How to Leak a Key

It’s been been interesting following the great Heartbleed crisis over the last week. The “Heartbleed Challenge” set up by Cloudflare established that you could get the server private key, specifically, the primes used to generate the key would show up in the Heartbleed data occasionally. Doing a memory dump of a simple OpenSSL server application ( showed that after reading the keys at startup, it later copies the prime data to higher locations in the heap where they become more likely to be included in a Heartbeat response.

An extract from a hex dump of the heap after making a single request shows:

000273f0  00 00 00 00 01 00 00 00  51 00 00 00 00 00 00 00  # Prime1
00027400  85 52 42 a0 71 54 76 bd  98 49 4f 25 fc d3 e1 89
00027460  00 00 00 00 01 00 00 00  51 00 00 00 00 00 00 00  # Prime2
00027470  3b c8 8b 7d 52 74 db e0  af 9e 13 d7 a2 0a fc ea
0002d020  60 55 00 00 00 00 00 00  50 45 00 00 00 00 00 00  # Input buffer
0002d030  16 03 01 16 03 03 00 28  e8 d6 b3 68 87 08 d4 7f
00031570  00 00 00 00 00 00 00 00  c1 44 00 00 00 00 00 00  # Output buffer
00031580  00 00 00 16 03 03 00 28  82 c9 a8 9a 87 b5 cf e7
00038990  90 00 00 00 00 00 00 00  50 00 00 00 00 00 00 00  # Prime 2 again
000389a0  3b c8 8b 7d 52 74 db e0  af 9e 13 d7 a2 0a fc ea
00038b50  30 00 00 00 00 00 00 00  50 00 00 00 00 00 00 00  # Prime 1 again
00038b60  85 52 42 a0 71 54 76 bd  98 49 4f 25 fc d3 e1 89  

The first line in each section is the malloc header, the second line is part of the data; also, all of these malloced blocks are in use (otherwise some of the data would be overwritten with free list data).

The first two primes are from the initial processing of the private key, the next two items are the input and output buffers for the connection, then we have two more copies of the primes, which only appear after a connection has been made.

So where are these persistent copies of the primes coming from? Some investigation shows that the first time OpenSSL actually uses an RSA key for a private operation, it calculates various auxiliary values for use in Montgomery arithmetic and stores them away in the key structure. Unfortunately, one of the values that gets stored away is the modulus used, which in this case is one of the primes:


static int RSA_eay_mod_exp(BIGNUM *r0,
                           const BIGNUM *I,
                           RSA *rsa,
                           BN_CTX *ctx)
  if (rsa->flags & RSA_FLAG_CACHE_PRIVATE)
    if (!BN_MONT_CTX_set_locked(&rsa->_method_mod_p,
                                p, ctx))
      goto err;
    if (!BN_MONT_CTX_set_locked(&rsa->_method_mod_q,
                                q, ctx))
      goto err;


BN_MONT_CTX *BN_MONT_CTX_set_locked(BN_MONT_CTX **pmont,
                                    int lock,
                                    const BIGNUM *mod,
                                    BN_CTX *ctx)
  if (!*pmont) {
    ret = BN_MONT_CTX_new();
    if (ret && !BN_MONT_CTX_set(ret, mod, ctx))
int BN_MONT_CTX_set(BN_MONT_CTX *mont,
                    const BIGNUM *mod,
                    BN_CTX *ctx)
  int ret = 0;
  BIGNUM *Ri,*R;
  if((Ri = BN_CTX_get(ctx)) == NULL) goto err;
  R= &(mont->RR); /* grab RR as a temp */
  if (!BN_copy(&(mont->N),mod)) goto err; /* Set N */

That BN_copy in the last line is the culprit. Once allocated, these values stay in memory until the RSA key is deleted. So even if the original key data is stored in protected or hard to get at memory, a Heartbleed attack may still be able to get to the primes.

This also explains why, as far as I know, no one has managed to capture the CRT parameters or the private exponent itself – they stay safely tucked away in the low part of the heap. Also, depending on the application startup behaviour, it might be that the primes end up below the input buffers (by default, OpenSSL caches up to 32 previously allocated buffers of each type) in which case they won’t be visible to Heartbleed – this might explain why in the Cloudflare challenge, only one of the primes seemed to be turning up.

Update 20/04/14: actually, it’s worse than this – bignum resizing can also leak private data onto the heap in unpredictable ways, more details to follow.


Loopless Ruler

This function (based on Algorithm L for loopless Gray code generation in TAOCP

void ruler(int n)
  int f[n+1];
  for (int j = 0; j <= n; j++) {
    f[j] = j;
  while(true) {
    int j = f[0];
    f[0] = 0;
    if (j == n) break;
    f[j] = f[j+1];
    f[j+1] = j+1;

enables us to compute the “ruler function”, ρ for 1, 2, 3…:

0 1 0 2 0 1 0 3 0 1 0 2 0 1 0...

It’s a “loopless” algorithm and uses f, an array of “focus pointers” to keep track of what’s going on.

The ruler function ρ(n) is essentially the number of trailing 0’s in the binary representation of n.

