtype = TREE_TYPE(field);
> +                               if (TREE_CODE(fieldtype) == INTEGER_TYPE)
> +                                       continue;
> +
> +                               *no_add_attrs = true;
> +                               error("structure variable %qD with %qE attribute has a non-integer field %qE", *node, name, field);
> +                               break;
> +                       }
> +
> +                       if (field)
> +                               break;
> +
> +#if BUILDING_GCC_VERSION <= 4007
> +                       vals = VEC_alloc(constructor_elt, gc, nelt);
> +#else
> +                       vec_alloc(vals, nelt);
> +#endif
> +
> +                       for (field = TYPE_FIELDS(type); field; field = TREE_CHAIN(field)) {
> +                               tree fieldtype;
> +
> +                               fieldtype = TREE_TYPE(field);
> +                               mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(fieldtype)) - 1);
> +                               mask = 2 * (mask - 1) + 1;
> +
> +                               if (TYPE_UNSIGNED(fieldtype))
> +                                       CONSTRUCTOR_APPEND_ELT(vals, field, build_int_cstu(fieldtype, mask & get_random_const()));
> +                               else
> +                                       CONSTRUCTOR_APPEND_ELT(vals, field, build_int_cst(fieldtype, mask & get_random_const()));
> +                       }
> +
> +                       DECL_INITIAL(*node) = build_constructor(type, vals);
> +                       break;
> +               }
> +
> +               case INTEGER_TYPE:
> +                       mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(type)) - 1);
> +                       mask = 2 * (mask - 1) + 1;
> +
> +                       if (TYPE_UNSIGNED(type))
> +                               DECL_INITIAL(*node) = build_int_cstu(type, mask & get_random_const());
> +                       else
> +                               DECL_INITIAL(*node) = build_int_cst(type, mask & get_random_const());
> +                       break;

What is happening here? Is this populating integers with the random
const? (I assume the ARRAY_TYPE version of this is the same thing,
only multiple times. Could that be made into a function instead of
cut/paste with a loop in the ARRAY_TYPE case below?

> +
> +               case ARRAY_TYPE: {
> +                       tree elt_type, array_size, elt_size;
> +                       unsigned int i, nelt;
> +
> +                       elt_type = TREE_TYPE(type);
> +                       elt_size = TYPE_SIZE_UNIT(TREE_TYPE(type));
> +                       array_size = TYPE_SIZE_UNIT(type);
> +
> +                       if (TREE_CODE(elt_type) != INTEGER_TYPE || !array_size || TREE_CODE(array_size) != INTEGER_CST) {
> +                               *no_add_attrs = true;
> +                               error("array variable %qD with %qE attribute must be a fixed length integer array type", *node, name);
> +                               break;
> +                       }
> +
> +                       nelt = TREE_INT_CST_LOW(array_size) / TREE_INT_CST_LOW(elt_size);
> +#if BUILDING_GCC_VERSION <= 4007
> +                       vals = VEC_alloc(constructor_elt, gc, nelt);
> +#else
> +                       vec_alloc(vals, nelt);
> +#endif
> +
> +                       mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(elt_type)) - 1);
> +                       mask = 2 * (mask - 1) + 1;

This mask calculation appears to be cut/pasted. Should it be a macro
or function taking "type" instead?

> +
> +                       for (i = 0; i < nelt; i++)
> +                               if (TYPE_UNSIGNED(elt_type))
> +                                       CONSTRUCTOR_APPEND_ELT(vals, size_int(i), build_int_cstu(elt_type, mask & get_random_const()));
> +                               else
> +                                       CONSTRUCTOR_APPEND_ELT(vals, size_int(i), build_int_cst(elt_type, mask & get_random_const()));
> +
> +                       DECL_INITIAL(*node) = build_constructor(type, vals);
> +                       break;
> +               }
> +               }
> +               break;
> +
> +       case FUNCTION_DECL:
> +               break;
> +       }
> +
> +       return NULL_TREE;
> +}
> +
> +static struct attribute_spec latent_entropy_attr = {
> +       .name                           = "latent_entropy",
> +       .min_length                     = 0,
> +       .max_length                     = 0,
> +       .decl_required                  = true,
> +       .type_required                  = false,
> +       .function_type_required         = false,
> +       .handler                        = handle_latent_entropy_attribute,
> +#if BUILDING_GCC_VERSION >= 4007
> +       .affects_type_identity          = false
> +#endif
> +};
> +
> +static void register_attributes(void *event_data __unused, void *data __unused)
> +{
> +       register_attribute(&latent_entropy_attr);
> +}
> +
> +static bool latent_entropy_gate(void)
> +{
> +       /* don't bother with noreturn functions for now */
> +       if (TREE_THIS_VOLATILE(current_function_decl))
> +               return false;
> +
> +       /* gcc-4.5 doesn't discover some trivial noreturn functions */
> +       if (EDGE_COUNT(EXIT_BLOCK_PTR_FOR_FN(cfun)->preds) == 0)
> +               return false;
> +
> +       return lookup_attribute("latent_entropy", DECL_ATTRIBUTES(current_function_decl)) != NULL_TREE;
> +}
> +
> +static tree create_a_tmp_var(tree type, const char *name)
> +{
> +       tree var;
> +
> +       var = create_tmp_var(type, name);
> +       add_referenced_var(var);
> +       mark_sym_for_renaming(var);
> +       return var;
> +}
> +
> +static enum tree_code get_op(tree *rhs)

Please describe this state machine, and why it does what it does. :)

> +{
> +       static enum tree_code op;
> +       unsigned HOST_WIDE_INT random_const;
> +
> +       random_const = get_random_const();
> +
> +       switch (op) {
> +       case BIT_XOR_EXPR:
> +               op = PLUS_EXPR;
> +               break;
> +
> +       case PLUS_EXPR:
> +               if (rhs) {
> +                       op = LROTATE_EXPR;
> +                       random_const &= HOST_BITS_PER_WIDE_INT - 1;
> +                       break;
> +               }

What's happening here with the random_const?

> +
> +       case LROTATE_EXPR:
> +       default:
> +               op = BIT_XOR_EXPR;
> +               break;
> +       }
> +       if (rhs)
> +               *rhs = build_int_cstu(unsigned_intDI_type_node, random_const);
> +       return op;
> +}
> +
> +static void perturb_local_entropy(basic_block bb, tree local_entropy)

What effect does this function have on the resulting code output?

