gistfile1.txt

root@luke:~/tmpfs# cat foo.sh 
rfile=$(mktemp)
size=$1
gcc -x c -O3 -o $rfile -  <<__EOF__

#include <stdlib.h>

int main(void) {
	int i;
	long int a;
	char *b;
	char buff[100010];
	int buflen=0;
	int intcounter;
	srandom(time(NULL));
	b = (char *) &a;
	for (i=0;i < ($1 / sizeof(long int));i++) {
		a = random();
		for (intcounter=0; intcounter<sizeof(long int); intcounter++)
			buff[buflen+intcounter] = b[intcounter];
		buflen+=intcounter;

		if (buflen > 100000){
			write(1,buff,buflen);
			buflen=0;
		}
	}
	write(1,buff,buflen);
	return 0;
}
__EOF__

$rfile > tmpfile; ls -lh tmpfile
rm $rfile
root@luke:~/tmpfs# time ./foo.sh 40000000
-rw-r--r-- 1 root root 39M Nov  3 10:10 tmpfile

real	0m0.144s
user	0m0.116s
sys	0m0.032s
root@luke:~/tmpfs# 

Or for something bigger?

root@luke:~/tmpfs# time ./foo.sh 4000000000
-rw-r--r-- 1 root root 3.8G Nov 3 10:13 tmpfile

real 0m8.097s
user 0m6.828s
sys 0m1.280s

Why not just do:

dd if=/dev/random of=tmpfile bs=1000 count=4000

or /dev/urandom if you're on a machine without a lot of entropy.

I do like the evilness, but, seems like a lot of overkill. :)

Because urandom is ridiculously slow. On my rather fast system /dev/urandom (Which is the FAST one) pulls 7.5MB/s.

Oh, it's also obviously mostly a joke now. A far better solution is just using shred, which gives comparable speeds. But I still like the vileness :P