This is what’s printed by ruler(4):

0 1 2 3 4 
1 1 2 3 4 
0 2 2 3 4 
2 1 2 3 4 
0 1 3 3 4 
1 1 3 3 4 
0 3 2 3 4 
3 1 2 3 4 
0 1 2 4 4 
1 1 2 4 4 
0 2 2 4 4 
2 1 2 4 4 
0 1 4 3 4 
1 1 4 3 4 
0 4 2 3 4 
4 1 2 3 4 

The left hand side is the ruler function of course, what are the other values?

It’s clearer if we print f in reverse and match up with binary representations:


If n[j] is the jth binary digit of n, then f[j] is just j, unless n[j] is the first of a block of 1’s, in which case f[j] is the index of the first 0 after that block.

For example, 01101, has block starting at 0 and ending at 1, so f[0] = 1, and another block starting at 2, ending at 4, so f[2] = 4; for other j, f[j] = j, so the focus pointers are 43411.

If n is of the form:

j   k

with f[0] = k, f[k] = k, f[k+1] = j

0n incrementing we get:

j   k

Now f[0] = 0, f[k] = j and f[k+1] = k+1, other values of f are the same

If k is 0, then we go from:

j   k

with f[0] = 0, f[1] = j


j   k

with f[0] = 0, f[1] = 1

Either way, we can see that the inner loop of ruler is correct.

Ringing the Changes

Apart from a brief visit many years ago to Macclesfield church with my friend Jo, I have never taken part in the ancient English art of bell-ringing, but permutation generation has always seemed a fascinating topic.

In TAOCP, Knuth has Algorithm P (“plain changes”) for generating all permutations using adjacent interchanges:

void taocp1(int *a, int n)
  int c[n+1];
  int o[n+1];
  for (int j = 1; j <= n; j++) {
    c[j] = 0; o[j] = 1;
  int j = n; int s = 0;
  int q = c[j] + o[j];
  if (q < 0) goto P7;
  if (q == j) goto P6;
  c[j] = q;
  goto P2;
  if (j == 1) return;
  s = s+1;
  o[j] = -o[j];
  j = j-1;
  goto P4;

It works just fine, but I find it a little hard to follow and let’s face it, we just don’t write code like that anymore, with all those gotos, 1-based arrays, etc.

0-basing the arrays isn’t too much of a problem, for eliminating the gotos we could take the functional approach:

void taocp2(int *a, int n)
  int c[n];
  int o[n];
  for (int j = 0; j < n; j++) {
    c[j] = 0; o[j] = 1;

void P2(int *a, int *c, int *o, int n) {

void P3(int *a, int *c, int *o, int n) {

void P4(int *a, int *c, int *o, int n, int j, int s) {
  int q = c[j] + o[j];
  if (q < 0) P7(a,c,o,n,j,s);
  else if (q == j+1) P6(a,c,o,n,j,s);
  else P5(a,c,o,n,j,s,q);

void P5(int *a, int *c, int *o, int n, int j, int s, int q) {
  c[j] = q;

void P6(int *a, int *c, int *o, int n, int j, int s) {
  if (j != 1) {

void P7(int *a, int *c, int *o, int n, int j, int s) {
  o[j] = -o[j];

but somehow that seems missing the point (I don’t think that the worst thing about that code is that it still uses modifiable state).

There are actually two component parts to the algorithm: generating a “mixed radix reflected Gray code” which defines the progressive inversions of the array elements, and going from the inversion changes to the actual indexes of the swapped objects.

For generating the Gray codes, with a bit of rewriting, 0-basing our arrays etc. we get:

void gray(int n)
  int j, c[n], o[n];
  for (int j = 0; j < n; j++) {
    c[j] = 0; o[j] = 1;
  do {
    for (j = n-1; j > 0; j--) {
      int q = c[j] + o[j];
      if (q < 0) {
        o[j] = -o[j];
      } else if (q == j+1) {
        o[j] = -o[j];
      } else {
        c[j] = q;
  } while (j != 0);

Each time around, find the highest element whose inversion count can be changed in the desired direction. For elements whose inversion count can’t be changed, change direction. If no element can be changed we are done.

The next step is to calculate the position of element j – but this is just j less the number of elements less than j that appear after j (ie. the number of inversions, c[j]), plus the number of elements greater than j that appear before j – but this is just the number of elements we have been passed over in the “if (q == j+1)” step above, so we can now add in the rest of algorithm P:

void taocp3(int *a, int n)
  int c[n];
  int o[n];
  for (int j = 0; j < n; j++) {
    c[j] = 0; o[j] = 1;
  int j;
  do {
    int s = 0;
    for (j = n-1; j > 0; j--) {
      int q = c[j] + o[j];
      if (q < 0) {
        o[j] = -o[j];
      } else if (q == j+1) {
        o[j] = -o[j];
        s++;          // This element will be before element j
      } else {
        c[j] = q;
  } while (j != 0);

Knuth’s algorithm essentially Dijkstra’s:

which explains it very lucidly: there is a 1-1 correspondence between inversion counts & permutations – and a Gray enumeration of inversion gives us a sequence of permutations where only 1 element at a time changes its inversion count, and only by 1 or -1, which can only be if we exchange adjacent elements.