> +{
> +       gimple_stmt_iterator gsi;
> +       gimple assign;
> +       tree rhs;
> +       enum tree_code subcode;
> +
> +       subcode = get_op(&rhs);
> +       assign = gimple_build_assign_with_ops(subcode, local_entropy, local_entropy, rhs);
> +       gsi = gsi_after_labels(bb);
> +       gsi_insert_before(&gsi, assign, GSI_NEW_STMT);
> +       update_stmt(assign);
> +}
> +
> +static void perturb_latent_entropy(basic_block bb, tree rhs)

Same for this. I assume this is effectively:

   u64 temp_latent_entropy;

   temp_latent_entropy = latent_entropy;
   temp_latent_entropy = temp_latent_entropy OP rhs
   latent_entropy = temp_latent_entropy;

Where does rhs come from? (Is this the "local_entropy" below?)

> +{
> +       gimple_stmt_iterator gsi;
> +       gimple assign;
> +       tree temp;
> +       enum tree_code subcode;
> +
> +       /* create temporary copy of latent_entropy */
> +       temp = create_a_tmp_var(unsigned_intDI_type_node, "temp_latent_entropy");
> +
> +       gsi = gsi_last_bb(bb);
> +
> +       /* 1. read... */
> +       add_referenced_var(latent_entropy_decl);
> +       mark_sym_for_renaming(latent_entropy_decl);
> +       assign = gimple_build_assign(temp, latent_entropy_decl);
> +       gsi_insert_before(&gsi, assign, GSI_NEW_STMT);
> +       update_stmt(assign);
> +
> +       /* 2. ...modify... */
> +       subcode = get_op(NULL);
> +       assign = gimple_build_assign_with_ops(subcode, temp, temp, rhs);
> +       gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
> +       update_stmt(assign);
> +
> +       /* 3. ...write latent_entropy */
> +       assign = gimple_build_assign(latent_entropy_decl, temp);
> +       gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
> +       update_stmt(assign);
> +}
> +
> +static void mix_in_sp(basic_block bb, tree local_entropy)

What is the stack pointer mixed into?

> +{
> +       gimple assign, call;
> +       tree frame_addr, rand_const;
> +       gimple_stmt_iterator gsi = gsi_after_labels(bb);
> +
> +       frame_addr = create_a_tmp_var(ptr_type_node, "local_entropy_frame_addr");
> +
> +       call = gimple_build_call(builtin_decl_implicit(BUILT_IN_FRAME_ADDRESS), 1, integer_zero_node);
> +       gimple_call_set_lhs(call, frame_addr);
> +       gsi_insert_before(&gsi, call, GSI_NEW_STMT);
> +       update_stmt(call);
> +
> +       assign = gimple_build_assign(local_entropy, fold_convert(unsigned_intDI_type_node, frame_addr));
> +       gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
> +       update_stmt(assign);
> +
> +       rand_const = build_int_cstu(unsigned_intDI_type_node, get_random_const());
> +       assign = gimple_build_assign_with_ops(BIT_XOR_EXPR, local_entropy, local_entropy, rand_const);
> +       gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
> +       update_stmt(assign);
> +}
> +
> +static unsigned int latent_entropy_execute(void)
> +{
> +       basic_block bb;
> +       tree local_entropy;
> +
> +       if (!latent_entropy_decl) {
> +               varpool_node_ptr node;
> +
> +               FOR_EACH_VARIABLE(node) {
> +                       tree var = NODE_DECL(node);
> +
> +                       if (DECL_NAME_LENGTH(var) < sizeof("latent_entropy") - 1)
> +                               continue;
> +                       if (strcmp(IDENTIFIER_POINTER(DECL_NAME(var)), "latent_entropy"))
> +                               continue;
> +                       latent_entropy_decl = var;
> +                       break;
> +               }
> +               if (!latent_entropy_decl)
> +                       return 0;
> +       }
> +
> +       gcc_assert(single_succ_p(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
> +       bb = single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun));
> +       if (!single_pred_p(bb)) {
> +               split_edge(single_succ_edge(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
> +               gcc_assert(single_succ_p(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
> +               bb = single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun));
> +       }
> +

This below needs a bit more detail in comments. IIUC, it's creating a
local (to the .o file? the basic block?) variable, initializing it
with the stack, perturbing it with random operations, then updating
the latent_entropy with it?

> +       /* create local entropy variable */
> +       local_entropy = create_a_tmp_var(unsigned_intDI_type_node, "local_entropy");

What value does local_entropy have initially?

> +
> +       /* 1. stack pointer */
> +       mix_in_sp(bb, local_entropy);
> +
> +       bb = bb->next_bb;
> +       /* 2. instrument each BB with an operation on the local entropy variable */
> +       while (bb != EXIT_BLOCK_PTR_FOR_FN(cfun)) {
> +               perturb_local_entropy(bb, local_entropy);
> +               bb = bb->next_bb;
> +       };
> +
> +       /* 3. mix local entropy into the global entropy variable */
> +       gcc_assert(single_pred_p(EXIT_BLOCK_PTR_FOR_FN(cfun)));
> +       perturb_latent_entropy(single_pred(EXIT_BLOCK_PTR_FOR_FN(cfun)), local_entropy);
> +       return 0;
> +}
> [...]

Thanks!

-Kees
Emese Revfy June 13, 2016, 9:49 p.m. UTC | #13
On Thu, 9 Jun 2016 14:51:45 -0700
Kees Cook <keescook@chromium.org> wrote:

> On Mon, May 30, 2016 at 4:31 PM, Emese Revfy <re.emese@gmail.com> wrote:
> > -  GCC_PLUGINS_CFLAGS := $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y))
> > +  GCC_PLUGINS_CFLAGS := $(strip $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y)) $(gcc-plugin-cflags-y))
> 
> Is this change part of latent_entropy, or a general fix to the gcc
> plugin infrastructure?

This is a new feature in the gcc plugin infrastructure. The latent_entropy plugin has an argument and
we must add it (with a space) to the cflags.

> > + * gcc plugin to help generate a little bit of entropy from program state,
> > + * used throughout the uptime of the kernel
> 
> I think this comment needs a lot of expanding. What are all the ways
> that this plugin makes changes to code? Things I think I see are:
> pre-filling data variables with randomness, creating a local_entropy
> variable (local to what?), mixing stack pointer (into what?), updating
> latent_entropy global.