Knuth also gives a “loopless” algorithm for generating reflected Gray sequences, and we could use a table of element positions to construct the permutations from this:

void gray2(int *m, int n)
  int a[n],f[n+1],o[n];
  for (int j = 0; j < n; j++) {
    a[j] = 0; f[j] = j; o[j] = 1;
  f[n] = n;
  while (true) {
    int j = f[0];
    f[0] = 0;
    if (j == n) break;
    a[j] += o[j];
    if (a[j] == 0 || a[j] == m[j]-1) {
      o[j] = -o[j];
      f[j] = f[j+1];
      f[j+1] = j+1;

The “classic” Johnson-Trotter algorithm for plain changes involves consideration of “mobile” elements: each element has a current direction and it is “mobile” if if it greater than the next adjacent element (if there is one) in that direction. The algorithm proceeds by finding the highest mobile element and moving it accordingly:

void jt(int *a, int n)
  int o[n]; // Direction of element i, 1 = left, 0 = right
  int c[n]; // Position of element i
  for (int j = 0; j < n; j++) {
    o[j] = 1; c[j] = j;
  int j; // Will be set to the highest mobile element
  do {
    // Search high to low. 0 is never mobile
    for (j = n-1; j > 0; j--) {
      int p = c[j]; // Position of element j
      if (o[j]) { // Going left
        int k = a[p-1];
        if (p > 0 && k < j) {
          // Swap and adjust positions
          a[p] = k; c[k] = p;
          a[p-1] = j; c[j] = p-1; 
      } else { // Going right
        int k = a[p+1];
        if (p < n-1 && k < j) {
          a[p] = k; c[k] = p;
          a[p+1] = j; c[j] = p+1; 
      o[j] = !o[j]; // Change direction & continue
  } while (j > 0);

Note that we can check elements from high to low directly & stop as soon as a mobile element is found – there is no need to check every element.

Johnson-Trotter and Algorithm P are essentially doing the same thing: when moving the mobile element, we are changing its inversion count in the same way as we do directly in P.

Performance-wise, the two algorithms are very similar though (somewhat surprisingly to me) the Johnson-Trotter version seems a little faster.

Either way, my laptop generates the 3628800 permutations of 10 elements in 0.03 seconds, outputting them (even to /dev/null) takes 4.5 seconds. It takes about 3 seconds to run through the 479001600 permutations of 12 elements (I haven’t tried outputting them). We can streamline the computation further by taking advantage of the fact that most of the time we are just moving the highest element in the same direction, but that can wait for another day.

Finally, here is a function that gives the next permutation in plain changes without maintaining any state:

bool nextperm(int *a, int n)
  int c[n], o[n];
  for (int i = 0; i < n; i++) {
    c[i] = o[i] = 0;
  // Count inversions
  for (int i = 0; i < n; i++) {
    for (int j = i+1; j < n; j++) {
      if (a[j] < a[i]) c[a[i]]++;
  // Find directions
  for (int i = 1, k = 0; i < n-1; i++) {
    k += c[i];
    o[i+1] = k%2;
  // Find mobile element and move it.
  for (int j = n-1, s = 0; j > 0; j--) {
    if (!o[j]) {
      if (c[j] < j) {
	return true;
    } else {
      if (c[j] > 0) {
	return true;
  return false;

void stateless(int *a, int n)
  do {
  } while (nextperm(a,n));

We use the Knuth/Dijkstra trick to avoid building a position table. I suspect that n-squared inversion counting can be improved, but we still come in at under second for all permutations of 10, though that’s about 30 times slower than the optimized version.



I’ve never written much BCPL, but it always seemed like a nice language. Many features ended up in C of course, but one that didn’t was VALOF, allowing the inclusion of a statement block in an expression, and using RESULTIS to return a value. Using slightly invented syntax:

        FOR I = 1 TO J DO $(
                IF I*I = J THEN RESULTIS 1
        RESULTIS 0

GCC defines the block-expression extension, allowing us to write:

int main(int argc, char *argv[])
  int j = atoi(argv[1]);
           bool found = false;
           for (int i = 1; i &lt; j; i++) {
             if (i*i == j) {
               found = true;
               goto end;

but having to use the last expression in the block as the return value can get a bit convoluted.

With C++11, we can do better with immediately applied lambda expressions, with return taking the role of RESULTIS:

int main(int argc, char *argv[])
  int j = atoi(argv[1]);
           for (int i = 1; i &lt; j; i++) {
             if (i*i == j) return true;
           return false;

If we don’t like the smiley syntax, we can even define a VALOF macro:

#define VALOF(e) [=](){e}()
int main(int argc, char *argv[])
  int j = atoi(argv[1]);
             for (int i = 1; i &lt; j; i++) {
               if (i*i == j) return true;
             return false;

though we’ll have to be a little bit careful with commas not enclosed in parentheses (and we might end up needing different macros for different variable capture methods).

Curiously, this idiom seems to be popular in Javascript.


Krazy Kompilation

Let’s take a simple factorial program:

int fact(int n)
  if (n == 0) return 1;
  else return n * fact(n-1);

and look at what gcc compiles it to (using Matthew Godbolt’s excellent GCC Explorer):

With -O1, a nice clean version of the recursive function:

	pushq	%rbx
	movl	%edi, %ebx
	movl	$1, %eax
	testl	%edi, %edi
	je	.L2
	leal	-1(%rdi), %edi
	call	fact(int)
	imull	%ebx, %eax
	popq	%rbx

With -O2, that’s impressive, it’s done some program transformation and now we have a loop:

	testl	%edi, %edi
	movl	$1, %eax
	je	.L4
	imull	%edi, %eax
	subl	$1, %edi
	jne	.L3

With -O3, whaaah!:

	testl	%edi, %edi
	je	.L7
	movl	%edi, %edx
	movl	%edi, %esi
	shrl	$2, %edx
	leal	0(,%rdx,4), %ecx
	testl	%ecx, %ecx
	je	.L8
	cmpl	$6, %edi
	jbe	.L8
	leal	-1(%rdi), %eax
	movl	%edi, -24(%rsp)
	movdqa	.LC1(%rip), %xmm4
	movl	%eax, -20(%rsp)
	leal	-2(%rdi), %eax
	movd	-20(%rsp), %xmm2
	movl	%eax, -16(%rsp)
	leal	-3(%rdi), %eax
	movd	-16(%rsp), %xmm1
	movl	%eax, -12(%rsp)
	xorl	%eax, %eax
	movd	-12(%rsp), %xmm0
	punpckldq	%xmm0, %xmm1
	movd	-24(%rsp), %xmm0
	punpckldq	%xmm2, %xmm0
	punpcklqdq	%xmm1, %xmm0
	movdqa	.LC0(%rip), %xmm1
	movdqa	%xmm1, %xmm3
	psrldq	$4, %xmm1
	addl	$1, %eax
	movdqa	%xmm0, %xmm2
	cmpl	%eax, %edx
	pmuludq	%xmm0, %xmm3
	paddd	%xmm4, %xmm0
	psrldq	$4, %xmm2
	pmuludq	%xmm1, %xmm2
	pshufd	$8, %xmm3, %xmm1
	pshufd	$8, %xmm2, %xmm2
	punpckldq	%xmm2, %xmm1
	ja	.L4
	movdqa	%xmm1, %xmm2
	subl	%ecx, %edi
	cmpl	%ecx, %esi
	psrldq	$8, %xmm2
	movdqa	%xmm2, %xmm0
	psrldq	$4, %xmm2
	pmuludq	%xmm1, %xmm0
	psrldq	$4, %xmm1
	pshufd	$8, %xmm0, %xmm0
	pmuludq	%xmm2, %xmm1
	pshufd	$8, %xmm1, %xmm1
	punpckldq	%xmm1, %xmm0
	movdqa	%xmm0, %xmm1
	psrldq	$4, %xmm1
	movdqa	%xmm1, %xmm2
	psrldq	$4, %xmm1
	pmuludq	%xmm0, %xmm2
	psrldq	$4, %xmm0
	pmuludq	%xmm0, %xmm1
	pshufd	$8, %xmm2, %xmm0
	pshufd	$8, %xmm1, %xmm1
	punpckldq	%xmm1, %xmm0
	movd	%xmm0, -24(%rsp)
	movl	-24(%rsp), %eax
	je	.L13
	imull	%edi, %eax
	subl	$1, %edi
	jne	.L9
	movl	$1, %eax
	movl	$1, %eax
	jmp	.L9

Well, OK, it’s unwound the loop and vectorized with SSE instructions though a more precise analysis will have to wait for a future post. Impressive optimization from GCC though and pretty groovy considering the biggest factorial we can fit in a 32-bit int is fact(15)!



Like, I suspect, many programmers, there are many software tools I have used on a regular basis for many years but remain woefully ignorant of their inner workings and true potential. One of these is the Unix command line.

It’s a common, for example, to want to make the error output of a program to appear as the normal output and trial and error, or Googling or looking at Stack Overflow leads to:

strace ls 2>&1 >/dev/null

which works fine but seems puzzling – we’ve told the program to send error output to normal output, then normal output to /dev/null so why doesn’t that discard everything, similar to:

strace ls >/dev/null 2>&1

This is because we don’t understand what is going on.

An indirection is actually a call to the dup2 system call. From the man page:

dup2() makes newfd be the copy of oldfd, closing newfd first if necessary'

So: n>&m does a dup2(m,n): close fd n if necessary, then make n be a copy of fd m, and n>file means: close n if necessary, open file as fd m, then do dup2(m,n).

Now it all makes sense:

strace ls 2>&1 1>/dev/null 

first of all makes 2 be a copy of 1, then changes 1 to point to /dev/null – the copying is done ‘by value’ as it were (despite the confusing, for C++ programmers anyway, use of ‘&’).