I demonstrated the details here:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13

> > +static unsigned HOST_WIDE_INT seed;
> > +static unsigned HOST_WIDE_INT get_random_const(void)
> > +{
> > +       unsigned int i;
> > +       unsigned HOST_WIDE_INT ret = 0;
> > +
> > +       for (i = 0; i < 8 * sizeof ret; i++) {
> > +               ret = (ret << 1) | (seed & 1);
> > +               seed >>= 1;
> > +               if (ret & 1)
> > +                       seed ^= 0xD800000000000000ULL;
> > +       }
> > +
> > +       return ret;
> > +}
> 
> Please add some comments above this function about why the seed is
> chosen this way, how it is expected to change over the lifetime of the
> plugin, etc.

You can see the comments here:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/4999276e866271c69186a8e3112c265b6a0f3205

> > +static tree handle_latent_entropy_attribute(tree *node, tree name, tree args __unused, int flags __unused, bool *no_add_attrs)
> 
> Can you add comments to each section below describing what's being
> checked for? Or describe above the function what specific situations
> are valid for using the attribute? (The latter patch says "functions",
> but also marks other kinds of things.)

I think the error messages already describe all the wrong situations.
What would you like to see in addition to the existing error messages?

You can find a description about the attribute here:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/f0ec66810682579109469b862ac5169aa2a743ca

> > +                       mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(type)) - 1);
> > +                       mask = 2 * (mask - 1) + 1;
> > +
> > +                       if (TYPE_UNSIGNED(type))
> > +                               DECL_INITIAL(*node) = build_int_cstu(type, mask & get_random_const());
> > +                       else
> > +                               DECL_INITIAL(*node) = build_int_cst(type, mask & get_random_const());
> > +                       break;
> 
> What is happening here? Is this populating integers with the random
> const? (I assume the ARRAY_TYPE version of this is the same thing,
> only multiple times. Could that be made into a function instead of
> cut/paste with a loop in the ARRAY_TYPE case below?

https://github.com/ephox-gcc-plugins/latent_entropy/commit/d65864b6ca2e61cc73cd28309ba0779fde75b4f2
 
> > +static enum tree_code get_op(tree *rhs)
> 
> Please describe this state machine, and why it does what it does. :)

https://github.com/ephox-gcc-plugins/latent_entropy/commit/c4cc18cfb5d37121fe62907bed6b5aaafb84fff8
 
> > +{
> > +       static enum tree_code op;
> > +       unsigned HOST_WIDE_INT random_const;
> > +
> > +       random_const = get_random_const();
> > +
> > +       switch (op) {
> > +       case BIT_XOR_EXPR:
> > +               op = PLUS_EXPR;
> > +               break;
> > +
> > +       case PLUS_EXPR:
> > +               if (rhs) {
> > +                       op = LROTATE_EXPR;
> > +                       random_const &= HOST_BITS_PER_WIDE_INT - 1;
> > +                       break;
> > +               }
> 
> What's happening here with the random_const?

I wrote a comment, you can find it here:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/da452fdbc0247095fdf2b1f52eb4ddd368fad640

> > +
> > +       case LROTATE_EXPR:
> > +       default:
> > +               op = BIT_XOR_EXPR;
> > +               break;
> > +       }
> > +       if (rhs)
> > +               *rhs = build_int_cstu(unsigned_intDI_type_node, random_const);
> > +       return op;
> > +}
> > +
> > +static void perturb_local_entropy(basic_block bb, tree local_entropy)
> 
> What effect does this function have on the resulting code output?

Would you like to see more on top of this comment:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13

> > +static void perturb_latent_entropy(basic_block bb, tree rhs)
> 
> Same for this. I assume this is effectively:
> 
>    u64 temp_latent_entropy;
> 
>    temp_latent_entropy = latent_entropy;
>    temp_latent_entropy = temp_latent_entropy OP rhs
>    latent_entropy = temp_latent_entropy;
> 
> Where does rhs come from? (Is this the "local_entropy" below?)

Sure, I'll rename the rhs parameter to local_entropy.

> > +static void mix_in_sp(basic_block bb, tree local_entropy)
> 
> What is the stack pointer mixed into?

I already wrote some comments:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/d2781819d774b4370c248cdb8d0dd2b47308b6f4

> This below needs a bit more detail in comments. IIUC, it's creating a
> local (to the .o file? the basic block?) variable, initializing it
> with the stack, perturbing it with random operations, then updating
> the latent_entropy with it?
> > +       /* create local entropy variable */
> > +       local_entropy = create_a_tmp_var(unsigned_intDI_type_node, "local_entropy");
> 
> What value does local_entropy have initially?

You can see it here:
https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13
Kees Cook June 14, 2016, 6:27 p.m. UTC | #14
On Mon, Jun 13, 2016 at 2:49 PM, Emese Revfy <re.emese@gmail.com> wrote:
> On Thu, 9 Jun 2016 14:51:45 -0700
> Kees Cook <keescook@chromium.org> wrote:
>
>> On Mon, May 30, 2016 at 4:31 PM, Emese Revfy <re.emese@gmail.com> wrote:
>> > -  GCC_PLUGINS_CFLAGS := $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y))
>> > +  GCC_PLUGINS_CFLAGS := $(strip $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y)) $(gcc-plugin-cflags-y))
>>
>> Is this change part of latent_entropy, or a general fix to the gcc
>> plugin infrastructure?
>
> This is a new feature in the gcc plugin infrastructure. The latent_entropy plugin has an argument and
> we must add it (with a space) to the cflags.

Okay, can you break this out into a separate commit?

>
>> > + * gcc plugin to help generate a little bit of entropy from program state,
>> > + * used throughout the uptime of the kernel
>>
>> I think this comment needs a lot of expanding. What are all the ways
>> that this plugin makes changes to code? Things I think I see are:
>> pre-filling data variables with randomness, creating a local_entropy
>> variable (local to what?), mixing stack pointer (into what?), updating
>> latent_entropy global.
>
> I demonstrated the details here:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13

That helps, thanks. Can you also mention how __latent_entropy changes
non-functions? (i.e. initializes them with random data.)

Also, I think this isn't accurate:

 * local_entropy ^= get_random_long();

Looking at the disassembly, it seems that static random values (i.e.
randomly chosen at gcc runtime) are added, rather than making calls to
the kernel's get_random_long() function.