Using strace here is not an accident, but used like this doesn’t tell us much: indirection is handled by the shell, not by the program, so we need to do something like this for further insight:

$ strace -f -etrace=clone,execve,open,dup2 bash -c 'ls >/dev/null 2>&1'
execve("/bin/bash", ["bash", "-c", "ls >/dev/null 2>&1"], [/* 46 vars */]) = 0
clone(Process 20454 attached
child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD, child_tidptr=0x7f643f4c09d0) = 20454
Process 20453 suspended
[pid 20454] open("/dev/null", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 3
[pid 20454] dup2(3, 1)                  = 1
[pid 20454] dup2(1, 2)                  = 2
[pid 20454] execve("/bin/ls", ["ls"], [/* 45 vars */]) = 0
Process 20453 resumed
Process 20454 detached
--- SIGCHLD (Child exited) @ 0 (0) ---

bash forks off a subprocess (a clone syscall these days rather than fork), which then sets up the input and output before calling execve to actually run the command.

We don’t have to limit ourselves to fds 0,1 and 2 that we get to start with:

$ runit () { echo stderr 1>&2; echo stdout; }
$ runit 1>/dev/null
$ runit 2>/dev/null
$ runit 3>&2 2>&1 1>&3
$ (runit 3>&2 2>&1 1>&3) 1> /dev/null
$ (runit 3>&2 2>&1 1>&3) 2> /dev/null

We duplicate 2 to 3, then 1 to 2, then 3 to 1, and we have swapped stdin and stderr.

We can also pass non-standard file descriptors in to programs, though this doesn’t seem to be a technique used much:

#include <unistd.h>
int main()
  char buffer[256];
  ssize_t n;
  while ((n = read(3,buffer,sizeof(buffer))) > 0) {

and do:

$ g++ -Wall cat34.cpp -o cat34
$ echo Hello World | ./cat34 3<&0 4>&1
Hello World

It’s interesting that this also works:

$ echo Hello World | ./cat34 3>&0 4<&1
Hello World

While we are this part of town, let’s talk briefly about named pipes, another feature that has been around for ever, but doesn’t seem to get used as much as it deserves. We can run in to problems though:

Suppose I want to capture an HTTP response from a web server, I can do this:

$ mkfifo f1
$ mkfifo f2
$ mkfifo f3
$ netcat -l 8888 <f1 >f2 &
$ netcat 80 <f2 >f3 &
$ tee foo.txt <f3 >f1 &

and try a download, but alas:

$ GET http://localhost:8888/ >/dev/null
Can't connect to localhost:8888 (Connection refused)

This doesn’t seem right, netcat should be listening on port 8888, I told it so myself! And checking with netstat shows no listener on 8888 and finally ps -aux shows no sign of any netcat processes – what is going on?

Once again, strace helps us see the true reality of things:

$ kill %1
$ strace netcat -l 8888 < f1 > f2

But strace tells us nothing – there is no output! Like the dog that didn’t bark in the night though, this is an important clue, and widening our area of investigation, we find:

$ strace -f -etrace=open,execve bash -c 'netcat -l 8888 < f1 > f2'
execve("/bin/bash", ["bash", "-c", "netcat -l 8888 < f1 > f2"], [/* 46 vars */]) = 0
Process 20621 attached
Process 20620 suspended
[pid 20621] open("f1", O_RDONLY ...

The shell is stalled trying to open the “f1” fifo, before it even gets around to starting the netcat program, which is why the first strace didn’t show anything. What we have forgotten is that opening a pipe blocks if there is no process with the other end open (it doesn’t have to actively reading or writing, it just has to be there). The shell handles redirections in the order they appear, so since our 3 processes are all opening their read fifo first, none have got around to opening their write fifo – we have deadlock, in fact, the classic dining philosophers problem, and a simple solution is for one philosopher to pick up the forks in a different order:

$ netcat -l 8888 >f2 <f1 &
$ netcat 80 <f2 >f3 &
$ tee foo.txt <f3 >f1 &
$ GET http://localhost:8888/ >/dev/null
$ cat foo.txt
HTTP/1.1 301 Moved Permanently
Server: Apache

We can, it should be noted, do this more easily with a normal pipeline and a single fifo, and avoid all these problems:

$ netcat 80 <f1 | tee foo.txt | netcat -l 8888 >f1

but that would be less fun and possibly less instructive.



In an Age of Highly Pipelined RISC Architectures, we might be tempted to write a basic ASCII lower-case function as:

int tolower1(int c)
  return c|-(c-0101U<032)&040;

and avoid any branching. We subtract ‘A’ from c and do an unsigned comparison with ‘Z’-‘A’, convert to a bitmask and use that to ‘or’ in the lower case bit where necessary.

A more quiche-eating approach would be the clear and obvious:

int tolower2(int c)
  if (c >='A' && c <= 'Z') return c+'a'-'A';
  else return c;

at which hard-nosed hacker types will sneer, however, an Age of Highly Pipelined RISC Architectures is also an Age of Conditional Move Instructions, and GCC generates the pleasantly concise:

	leal	-65(%rdi), %ecx
	leal	32(%rdi), %edx
	movl	%edi, %eax
	cmpl	$25, %ecx
	cmovbe	%edx, %eax

for our simple approach rather than the distinctly bloated:

	leal	-65(%rdi), %eax
	cmpl	$25, %eax
	setbe	%al
	movzbl	%al, %eax
	negl	%eax
	andl	$32, %eax
	orl	%edi, %eax

All is not lost though for the hard-nosed hacker, for this is also an Age of Vectorization and we can do something similar using SSE2 instructions to lower-case not one but 16 characters at a time:

#include <x86intrin.h>
void tolower128(char *p)
  __m128i v1 = _mm_loadu_si128((__m128i*)p);
  __m128i v2 = _mm_cmplt_epi8(v1,_mm_set1_epi8('A'));
  __m128i v3 = _mm_cmpgt_epi8(v1,_mm_set1_epi8('Z'));
  __m128i v4 = _mm_or_si128(v2,v3);
  __m128i v5 = _mm_andnot_si128(v4,_mm_set1_epi8(0x20));
  _mm_storeu_si128((__m128i*)p, _mm_or_si128(v1,v5));

Now everyone should be happy!


Markov Computation

Markov computation is an attractively simple definition of universal computability.

A program is a list of rewrite rules:

xxx => yyy

or a terminal rule:

xxx =>. yyy

where xxx and yyy are arbitrary strings over some alphabet.

A computation takes a string over that alphabet and repeatedly rewrites it until termination: each rewrite uses the first rule with a LHS occuring anywhere in the string, which is then replaced by the RHS of the rule. If the rule is terminal the computation stops, it also stops if no rule matches.

Since I had a need to brush up on my Perl recently, I came up with this:

#!/usr/bin/perl -w
use strict;

my $line = shift @ARGV;
my @rules;

while (<>) {
    if (/^\s*(\S*)\s*=>(\.?)\s*(\S*)/) {
        push @rules,[$1,length $1,$3,$2 ne ""];

print "$line\n";
for (my $i = 0; $i < @rules; ) {
     my ($lhs,$len,$rhs,$term) = @{$rules[$i++]};
     my $ix = index $line,$lhs;
     if ($ix >= 0) {
        substr($line,$ix,$len) = $rhs;
        print "$line\n";
        $i = $term?@rules:0;

The line to rewrite is given as a command line argument and the program is read from input (either stdin or from files given on the command line). We use some regexps to match the program lines and, while it’s tempting to use regexp rewriting for running the program, we want to avoid any string characters being interpreted specially – we can use quotemeta for this but here we just use the non-regexp string operators (note the slightly wacky Perl use of substr as an l-value).

Now we have our evaluator, let’s write some programs:

Here is a multiplication program:

$ cat mult.mkv
x1 => 1x
x| => |1
x => |1
+| => |
+1 => 1+x
1* => *+
*1 => *
*| =>

$ ./markov '111*11' mult.mkv

Add small Roman numerals:

$ cat roman.mkv
# Final change to subtractive form
@X => X@
@IIII =>. IV
@ =>.
# Initial expansion of subtractive form
# Simplification
VV => X
# Rules for *
I*X => XI*
I*V => VI*
V*X => XV*
* =>
# Move from LHS to RHS
I+I => II
V+I => VI
V+V => X
X+ => X
I+ => +I*
V+ => +V*
# Go to finish
+ => @

$ ./markov I+II+III+IV+V+VI+VII+VIII+IX+X < roman.mkv

And here is one for checking the Collatz conjecture:

$ cat collatz.mkv
>11 => 1>
>1 => <111
> =>
1< => <111111
< => 1
11 => >11

$ ./markov 11111 collatz.mkv

See the 1st edition of Mendelson’s excellent book on Mathematical Logic for a full treatment, including a demonstration of equivalence with other notions of computability.


IPv6 TUN reflector

We had a look as a simple network simulation using TUN a couple of posts ago:

Let’s now have a look at getting it all working with IPv6. Changing the address swapping code is fairly straightforward, and for some extra points we’ll add a facility for printing out the source and destination addresses of each packet forwarded, the correct function for doing this is now inet_ntop, which works for both v4 and v6 addresses.

Here are the main changes [see for full code]:

It’s convenient to define a 32-bit swap function:

void swap32(uint8_t *p, uint8_t *q)
  uint32_t t = get32(p);

and our main function is now:

#define SRC_OFFSET4 12
#define DST_OFFSET4 16
#define SRC_OFFSET6 8
#define DST_OFFSET6 24

void reflect(uint8_t *p, size_t nbytes)
  uint8_t version = p[0] >> 4;
  switch (version) {
  case 4:
    if (verbosity > 0) {
      char fromaddr[INET_ADDRSTRLEN];
      char toaddr[INET_ADDRSTRLEN];
      inet_ntop(AF_INET, p+SRC_OFFSET4, fromaddr, sizeof(fromaddr));
      inet_ntop(AF_INET, p+DST_OFFSET4, toaddr, sizeof(toaddr));
      printf("%zu: %s->%s\n", nbytes, fromaddr, toaddr);
    // Swap source and dest of an IPv4 packet
    // No checksum recalculation is necessary
  case 6:
    if (verbosity > 0) {
      char fromaddr[INET6_ADDRSTRLEN];
      char toaddr[INET6_ADDRSTRLEN];
      inet_ntop(AF_INET6, p+SRC_OFFSET6, fromaddr, sizeof(fromaddr));
      inet_ntop(AF_INET6, p+DST_OFFSET6, toaddr, sizeof(toaddr));
      printf("%zu: %s->%s\n", nbytes, fromaddr, toaddr);
    // Swap source and dest of an IPv6 packet
    // No checksum recalculation is necessary
    for (int i = 0; i < 4; i++) {
    fprintf(stderr, "Unknown protocol %u\n", version);