>> > +static unsigned HOST_WIDE_INT seed;
>> > +static unsigned HOST_WIDE_INT get_random_const(void)
>> > +{
>> > +       unsigned int i;
>> > +       unsigned HOST_WIDE_INT ret = 0;
>> > +
>> > +       for (i = 0; i < 8 * sizeof ret; i++) {
>> > +               ret = (ret << 1) | (seed & 1);
>> > +               seed >>= 1;
>> > +               if (ret & 1)
>> > +                       seed ^= 0xD800000000000000ULL;
>> > +       }
>> > +
>> > +       return ret;
>> > +}
>>
>> Please add some comments above this function about why the seed is
>> chosen this way, how it is expected to change over the lifetime of the
>> plugin, etc.
>
> You can see the comments here:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/4999276e866271c69186a8e3112c265b6a0f3205

Ah-ha, thanks. I missed the assignment of "seed" during the init phase.

>> > +static tree handle_latent_entropy_attribute(tree *node, tree name, tree args __unused, int flags __unused, bool *no_add_attrs)
>>
>> Can you add comments to each section below describing what's being
>> checked for? Or describe above the function what specific situations
>> are valid for using the attribute? (The latter patch says "functions",
>> but also marks other kinds of things.)
>
> I think the error messages already describe all the wrong situations.
> What would you like to see in addition to the existing error messages?
>
> You can find a description about the attribute here:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/f0ec66810682579109469b862ac5169aa2a743ca

I think that clears it up nicely.

>
>> > +                       mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(type)) - 1);
>> > +                       mask = 2 * (mask - 1) + 1;
>> > +
>> > +                       if (TYPE_UNSIGNED(type))
>> > +                               DECL_INITIAL(*node) = build_int_cstu(type, mask & get_random_const());
>> > +                       else
>> > +                               DECL_INITIAL(*node) = build_int_cst(type, mask & get_random_const());
>> > +                       break;
>>
>> What is happening here? Is this populating integers with the random
>> const? (I assume the ARRAY_TYPE version of this is the same thing,
>> only multiple times. Could that be made into a function instead of
>> cut/paste with a loop in the ARRAY_TYPE case below?
>
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/d65864b6ca2e61cc73cd28309ba0779fde75b4f2

Great! This is much more readable to me.

>> > +static enum tree_code get_op(tree *rhs)
>>
>> Please describe this state machine, and why it does what it does. :)
>
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/c4cc18cfb5d37121fe62907bed6b5aaafb84fff8

Great!

>
>> > +{
>> > +       static enum tree_code op;
>> > +       unsigned HOST_WIDE_INT random_const;
>> > +
>> > +       random_const = get_random_const();
>> > +
>> > +       switch (op) {
>> > +       case BIT_XOR_EXPR:
>> > +               op = PLUS_EXPR;
>> > +               break;
>> > +
>> > +       case PLUS_EXPR:
>> > +               if (rhs) {
>> > +                       op = LROTATE_EXPR;
>> > +                       random_const &= HOST_BITS_PER_WIDE_INT - 1;
>> > +                       break;
>> > +               }
>>
>> What's happening here with the random_const?
>
> I wrote a comment, you can find it here:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/da452fdbc0247095fdf2b1f52eb4ddd368fad640
>

Thanks!

>> > +
>> > +       case LROTATE_EXPR:
>> > +       default:
>> > +               op = BIT_XOR_EXPR;
>> > +               break;
>> > +       }
>> > +       if (rhs)
>> > +               *rhs = build_int_cstu(unsigned_intDI_type_node, random_const);
>> > +       return op;
>> > +}
>> > +
>> > +static void perturb_local_entropy(basic_block bb, tree local_entropy)
>>
>> What effect does this function have on the resulting code output?
>
> Would you like to see more on top of this comment:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13

I think that comment should be fine.

>> > +static void perturb_latent_entropy(basic_block bb, tree rhs)
>>
>> Same for this. I assume this is effectively:
>>
>>    u64 temp_latent_entropy;
>>
>>    temp_latent_entropy = latent_entropy;
>>    temp_latent_entropy = temp_latent_entropy OP rhs
>>    latent_entropy = temp_latent_entropy;
>>
>> Where does rhs come from? (Is this the "local_entropy" below?)
>
> Sure, I'll rename the rhs parameter to local_entropy.

Thanks.

>
>> > +static void mix_in_sp(basic_block bb, tree local_entropy)
>>
>> What is the stack pointer mixed into?
>
> I already wrote some comments:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/d2781819d774b4370c248cdb8d0dd2b47308b6f4

Awesome.

>> This below needs a bit more detail in comments. IIUC, it's creating a
>> local (to the .o file? the basic block?) variable, initializing it
>> with the stack, perturbing it with random operations, then updating
>> the latent_entropy with it?
>> > +       /* create local entropy variable */
>> > +       local_entropy = create_a_tmp_var(unsigned_intDI_type_node, "local_entropy");
>>
>> What value does local_entropy have initially?
>
> You can see it here:
> https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13

Got it, thanks. These changes look great. If you can send an updated
series for latent_entropy, I'll add it to -next.

-Kees
Emese Revfy June 14, 2016, 10:31 p.m. UTC | #15
On Tue, 14 Jun 2016 11:27:00 -0700
Kees Cook <keescook@chromium.org> wrote:

> On Mon, Jun 13, 2016 at 2:49 PM, Emese Revfy <re.emese@gmail.com> wrote:
> > On Thu, 9 Jun 2016 14:51:45 -0700
> > Kees Cook <keescook@chromium.org> wrote:
>
> >> > + * gcc plugin to help generate a little bit of entropy from program state,
> >> > + * used throughout the uptime of the kernel
> >>
> >> I think this comment needs a lot of expanding. What are all the ways
> >> that this plugin makes changes to code? Things I think I see are:
> >> pre-filling data variables with randomness, creating a local_entropy
> >> variable (local to what?), mixing stack pointer (into what?), updating
> >> latent_entropy global.
> >
> > I demonstrated the details here:
> > https://github.com/ephox-gcc-plugins/latent_entropy/commit/049acd9f478d47ee6526d8e93ab8cfcc3ff91b13
> 
> That helps, thanks. Can you also mention how __latent_entropy changes
> non-functions? (i.e. initializes them with random data.)
> 
> Also, I think this isn't accurate:
> 
>  * local_entropy ^= get_random_long();
> 
> Looking at the disassembly, it seems that static random values (i.e.
> randomly chosen at gcc runtime) are added, rather than making calls to
> the kernel's get_random_long() function.