Setting up the the v6 addresses for our new interface is a little different. As before, we bring the interface up:

$ ip link set tun0 up

Now, we need to add a link-local address, mandatory for all IPv6 interfaces:

$ ip -6 addr add fe80::1/64 dev tun0

We can use ping6 to try this out, localizing the request to the tun0 interface:

$ ping6 -I tun0 fe80::a617:31ff:fe5a:334f
PING fe80::a617:31ff:fe5a:334f(fe80::a617:31ff:fe5a:334f) from fe80::1 tun0: 56 data bytes
64 bytes from fe80::a617:31ff:fe5a:334f: icmp_seq=1 ttl=64 time=0.164 ms

This address is also the link-local address of my Wifi interface, but there is no ambiguity as we must specify which interface to use.

We can also add a private network address. IPv6 does not have the same concept of a private network as IPv4, instead we define Unique Local Addresses: append 0xfd to a random 10 digit hex global id and add an arbitrary 4 digit subnet identifier. Any random global id is fine – the idea is to ensure that any given network will have a different id from any other private network it is likely to come in contact with – we don’t need to worry about true global uniqueness though the Birthday Paradox tells us that we are likely to have a potential conflict with only about a million private networks (there might be lots of people out there with the same name and birthday as you, but you are unlikely to meet one of them at random).

We can generate our own random address, for example, using the method described in RFC4193, or use /dev/random:

$ hexdump -v -e '/1 "%02x"' -n 5 /dev/urandom; echo

or just copy a random sequence from somewhere on the Internet, for example, the one used here:

$ ip -6 route add fd2a:cd2c:8bc4:0::/64 dev tun0

This adds a local network with a global id of 2acd2c8bc4 and a subnet id of 0.

We can also define a larger subnet:

$ ip -6 route add fd2a:cd2c:8bc4:1100::/56 dev tun0

Now traffic to any IPv6 address of form fd2a:cd2c:8bc4:11xx:… will be sent to our TUN device:

$ ping6 fd2a:cd2c:8bc4:11ff::23
PING fd2a:cd2c:8bc4:11ff::23(fd2a:cd2c:8bc4:11ff::23) 56 data bytes
64 bytes from fd2a:cd2c:8bc4:11ff::23: icmp_seq=1 ttl=64 time=0.110 ms

Indeed, we can define all subnets for another global id:

$ hexdump -e '/1 "%02x"' -n 5 /dev/urandom; echo
$ sudo ip -6 route add fd40:bd2f:7ba0::/48 dev tun0

Just for interest, here’s our entire IPv6 routing table:

$ route -A inet6
Kernel IPv6 routing table
Destination Next Hop Flag Met Ref Use If
fd2a:cd2c:8bc4::/64 :: U 1024 0 0 tun0
fd2a:cd2c:8bc4:1100::/56 :: U 1024 0 0 tun0
fd40:bd2f:7ba0::/48 :: U 1024 0 0 tun0
fe80::/64 :: U 256 0 0 wlan0
fe80::/64 :: U 256 0 0 tun0
::/0 :: !n -1 1 524 lo
::1/128 :: Un 0 1 35 lo
fe80::1/128 :: Un 0 1 10 lo
fe80::a617:31ff:fe5a:334f/128 :: Un 0 1 7 lo
ff00::/8 :: U 256 0 0 wlan0
ff00::/8 :: U 256 0 0 tun0
::/0 :: !n -1 1 524 lo

Finally, to set up a simple service to use IPv6:

In one terminal:

$ nc -l -6 9901

In another:

$ nc -6 fd2a:cd2c:8bc4:11ff::23 9901

and our logging now looks like this:

$ ./reflect -v
Capability CAP_NET_ADMIN: 1 0 1
Created tun device tun0
48: fe80::1->ff02::2
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
48: fe80::1->ff02::2
48: fe80::1->ff02::2
80: fe80::1->fd2a:cd2c:8bc4:11ff::23
80: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
79: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23
72: fe80::1->fd2a:cd2c:8bc4:11ff::23

Those ff02::2 addresses are for IPv6 router discovery. The rest are two TCP flows, one in each direction (we can get more detail from Wireshark, in particular, the relevant port numbers, but this gives the general idea).


Fun with Flora

It’s been a while since I’ve played with AVR microcontrollers:

but I couldn’t resist getting myself an Adafruit Flora:

The tutorials etc. on the Adafruit website mainly cover using Windows and Apple development platforms, though there is much useful information in their forums on using Linux – I suppose we are meant to be able to work these things out for ourselves (actually, I suspect the real reason is that there is too much variation across different distributions to be able to give a single consistent set of instructions). Anyway, this is what worked for me to get everything working nicely on Ubuntu 12.10.