The plugin doesn't insert calls to the kernel's get_random_long().
That was just an example (the plugin instrumentation would look like this in the kernel).
I rewrote these calls to a random constant.

Patch
diff mbox

diff --git a/arch/Kconfig b/arch/Kconfig
index 5feadad..7115867 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -393,6 +393,24 @@  config GCC_PLUGIN_SANCOV
 	  gcc-4.5 on). It is based on the commit "Add fuzzing coverage support"
 	  by Dmitry Vyukov <dvyukov@google.com>.
 
+config GCC_PLUGIN_LATENT_ENTROPY
+	bool "Generate some entropy during boot and runtime"
+	depends on GCC_PLUGINS
+	help
+	  By saying Y here the kernel will instrument some kernel code to
+	  extract some entropy from both original and artificially created
+	  program state.  This will help especially embedded systems where
+	  there is little 'natural' source of entropy normally.  The cost
+	  is some slowdown of the boot process (about 0.5%) and fork and
+	  irq processing.
+
+	  Note that entropy extracted this way is not known to be cryptographically
+	  secure!
+
+	  This plugin was ported from grsecurity/PaX. More information at:
+	   * https://grsecurity.net/
+	   * https://pax.grsecurity.net/
+
 config HAVE_CC_STACKPROTECTOR
 	bool
 	help
diff --git a/arch/powerpc/kernel/Makefile b/arch/powerpc/kernel/Makefile
index 2da380f..6c7e448 100644
--- a/arch/powerpc/kernel/Makefile
+++ b/arch/powerpc/kernel/Makefile
@@ -16,10 +16,10 @@  endif
 
 ifdef CONFIG_FUNCTION_TRACER
 # Do not trace early boot code
-CFLAGS_REMOVE_cputable.o = -mno-sched-epilog $(CC_FLAGS_FTRACE)
-CFLAGS_REMOVE_prom_init.o = -mno-sched-epilog $(CC_FLAGS_FTRACE)
-CFLAGS_REMOVE_btext.o = -mno-sched-epilog $(CC_FLAGS_FTRACE)
-CFLAGS_REMOVE_prom.o = -mno-sched-epilog $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_cputable.o = -mno-sched-epilog $(CC_FLAGS_FTRACE) $(DISABLE_LATENT_ENTROPY_PLUGIN)
+CFLAGS_REMOVE_prom_init.o = -mno-sched-epilog $(CC_FLAGS_FTRACE) $(DISABLE_LATENT_ENTROPY_PLUGIN)
+CFLAGS_REMOVE_btext.o = -mno-sched-epilog $(CC_FLAGS_FTRACE) $(DISABLE_LATENT_ENTROPY_PLUGIN)
+CFLAGS_REMOVE_prom.o = -mno-sched-epilog $(CC_FLAGS_FTRACE) $(DISABLE_LATENT_ENTROPY_PLUGIN)
 # do not trace tracer code
 CFLAGS_REMOVE_ftrace.o = -mno-sched-epilog $(CC_FLAGS_FTRACE)
 # timers used by tracing
diff --git a/include/linux/random.h b/include/linux/random.h
index e47e533..4ea73ee 100644
--- a/include/linux/random.h
+++ b/include/linux/random.h
@@ -18,6 +18,16 @@  struct random_ready_callback {
 };
 
 extern void add_device_randomness(const void *, unsigned int);
+
+#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
+static inline void add_latent_entropy(void)
+{
+	add_device_randomness((const void *)&latent_entropy, sizeof(latent_entropy));
+}
+#else
+static inline void add_latent_entropy(void) {}
+#endif
+
 extern void add_input_randomness(unsigned int type, unsigned int code,
 				 unsigned int value);
 extern void add_interrupt_randomness(int irq, int irq_flags);
diff --git a/init/main.c b/init/main.c
index 4c17fda..07e4174 100644
--- a/init/main.c
+++ b/init/main.c
@@ -781,6 +781,7 @@  int __init_or_module do_one_initcall(initcall_t fn)
 	}
 	WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
 
+	add_latent_entropy();
 	return ret;
 }
 
diff --git a/kernel/fork.c b/kernel/fork.c
index cdf520f..d07d5a6 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1766,6 +1766,7 @@  long _do_fork(unsigned long clone_flags,
 
 	p = copy_process(clone_flags, stack_start, stack_size,
 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
+	add_latent_entropy();
 	/*
 	 * Do this prior waking up the new thread - the thread pointer
 	 * might get invalid after that point, if the thread exits quickly.
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index f8f3bfc..d10324e 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1234,6 +1234,11 @@  static void __free_pages_ok(struct page *page, unsigned int order)
 	local_irq_restore(flags);
 }
 
+#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
+volatile u64 latent_entropy;
+EXPORT_SYMBOL(latent_entropy);
+#endif
+
 static void __init __free_pages_boot_core(struct page *page, unsigned int order)
 {
 	unsigned int nr_pages = 1 << order;
diff --git a/scripts/Makefile.gcc-plugins b/scripts/Makefile.gcc-plugins
index 5e22b60..cd7902c 100644
--- a/scripts/Makefile.gcc-plugins
+++ b/scripts/Makefile.gcc-plugins
@@ -6,6 +6,12 @@  ifdef CONFIG_GCC_PLUGINS
 
   gcc-plugin-$(CONFIG_GCC_PLUGIN_CYC_COMPLEXITY)	+= cyc_complexity_plugin.so
 
+  gcc-plugin-$(CONFIG_GCC_PLUGIN_LATENT_ENTROPY)	+= latent_entropy_plugin.so
+  gcc-plugin-cflags-$(CONFIG_GCC_PLUGIN_LATENT_ENTROPY)	+= -DLATENT_ENTROPY_PLUGIN
+  ifdef CONFIG_PAX_LATENT_ENTROPY
+    DISABLE_LATENT_ENTROPY_PLUGIN			+= -fplugin-arg-latent_entropy_plugin-disable
+  endif
+
   ifdef CONFIG_GCC_PLUGIN_SANCOV
     ifeq ($(CFLAGS_KCOV),)
       # It is needed because of the gcc-plugin.sh and gcc version checks.
@@ -19,9 +25,9 @@  ifdef CONFIG_GCC_PLUGINS
     endif
   endif
 
-  GCC_PLUGINS_CFLAGS := $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y))
+  GCC_PLUGINS_CFLAGS := $(strip $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y)) $(gcc-plugin-cflags-y))
 
-  export PLUGINCC GCC_PLUGINS_CFLAGS GCC_PLUGIN SANCOV_PLUGIN
+  export PLUGINCC GCC_PLUGINS_CFLAGS GCC_PLUGIN SANCOV_PLUGIN DISABLE_LATENT_ENTROPY_PLUGIN
 
   ifeq ($(PLUGINCC),)
     ifneq ($(GCC_PLUGINS_CFLAGS),)
diff --git a/scripts/gcc-plugins/Makefile b/scripts/gcc-plugins/Makefile
index 88c8ec4..a3f8ca4a 100644
--- a/scripts/gcc-plugins/Makefile
+++ b/scripts/gcc-plugins/Makefile
@@ -23,5 +23,6 @@  always := $($(HOSTLIBS)-y)
 
 cyc_complexity_plugin-objs := cyc_complexity_plugin.o
 sancov_plugin-objs := sancov_plugin.o
+latent_entropy_plugin-objs := latent_entropy_plugin.o
 
 clean-files += *.so
diff --git a/scripts/gcc-plugins/latent_entropy_plugin.c b/scripts/gcc-plugins/latent_entropy_plugin.c
new file mode 100644
index 0000000..f606caa
--- /dev/null
+++ b/scripts/gcc-plugins/latent_entropy_plugin.c
@@ -0,0 +1,454 @@ 
+/*
+ * Copyright 2012-2016 by the PaX Team <pageexec@freemail.hu>
+ * Copyright 2016 by Emese Revfy <re.emese@gmail.com>
+ * Licensed under the GPL v2
+ *
+ * Note: the choice of the license means that the compilation process is
+ *       NOT 'eligible' as defined by gcc's library exception to the GPL v3,
+ *       but for the kernel it doesn't matter since it doesn't link against
+ *       any of the gcc libraries
+ *
+ * gcc plugin to help generate a little bit of entropy from program state,
+ * used throughout the uptime of the kernel
+ *
+ * TODO:
+ * - add ipa pass to identify not explicitly marked candidate functions
+ * - mix in more program state (function arguments/return values, loop variables, etc)
+ * - more instrumentation control via attribute parameters
+ *
+ * BUGS:
+ * - none known
+ *
+ * Options:
+ * -fplugin-arg-latent_entropy_plugin-disable
+ *
+ * Attribute: __attribute__((latent_entropy))
+ *  The latent_entropy gcc attribute can be only on functions and variables.
+ *  If it is on a function then the plugin will instrument it. If the attribute
+ *  is on a variable then the plugin will initialize it with a random value.
+ *  The variable must be an integer, an integer array type or a structure with integer fields.
+ */
+
+#include "gcc-common.h"
+
+int plugin_is_GPL_compatible;
+
+static GTY(()) tree latent_entropy_decl;
+
+static struct plugin_info latent_entropy_plugin_info = {
+	.version	= "201605292100",
+	.help		= "disable\tturn off latent entropy instrumentation\n",
+};
+
+static unsigned HOST_WIDE_INT seed;
+static unsigned HOST_WIDE_INT get_random_const(void)
+{
+	unsigned int i;
+	unsigned HOST_WIDE_INT ret = 0;
+
+	for (i = 0; i < 8 * sizeof ret; i++) {
+		ret = (ret << 1) | (seed & 1);
+		seed >>= 1;
+		if (ret & 1)
+			seed ^= 0xD800000000000000ULL;
+	}
+
+	return ret;
+}
+
+static tree handle_latent_entropy_attribute(tree *node, tree name, tree args __unused, int flags __unused, bool *no_add_attrs)
+{
+	tree type;
+	unsigned long long mask;
+#if BUILDING_GCC_VERSION <= 4007
+	VEC(constructor_elt, gc) *vals;
+#else
+	vec<constructor_elt, va_gc> *vals;
+#endif
+
+	switch (TREE_CODE(*node)) {
+	default:
+		*no_add_attrs = true;
+		error("%qE attribute only applies to functions and variables", name);
+		break;
+
+	case VAR_DECL:
+		if (DECL_INITIAL(*node)) {
+			*no_add_attrs = true;
+			error("variable %qD with %qE attribute must not be initialized", *node, name);
+			break;
+		}
+
+		if (!TREE_STATIC(*node)) {
+			*no_add_attrs = true;
+			error("variable %qD with %qE attribute must not be local", *node, name);
+			break;
+		}
+
+		type = TREE_TYPE(*node);
+		switch (TREE_CODE(type)) {
+		default:
+			*no_add_attrs = true;
+			error("variable %qD with %qE attribute must be an integer or a fixed length integer array type"
+				"or a fixed sized structure with integer fields", *node, name);
+			break;
+
+		case RECORD_TYPE: {
+			tree field;
+			unsigned int nelt = 0;
+
+			for (field = TYPE_FIELDS(type); field; nelt++, field = TREE_CHAIN(field)) {
+				tree fieldtype;
+
+				fieldtype = TREE_TYPE(field);
+				if (TREE_CODE(fieldtype) == INTEGER_TYPE)
+					continue;
+
+				*no_add_attrs = true;
+				error("structure variable %qD with %qE attribute has a non-integer field %qE", *node, name, field);
+				break;
+			}
+
+			if (field)
+				break;
+
+#if BUILDING_GCC_VERSION <= 4007
+			vals = VEC_alloc(constructor_elt, gc, nelt);
+#else
+			vec_alloc(vals, nelt);
+#endif
+
+			for (field = TYPE_FIELDS(type); field; field = TREE_CHAIN(field)) {
+				tree fieldtype;
+
+				fieldtype = TREE_TYPE(field);
+				mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(fieldtype)) - 1);
+				mask = 2 * (mask - 1) + 1;
+
+				if (TYPE_UNSIGNED(fieldtype))
+					CONSTRUCTOR_APPEND_ELT(vals, field, build_int_cstu(fieldtype, mask & get_random_const()));
+				else