Firstly, set up your IDE. The standard 1.0.4 installation is what is recommended by Adafruit, so we’ll start there (see later for 1.5.2). We need a Flora entry in the boards.txt file, and a board specific header file, to go in variants/flora, telling the compiler where the pins etc. are. These files can be extracted from the “official” Adafruit Arduino IDE for the Flora, or the excellent kkolbo has put them in a handy zip file, referenced here:

Now we have a Flora entry on the Tools/Board menu and we can try downloading a “sketch”, as Arduino programs apparently are called. My Flora already has a blinkenlights program on it, so we’ll be wanting to change the timing or something so we can see that we really are running a new program. So, here we have the “Hello, world” of the embedded world (sketch taken directly from Adafruit tutorial):

// Pin D7 has an LED connected on FLORA.
// give it a name:
int led = 7;

// the setup routine runs once when you press reset:
void setup() {                
  // initialize the digital pin as an output.
  pinMode(led, OUTPUT);     

// the loop routine runs over and over again forever:
void loop() {
  digitalWrite(led, HIGH);   // turn the LED on
  digitalWrite(led, LOW);    // turn the LED off

Nice, a little way from a Stratum-1 NTP server, but it’s a start.

To get our sketch running on the Flora, we must bear some things in mind. First thing (this held me up for a while), you upload with “Upload” and not with “Upload Using Programmer” – I suppose you only do that if you have an actual Programmer, pretty obvious really, secondly there is a “modem-manager” program, part of the Linux network manager that sometimes does strange things to the USB-pretending-to-be-a-serial-port that we are attempting to program over, and thirdly, you must, of course, have the right permissions to access said USB/serial device – for a serial over USB link, you will be looking for a device called /dev/ttyACMx and this should appear in the “Tools/Serial Ports” menu item – if no port is shown there but a suitable device exists in /dev then you probably have a permissions problem. Make sure you have the port selected too, with a tick next it. You can run the IDE as root or fiddle with group membership, but the proper way of dealing with this is with the udev device manager (which also allows us to tell modem-manager to keep its hands off).

Like the Leonardo, the Flora has one USB port, used both by the program, and for programming and it seems that the device reports different USB product ids depending on what mode it’s in – just after a reset, it reports product id 0004, then after going to “normal” mode, the product id becomes 8004 (and some of the USB properties change), and we need to set udev rules for both. (“lsusb” and “udevadm info -a -n /dev/ttyACMx” are useful here to find out what is going on).

The upshot is that we need two udev lines, for the different states of the USB connection and the different idProducts that are reported:

SUBSYSTEMS=="usb", ATTRS{idVendor}=="239a", ATTRS{idProduct}=="0004", MODE="0660", GROUP="plugdev", ENV{ID_MM_DEVICE_IGNORE}="1"
SUBSYSTEMS=="usb", ATTRS{idVendor}=="239a", ATTRS{idProduct}=="8004", MODE="0660", GROUP="plugdev", ENV{ID_MM_DEVICE_IGNORE}="1"

The group setting ensures normal users (which Ubuntu adds to the plugdev group) can access the port, and the ID_MM_DEVICE_IGNORE tells modem-manager to ignore this device, so we no longer need to disable modem-manager while programming. There are some suggestions on the internet of using a SYMLINK+=”ttyACM%n” entry too, but that doesn’t seem to be necessary for our purposes (actually, adding “SYMLINK+=ttyUSB0” gives us a means of adding a consistent device name that is recognized by the Arduino IDE, but doesn’t clash with the kernel assigned ACM names). Add these lines into, say, /etc/udev/rules/90-flora.rules, restart udev with service udev restart and reconnect or reset your Flora. You should see something like:

$ ls -l /dev/ttyACM*
crw-rw---- 1 root plugdev 166, 1 May 18 15:59 /dev/ttyACM1

and programming should now work just fine as a normal user (assuming they are in the plugdev group). Sometimes the soft reset that the programmer does to initiate upload doesn’t work and you have to do a hard reset, but that seems to be a standard problem with Leonardo/Flora. Most of the time it seems to work just fine with the software reset.

A couple of observations: when programming, the software reset of the serial port fails if the port isn’t selected in the Serial Port menu, though programming after a hard reset will still suceed. Also sometimes if the upload fails, the avrdude process isn’t shut down properly and keeps the serial port device open, so when the USB connection is reestablished, the device minor number goes up (ie. we now have /dev/ttyACM1 rather than /dev/ttyACM0). If this happens, “killall avrdude” will restore the status quo.

Finally, I had originally tried using an Arduino 1.5.2 installation I set up when I was thinking of having a play with the ARM-based Arduino Due. Initially this didn’t work too well for the Flora, but after sorting out the USB device problems as above and copying the relevant seeming lines from the Leonardo section of the boards.txt file to the Flora section (and copying the variants/flora directory), it was just fine: Flora
flora.bootloader.lock_bits=0x2F{build.vid} -DUSB_PID={}

(I’m not sure that I’ve got that bootloader, but I don’t need it at the moment).

Anyway, a nice bit of kit, good work from Limor Fried and all at Adafruit, and now I can make the LED flash just as I like, time to think of some more interesting applications. Watch this space…