+					CONSTRUCTOR_APPEND_ELT(vals, field, build_int_cst(fieldtype, mask & get_random_const()));
+			}
+
+			DECL_INITIAL(*node) = build_constructor(type, vals);
+			break;
+		}
+
+		case INTEGER_TYPE:
+			mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(type)) - 1);
+			mask = 2 * (mask - 1) + 1;
+
+			if (TYPE_UNSIGNED(type))
+				DECL_INITIAL(*node) = build_int_cstu(type, mask & get_random_const());
+			else
+				DECL_INITIAL(*node) = build_int_cst(type, mask & get_random_const());
+			break;
+
+		case ARRAY_TYPE: {
+			tree elt_type, array_size, elt_size;
+			unsigned int i, nelt;
+
+			elt_type = TREE_TYPE(type);
+			elt_size = TYPE_SIZE_UNIT(TREE_TYPE(type));
+			array_size = TYPE_SIZE_UNIT(type);
+
+			if (TREE_CODE(elt_type) != INTEGER_TYPE || !array_size || TREE_CODE(array_size) != INTEGER_CST) {
+				*no_add_attrs = true;
+				error("array variable %qD with %qE attribute must be a fixed length integer array type", *node, name);
+				break;
+			}
+
+			nelt = TREE_INT_CST_LOW(array_size) / TREE_INT_CST_LOW(elt_size);
+#if BUILDING_GCC_VERSION <= 4007
+			vals = VEC_alloc(constructor_elt, gc, nelt);
+#else
+			vec_alloc(vals, nelt);
+#endif
+
+			mask = 1ULL << (TREE_INT_CST_LOW(TYPE_SIZE(elt_type)) - 1);
+			mask = 2 * (mask - 1) + 1;
+
+			for (i = 0; i < nelt; i++)
+				if (TYPE_UNSIGNED(elt_type))
+					CONSTRUCTOR_APPEND_ELT(vals, size_int(i), build_int_cstu(elt_type, mask & get_random_const()));
+				else
+					CONSTRUCTOR_APPEND_ELT(vals, size_int(i), build_int_cst(elt_type, mask & get_random_const()));
+
+			DECL_INITIAL(*node) = build_constructor(type, vals);
+			break;
+		}
+		}
+		break;
+
+	case FUNCTION_DECL:
+		break;
+	}
+
+	return NULL_TREE;
+}
+
+static struct attribute_spec latent_entropy_attr = {
+	.name				= "latent_entropy",
+	.min_length			= 0,
+	.max_length			= 0,
+	.decl_required			= true,
+	.type_required			= false,
+	.function_type_required		= false,
+	.handler			= handle_latent_entropy_attribute,
+#if BUILDING_GCC_VERSION >= 4007
+	.affects_type_identity		= false
+#endif
+};
+
+static void register_attributes(void *event_data __unused, void *data __unused)
+{
+	register_attribute(&latent_entropy_attr);
+}
+
+static bool latent_entropy_gate(void)
+{
+	/* don't bother with noreturn functions for now */
+	if (TREE_THIS_VOLATILE(current_function_decl))
+		return false;
+
+	/* gcc-4.5 doesn't discover some trivial noreturn functions */
+	if (EDGE_COUNT(EXIT_BLOCK_PTR_FOR_FN(cfun)->preds) == 0)
+		return false;
+
+	return lookup_attribute("latent_entropy", DECL_ATTRIBUTES(current_function_decl)) != NULL_TREE;
+}
+
+static tree create_a_tmp_var(tree type, const char *name)
+{
+	tree var;
+
+	var = create_tmp_var(type, name);
+	add_referenced_var(var);
+	mark_sym_for_renaming(var);
+	return var;
+}
+
+static enum tree_code get_op(tree *rhs)
+{
+	static enum tree_code op;
+	unsigned HOST_WIDE_INT random_const;
+
+	random_const = get_random_const();
+
+	switch (op) {
+	case BIT_XOR_EXPR:
+		op = PLUS_EXPR;
+		break;
+
+	case PLUS_EXPR:
+		if (rhs) {
+			op = LROTATE_EXPR;
+			random_const &= HOST_BITS_PER_WIDE_INT - 1;
+			break;
+		}
+
+	case LROTATE_EXPR:
+	default:
+		op = BIT_XOR_EXPR;
+		break;
+	}
+	if (rhs)
+		*rhs = build_int_cstu(unsigned_intDI_type_node, random_const);
+	return op;
+}
+
+static void perturb_local_entropy(basic_block bb, tree local_entropy)
+{
+	gimple_stmt_iterator gsi;
+	gimple assign;
+	tree rhs;
+	enum tree_code subcode;
+
+	subcode = get_op(&rhs);
+	assign = gimple_build_assign_with_ops(subcode, local_entropy, local_entropy, rhs);
+	gsi = gsi_after_labels(bb);
+	gsi_insert_before(&gsi, assign, GSI_NEW_STMT);
+	update_stmt(assign);
+}
+
+static void perturb_latent_entropy(basic_block bb, tree rhs)
+{
+	gimple_stmt_iterator gsi;
+	gimple assign;
+	tree temp;
+	enum tree_code subcode;
+
+	/* create temporary copy of latent_entropy */
+	temp = create_a_tmp_var(unsigned_intDI_type_node, "temp_latent_entropy");
+
+	gsi = gsi_last_bb(bb);
+
+	/* 1. read... */
+	add_referenced_var(latent_entropy_decl);
+	mark_sym_for_renaming(latent_entropy_decl);
+	assign = gimple_build_assign(temp, latent_entropy_decl);
+	gsi_insert_before(&gsi, assign, GSI_NEW_STMT);
+	update_stmt(assign);
+
+	/* 2. ...modify... */
+	subcode = get_op(NULL);
+	assign = gimple_build_assign_with_ops(subcode, temp, temp, rhs);
+	gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
+	update_stmt(assign);
+
+	/* 3. ...write latent_entropy */
+	assign = gimple_build_assign(latent_entropy_decl, temp);
+	gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
+	update_stmt(assign);
+}
+
+static void mix_in_sp(basic_block bb, tree local_entropy)
+{
+	gimple assign, call;
+	tree frame_addr, rand_const;
+	gimple_stmt_iterator gsi = gsi_after_labels(bb);
+
+	frame_addr = create_a_tmp_var(ptr_type_node, "local_entropy_frame_addr");
+
+	call = gimple_build_call(builtin_decl_implicit(BUILT_IN_FRAME_ADDRESS), 1, integer_zero_node);
+	gimple_call_set_lhs(call, frame_addr);
+	gsi_insert_before(&gsi, call, GSI_NEW_STMT);
+	update_stmt(call);
+
+	assign = gimple_build_assign(local_entropy, fold_convert(unsigned_intDI_type_node, frame_addr));
+	gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
+	update_stmt(assign);
+
+	rand_const = build_int_cstu(unsigned_intDI_type_node, get_random_const());
+	assign = gimple_build_assign_with_ops(BIT_XOR_EXPR, local_entropy, local_entropy, rand_const);
+	gsi_insert_after(&gsi, assign, GSI_NEW_STMT);
+	update_stmt(assign);
+}
+
+static unsigned int latent_entropy_execute(void)
+{
+	basic_block bb;
+	tree local_entropy;
+
+	if (!latent_entropy_decl) {
+		varpool_node_ptr node;
+
+		FOR_EACH_VARIABLE(node) {
+			tree var = NODE_DECL(node);
+
+			if (DECL_NAME_LENGTH(var) < sizeof("latent_entropy") - 1)
+				continue;
+			if (strcmp(IDENTIFIER_POINTER(DECL_NAME(var)), "latent_entropy"))
+				continue;
+			latent_entropy_decl = var;
+			break;
+		}
+		if (!latent_entropy_decl)
+			return 0;
+	}
+
+	gcc_assert(single_succ_p(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
+	bb = single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun));
+	if (!single_pred_p(bb)) {
+		split_edge(single_succ_edge(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
+		gcc_assert(single_succ_p(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
+		bb = single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun));
+	}
+
+	/* create local entropy variable */
+	local_entropy = create_a_tmp_var(unsigned_intDI_type_node, "local_entropy");
+
+	/* 1. stack pointer */
+	mix_in_sp(bb, local_entropy);
+
+	bb = bb->next_bb;
+	/* 2. instrument each BB with an operation on the local entropy variable */
+	while (bb != EXIT_BLOCK_PTR_FOR_FN(cfun)) {
+		perturb_local_entropy(bb, local_entropy);
+		bb = bb->next_bb;
+	};
+
+	/* 3. mix local entropy into the global entropy variable */
+	gcc_assert(single_pred_p(EXIT_BLOCK_PTR_FOR_FN(cfun)));
+	perturb_latent_entropy(single_pred(EXIT_BLOCK_PTR_FOR_FN(cfun)), local_entropy);
+	return 0;
+}
+
+static void latent_entropy_start_unit(void *gcc_data __unused, void *user_data __unused)
+{
+	tree latent_entropy_type;
+
+	seed = get_random_seed(false);
+
+	if (in_lto_p)
+		return;
+
+	/* extern volatile u64 latent_entropy */
+	gcc_assert(TYPE_PRECISION(long_long_unsigned_type_node) == 64);
+	latent_entropy_type = build_qualified_type(long_long_unsigned_type_node, TYPE_QUALS(long_long_unsigned_type_node) | TYPE_QUAL_VOLATILE);
+	latent_entropy_decl = build_decl(UNKNOWN_LOCATION, VAR_DECL, get_identifier("latent_entropy"), latent_entropy_type);
+
+	TREE_STATIC(latent_entropy_decl) = 1;
+	TREE_PUBLIC(latent_entropy_decl) = 1;
+	TREE_USED(latent_entropy_decl) = 1;
+	DECL_PRESERVE_P(latent_entropy_decl) = 1;
+	TREE_THIS_VOLATILE(latent_entropy_decl) = 1;
+	DECL_EXTERNAL(latent_entropy_decl) = 1;
+	DECL_ARTIFICIAL(latent_entropy_decl) = 1;
+	lang_hooks.decls.pushdecl(latent_entropy_decl);
+}
+
+#define PASS_NAME latent_entropy
+#define PROPERTIES_REQUIRED PROP_gimple_leh | PROP_cfg
+#define TODO_FLAGS_FINISH TODO_verify_ssa | TODO_verify_stmts | TODO_dump_func | TODO_update_ssa
+#include "gcc-generate-gimple-pass.h"
+
+int plugin_init(struct plugin_name_args *plugin_info, struct plugin_gcc_version *version)
+{
+	bool enabled = true;
+	const char * const plugin_name = plugin_info->base_name;
+	const int argc = plugin_info->argc;
+	const struct plugin_argument * const argv = plugin_info->argv;
+	int i;
+
+	struct register_pass_info latent_entropy_pass_info;
+
+	latent_entropy_pass_info.pass				= make_latent_entropy_pass();
+	latent_entropy_pass_info.reference_pass_name		= "optimized";
+	latent_entropy_pass_info.ref_pass_instance_number	= 1;
+	latent_entropy_pass_info.pos_op				= PASS_POS_INSERT_BEFORE;
+	static const struct ggc_root_tab gt_ggc_r_gt_latent_entropy[] = {
+		{
+			.base = &latent_entropy_decl,
+			.nelt = 1,
+			.stride = sizeof(latent_entropy_decl),
+			.cb = &gt_ggc_mx_tree_node,
+			.pchw = &gt_pch_nx_tree_node
+		},
+		LAST_GGC_ROOT_TAB
+	};
+
+	if (!plugin_default_version_check(version, &gcc_version)) {
+		error(G_("incompatible gcc/plugin versions"));
+		return 1;
+	}
+
+	for (i = 0; i < argc; ++i) {
+		if (!(strcmp(argv[i].key, "disable"))) {
+			enabled = false;
+			continue;
+		}
+		error(G_("unkown option '-fplugin-arg-%s-%s'"), plugin_name, argv[i].key);
+	}
+
+	register_callback(plugin_name, PLUGIN_INFO, NULL, &latent_entropy_plugin_info);
+	if (enabled) {
+		register_callback(plugin_name, PLUGIN_START_UNIT, &latent_entropy_start_unit, NULL);
+		register_callback(plugin_name, PLUGIN_REGISTER_GGC_ROOTS, NULL, (void *)&gt_ggc_r_gt_latent_entropy);
+		register_callback(plugin_name, PLUGIN_PASS_MANAGER_SETUP, NULL, &latent_entropy_pass_info);
+	}
+	register_callback(plugin_name, PLUGIN_ATTRIBUTES, register_attributes, NULL);
+
+	return 0;
+}